WORLD BANK REG IONAL AND SECTO RAL STUDI ES 21994 March 2001 Uganda's Recovery The Role of FPrmis, Firms, and Governmnent EDITED BY RITVA REINIKKA PAUL COLLIER 159"H I= ' Uganda's Recovery Uganda's Recovery The Role of Farms, Firms, and Government Edited by Ritva Reinikka Paul Collier The World Bank Washington, D.C. Copyright © 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 1818 H Street, N.W., Washington, D.C. 20433, USA All rights reserved Manufactured in the United States of America First printing March 2001 The World Bank Regional and Sectoral Studies series provides an outlet for work that is relatively focused in its subject matter or geographic coverage and that contributes to the intellectual foundations of development operations and policy formulation. Some sources cited in this publication may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this book are entirely those of the authors and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. Permission to photocopy items for internal or personal use, for the internal or personal use of specific clients, or for educational classroom use is granted by the World Bank, provided that the appropriate fee is paid directly to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone 978-750-8400, fax 978-750-4470. Please contact the Copyright Clearance Center before photocopying items. For permission to reprint individual articles or chapters, please fax a request with complete information to the Republication Department, Copyright Clearance Center, fax 978-750-4470. All other queries on rights and licenses should be addressed to the Office of the Publisher, World Bank, at the address above or faxed to 202-522-2422. Cover art: Hand shield, Uganda. Wood, pigment. Pastoral nilotic peoples of Uganda. This work is included in the collection of the World Bank Art Program, which makes particular efforts to identify artists from developing nations and to make their work available to a wider audience. Library of Congress Cataloging-in-Publication Data Uganda's recovery: the role of farms, firms, and government / edited by Ritva Reinikka, Paul Collier. p. cm. Includes bibliographical references and index. ISBN 0-8213-4664-4 1. Uganda--Economic conditions--1979- 2. Uganda--Economic policy. 3. Uganda-- Politics and government--1979- I. Reinikka, Ritva. II. Collier, Paul. HC870.U45 2001 338.96761--dc2l 00-049791 Contents Acknowledgments .................................................. ix Contributors .................................................. xi Foreword .................................................. xiii Map of Uganda .................................................. xvi 1. Introduction .1 Paul Collier and Ritva Reinikka Postconflict Recovery and Macroeconornic Reforms .5 Households .6 Firms .7 Governent .8 Sustainability and Lessons .10 References .11 Part I. Postconflict Recovery and Macroeconomic Reforms ....... 13 2. Reconstruction and Liberalization: An Overview .15 Paul Collier and Ritva Reinikka The Inheritance of Disorder .16 The Restoration of Peace .21 Growth Policies in the Context of the Postconflict Economy . 24 Economic Liberalization .31 Impact on Investment and Exports .38 The Role of Aid .39 v vi Contents Conclusions .............................................. 44 References .............................................. 45 3. Exchange Reforms, Stabilization, and Fiscal Management ............. 49 Mark Henstridge and Louis Kasekende Exchange Reforms .............................................. 50 Legalizing the Parallel Market and Exchange Rate Unification ........ 52 The Achievement of Macroeconomic Stability ..................................... 56 Planning and Implementing Fiscal Policy ............................................. 58 Targets, Tradeoffs, and Costs in Macroeconomic Management ......... 71 Conclusions .............................................. 76 References .............................................. 77 Part II. Household Responses and Constraints .81 4. Changes in Poverty and Inequality .............................. ............. 83 Simon Appleton Changes in Mean Consumption Per Capita .......................................... 87 Defining an Absolute Poverty Line for Uganda ................................... 89 Sectoral Decomposition of Poverty Changes ...................................... 105 Summary and Conclusions ........................................... 111 Annex 4.1. Methodology ........................................... 113 References .................................................. 119 5. Rural Households: Incomes, Productivity, and Nonfarm Enterprises ........................................... 123 Klaus Deininger and John Okidi A Panorama of Rural Uganda ................. .......................... 124 Intertemporal Changes in Household Income ................................... 137 Agricultural Productivity and Nonfarm Enterprises ............... ......... 143 Conclusions ........................................... 152 Annex 5.1. Tables of Estimation Results ........................................... 154 References ........................................... 174 6. Crop Markets and Household Participation ..................................... 177 Donald Larson and Klaus Deininger Market Participation in the Early 1990s ........................................... 178 A Market Model for Community Trade ........................................... 180 The Determinants of Market Participation .......................................... 191 The Effects of Price Changes on Household Welfare ................ ......... 194 Are Crop Markets Developing? ........................................... 195 Conclusions and Policy Implications ......................................... .. 202 Annex 6.1. Calculating Household Welfare ........................................ 203 References ........................................... 203 Contents vii Part III. Firm Responses and Constraints ............................. 205 7. Confronting Competition: Investment, Profit, and Risk .207 Ritva Reinikka and Jakob Svensson Investment Response .209 Constraints to Investment .216 Conclusions and Policy Recommendations .226 Annex 7.1. Data and Estimation Results .228 Annex 7.2. Derivation of the Investment Equation .231 References .232 8. Productivity and Exports .235 Bernard Gauthier Trade Liberalization, Exports, and Productivity .236 Enterprise Responses to Changing Incentives .238 Export Response .246 Conclusions .252 Annex 8.1. Productivity Measures .253 References .265 Part IV Government Performancefrom a Beneficiary Perspective.....................................................................269 9. A Quest for Revenue and Tax Incidence .271 Duanjie Chen, John Matovu, and Ritva Reinikka Revenue Trends and Tax Reforms .276 Method and Data for Tax Incidence Analysis .278 Tax Incidence on Households .280 Marginal Effective Tax Rate for Firms .281 Cross-Border Comparison for Foreign Firms .288 Compliance and Tax Administration .292 Conclusions .295 Annex 9.1. Household Incidence Analysis and the Concept of Welfare Dominance .296 Annex 9.2. Marginal Effective Tax Rate .298 Annex 9.3. Figures and Tables for Household Incidence and METR .302 References .316 10. The Cost of Doing Business: Finns' Experience with Corruption .319 Jakob Svensson The Data .321 Incidence, Level, and Effects of Corruption .321 Case Studies .331 viii Contents Conclusions ............................................ 334 Annex 10.1. Ranking of Constraints and Payment of Bribes ............ 336 References ............................................ 340 11. Recovery in Service Delivery: Evidence from Schools and Health Centers ............................................ 343 Ritva Reinikka Diagnostic Survey .............................................. 346 Education and Public Spending ............................................... . 347 Health Care and Public Spending ......................................... ...... 363 Conclusions and Policy Changes ................................... ............ 366 References ............................................... 368 12. What Can We Expect from Universal Primary Education? ............ 371 Simon Appleton Access to Education Prior to the UPE Initiative ...................... ........... 373 Returns to Education: Productivity and Labor Allocation Effects ... 378 Effects of UPE on School Quality ............................................ 395 Summary and Conclusions ........................................... 400 Annex 12.1. Models ........................................... 401 References ........................................... 402 13. Combating Illness .407 Paul Hutchinson Health Policy and Access to Services .409 Burden of Disease .414 Demand for Curative and Preventive Services .423 Conclusions and the Way Forward .430 Annex 13.1. Data and Estimation Results .433 References .444 Part V Sustainability and Lessons ......... ................ 451 14. Beyond Recovery ............................. 453 Paul Collier and Ritva Reinikka References ............................. 460 Appendixes A. Household Surveys ............................. 463 B. The Uganda Enterprise Survey ............................. 467 References ............................. 473 List of Tables, Figures, and Boxes ............................. 475 Index ............................. 481 Acknowledgments Much of the credit for inspiring this book must go to the Ugandan economic team led by Emmanuel Tumusiime-Mutebile, permanent secretary and sec- retary to the Treasury, which has made a tremendous effort to turn around the Ugandan economy since the mid-1980s. The Bureau of Statistics, the Pri- vate Sector Foundation, and the Uganda Manufacturers Association's Infor- mation and Consultancy Service were instrumental in obtaining the microeconomic data on households and firms used in most of the chapters of this book. For their encouragement and support to the project we are grate- ful to James Adams and Shantayanan Devarajan at the World Bank and Peter Miovic and Joseph Stiglitz. Drafts of the chapters were presented and discussed at a conference on the Comprehensive Development Framework in October 1999 in Kampala. We would like to thank participants for the stimulating and open discussion, and the World Bank Institute for financial and other support for the conference. The govemments of Austria, Japan, Sweden, and the United Kingdom have supported parts of the survey work and analysis contained in this book. Financial support was also received from the Bank's Poverty Reduction and Economic Management network in the form of a PREM fellowship. We would like to acknowledge the World Bank's Editorial Comnmittee and four anonymous referees for comments on the manuscript. Finally, special thanks go to the World Bank publications team and to Hedy Sladovich who worked diligently and with good humor to bring this book to its final form. ix Contributors Simon Appleton Lecturer, University of Nottingham; Research Associate, Centre for the Study of African Econo- mies, University of Oxford, United Kingdom Duanjie Chen Associate Director, International Tax Program, Institute for International Business, University of Toronto, Canada Paul Collier Director, Development Research Group, World Bank, Washington, D.C. Klaus Deininger Senior Economist, Development Research Group, World Bank, Washington, D.C. Beemard Gauthier Professor, Institut d'tconomie Appliquee, Ecole des Hautes Etudes Commerciales, Montreal, Canada Mark Henstridge Economist, African Department, International Monetary Fund, Washington, D.C. Paul Hutchinson Researcher, Economics Department, University of North Carolina, Chapel Hill Louis Kasekende Deputy Governor, Bank of Uganda, Kampala xi xii Contributors Donald Larson Senior Economist, Development Research Group, World Bank, Washington, D.C. John Matovu Economist, IMF Institute, International Monetary Fund, Washington, D.C. John Okidi Senior Research Fellow, Economic Policy Research Centre, Kampala, Uganda Ritva Reinikka Research Manager, Development Research Group, World Bank, Washington, D.C. Jakob Svensson Assistant Professor, Institute for International Economic Studies, Stockholm University, Sweden; Senior Economist, Development Research Group, World Bank, Washington, D.C. Foreword The economic reforms implemented in Uganda under the leadership of Presi- dent Museveni and the economic recovery that these reforms have gener- ated have justifiably attracted a great deal of attention among development practitioners and academics around the world. Uganda is rightly regarded as a pioneer of macroeconomic stabilization and structural adjustment in Sub- Saharan Africa for two reasons: first, because of the extent and consistency of its economic reform program, especially in the areas of fiscal policy, exchange rate reforms, trade policy, and the use of debt relief to enhance public expen- diture on basic social services; and second, because of the reform program's success in restoring macroeconomic stability, boosting the economic growth rate, and reducing poverty. The reform program followed a prolonged period of economic decline and civil strife that devastated human and physical capital and destroyed the economy's formal sectors, not least because this period witnessed signifi- cant erosion of much of the institutional framework that is required to sup- port transactions in a modern economy. Sadly, too many countries have en- tered the new millennium suffering from a collapse of the economy, of state capacities, of social capital, and of law and order similar to that which af- flicted Uganda in the 1970s and 1980s. Understanding what Uganda has achieved and the strengths and weak- nesses of its economic reform program is especially important. Exploring the challenges that the country faces as it attempts to sustain its recovery by raising private investment levels and improving human resource capacities is also important. This volume is unique in providing such a comprehensive analysis of policy reform in a Sub-Saharan African country and the lessons that can be learned from this analysis for future policy reforms, both in Uganda and in other countries. The volume will have played an invaluable role if it xiii xiv Foreword brings these lessons to a wide audience and can stimulate debate among development practitioners. This book consists of a series of studies written by a range of specialists, all with considerable expertise in their respective fields, that analyze the re- sponses of private sector agents-households, farms, and firms-and of the government itself, to the macroeconomic and structural reforms implemented since the late 1980s in a society recovering from a traumatic civil conflict. The importance of this line of enquiry cannot be underestimated, because the success or failure of market-oriented reforms depends crucially on just how private sector agents are able to respond to the incentives and opportunities created by the reforms. In this context, the consistency of government policy over time has become an invaluable national asset. Supporters of market- oriented reforms argue that they can stimulate increased production and in- vestment and a more efficient allocation of resources that, over time, will boost incomes, enhance welfare, and reduce poverty. By contrast, critics of market-oriented reforms argue that market imperfections and structural or institutional constraints prevent a positive response from private sector agents. Resolving these contentious issues requires detailed empirical analysis of the type presented in this volume. The analysis in this book draws on a wealth of quantitative data derived from a series of household surveys and from surveys of firms conducted in the 1990s and more recently in 1999/2000. The household surveys, conducted at intervals between 1992 and 1997, permit analysis of the evolution of in- come, expenditures, and poverty during this period. The impact of reforms on rural factor markets, on crop and livestock production decisions, and on firms' investment decisions are also among the issues researched in this vol- ume. It is the solid quantitative database on which so much of the empirical analysis is based that makes this volume so important: few other studies of structural adjustment reforms have drawn on such a comprehensive data- base that has been compiled using state-of-the-art data collection techniques in developing countries. Unfortunately, rigorous empirical analysis of reforms in Sub-Saharan Africa has often been impeded by a lack of hard data, a defi- ciency that this volume helps to rectify. While this book praises Uganda's achievements where warranted, it pro- vides an objective assessment of the reforms and does not shy away from identifying areas where policy mistakes were made, for example, where implementing reforms earlier might have generated higher rates of return and alleviated bottlenecks to private sector production. It points out where major weaknesses still exist, notably, the corruption in the public sector, which raises the cost of doing business in Uganda and undermines the quality of public services, the still poor enforcement of contracts, and the deficiencies in the physical infrastructure. While reforms created economic opportuni- ties that led to reductions in poverty among most groups of poor house- holds, a notable exception is households with nonworking heads, which demonstrates the need for more effective transfer mechanisms to support Foreword xv vulnerable households. The objectivity and clarity of the findings in this volume provide a valuable service to those in Uganda who are striving to deepen the reform program and to move ahead to tackle the more difficult institutional reforms that are needed to reduce the cost of doing business in Uganda and to improve the incentives for saving, investment, and trade. As someone who has been closely involved in designing and imple- menting economic reforms in Uganda for more than a decade, I believe that these reforms deserve the type of comprehensive evaluation, based on a rigorous analysis of quantitative data, that this volume provides. I hope that it will enable the lessons that can be learned from our efforts to imple- ment reforms in Uganda to be disseminated to a wider audience and that these lessons will be of benefit to others in the developing world who are working to reform their economies. Emmanuel Tumusiime-Mutebile Permanent Secretary/Secretary to the Treasury Ministry of Finance, Planning, and Economic Development, Uganda Map of Uganda _______________________________________________________________________ IEBR 31212 <>t ~~~S U D A N > UGANDA . vSUDAN G.6~~~~~~~~~~- a DISTRICI CAPITA'S J , 3' , S I D STRIB-U DARES' , M REGION BOUNDARIES t _ - INTE3NATIONAL BOUNDAERiES f. DEMOCRATIC 1 OF CONGOf5 A *s N 4 E N A 4zwb-S;sE E R if 1! Mi 3 _RE3 o ...... < ' I f / w7 fW E S iU ,EbYliRoN , AO MS3iS DJ-9ao;5 g ;9:0Tą;;5 ffjr K E N Y A D AE AN _ _ _ _ _- I .' "o °' A ;' AADr Ai eb NI CII '5? S;~ ;0t$0;00t TA0i N0 3 ZA NA *tSP_~~AAE~A C / \ \ , 600 - Counterfactual 500- 0400- E 3 0 0 . Budgeted 300 - 200 ~ l l l l l l l l l l l l l l l l l l l l l l lActial 100 July January June January June 1994 1995 1995 1996 1996 Source: Ministry of Finance, Planning, and Economic Development and Bank of Uganda data. that achieving low inflation largely depended on careful budget implemen- tation through the use of the cash flow. Fiscal Shocks The experience of resources falling short of budget projections was the main reason for adopting a monthly cash flow system. However, since 1992 the main fiscal shocks have come from persistent demands within the government to increase spending within each fiscal year. To understand the potential impact of these demands for extra or supplementary expenditures, see figure 3.2. The top graph in this figure shows actual expenditures, the monthly average of budgeted expenditures, and counterfactual expenditures for 1994/95 and 1995/ 96. The latter are equal to the total supplementaries approved by parliament evenly distributed over the last nine months of each fiscal year.8 Had all else 8. The amounts finally approved by parliament in 1994/95 and 1995/96 were much smaller than those originally requested by spending ministries; in figure 3.2, Exchange Reforms, Stabilization, and Fiscal Management 67 remained the same, the supplementaries would have been financed by increased credit to government, leading to more base money, as shown in the bottom graph of figure 3.2. Counterfactual base money peaks at U Sh 560 billion, which is about twice both the actual and programmed levels. Assuming that prices would have increased in proportion to the excess increase in the supply of money, inflation would have risen to an annual rate of at least 25 percent.9 This figure understates the consequences of such monetary expansion, because an- nual inflation rates higher than about 10 percent are likely to lead to a sharp reduction in the private sector's demand for real money balances. Therefore, higher inflation would have sharply increased the velocity of money, and in- flation would rise much more than implied by the counterfactual series on base money. Given these demands for extra spending, the Ministry of Finance faced a tradeoff. It could either keep the allocations in the original budget intact, adding the extra supplementary spending to total spending, or it had to cut other expenditures to retain control of total spending. As the ministry was charged by the president to maintain price stability, it decided to keep total expenditures within the resource envelope, and balance the supplementary spending by cuts elsewhere in the budget. Following the introduction of the cash flow system in 1992, there was no shortfall in donor resources-although it was still difficult to predict the timing of disbursements. A small shortfall of revenue occurred in 1993/94, but it was not until 1996/97 that actual receipts were significantly less than projected, in this case by the equivalent of 1.4 percent of GDP.10 The ability to monitor the budget implementation through the cash flow system enabled the government to cut expenditures (on a cash basis) by 0.76 percent of GDP, with the remaining gap more than offset by increased foreign grants. At the same time, however, higher expenditures (on a commitment basis) were financed by increased do- mestic arrears. Arrears were the result of a lack of control over the line minis- tries' ability to enter into expenditure commitments. Despite the increased do- mestic arrears, the adjustments made to cash expenditure in 1996/97 delivered low underlying inflation of 2 percent, despite lower than projected revenue."' This is quite an achievement compared with the underlying inflation of 50 per- cent that followed the failure to adjust to a revenue shock in 1991/92. In the face of a shock to resources, maintaining low inflation-although not impossible in the absence of the cash flow system-was certainly facilitated by it. they were allocated over the last nine months of the fiscal year because parliamentary approval was not forthcoming until three months into the year. 9. Assuming a constant demand for real money balances. 10. Revenues were U Sh 90 billion less than projected, owing to sluggish imports and the difficulties in implementing a new value added tax. 11. The underlying inflation index excluded food crop prices (but included pro- cessed food), and was not, therefore, sensitive to the possibility that dry weather would push up food crop prices. 68 Mark Henstridge and Louis Kasekende External Shocks In addition to absorbing within-year fiscal shocks, the cash flow system helped the government respond to macroeconomic shocks that could otherwise have led to higher inflation. As discussed previously, Uganda experienced a cof- fee boom in 1994-96 coupled with significant inflows of foreign capital. Both threatened to lead to a major appreciation of the exchange rate. Generally, in formulating fiscal policy a tradeoff was perceived between increasing donor budget support and the concern that doing so would con- tribute to an overvaluation of the real exchange rate. This tradeoff also sur- faced when the government's economic team tried to dampen nominal ex- change rate appreciation during the course of the fiscal year. To offset nominal appreciation, increased foreign reserves were projected arising from the Bank of Uganda intervention in the foreign exchange market; these interventions were to be offset by reductions in net credit to government. This, of course, implied a tighter fiscal policy."2 Similarly, if the corresponding monetary in- jection were perceived to threaten price stability, then an offsetting reduction in net domestic assets through fiscal tightening had to be made. The theory behind, and actual experience of, temporary trade shocks sug- gest that they lead to an appreciation of the real exchange rate, either through an appreciation of the nominal exchange rate, an increase in the price level, or some combination of both (see Bevan, Collier, and Gunning 1989, 1990; Collier and Gunning 1996). The appropriate fiscal response to a temporary trade shock and to increased inflows of foreign capital is to increase public savings, which is what the Ugandan government did, as reflected in the nega- tive central bank financing from 1993/94 onward (table 3.2). These reduc- tions in central bank credit to the government were achieved in part by set- ting a tight budget, but were also sustained through the cash flow system. In addition, the government imposed a coffee stabilization tax, which was a hotly debated topic in Uganda at the time. Arguments for and against the coffee tax are presented in boxes 3.2 and 3.3. The role of the cash flow in managing the consequences of the coffee boom is illustrated in figure 3.3. The top graph shows that the increase in the terms of trade was followed, with a lag of one quarter, by an increase in the producer price of coffee. With a lag of another quarter, there is a sustained increase in underlying inflation (here a weighted average of the quarterly underlying in- flation index).'3 The bottom graph shows that underlying inflation increased sharply during 1995 and shows the discrepancy between the planned budget and the implemented fiscal stance (as proxied by the budgeted and actual 12. Bevan (1998) concludes that the impact on the exchange rate of additional inflows is at worst ambiguous, at best benign. 13. The means and ranges of these series have been adjusted to maximize visual correlation, and the left-hand scale is therefore that of the terms of trade index. Exchange Reforms, Stabilization, and Fiscal Management 69 Box 3.2. Arguments for the Coffee Stabilization Tax In June 1994, frost in Brazil triggered a sharp, but temporary, increase in interna- tional coffee prices. Uganda expected to earn US$500 million over the year to September 1995, compared with coffee export earnings of US$180 million over the year to September 1994. The increase in coffee export earnings was expected to be equivalent to more than 70 percent of the stock of broad money at the end of December 1994, and thus presented a potentially serious threat to monetary sta- bility. The rational private response, that is, to accurnulate domestic financial assets, would look the same in the monetary statistics as an inflationary mon- etary expansion, the difference being that the latter would quickly lead to infla- tion. The risk was that the magnitude of monetary expansion involved could have been so large that little could have been done in the event of an unwar- ranted monetary expansion regardless of the source to rescue price stability. It was not clear that the Kenyan experience during the 1970s coffee boom would be replicated (see Bevan, Collier, and Gunning 1990). The structure of the coffee export sector was different, being private and liberalized in the 1990s in Uganda. Payments were made largely in cash, whereas in Kenya they were deposited in farmer's accounts. Finally, far fewer bank branches were within easy reach of most Ugandan coffee farmers, raising doubts about the likeli- hood of the coffee windfall being held in domestic financial assets (other than cash) to the same extent as had been possible in Kenya. Faced with these risks, the government decided to introduce a graduated tax on coffee export earnings above a threshold. The rates and threshold were decided following extensive consultations with the coffee exporters. The coffee stabilization tax was set at 20 percent on receipts above a threshold of U Sh 1,100 per kilogram, and 40 percent on receipts above U Sh 2,200 per kilogram. The lower threshold was determined relative to a normal rate of net profit. The tax was specifically designed so that the government would save rather than spend the money during the boom, allowing the Bank of Uganda to purchase foreign exchange from the market and thus ease some of the pressures on the exchange rate without increasing reserve money. The coffee stabilization tax came into force late in 1994. Collections amounted to US$15 million in fiscal year 1994/95, and US$13 million in 1995/96, the sec- ond and final year of the boom. Ex post, while the coffee tax revenue did afford modest room for intervention in the foreign exchange market, much more was gained through larger savings from general budgetary operations. In the event, the risks-reasonably perceived ex ante-did not materialize. change in credit to government). Between the second quarter of 1994 and the first quarter of 1996, government savings with the Bank of Uganda increased each quarter as the cash flow system enabled the government to implement a tighter than budgeted fiscal policy in response to the increased inflation. The short-term fiscal response also manifests itself in reduced volatility of actual central bank financing relative to the budget. In addition, from the beginning 70 Mark Henstridge and Louis Kasekende Box 3.3. Arguments Against the Coffee Stabilization Tax While windfall taxation was a reasonable response to the perceived risks the coffee boom posed for macroeconomic stability, the government's argument that it was necessary to reduce exchange rate appreciation during the boom was based on a misreading of private responses to a temporary income windfall. The in- crease in coffee prices, triggered by a frost in Brazil, was not a unique event. Ugandan coffee farmers clearly remembered the previous boom of the late 1970s and thus would have likely understood that their income gains would be tempo- rary. Faced with such an income surge, the rational response is to use the money to boost savings and investment rates. Initially, these savings would be liquid and then gradually converted into fixed assets over time. In aggregate, the only liquid asset that the private sector could acquire was claims on the government in the form of cash, since other financial claims net out. Hence, the coffee boom would be expected to cause an initially large increase in the demand for real money balances, followed by a decline and a surge in fixed investment. The task of the authorities was to accommodate this private sector savings and investment strategy rather than to nullify it through taxation. Indeed, since the private sector had not invested in Uganda for more than 20 years, a private investment boom was socially highly desirable rather than an appropriate ob- ject for taxation. As so little revenue was collected by the tax, its effects-good or bad-on the progress of the boom were marginal. The monetary and real effects of the windfall proceeded as described above. Initially, the demand for real money balances increased sharply. Because the Ugandan economy is characterized by highly flexible prices (a legacy of the demise of long-term contracts under the stress of volatility), the private sector could achieve desired real money bal- ances through changes in the price level. In real terms, private fixed investment rose by 38 percent in the first year of the boom (from July 1994 to June 1995) and by an additional 17 percent in the following year. The investment rate out of the private income windfall from the coffee boom was probably well over 50 percent. Finally, by the early 1990s, Ugan- dan coffee farmers had below average household incomes (see chapter 4 in this volume). Hence, as shown in chapter 9, a stronger coffee tax would have been more regressive, as well as hitting private investment harder. It proved to have been unnecessary as a stabilization measure, having made only a small contribu- tion to the significant fiscal savings that were accumulated anyway. It also had the potential to discourage coffee planting in the long term if coffee farmers antici- pate that during the next coffee boom the tax will be reintroduced. This was why it was so critical to remove the tax from the statute book, even though by 1996 the price of coffee had fallen so that revenue was no longer being generated. of 1994 the nominal exchange rate appreciated, which was partially countered by intervention in the foreign exchange market. Reductions in net domestic assets through tighter fiscal policy were intended to offset the monetary im- pact of increased net foreign assets. Exchange Reforms, Stabilization, and Fiscal Management 71 Figure 3.3. Short-Term Fiscal Response to Increased Inflation, Fiscal Years 1992/93-1997/98 The producer price of coffee and inflation :' 700 - 0.025 6 600 - - 0.020 . 5 500 -- 400 - - 0.015 300 - 0.010 ~ 200 0. 100 u 0 0~~~~~~~~~~~~~~~~~~ 1993/94 1994/95 1995/96 1996/97 1997/98 Inflation, budgeted, and actual Bank of Uganda financing i 30,000 0.03 0 ....................0 b -30,000 -0.03 2 v-60,000- I ' 1 --0.06 1 1993/94 1994/95 1995/96 1996/97 1997/98 Source: Ministry of Finance, Planning, and Economic Development data. Targets, Tradeoffs, and Costs in Macroeconomic Management The operation of the cash flow system raises a number of questions. Why target inflation when implementing the budget rather than the monetary aggregates that are used to prepare the budget? Why emphasize short-term fiscal adjustment rather than a more active use of monetary policy? What were the costs of short-term fiscal adjustments in terms of reduced quality of public spending? Inflation Targeting Inflation was targeted for two reasons. First, prices were flexible and responded quickly to changes in the money supply. Second, there was a three-month lag in the compilation of monetary statistics, while the consumer price index was available at the end of each month. Taken together, this meant that changes in monetary conditions showed up in prices at about the same time that they appeared in the statistics for broad money. Table 3.3 shows the rolling time- table for the compilation of cash flow and the information lags involved in 72 Mark Henstridge and Louis Kasekende tracking the implementation of the budget and the evolution of money and prices. Looking directly at the price data sidestepped the difficulties of sepa- rating signal from noise in the monetary data, especially for the unpredictable short-run changes in money demand that are characteristic of a remonetizing economy like the one in Uganda. Price flexibility is illustrated by inflation having stopped within one quarter in 1992. It is not that the demand for money in Uganda is fundamentally different from most other economies (see Henstridge 1999). Indeed, the long-run stabil- ity of money demand was central to the financial programming used to con- struct the budget and in the fiscal program agreed with the IMF. But the use of a formal money demand relationship in short-term macroeconomic manage- ment was difficult for two reasons. First, the demand for money relationships estimated for Uganda used quarterly data, which were not available quickly enough to be incorporated into the analysis behind short-term adjustments. Because many of the data were produced with a long lag, the breadth of as- sumptions that would have to be made in order to project money demand were more likely to be hostages to fortune than accurate inputs into short-term policy decisions. Second, in a shock prone, flexible price economy that was in the process of remonetizing, there were unpredictable short-term shifts in money demand. Even if the path of real money balances were to be convinc- ingly projected on a quarterly basis, a judgment on whether monthly fluctua- tions in monetary conditions are out of line with the quarterly projections would still have to be made. The most timely and reliable data to inform such a judg- ment would come from current inflation, which therefore might as well be targeted directly. Trying to follow a short-term monetary program would largely have consisted of trying to work out why there was price stability when the program appeared to be off-track. Indeed, the Bank of Uganda's experience with a reserve money program, which is predicated on a stable relationship between the monetary base and broad money, has found that the remonetization of the economy and increasing confidence in financial instruments has made it difficult to sustain the assumption of a stable money multiplier. With wild fluc- tuations observed in the multiplier, the central bank relies on a broad range of indicators of monetary conditions, including the excess reserves held by com- mercial banks and the trends in the discount rates on treasury bills. Short-Term Fiscal Adjustment versus Monetary Policy This chapter has emphasized the role of the cash flow system and short-term fiscal adjustments in the implementation of macroeconomic policy. Fiscal adjustment both reduced demand directly and led to a reduction in credit to the government and, therefore, to a reduction in net domestic assets and base money. Fiscal adjustments were made because the scope for independent monetary policy has been too limited, even though the budgets have been consistent with macroeconomic stability since 1992. Two monetary policy instruments were available: changing the reserve requirements of commercial banks and issuing treasury bills. In addition, Exchange Reforms, Stabilization, and Fiscal Management 73 the Bank of Uganda retained control of the rediscount rate when interest rates were liberalized. However, in an uncompetitive, segmented commer- cial banking sector with excess liquidity, changes in the rediscount rate had little bite. Changing the banks' reserve requirements would, in principle, reduce lending at the margin, and hence lower net domestic assets and money in the economy. The central bank was unwilling, however, to change reserve requirements, because several of the smaller commercial banks were too fragile to comply without becoming bankrupt. While the argument against active use of reserve requirements might have been reasonable with regard to the immediate stability of the financial sector, it neutered one instrument of monetary policy. What remained was the treasury bill. However, the government's judg- ment was that a short-term fiscal adjustment provided more monetary bite- both through a reduction in base money and a reduction in aggregate de- mand-for a given shilling cost, than did increased issues of treasury bills. The volume of sales of treasury bills in the primary auction managed by the Bank of Uganda was not determined by the government's financing re- quirements. Since June 1992, the government has set budgets with no planned increase in domestic borrowing. The main reason for these sales was to de- velop a capital market, specifically a secondary market in treasury bills, within which the Bank of Uganda could conduct open market operations. Until 1996/ 97, however, there had been virtually no secondary trading of treasury bills. Banks held the bulk of outstanding bills (83 percent of a total of U Sh 88 billion in June 1997), so the treasury bill as an instrument of monetary policy was limited to its influence on the composition of the banking system's as- sets. Many commercial banks held large cash reserves because they were underlent and had a strong liquidity preference. The net effect of additional sales of treasury bills on banks' liquidity depended on whether any of these excess reserves would have gone into increased lending instead of additional treasury bills. If so, then there was some dampening of liquidity. This damp- ening was unlikely, however, because the reasons why banks had excess re- serves in the first place had not changed. From a commercial bank's perspec- tive, lending to the private sector was much like an equity investment, because audits had little credibility, making systematic risk assessment difficult; and foreclosure or loan recovery through the courts was difficult, if not impos- sible. If, for these reasons, additional treasury bills did not substitute for lend- ing, then sales in the primary auction served as substitutes for otherwise unremunerated excess reserves, and had no impact on monetary conditions- a classic liquidity trap. The banks also held excess reserves because of a strong liquidity prefer- ence in the absence of an interbank or overnight market. Beyond the point where additional treasury bills served to remunerate excess reserves, banks required sharply higher interest rates to compensate for the increased risk of reduced liquidity. As a result, increased primary issues of treasury bills be- yond the point where they started to have a monetary impact led to a sharply higher interest cost for the government budget. As mentioned before, in the 74 Mark Henstridge and Louis Kasekende government's view a short-term fiscal adjustment provided more monetary bite for a given shilling cost than did increased issues of treasury bills. Hence, fiscal adjustment was the preferred instrument of short-term management. Largely as a result of a combination of macroeconomic stability and im- provements in financial oversight and bank supervision, financial markets in Uganda have become more sophisticated in recent years. The level of monetization (the ratio of broad money to GDP) increased to 13 percent of GDP in 1998, compared with 10 percent in 1994. Financial savings as a ratio to GDP increased to 35 percent in 1998, compared with 27 percent in 1994. The ratio of the outstanding stock of treasury bills to broad money is 23 percent, compared with 11 percent in 1994, and bank holdings of excess reserves have been reduced. These developments have increased the po- tency of monetary policy and strengthened the link between the foreign exchange market and the domestic financial markets. During an episode of sharp depreciation in 1999, the central bank tightened monetary conditions using indirect monetary instruments (the rediscount rate and issues of trea- sury bills) with some success. Costs of Budget Discipline Making within-year adjustments to the releases of funds clearly disrupts original spending plans, partly because it increases volatility, and-argu- ably more seriously-because it weakens the budget as the instrument for allocating public resources.'4 These are common objections to the use of spending cuts in implementing macroeconomic policy and to the use of cash budgets. However, the extent to which the use of the cash flow system for macro- economic management can be blamed for the costs is limited. Ministries could face within-year cuts to their approved budgets for two reasons. In one case, they may need to cut aggregate total spending to make up for a shortfall in the resource envelope. In another, they might need to cut one ministry's bud- get to accommodate extra spending by another ministry to keep total spend- ing under control. The accommodation of powerful ministries' overspend- ing-usually those close to the president-accounted for two-thirds of the variation between the budgeted allocations for each ministry and the even- tual out-turn between 1992 and 1997 (Moon 1997). The cash flow system was not responsible for the prevalence or persistence of demands for supplemen- tary spending-indeed, Stasavage and Moyo (1999) argue that such ill-disci- pline was at least as rampant prior to the adoption of the cash flow system- and in fact, it was essential for containing the damage. The combination of 14. Although assessing the microeconomic costs would be difficult, such assess- ments could strengthen the apparently weak connection between the release of funds and the achievement of the desired delivery of public services at the facility level (see chapter 11 in this volume). Exchange Reforms, Stabilization, and Fiscal Management 75 resource-constrained budgeting ex ante and its implementation ex post us- ing cash flow simply made abysmal budget discipline more obvious. The public spending program most frequently suffered from unexpected cuts because supplementary expenditures were approved, rather than as a result of the operation of cash flow per se. Furthermore, when the loss of expenditure control results in sharply in- creased inflation, the real purchasing power of the cash allocations in the original budget would be diminished anyway. Therefore, cutting nominal allocations to control inflation may not be as costly for spending programs as it first appears. In any case, the appropriate tradeoff between the costs and benefits of mak- ing expenditure cuts should extend beyond the confines of the government's expenditure program, across the entire economy. If the government decides not to cut expenditure, the benefits to spending ministries of receiving their nominal allocations-despite increased inflation-and the benefits to the in- tegrity of the budget process have to be weighed against the costs of the in- creased inflation to the private sector. The indicators shown in table 3.1 imply that these costs would have been large. The impact on expenditures for social programs that would help reduce poverty (such as health and education) was minimized by designating this spending as "priority program area" expendi- tures, which would not be subject to cuts. In effect, protecting the priority pro- gram areas from cuts was equivalent to a contingent decision on which items would be hit if, or in most years when, budget discipline broke down. As mentioned previously, a drawback of using intrayear adjustment of spending was that the cash flow system imposed discipline on the total bud- get aggregates, which undermined the budget formulation process as some ministries became increasingly skeptical whether their allocations, approved by parliament, would survive the first few months of supplementaries. Evi- dently, the solution to this problem is not to relax fiscal policy, but to strengthen budget discipline through political will."5 Commitments and Arrears Achieving restraint through control of cash spending does not prevent an accumulation of commitments to spend. If, in the face of a commitment to spend, the cash for payment is not released, the spending has not been cut, it has instead been financed by the accumulation of arrears. As well as dis- torting the allocation of public resources, the accumulation of arrears can also lead to inflated contracts, as suppliers factor into their prices the likely delay in payment as well as a premium for the risk of not getting paid at all. A system of local letters of credit was set up to prevent suppliers from 15. The costs of poor budget discipline had been a feature of the BFP since 1995/ 96. From 1997/98 onward, improvements in discipline have been addressed directly in the BFP as part of the process of compiling the budget. 76 Mark Henstridge and Louis Kasekende contracting to deliver or actually delivering goods and services unless the means of payment had been earmarked in a government account at the Bank of Uganda. This system did not prevent arrears from accumulating to a total of about U Sh 260 billion by 1999 (28 percent of domestic revenue), partly due to a cumbersome system, but mainly because suppliers appeared willing to risk delayed payment. In 1997 it was established that no commitment was valid unless approved by the Ministry of Finance. In 1999/2000 a new system of commitment con- trol was implemented (with technical support from the IMF) across all cen- tral government spending agencies for nonwage expenditures. As a result, overcommitments have been sharply reduced. During the first two quarters of 1999/2000, overcommitments are estimated to have totaled only U Sh 6.4 billion, compared with an annual accumulation of domestic arrears of more than U Sh 100 billion, on average, over the previous three years. Conclusions The essential foundations both for the subsequent reforms and the resur- gence of growth and decline in poverty were the legalization of the parallel market for foreign exchange and the achievement of macroeconomic stabil- ity. These reforms originated not from conditions imposed by the World Bank or the IMF, but from within the govemment. The debate about the exchange rate in 1989 and 1990, as well as the con- trasting views on the gains from low inflation prior to 1992, show that the NRM govemment was not a monolith. A monolithic model of government cannot account for the shift away from the early experiment with revival of the control economy and revaluation in 1986/87. Once the debate on the di- rection of the exchange rate was settled by the early 1990s, reforms were profound and decisive. A perhaps unusual, but effective, approach to stabilization combined a resource-constrained budget and its implementation through a monthly cash flow system. The latter gave the government the ability to respond to shocks and ensure that the intended outcome of low inflation was indeed achieved. The cash flow approach used in Uganda has proved more flexible and rela- tively less costly than the monthly cash budget used, for example, in Tanza- nia and Zambia. The cash flow system enabled annual budget implementa- tion to be more flexibly adjusted to keep inflation low in response to changing macroeconomic conditions, a shortfall in revenue, or within-year demands for extra spending from some parts of the government. Targeting inflation through short-term fiscal adjustment, however, in- volved some tradeoffs and costs. First, the demand for additional spending within the budget year confronted the Ministry of Finance with a tradeoff between preserving the budget aggregates and preserving the original bud- get allocations to other ministries. The resolution of this dilemma was clear: the aggregates-and macroeconomic stability-were preserved. Although this Exchange Reforms, Stabilization, and Fiscal Management 77 decision disrupted some expenditure programs, the costs to the spending programs were clearly outweighed by the benefits of low inflation to the rest of the economy. This chapter argues that poor budget discipline-rather than the cash flow system that was used to manage the budget-is mostly respon- sible for the disruption in the spending programs. Second, in planning fiscal policy, particularly during the period of high coffee prices and increased foreign inflows in the mid-1990s, the increased donor financing of the budget and the desire to contain an appreciation of the exchange rate posed another tradeoff. The Bank of Uganda had to bal- ance the intervention in the foreign exchange market to contain the nominal appreciation against the fiscal adjustment needed to keep the domestic price level stable (and thereby avoid a real appreciation). This tradeoff was man- aged by maintaining the target of low inflation as the primary objective. Drawbacks to a reliance on short-term fiscal adjustments include-in addition to the disruption to spending programs-the potential for accu- mulation of domestic arrears. However, tackling both budget discipline and arrears will become easier as the constraints that led to the adoption of a cash flow system in the first place-the limited scope for independent mon- etary policy and the strong likelihood of shocks-ease, and the horizon over which policy is implemented can stretch over periods longer than a month or a quarter. The conclusion that cash flow was a vital part of the maintenance of mac- roeconomic stability in Uganda echoes Treasury Secretary Tumusiime- Mutebile's (1998) contention that "The Budget Framework Process and cash flow have been the two most important technical instruments in maintain- ing stability since 1992." Economic conditions in Uganda since 1992, includ- ing liberalized, flexible prices, thin or nonexistent financial markets, trade shocks, and flows of foreign capital, supported the use of the cash flow sys- tem as part of macroeconomic management. Moreover, Honohan and O'Connell's (1997) review of the development of monetary regimes across Africa over the last 30 years shows a trend toward such conditions across the continent. They conclude that greater fiscal restraint is needed to make the transition toward a more open and flexible economy, which, in turn, exacerbate problems of policy credibility and macroeconomic management. To the extent that Uganda's monetary conditions exist elsewhere, its suc- cessful experience with macroeconomic management may be a relevant model for other countries as well. References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Adam, Christopher S., and David L. Bevan. 1997. "Fiscal Restraint and the Cash Budget in Zambia." University of Oxford, Centre for the Study of African Economies, U.K. Processed. 78 Mark Henstridge and Louis Kasekende Bevan, David L. 1998. "Uganda Public Expenditure Review: Macroeconomic Options in the Medium Term." University of Oxford, St. John's Col- lege, U.K. Processed. Bevan, David L., Paul Collier, and Jan-Willem Gunning. 1989. Peasants and Governments: An Economic Analysis. Oxford, U.K.: Clarendon Press. _.___ 1990. Controlled Open Economies: A Neoclassical Approach to Structural- ism. Oxford, U.K.: Clarendon Press. Collier, Paul, and Jan-Willem Gunning. 1996. "Policy Towards Commodity Shocks in Developing Countries." Working Paper no. 96/84. Interna- tional Monetary Fund, Washington, D.C. Henstridge, Mark. 1997. "Implementing Fiscal Adjustment: Uganda's Cash Flow." University of Oxford, Centre for the Study of African Econo- mies, U.K. Processed. .1999. "De-monetization, Inflation, and Coffee: The Demand for Money in Uganda." Journal of African Economies 8(3): 345-85. Forthcoming. "Fiscal Management and Macroeconomic Stability in Uganda." Working paper. International Monetary Fund, Washington, D.C. Honohan, Patrick, and Stephen A. O'Connell. 1997. "Contrasting Monetary Regimes in Africa." Working Paper no. 97/64. International Monetary Fund, Washington, D.C. Kasekende, Louis A., and Moazzam Malik. 1994. "Dual Exchange Regimes, Unification, and Development: The Case of Uganda." Paper presented at the Conference on Adjustment and Poverty in Sub-Saharan Africa, March, Accra, Ghana. Kasekende, Louis A., and Germina Ssemogerere. 1994. "Exchange Rate Uni- fication and Economic Development: The Case of Uganda, 1987-92." World Development 22(8): 1183-98. Kasekende, Louis A., Damoni Kitabire, and Matthew Martin. 1996. "Capital Inflows and Macroeconomic Policy in Sub-Saharan Africa." In Inter- national Monetary and Financial Issues for the 1990s, vol. 8. New York: United Nations. Kharas, Homi, and Brian Pinto. 1989. "Exchange Rate Rules, Black Market Premia, and Fiscal Deficits: The Bolivian Hyperinflation." Review of Economic Studies 56(3): 435-47.' Lizondo, Jose Saul. 1987. "Unification of Dual Exchange Markets." Journal of International Economics 22: 57-77. Moon, Allister J. 1997. "Uganda's Budget Framework: Presentation to the Parliament of Uganda." Uganda Public Expenditure Review FY 97/ 98. World Bank, Washington, D.C. Processed. Exchange Reforms, Stabilization, and Fiscal Management 79 Morris, Stephen. 1989a. "Macroeconomic Features of the Uganda Economy and Some Policy Implications. Part One: The Relationship between Money Prices and the Parallel Market Exchange Rate." Discussion Paper no. 1. Ministry of Planning and Economic Development, Kampala. .1989b. "Macroeconomic Features of the Ugandan Economy and Some Policy Implications. Part Two: The Impact of Official Exchange Rate Devaluation on Uganda." Discussion Paper no. 2. Ministry of Plan- ning and Economic Development, Kampala. . 1995. "Inflation Dynamics and the Parallel Market for Foreign Ex- change." Journal of Development Economics 46: 295-316. Obbo, Sam, and John Baptist Waswa. 1992 "Budget aims at 15 Percent Infla- tion." The New Vision 7(154): 1-16. Pinto, Brian. 1988. "Black Markets for Foreign Exchange, Real Exchange Rates, and Inflation: Overnight Versus Gradual Reform." Policy Research Working Paper no. 84. World Bank, Development Research Group, Washington, D.C. .1989. "Black Market Premia, Exchange Rate Unification, and Inflation in Sub-Saharan Africa." World Bank Economic Review 3(3): 321-38. Republic of Uganda. 1992. "The 'Way Forward I.' Macroeconomic Strategy, 1990-1995." Ministry of Planning and Economic Development, Kampala. . 2000. "Poverty Eradication Action Plan. Second Draft." Ministry of Finance, Planning and Economic Development, Kampala. _.__ -Various years. "Background to the Budget." Ministry of Finance, Plan- ning and Economic Development. Kampala. Stasavage, David, and Dambisa Moyo. 1999. "Are Cash Budgets a Cure for Excess Fiscal Deficits (and at What Cost)?" Working paper no. 99.11. University of Oxford Centre for the Study of African Economies, U.K. Tumusiime-Mutebile, Emmanuel. 1998. "Opening the Budget to Stakehold- ers: Use of MTEF/PER Process as a Catalyst for Enhanced Account- ability." Paper presented at the Budget Process and Foreign Aid Work- shop, November, World Bank, Washington, D.C. Processed. . 1999. "Uganda's Experience with the Medium-Term Expenditure Framework." Ministry of Finance, Planning and Economic Develop- ment, Kampala. Processed. Part II Household Responses and Constraints 4 Changes in Poverty and Inequality Simon Appleton According to macroeconomic data, Uganda in the 1990s was a rare economic success story in Sub-Saharan Africa. However, it has been questioned whether the growth recorded in official statistics was reflected in rising living stan- dards, particularly for the poor. This concern was, for example, voiced in the 1997 Human Development Report: "The perennial concern is that the benefits of strong growth have yet to translate into measurable improvements in the standard of living for the majority of people" (UNDP 1997, p. 2). Widespread concern about unchanging poverty levels is also reflected in the report from the Uganda Participatory Poverty Assessment Project (UPPAP), a major attempt by the Ugandan government to consult the people and hear "the voices of the poor." A summary of UPPAP's major findings based on seven districts concluded: "Through analysis of long-term trends in poverty, many local people felt that poverty was worsening in their communities.. . Local people reported more movement into poverty than out of it" (Republic of Uganda 1999, p. 10). Given these perceptions one might question whether the growth appar- ent in the national accounts led to substantial reductions in poverty. Aside from household surveys, national accounts can draw upon little hard informa- tion on incomes from nonexport agriculture and informal sector activities, Uganda's most important income sources. Moreover, national accounts tell us nothing about how incomes are distributed. The author is grateful to the Ugandan Bureau of Statistics for access to the data. Tom Emwanu, Johnson Kagugube, Margaret Kakande, and James Muwonge helped with some of the analysis. In addition to the editors of this volume, Lionel Demery, Jesko Hentschel, John Mackinnon, Francis Teal, and participants in a semi- nar at the University of Oxford provided valuable comments. 83 84 Simon Appleton Fortunately, Uganda is one of the few Sub-Saharan countries that can con- vincingly address the question of what happened to poverty-as measured by private consumption-in the 1990s. This is because of a large-scale household survey program that began in 1992 with the integrated household survey (IHS). This baseline survey was followed by four monitoring surveys (MS-1, MS-2, MS-3, and MS-4) designed to monitor living standards on virtually an annual basis. The surveys have large samples-typically 5,000 households-but are particularly impressive in the number of communities sampled, typically 500. The surveys are designed to be nationally representative, although a few inse- cure areas were excluded (see appendix A at the end of the book for details). All five surveys rely on similar sampling procedures and questionnaires (see annex 4.1 for further details on the surveys and the adjustments needed to compare real private consumption over time).1 This chapter uses the household data from these surveys to estimate changes in average living standards, poverty, and inequality from 1992 to 1997/98.2 During this period, the growth in mean consumption per capita estimated from the household surveys matches that reported in the national accounts. More- over, at all points of the income distribution, households are better off in 1998 than 1992. This implies that-regardless of where the poverty line is set- poverty was reduced in the period. Indeed, at the lower points of the income distribution, living standards grew more than at the mean. Consequently, in- equality was reduced. Both overall growth and falling inequality contributed 1. An earlier attempt to compare the IHS with an earlier survey, the household budget survey (HBS) of 1989, was unsuccessful (Appleton 1996). The HBS was reana- lyzed as part of the preparation for this chapter, but still produced apparently incom- parable results with the IHS and the monitoring surveys. Consumption in the HBS appears too high relative to the subsequent surveys. Appleton (1996) suggested that the incomparability arose from questionnaire design problems with the IHS. How- ever, this suggestion appears less plausible given the evidence in this chapter of the comparability of the IHS results and those from the monitoring surveys. These moni- toring surveys were not subject to the same supposed problems of questionnaire de- sign as was the IHS. Sampling problems with the HBS may be a more likely explana- tion for the incomparability of results. Mean household size was one person higher in the HBS than in the census of 1991 and the subsequent household surveys. 2. This analysis complements the poverty study carried out using the surveys conducted by the government's Coordination of Poverty Eradication Project and Department of Statistics (Republic of Uganda 1997b). The earlier study was conducted before the release of the third and fourth monitoring surveys and used a poverty line defined as two-thirds of mean consumption per adult equivalent in the IHS. This chapter derives a poverty line based on calorie requirements and updates the analy- sis. It also includes some additional adjustments to ensure comparability of the con- sumption data, together with some further decompositions of interest. Note, how- ever, that the two studies, despite rather different methods, agree on the general direction of poverty trends in Uganda during the period. Changes in Poverty and Inequality 85 to poverty reduction, although growth in incomes contributed the most. That is not to say that some households did not became poorer during the period, just that such adverse movements were more than offset by other households escaping poverty. We derive an absolute poverty line for 1992 that defines around 56 percent of the population as poor (figure 4.1). By 1997/98, this pro- portion had fallen to 44 percent. Clearly, this is a substantial poverty reduc- tion-more than 20 percent in just five years. The reduction in poverty was neither uniform across all regions, nor the same each year. Poverty fell most in the central region and least in the eastern region. Poverty reduction was most marked at the beginning and at the end of the period, with no gains for the poorest indicated during the first three monitoring surveys (1993/94 to 1995/ 96). However, poverty did fall in every region between 1992 and 1997/98; na- tionally, it also fell between every survey. An apparent discrepancy exists between the findings from the house- hold surveys and the perceptions of increased poverty as reported in the UPPAP and more widely in Uganda. It should be noted that the UPPAP uses a wider concept of poverty. This concept goes beyond the private con- sumption measured in the household surveys and includes insecurity and poor government services. Security is better than in the early 1980s, but the situation deteriorated in parts of the country during the household survey periods, as shown by the increasing number of districts excluded from the Figure 4.1. Poverty in Uganda, 1992 and 1997/98 80 - 60 - 40 - I- 20 National Rural Urban Central Westem Eastern Northern * 1992 ] 1997/98 Source: Author's calculations. 86 Simon Appleton sample during the course of the program. For public services, evidence ex- ists-for example, from the Ugandan school surveys of 1996 and 2000-of an improvement, although the level of provision still remains poor. As dis- cussed elsewhere in this volume, the universal primary education initia- tive of 1997 dramatically widened access to education, although perhaps at some cost to quality. The spread of AIDS constitutes a grave threat to health, although there is some evidence of falling child mortality based on the Demographic and Health Surveys of 1988 and 1995. Aside from the differing conceptions of poverty, other methodological issues might explain the apparent discrepancy between the qualitative and quantitative evidence. One is a possible difference in the time horizon. In part, perceptions of increasing poverty may refer to trends over a longer time horizon than the five-year interval covered by the household surveys.3 More- over, the UPPAP report had the difficult task of drawing general conclusions about poverty trends based on a mass of qualitative data. As already stated, the time trend analysis for the Iboa community tried to quantify eight factors related to poverty for seven time periods. This must be combined with simi- larly disaggregated data for the other 35 communities. It seems doubtful that the UPPAP attempted a quantitative aggregation of this disparate data and, indeed, at times the conclusions seem to sacrifice rigor for comprehensive- ness. Moreover, as far as can be ascertained, the time trend analysis in the UPPAP provides no information about distribution within communities. The eight factors identified as related to poverty in the Iboa community-such as food availability or access to health services-appear to pertain to the com- munity as a whole and not specifically to its poorer members. More work 3. In the UPPAP, time trend analysis was done in five-year intervals, often start- ing in 1970-74 and going through to 1995-99. For each five-year interval, members of local communities placed pebbles to quantify indicators or causes of poverty. For example, in the time trend diagram illustrated in the UPPAP report, the chosen com- munity (Iboa in the Moyo district) placed 10 stones for food availability in 1970-74 and only 4 for food availability in 1995-99 (Republic of Uganda 1999, figure 2.5). This suggests a perception of greatly increasing poverty over the long term. However, for the subperiod covered by the surveys, food availability is reported not to deterio- rate-remaining at four stones throughout the 1990s (up from one stone in 1985- 89)-and community assessments for 2000 and beyond are relatively optimistic (eight stones). Food availability is just one of eight factors presented in the illustrative time trend diagram. However, for most of the other factors, there is no deterioration be- tween 1990-94 and 1995-99; indeed there is substantial improvement in most factors compared with 1985-89. The illustration of the time trend analysis for the Iboa com- munity raises questions about the conclusion of the UPPAP report that there was increasing poverty in the Moyo district. The Iboa community was one of probably only four communities visited by UPPAP in Moyo (36 communities were visited from 9 districts). However, the time trend analysis presented for the Iboa community pro- vides no indication that poverty was increasing in the district in the 1990s. Changes in Poverty and Inequality 87 needs to be done to reconcile the findings of the UPPAP and the survey data. At this stage, the extent to which there is a genuine contradiction is not clear, still less whether the UPPAP's analysis of poverty trends is to be preferred. The remainder of this chapter is organized as follows. First, it outlines the growth in mean consumption per capita as reported by the household sur- veys between 1992 and 1997/98 and shows this growth to be close to that estimated in the national accounts. It then derives an absolute poverty line for Uganda and uses it to show the substantial reduction in poverty during the surveys. Falls in inequality are also documented, although they are shown to explain only a small part of the reduction in poverty relative to general economic growth. Poverty reduction is decomposed by economic sectors, with the cash crop sector shown to account for more than half of the fall in poverty. The chapter ends by summarizing the main results. Those inter- ested in the methodological details of the analysis should refer to annex 4.1. Changes in Mean Consumption Per Capita The focus of this chapter is household private consumption per capita as a measure of individual welfare. This is not to deny that other dimensions of well-being are important. However, private consumption-including consump- tion of home-produced food-is a central indicator of economic welfare. We do not address the issue of intrahousehold allocation, but see Appleton, Chessa, and Hoddinott (1999) for suggestive evidence of boy-girl discrimination based on the same data. We adjust the consumption figures reported in the house- hold surveys in a number of ways detailed in annex 4.1, perhaps the most important being to take account of regional differences in food prices. After the adjustments, the household surveys imply that real consumption per capita rose by 16.5 percent between 1992 and 1997/98 (see table 4.1). For both rural and urban areas separately, the rise was slightly less, measuring 15.9 and 11.4 percent, respectively This discrepancy between the national and disaggregated figures can be explained by the increase in the estimate of the relative size of the urban population from 12.4 percent in the IHS to 13.3 percent in MS-4. The overall rise in mean consumption in the bottom line, fully adjusted figures is not driven by the adjustments to the data. Indeed, the adjustments lead to a downward revision of the growth implied in the unadjusted figures. The average growth in mean consumption per capita from the first to last household survey is close to that estimated in the national accounts.4 Making precise comparisons is hard, because the national accounts data are reported in fiscal years (uly to June), whereas the IHS and MS-4 covered something closer to calendar years (see book appendix A). The surveys from the IHS to 4. Note that the figures in table 4.1 differ from constant price consumption as reported in the national accounts, because they use the consumer price index rather than gross domestic product deflator. 88 Simon Appleton Table 4.1. Estimates of Private Consumption Per Capita National accounts Nominal Real Percentage Fiscal year (U Sh/month) (1989 U Sh prices) growth (real) 1991/92 12,094 6,205 n.a. 1992/93 16,167 6,380 2.8 1993/94 16,949 6,275 -1.6 1994/95 19,824 6,917 10.2 1995/96 22,151 7,192 4.0 1996/97 24,070 7,243 0.7 1997/98 26,067 7,414 2.4 Household surveys (U Sh per month, 1989 prices) Year National Rural Urban IHS (1992) 5,452 4,735 10,752 MS-1 (1993/94) 5,718 4,862 11,645 MS-2 (1994/95) 6,058 5,206 12,067 MS-3 (1995/96) 6,187 5,242 12,246 MS-4 (1997/98) 6,353 5,488 11,979 n.a. Not applicable. Note: National accounts data are in fiscal years (July 1-june 30). Real consumption is obtained using the consumer price index as a deflator. Source: Author's calculations based on household survey data; national accounts data from the Uganda Bureau of Statistics. MS-4 span an interval of almost exactly five years, with the mid-points of both surveys falling around August. However, in the five-year interval from fiscal year 1992/93 to fiscal 1997/98, real private consumption per capita in the na- tional accounts rose by 16.1 percent. These figures are remarkably close to the 16.5 percent figure derived from the household surveys. The two estimates may not be strictly independent, because the household survey data were one source used in estimating consumption in the national accounts. However, some of the monitoring surveys may not have been used for the national ac- counts estimates, as there was a substantial lag in cleaning the data and writ- ing the official survey reports. Moreover, both the level of consumption and the patterns of year-on-year changes were different in the macro and micro data. The household surveys reported substantially lower levels of private con- sumption than did the national accounts: in some cases, the discrepancy is almost a third. The household survey data also showed a smoother pattern of growth than did the macro figures. Mean consumption per capita rose strongly between each survey: by 4.9 percent between IHS and MS-1, by 5.9 percent Changes in Poverty and Inequality 89 between MS-1 and MS-2, by 2.1 percent between MS-2 and MS-3, and by 2.7 percent between MS-3 and MS-4. The phasing of the increases in consumption over the five years was very different for urban and rural areas. In line with the national pattern, living standards in rural areas grew fairly steadily between each survey. In urban areas, most of the growth in the period occurred be- tween 1992 and 1993/4, when real consumption per capita rose by 8.3 percent. In summary, the household surveys broadly corroborated the improve- ment in living standards recorded in the national accounts. One new piece of information they provided was the rural-urban breakdown. In particular, they showed that rural areas enjoyed growth comparable to that experienced in urban areas. However, the discussion so far has been in terms of develop- ments at the mean. To measure the reduction in poverty, we need to go fur- ther, beginning with setting a poverty line. Defining an Absolute Poverty Line for Uganda Uganda currently does not have an officially approved poverty line.5 This section constructs an absolute poverty line, reflecting the monetary cost of meeting certain basic needs. When using a poverty line to evaluate improve- ments in living standards of the poor over time, it is desirable to fix the pov- erty line in real terms. If the poverty line is made relative and allowed to rise with improvements in general living standards, then it is possible that pov- erty will rise despite improvement in the living standards of the poor. While such an increase in relative poverty may be interesting, our focus in this chap- ter is on whether poorer people have become materially better off. This is not to deny that poverty ultimately has an important relative aspect and that countries may want to set higher poverty lines as they become more affluent. However, for the relatively short period analyzed here, this does not seem to be a relevant issue. We set the poverty line by following the approach of Ravallion and Bidani (1994). Further information on the derivation of the line is in annex 4.1. Here 5. Kikafunda, Serunjogi, and Migadde (1992) have estimated a nutrition-based absolute poverty line for Uganda using the 1989/90 HBS. They arrive at a figure of U Sh 6,745 per month per person. This is somewhat higher than the estimate in this chapter of U Sh 6,252 (1989 prices) per adult equivalent per month. The use of the HBS, which excludes much of the north region and has higher consumption esti- mates, may partially account for the discrepancy. There are also differences in method: Kikafunda, Serunjogi, and Migadde (1992) used regional food baskets and appear to have allowed for heavy levels of meat consumption. In the western rural and many other areas, their baskets allow the poor to eat 64 percent as much meat (in kilo- grams) as matooke (plantain) (see table 7, p.38). By contrast, in this chapter, the ratio of meat to matooke weights in the food basket is 1.6 percent. 90 Simon Appleton we provide merely an overview. Note that a large degree of judgment is in- volved in setting a poverty line. Consequently, one should not attach too much importance to the estimates we derive for the level of poverty in any single year or place. Instead, attention should be focused on poverty com- parisons, and indeed, on whether they are robust with regard to setting the poverty line. In common with most of the literature, the Ravallion and Bidani approach focuses on defining food-related needs and only indirectly esti- mates nonfood requirements. Specifically, it first defines a food poverty line based on the cost of obtaining sufficient calories given the typical food bas- ket of the poor. This approach contains an element of circularity, as one does not know which people are poor before the poverty line has been set. How- ever, in the case of Uganda, we focused on the food-basket of the poorest 50 percent of the population, because previous studies using similar methods had identified around half of the population as poor (Republic of Uganda 1996). We initially focus on the cost of obtaining 3,000 calories, which roughly corresponds to the energy requirements of male subsistence farmers accord- ing to the principles set by the World Health Organization (WHO 1985). Table 4.2 shows the derivation of the food poverty line, defined as the cost of ob- taining 3,000 calories, using the food basket of the poorest 50 percent of the population in the first monitoring survey (MS-i1). The food basket is valued in constant prices, namely, the median food prices reported in the survey. In practice, food prices differ between regions, and this variation implies differ- ent food poverty lines in nominal prices (see table 4.3). Under the Ravallion and Bidani approach, nonfood requirements are es- timated as the nonfood spending of households whose total consumption is just equal to the food poverty line. The rationale for using nonfood spending is that if households are sacrificing the food expenditure needed to meet their calorie requirements for nonfood spending, then nonfood spending should also be considered as needed. As the surveys do not provide information about nonfood prices, we allow nonfood requirements to vary by region and by rural-urban location. Table 4.3 reports the predicted food share of those whose total consumption was just equal to the food poverty line in various locations and the implied total poverty lines. We also report a national aver- age poverty line, although in the analysis we used the poverty line specific to the location of the household. The poverty lines reported in table 4.3 are the expenditures required to obtain 3,000 calories and meet nonfood requirements. In practice, many house- hold members do not need 3,000 calories, as calorie requirements vary by age and by sex. We use equivalence scales to allow for this (see table A4.2). For example, children are assigned equivalence scales equal to their calorie require- ments divided by 3,000. This implicitly assumes that children's nonfood re- quirements are lower than those of men in the same proportion as their calorie requirements. For women, such an assumption is less defensible, and instead we assume that their nonfood needs are the same as men's. Given these equiva- lence scales, members of a household are said to be poor when their total con- sumption per adult equivalent falls below the poverty line. Changes in Poverty and Inequality 91 Table 4.2. Derivation of the Food Poverty Line (MS-1 prices) Cost per month Quantity (U Sh (kg per Price Caloriesl Retention Calories 1993 Food item month) (U Sh/lkg) kg ratio per day prices) Matooke (plantain) 28.54 67 770 0.50 366 1,903 Sweet potatoes 34.12 63 1,020 0.70 812 2,133 Cassava 9.02 200 2,557 0.89 684 1,804 Irish potatoes 0.36 250 750 0.85 8 89 Rice 0.06 700 3,600 1.00 7 42 Maize (grain) 0.30 400 3,470 0.90 32 121 Maize (flour) 1.54 350 3,540 1.00 181 538 Bread 0.02 1,300 2,490 1.00 1 20 Millet 2.25 300 3,231 0.65 158 676 Sorghum 1.57 200 3,450 0.90 163 314 Beef 0.31 1,100 2,340 0.80 19 339 Other meat 0.05 1,000 2,340 0.75 3 52 Chicken 0.09 1,167 1,460 0.61 3 111 Fresh fish 0.62 467 1,030 0.60 13 290 Smoked fish 0.39 583 3,005 0.70 28 229 Eggs 0.00 2,000 1,490 0.88 0 8 Milk 0.55 400 640 1.00 12 219 Cooking oil/ghee 0.06 1,400 8,570 1.00 18 89 Passion fruit 0.10 382 920 0.75 2 37 Sweet bananas 2.34 50 1,160 0.56 51 117 Tomatoes 0.70 192 200 0.95 4 134 Cabbages 0.33 125 230 0.78 2 41 Beans (fresh) 0.73 400 1,040 0.75 19 292 Beans (dry) 2.86 350 3,300 0.75 236 1,002 Groundnuts 0.59 600 2,350 0.93 43 355 Sim-sim 0.45 222 5,930 1.00 89 100 Sugar 0.35 1,000 3,750 1.00 44 352 Total 3,000 11,463 Source: Author's calculations using data from MS-1 provided by the Uganda Bureau of Statistics. Poverty Trends Tables 4.4 to 4.8 present the poverty statistics for the five surveys (see an- nex 4.1 for definitions). Data are disaggregated by location, both by urban- rural and by the four regions of the country. Along with the poverty statis- tics, we report the percentage of people in each location and their mean household consumption per adult equivalent. We also report the contribu- tion each location makes to each poverty statistic (that is, what percentage 92 Simon Appleton Table 4.3. Poverty Lines (1993 prices) Predicted Poverty line Food poverty Poverty line Region food share (constant prices) line (nominal) (nominal) Central rural 0.609 15,947 13,971 19,435 Central urban 0.490 17,314 14,837 22,409 Eastern rural 0.653 15,446 8,832 11,900 Eastern urban 0.557 16,548 11,300 16,312 Western rural 0.675 15,189 8,209 10,877 Western urban 0.589 16,174 9,245 13,043 Northern rural 0.638 15,610 8,410 11,452 Northern urban 0.578 16,304 9,433 13,417 National (average) 0.566 16,443 11,463 16,443 Note: Nominal and national lines are shown for information only and are not used in the analysis. Source: Author's calculations using data from MS-1 provided by the Uganda Bureau of Statistics. of national poverty is attributable to each location). Given that poverty sta- tistics are estimates, it is useful to test whether changes in their values are statistically significant (Kakwani 1993). We report tests of the significance of the changes in the poverty statistics between IHS and MS-4 in table 4.9. We report three types of poverty statistics, termed P0, Pl, and P2. All three statistics are from the so-called Pc or Foster, Greer, and Thorbecke (1984) class of indicators. The P0 statistic is simply the proportion of Ugandans esti- mated to live below the poverty line. In the first survey, the IHS, the propor- tion was 56 percent. This statistic shows that absolute poverty levels were very high in Uganda. Most people did not have enough money to meet our estimate of their basic needs. Indeed, 36 percent of Ugandans lived below the food poverty line (statistics on food poverty are not reported in the tables but are available from the author on request). To restate these findings, more than one-third did not have enough even to meet only their calorie require- ments, let alone any other needs.6 These high poverty rates are perhaps not surprising given the country's low national income (ranked sixth lowest in the world in 1992 by the World Bank [1994]). Poverty rates in urban areas were much lower than in rural areas, but were nonetheless substantial: 28 percent of urban people lived below the poverty line and 11 percent lived 6. This finding raises the question of what happens to those Ugandans who we estimate are not getting enough calories. Note that we probably overestimate malnu- trition and poverty because of general measurement error (as some people may un- derreport consumption), and because we estimate calories obtained from food pur- chases made over a short recall period (some people may be living on stocks of Table 4.4. Poverty in the Integrated Household Survey Population Contribution to Location share (percent) Mean CPAE P0 P P2 p1 PI 2 National 100.0 6,900 55.7 20.3 9.90 100.0 100.0 100.0 Rural 87.6 6,091 59.7 22.0 10.81 93.8 94.9 95.6 Urban 12.4 12,608 27.8 8.3 3.48 6.2 5.1 4.4 Central 30.6 8,865 45.6 15.3 7.04 25.1 23.1 21.8 East 27.9 6,115 58.8 22.0 10.85 29.4 30.3 30.6 West 24.2 6,449 53.1 18.7 9.01 23.0 22.3 22.0 North 17.3 5,317 72.2 28.6 14.64 22.4 24.4 25.6 Central rural 22.7 6,861 54.3 18.7 8.76 22.1 20.8 20.1 Central urban 8.0 14,564 20.8 5.7 2.16 3.0 2.2 1.7 East rural 25.4 5,866 60.6 23.0 11.38 27.6 28.7 29.2 East urban 2.5 8,633 40.4 12.6 5.52 1.8 1.6 1.4 West rural 23.1 6,223 54.3 19.2 9.31 22.5 21.9 21.7 West urban 1.1 11,299 28.9 7.3 2.60 0.6 0.4 0.3 North rural 16.5 5,195 73.0 29.0 14.83 21.6 23.5 24.7 North urban 0.8 7,677 55.2 21.2 10.92 0.8 0.9 0.9 CPAE Consumption per adult per equivalent (1989 Uganda shillings per month). Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Table 4.5. Poverty Rates in MS-1 Population Contribution to Location share (percent) Mean CPAE P pI PI2 PO P2 National 100.0 7,281 51.2 16.9 7.48 100.0 100.0 100.0 Rural 87.4 6,327 55.6 18.6 8.27 94.8 95.9 96.6 Urban 12.6 13,885 21.0 5.5 2.02 5.2 4.1 3.4 Central 31.4 9,860 34.5 10.4 4.26 21.2 19.3 17.9 East 26.5 6,085 57.6 19.7 9.06 29.9 30.9 32.1 West 26.3 6,527 53.9 17.4 7.31 27.7 27.0 25.7 North 15.7 5,403 69.3 24.6 11.57 21.2 22.8 24.3 Central rural 23.1 7,635 41.9 12.9 5.39 18.9 17.6 16.6 Central urban 8.3 16,044 13.9 3.3 1.10 2.2 1.6 1.2 East rural 24.5 5,783 59.8 20.6 9.56 28.6 29.9 31.3 East urban 2.0 9,765 31.4 8.1 3.00 1.2 1.0 0.8 West rural 25.2 6,307 55.3 17.8 7.52 27.2 26.5 25.3 West urban 1.2 11,219 24.7 7.4 2.73 0.6 0.5 0.4 North rural 14.6 5,203 70.7 25.3 11.97 20.1 21.8 23.3 North urban 1.1 8,029 51.4 15.2 6.33 1.1 1.0 0.9 CPAE Consumption per adult per equivalent (1989 Uganda shillings per month). Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Table 4.6. Poverty Rates in MS-2 Population Contribution to Location share (percent) Mean CPAE P0 P P2 pI P2 National 100.0 7,659 50.2 16.3 7.25 100.0 100.0 100.0 Rural 87.6 6,712 54.3 17.7 7.90 94.7 95.2 95.4 Urban 12.4 14,342 21.5 6.3 2.69 5.3 4.8 4.6 Central 31.8 10,983 30.3 8.3 3.38 19.1 16.2 14.8 East 28.5 5,681 65.3 23.4 11.10 37.0 41.0 43.6 I'D West 25.3 6,839 50.9 15.2 6.41 25.6 23.6 22.4 cn North 14.5 5,677 63.5 21.5 9.67 18.2 19.1 19.3 Central rural 23.7 8,995 36.3 9.9 4.01 17.1 14.4 13.1 Central urban 8.1 16,815 12.6 3.8 1.52 2.0 1.9 1.7 East rural 26.3 5,411 67.1 24.4 11.61 35.1 39.4 42.1 East urban 2.2 8,945 43.4 11.9 4.91 1.9 1.6 1.5 West rural 24.1 6,563 52.1 15.6 6.60 25.0 23.1 21.9 West urban 1.2 12,264 25.6 6.6 2.62 0.6 0.5 0.4 North rural 13.5 5,506 64.9 21.9 9.80 17.5 18.2 18.3 North urban 0.9 8,181 41.8 15.6 7.75 0.8 0.9 1.0 CPAE Consumption per adult per equivalent (1989 Uganda shillings per month). Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Table 4.7. Poverty Rates in MS-3 Population Contribution to Location share (percent) Mean CPAE P0 Pl P2 po P P2 National 100.0 7,759 49.1 16.4 7.64 100.0 100.0 100.0 Rural 86.5 6,742 53.7 18.1 8.49 94.5 95.4 96.1 Urban 13.5 14,273 19.8 5.6 2.23 5.5 4.6 3.9 Central 28.8 10,672 30.4 8.2 3.16 17.8 14.4 11.9 East 30.8 6,463 58.4 21.4 10.83 36.6 40.0 43.6 West 25.1 7,371 46.3 14.5 6.29 23.6 22.1 20.7 > North 15.4 5,525 70.2 25.1 11.84 22.0 23.4 23.8 Central rural 19.8 8,383 37.4 10.2 3.94 15.1 12.3 10.2 Central urban 9.0 15,731 14.8 3.8 1.44 2.7 2.1 1.7 East rural 28.7 6,066 60.4 22.3 11.35 35.3 38.8 42.6 East urban 2.1 11,877 31.6 9.2 3.74 1.4 1.2 1.0 West rural 23.8 7,066 47.9 15.0 6.54 23.2 21.8 20.4 West urban 1.3 13,014 16.8 4.3 1.66 0.4 0.3 0.3 North rural 14.2 5,276 72.5 25.9 12.28 21.0 22.5 22.8 North urban 1.1 8,633 41.2 14.0 6.34 1.0 1.0 0.9 CPAE Consumption per adult per equivalent (1989 Uganda shillings per month). Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Table 4.8. Poverty Rates in MS-4 Population Contribution to Location share (percent) Mean CPAE P0 P P2 p0 PI 2 National 100.0 8,078 44.4 13.7 5.91 100.0 100.0 100.0 Rural 86.7 7,127 48.7 15.2 6.56 95.0 95.8 96.3 Urban 13.3 14,264 16.7 4.3 1.65 5.0 4.2 3.7 Central 30.0 10,958 27.9 7.6 3.04 18.9 16.7 15.5 East 28.5 6,739 54.3 18.3 8.20 34.9 38.0 39.6 I' West 24.9 7,369 42.8 11.0 4.03 24.0 20.1 17.0 '4 North 16.5 6,226 59.8 21.0 10.00 22.2 25.2 27.9 Central rural 21.3 8,957 34.5 9.6 3.91 16.6 15.0 14.1 Central urban 8.7 15,874 11.8 2.7 0.91 2.3 1.7 1.3 East rural 26.3 6,336 56.8 19.2 8.67 33.6 36.8 38.6 East urban 2.2 11,455 25.2 7.1 2.74 1.3 1.2 1.0 West rural 23.7 7,097 44.0 11.4 4.15 23.5 19.7 16.7 West urban 1.2 12,589 19.7 4.6 1.57 0.5 0.4 0.3 North rural 15.4 5,988 61.8 21.7 10.36 21.4 24.3 26.9 North urban 1.2 9,406 34.0 11.0 5.19 0.9 0.9 1.0 CPAE Consumption per adult per equivalent (1989 Uganda shillings per month). Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. 98 Simon Appleton Table 4.9. T-Test Statistics for Hypothesis of Equality of Poverty Statistics in IHS and MS-4 Location P0 PI 2 National 13.98 18.62 18.67 Rural 10.37 14.50 14.80 Urban 10.64 10.85 9.17 Central 12.61 14.04 12.86 East 2.99 5.33 6.13 West 6.37 11.75 13.13 North 6.69 8.05 7.44 Central rural 9.83 11.15 10.37 Central urban 5.70 6.17 5.03 East rural 1.97 4.03 4.74 East urban 6.80 6.48 6.18 West rural 4.98 9.51 10.74 West urban 4.02 3.85 3.12 North rural 4.86 6.08 5.69 North urban 6.78 6.93 5.96 Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. below the food poverty line. Poverty rates showed pronounced regional dif- ferences. In the poorest area, northern rural, 73 percent lived below the pov- erty line. However, poverty was widespread in all areas. Even in the most prosperous location, central urban, more than one in five people lived below the poverty line. The other poverty indicators, P1 and P2. show similar pat- terns across the country. The P, index can be interpreted as the per capita aggregate poverty gap, that is, the mean shortfall of the welfare of the poor from the poverty line, expressed as a proportion of the poverty line and aver- aged across the population as a whole. The P2 index is the per capita aggre- gate poverty gap squared.7 Absolute poverty remained pervasive at the end of the four surveys. How- ever, it did fall substantially. In MS-4, 44 percent people were poor compared food obtained earlier). Moreover, people who are genuinely not obtaining enough calories may still be able to function adequately, because a safety margin is built into the World Health Organization estimates of calorie requirements. However, not ob- taining enough calories may prevent the very poor from engaging in energy- intensive activities, leading to destitution or worse. High rates of child stunting in the country suggest that undernutrition is a genuine and widespread problem. 7. The advantage of the P, indicator over P0 is that it reflects how far below the poverty line the poor are. The advantage of the P2 indicator over P1 is that it will fall if income is redistributed from those who are poor to those who are even poorer. Both Changes in Poverty and Inequality 99 with 56 percent in the IHS. The 21 percent fall in the headcount was accompa- nied by a 17 percent rise in mean consumption per adult equivalent. This im- plies an elasticity of poverty with respect to growth of approximately -1.24. This elasticity is rather low (in absolute terms): for example, in Nigeria, the figure was estimated to be -1.45, while in Ghana it was put at -1.73 (World Bank 1995). However, this seems to reflect the high level of the poverty line rather than any regressive aspect of Uganda's pattern of growth. Using a lower poverty line, the food poverty line, the growth elasticity is higher, at -1.8. This reflects the larger proportionate fall in the number of people living below the food poverty from 36 percent to 25 percent during the period. The other Pa indices also show marked declines, especially the P2 index. Whereas the P0 indi- cator fell by 21 percent, P, fell by 33 percent and P2 by 40 percent. By any stan- dards, the fall in poverty over a period of only five years has been substantial. As poverty rates have fallen, the cost of interventions to reduce poverty has also fallen (although this is somewhat offset by population growth). The Pi index is proportional to the cost (per adult equivalent) of eliminating pov- erty through perfectly targeted transfers. Our estimates imply that the mini- mum estimate of the cost of eliminating poverty through transfers has fallen by more than a quarter.8 The P1 index for the IHS implies a total annual cost of eliminating poverty through perfect transfers ("the simple sum") of U Sh 711,419 million (1993/4 prices) or US$594 million (using the 1993 official ex- change rate). The corresponding figures for 1997/98 are U Sh 555,378 million (1993/4 prices) and US$464 million.9 Poverty fell in both rural and urban areas during the five-year period. Mean living standards rose faster in rural areas: the mean rise in consumption per advantages come at the cost of having indicators that are less immediately intuitive than the simple headcount PO. For this reason, we often refer to the PO when all three indicators yield qualitatively similar results. 8. The total cost of eliminating poverty through perfect targeting is given by n*P,*Z, where n is the population and Z the poverty line. We include Bundibugyo, Gulu, Kasese, and Kitgum in the population, although they were excluded from the estimate of Pl. As these districts are poorer than Uganda as a whole, we will have understated the cost (by around two percentage points in 1992). 9. It is tempting to compare these figures with Uganda's extemal assistance in 1993 of US$531 million. Uganda's present external assistance is roughly equal to the cost of eliminating poverty through perfect targeting. However, one cannot assume that poverty could be eradicated by channeling external assistance into transfers to the poor. As shown in chapter 2 in this volume, the assistance currently has a consid- erable impact in reducing poverty and thus, channeling it to transfers would worsen the poverty gap that had to be filled by transfers, given that transfers are unlikely to be perfect. An alternative assumption is that targeting is infeasible, in which case transfers must be uniform. The P1 measure gives a ratio of the cost of eliminating poverty through perfectly targeted transfers relative to that of uniform transfers. In 1997/98, it would have cost US$3,387 million to eradicate poverty through 100 Simon Appleton adult equivalent was higher in rural areas than in urban areas (17 percent com- pared with 13 percent). However, focusing on the urban mean may be mislead- ing. Poverty statistics fell proportionately more in urban than in rural areas. The headcount fell by two-fifths in urban areas, and the proportionate fall in rural areas was less than one-fifth. Perhaps surprisingly, living standards in central urban areas grew modestly, by 9 percent, between the first and last sur- veys. This may be partly a consequence of in-migration: the estimated share of the country's population in these areas rose by 0.7 percent, a proportionate in- crease in the size of the central urban population of 9 percent. Other urban areas experienced large improvements in living standards, with northern and east- ern towns seeing rises in mean consumption of 23 and 33 percent, respectively. All regions had lower poverty in 1997/8 than in 1992, regardless of which Pa statistic is used or whether the poverty is measured relative to the total poverty line or just the food poverty line. Furthermore, all these reductions in poverty are statistically significant (see table 4.9). However, the magni- tude of the falls varied greatly. Mean consumption per adult equivalent rose most strongly in the central region, by 24 percent, and most modestly in the eastern region, by 10 percent. The corresponding figures for the westem and northem regions were 14 and 17 percent, respectively. These movements in average living standards are reflected in the changes in the poverty statistics. The central region saw the sharpest fall in poverty, with the headcount fall- ing by more than a third, from 46 to 28 percent. In the east, the headcount fell by only five percentage points. In the north and west, the headcount fell by 10 and 13 percentage points, respectively. The poverty gap, Pl, was halved in the central region, but fell by only 17 percent in the eastem region. One measure of the severity of poverty, P2. fell by 57 percent in the central region, but only 24 percent in the eastern. The net effect of these regional disparities was to widen the gap in living standards between the central and eastem regions. In 1992, the central region accounted for 25 percent of the poor and the eastern region accounted for 29 percent. By 1997/98, the central region accounted for only 17 percent of the poor com- pared with the eastern region, which accounted for 38 percent. Defining uniform transfers to all Ugandans (assuming no administrative costs). Furthermore, if the transfers were used to fund private consumption, they would have to be per- petual. Poverty would be eliminated in one year but would return in the next. One- off transfers may have permanent benefits to the extent that they are saved and in- vested, but such saving would imply transfers would have to be correspondingly higher to raise the consumption of the poor to the poverty line. Substantial extemal assistance is likely to continue in the fairly long term, but donors are unlikely to pay indefinitely. Finally, part of the external assistance is made as loans rather than grants or tied to particular imports. Nonetheless, it remains a legitimate question whether external assistance could make a larger impact on poverty if it were channeled more directly to the poor. Changes in Poverty and Inequality 101 poverty relative to the food poverty line only, the contrast is even starker. Although the eastern and northern regions account for less than half the popu- lation surveyed in 1997/98, they accounted for three-fifths of those whose total consumption was insufficient even to meet their calorie needs. (In 1992, they accounted for just over half.) It is noteworthy that this occurred during a time of administrative and fiscal decentralization. These institutional changes are surely not responsible for the increasing spatial disparity in wel- fare and poverty; however, the widening geographic inequalities may war- rant greater government redistribution between regions. The conclusion that poverty fell between the IHS and MS-4 is robust with regard to the choice of poverty line. Figure 4.2 shows the results of dominance analysis by plotting the poverty incidence curves for the five surveys. The pov- erty incidence curves plot the headcount indices on the y axis against different poverty lines (expressed as multiples of the original poverty line) on the x axis. As the poverty incidence curve for the IHS is above that for the MS-4, for all poverty lines there would be a higher headcount in the IHS than in the MS-4. Given such first-order dominance, it also follows that poverty would be higher in the IHS than MS-4 for all absolute poverty lines and for all Pa statistics other than PO. By contrast, the poverty incidence curve for MS-3 intersects that for MS-1 and MS-2 at several points, implying that the MS-3 curve does not wholly dominate them. In particular, for very low poverty lines-those around 50 Figure 4.2. Poverty Incidence Curves, 1992-97 100 - 0 80 0 40 20 ~ 0 u 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Consumption as multiples of the poverty line IHS -.---------.MS-1 MS-2 ----- MS-3 MS4 Source: Author's calculations. 102 Simon Appleton percent of the poverty line-the headcount is higher for MS-3 than for MS-1 and MS-2. This implies that the position of the very poorest households may have deteriorated between MS-1 and MS-3. The emphasis of the discussion in this section and, indeed, in most of the chapter is on comparing the first and last surveys. Movements in living standards during the intervening surveys are not stressed. However, being able to track changes in living standards on a yearly basis is useful in deter- mining whether the change appears to be incremental or merely driven by one or another survey year being somehow atypical, for example, having an exceptionally good or bad harvest. As already noted, mean living stan- dards grew between each survey, although the growth was strongest be- tween the first three surveys (IHS to MS-2). The headcount index at the national level also fell between each survey. This fairly steady year-on-year growth and poverty reduction is reassuring, because it implies that the improvement in living standards in the last survey, as compared with the first, is not driven by atypical conditions such as a year of abnormal weather conditions. That established, there has been considerable variation in growth and poverty reduction between each of the five surveys, especially at the regional level. For example, whether poverty was reduced between MS-2 and MS-3 is questionable. The P2 index actually worsened, while the P1 index remained constant. The time path of poverty reduction has varied particularly at the regional level. Some poverty indicators worsened for the western region between IHS and MS-1. The west appears to bounce back between MS-1 and MS-2, but the eastern region and other urban areas experienced worsening poverty. Between MS-1 and MS-2, poverty indica- tors worsened in the north. The headcount in the central region rose be- tween MS-3 and MS-4. Clearly poverty reduction was not smooth and con- tinuous across all regions throughout the period. Inequality and Growth Poverty statistics focus only on the lower part of the distribution of welfare. Even within that part the statistics can mask important features due to the aggregation involved. It is therefore more informative to look at the distri- bution in its entirety, which figure 4.2 does. A nongraphical way of present- ing the distribution is to report the values of consumption per adult equiva- lent at the median and at other deciles (table 4.10). The median rise in living standards between the IHS and MS-4 was 19 percent, two percentage points higher than the rise in the mean. As implied by the dominance analysis, consumption per adult equivalent was higher in MS-4 than in the IHS at all deciles. Comparing the relative gains (not tabulated), the lower (poorer) deciles tend to experience a greater rise in living standards. The rise in con- sumption per adult equivalent is 27 percent at the bottom decile, 23 percent at the second decile, and 21 percent for the third poorest. Disaggregating Changes in Poverty and Inequality 103 Table 4.10. Consumption Per Adult Equivalent at Each Decile (1989 U Sh per month) Decile IHS MS-1 MS-2 MS-3 MS-4 National 1 2,487 2,900 2,901 2,792 3,162 2 3,235 3,640 3,657 3,664 3,992 3 3,958 4,355 4,415 4,472 4,799 4 4,667 4,995 5,195 5,227 5,575 5 5,459 5,847 5,936 6,011 6,478 6 6,394 6,732 6,793 7,053 7,406 7 7,591 7,923 8,038 8,370 8,971 8 9,182 9,708 9,892 10,461 10,786 9 12,233 12,796 13,641 14,174 14,170 Rural 1 2,393 2,809 2,800 2,694 3,061 2 3,107 3,452 3,552 3,494 3,833 3 3,777 4,163 4,216 4,226 4,534 4 4,430 4,744 4,881 4,959 5,267 5 5,183 5,423 5,589 5,661 6,016 6 5,917 6,267 6,359 6,487 6,898 7 6,975 7,292 7,361 7,509 7,922 8 8,320 8,516 8,854 9,135 9,729 9 10,552 10,680 11,376 12,194 12,126 Urban 1 4,102 4,746 4,519 4,596 5,172 2 5,368 6,247 6,129 6,338 6,972 3 6,715 7,789 7,376 8,103 8,570 4 8,147 9,455 8,804 9,375 9,894 5 9,523 11,365 10,313 11,544 11,408 6 11,162 13,103 12,906 13,402 13,231 7 13,424 15,001 16,158 15,364 15,795 8 16,623 18,447 20,147 18,752 20,143 9 22,003 24,577 27,696 26,207 27,372 Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. into rural and urban areas, the pattern in rural areas is close to that in the country as a whole. However, in urban areas, the picture is rather different. For a start, the median rise in consumption per adult equivalent in urban areas during the period was 20 percent, substantially more than the 13 per- cent rise in mean consumption per adult equivalent. Similarly, large rises were apparent for all urban deciles. Consequently, mean consumption per adult equivalent provides a misleading picture of the overall improvement 104 Simon Appleton in living standards of the urban population.'" Using the median rather than the mean implies that urban living standards have risen faster than rural ones. In urban areas as in rural areas, there was again a tendency for con- sumption to rise more at the lower deciles. For example, consumption rose by 26 percent at the bottom decile; for the second and third deciles the rise was 39 and 28 percent, respectively. Focusing on growth at the median and at each decile implies a rather dif- ferent time path from that implied by growth at the mean. Both perspectives agree that there was substantial growth of more than 5 percent between the IHS and MS-1. However, growth between MS-3 and MS-4 is also high (more than 5 percent) at the median, but less than half that at the mean. Viewed at the median, growth between MS-1 and MS-3 was modest (2.8 percent). In- deed, during this period, the poorest 20 percent of the population did not experience noticeable improvements in living standards and the poorest got poorer. Consumption per adult equivalent at the bottom decile was 4 percent lower in MS-3 than in MS-1, while for the second decile, living standards were essentially unchanged. Table 4.11 reports the Gini coefficients for the surveys as a measure of the overall inequality in consumption per capita. The Gini coefficient, and hence inequality, falls between the first and last surveys. This indicates that the lower deciles saw greater rises in living standards than the more afflu- ent. The improvement in the progressivity of the distribution is most marked in urban areas." The fall in inequality within Uganda has made some contribution to pov- erty reduction, but most of the gains can be attributed to overall growth. This is shown by a decomposition of the change in poverty statistics between IHS and MS-4 following Datt and Ravallion (1992). We decompose the change in a poverty indicator P between two years, t, and t2 into three components: growth, G; distribution, D; and a residual, R: P -P =G+D+R. The growth component, G, is the difference between the initial poverty indicator and what would have arisen from distributionally neutral growth. 10. Consumption seems to have fallen during the surveys for some households who were in the top 10th of the urban population. Because these households have high consumption, their fortunes are influential in determining the mean rise in con- sumption per adult equivalent, which is calculated in the macroeconomic way by summing consumption across all households and dividing that sum by the popula- tion. Whether the apparent fall in the living standards of the top 10th of the urban population is genuine requires further investigation. However, it is not central to this chapter given our focus on the poor. 11. Like the discrepancy between mean and median growth in urban areas, this was partly driven by the apparent fall in consumption among the top 10 percent of the urban population. Changes in Poverty and Inequality 105 Table 4.11. Gini Coefficients for Uganda Survey Rural Urban National IHS 0.326 0.394 0.364 MS-1 0.296 0.365 0.345 MS-2 0.320 0.396 0.365 MS-3 0.325 0.373 0.366 MS-4 0.311 0.345 0.347 Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. That is to say, if there was the same mean per capita consumption, M, as in year t2, but the same relative distribution (Lorenz curve, L) as in t1 , then G = P(M1, L,) - P,. 2 1 1 The distribution component, D, is the difference between the initial pov- erty indicator and what would have arisen from a pure distributional change; that is, if there was the same mean per capita consumption as in year t, but the same relative distribution as in t2, then D = P(M, , L, ) - P1. Figure 4.3 shows the results of this decomposition. Growth accounts for 95 percent of the fall in the percentage of Ugandans in poverty. Improve- ments in distribution account for only 3 percent of the fall. For the other poverty indices, the contribution of shifts in the distribution of welfare rises relative to that of growth, but remains secondary. For the P1 index, distribu- tional changes account for a fifth of the fall in poverty, while for the P2 index, they account for 29 percent. Sectoral Decomposition of Poverty Changes Poverty statistics can be disaggregated in many ways. One interesting disag- gregation is by economic sector, as this provides a potential link between macroeconomic events and welfare of households. Table 4.12 classifies house- holds into mutually exclusive sectors roughly corresponding to those used in standard national accounts. With two exceptions, the classification is based on the main industry in which the household head works. One exception is for households that grew coffee. These households are defined as cash crop households, regardless of the head of the household's occupation. Typically, such households will obtain more revenue from food crops, but are still as- signed to the cash crop sector. The other exception is for households where the head is not working (mainly households with retired heads). These house- holds were placed into a separate category "not working," although some members may be generating income. The classification is a convenience designed to obtain mutually exclusive assignments of households to sectors 106 Simon Appleton Figure 4.3. Growth and Redistribution Decomposition, 1992-1997/98 *> 100 P. 80- 60 I - 200 -; 20 PO P, P2 *Growth Distribution EResidual Source: Author's calculations. given the data constraints (which include the absence of data on income by sector in the monitoring surveys). In reality, households may work in many industries, and in some cases the main industry in which the head works may not be the household's most important source of income. We disaggregated poverty by sector for the IHS and for MS-3 (tables 4.12 and 4.13). We could not carry out the disaggregation in MS-4 because the survey did not identify which crop farmers grew cash crops. In 1992 most Ugandans (70 percent) lived in households where the head's main activity was crop farming."2 Around one-third of those individuals lived in house- h-olds growing some nonfood cash crop. This reflects the fact that coffee grow- ing was widespread, despite the fact that in 1992/93 it accounted for only around 3 to 4 percent of total crop agricultural revenue (World Bank 1996). There is some evidence of movement infto cash crops during the period of the surveys; for example, the size of the sector increased from covering 23 per- cent of people in the IHS to covering 27 percent in MS-3. However, there is no evidence of a movement out of agrculture; indeed, the sector grew slightly 12. Henceforth, for ease of expression, we will refer to people as being in a sector if their head's main activity is in that sector. This should not be taken to imply that all the people said to be in the sector actually work in the sector. Table 4.12. Poverty by Sector of Household Head, IHS Population Contribution to Sector share (percent) Mean CPAE P0 P P2 p0 P1 P2 National 100.0 6,900 55.7 20.3 9.90 100.0 100.0 100.0 Food crop 47.2 5,649 64.1 24.5 12.32 54.3 57.0 58.8 Cash crop 23.4 6,027 60.4 20.7 9.59 25.4 23.8 22.7 Noncrop agriculture 2.7 6,642 55.0 22.2 11.31 2.6 2.9 3.1 Mining 0.1 9,418 31.5 2.6 0.21 0.0 0.0 0.0 Manufacturing 3.7 8,009 43.6 15.8 7.63 2.9 2.9 2.9 Public utilities 0.1 9,089 33.6 5.6 1.62 0.1 0.0 0.0 Construction 1.3 10,656 36.4 11.5 4.58 0.9 0.8 0.6 Trade 6.7 11,864 25.2 7.2 3.11 3.0 2.4 2.1 Hotels 0.5 10,054 25.8 8.1 3.30 0.2 0.2 0.2 Transport and communication 1.5 9,787 31.5 11.0 5.05 0.9 0.8 0.8 Miscellaneous services 1.6 12,561 26.6 10.2 5.03 0.8 0.8 0.8 Govermment services 6.8 10,104 36.2 10.5 4.49 4.4 3.5 3.1 Not working 4.3 6,929 58.2 22.9 11.65 4.5 4.8 5.0 CPAE Consumption per adult equivalent. Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Table 4.13. Poverty by Sector of Household Head, MS-3 Population Contribution to Sector share (percent) Mean CPAE P0 P P2 P0 P P National 100.0 7,764 49.1 16.4 7.64 100.0 100.0 100.0 Food crop 44.2 5,813 62.1 22.5 10.99 55.8 60.6 63.5 Cash crop 26.7 7,519 46.0 11.9 4.52 25.0 19.4 15.8 Noncrop agriculture 2.1 8,197 40.2 14.5 6.89 1.7 1.8 1.9 C Mining 0.2 5,974 74.2 12.7 2.85 0.3 0.1 0.1 Manufacturing 3.3 10,181 27.1 8.7 3.60 1.8 1.7 1.6 Public utilities 0.1 13,192 11.3 1.5 0.19 0.0 0.0 0.0 Construction 1.1 9,695 35.0 8.7 3.02 0.8 0.6 0.4 Trade 6.9 13,248 20.0 4.5 1.66 2.8 1.9 1.5 Hotels 1.0 11,972 19.3 5.1 1.65 0.4 0.3 0.2 Transport and communication 1.9 14,084 14.8 6.6 3.23 0.6 0.8 0.8 Miscellaneous services 2.2 11,428 27.9 10.8 4.88 1.2 1.4 1.4 Goverrnment services 5.5 11,387 29.5 7.7 2.90 3.3 2.6 2.1 Not working 4.9 7,662 62.1 29.0 16.60 6.3 8.7 10.8 CPAE Consumption per adult equivalent. Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Changes in Poverty and Inequality 109 in terms of population share during the surveys. Trade and government ser- vices were the next most populous sectors, each covering 7 percent of Ugan- dans. Trade did not change in size during the surveys, although households in the government sector decreased from 6.8 percent in the IHS to 5.5 percent in the MS-3. The "not working" sector was the next largest sector, growing from 4.3 to 4.9 percent. Manufacturing remained fairly constant at around 3.5 of the population. Other sectors covered 2 percent or less of the popula- tion, with some sign of growth in the size of the service sector. The food crop sector was the poorest of the major sectors in 1992/93, and this sector experienced only relatively modest declines in poverty. The P0 and Pi indicators fell by less-both absolutely and relatively-than those for the country as a whole. Cash crop farming was the second poorest sector in the IHS, but this sector experienced dramatic declines in poverty between the IHS and MS-3. Regardless of the poverty indicator used, the reduction in poverty in the cash crop sector was more than twice as large as that in the country as a whole.'3 These improvements in poverty were driven by an above average rises in mean consumption per adult equivalent, which rose by a third in the sector. One factor underlying these gains was the rise in the world price for coffee during the period. The unit price of Ugandan coffee exports was as follows: Fiscal year Price (US$/kg) 1991/92 0.86 1992/93 0.82 1993/94 1.14 1994/95 2.55 1995/96 1.72 1996/97 1.38 1997/98 1.57 Source: Republic of Uganda (1994, 1997a, 2000). At the height of the coffee boom, Uganda was receiving export prices for coffee that were triple those in 1992. Other factors were also important. Poor weather conditions in coffee growing areas depressed output in 1991/92. Output was also likely to have been enhanced by the price and market liber- alization policies in the coffee subsector. Although these were initiated in 1990, a lagged response in output was likely because of the time needed for newly planted coffee trees to bear fruit. 13. This comparison is fairly straightforward, because poverty rates in the cash crop sector were of a sinilar magnitude to those in the country as a whole in the IHS. Between the IHS and MS-3, the headcount fell by 24 percent for the cash crop sector; for the country it fell only half as much-by 12 percent. 110 Simon Appleton Poverty fell in nearly all sectors. One exception was mining, although this result may be questionable, given the very small sample size. In addi- tion, there was an increase in the headcount defined relative to the food pov- erty line in the miscellaneous service sector (and mean consumption per adult equivalent fell), but other poverty statistics for that sector improved slightly. However, perhaps the most notable exception to the generally favorable trends was in the nonworking sector, where all poverty indicators worsened de- spite rising mean consumption per adult equivalent. The headcount rose only slightly, but this rise masks more serious deterioration in other indicators. The Pi statistic rose by more than a quarter and the P2 statistic by two-fifths. The cash crop sector was not the only one to experience reductions in poverty much above the national trend. Although manufacturing and trade started from much lower initial levels of poverty, both saw greater propor- tionate reductions. Hotels, construction, and transport and communications also performed strongly. The government sector lagged somewhat behind the country as a whole in terms of growth in mean per capita consumption, although poverty rates fell comparably. It is possible to decompose the national change in poverty into the effects of changes in poverty within sectors and movements between sectors (Ravallion and Huppi 1991). This allows one to assess whether poverty has fallen because people within certain sectors have become better off or be- cause people have moved to more affluent sectors. If P,1 is a poverty indica- tor for time t,, then: P - Pt, = Y(Pi2 - Pi,,)ni,, intrasectoral effects + X(nt2 - nit,)Pitl intersectoral effects + 2 (P"2 - Pj,)(ni,2 - n1,l) interaction effects, where nit2 is the proportion of the population in sector i at time t, and Pi,, is the poverty indicator for sector i at time tl. The interaction effects tell us whether people switched out of or into sectors where poverty was falling (if positive, people moved into sectors where poverty was falling). Applying this methodology to Uganda, table 4.14 shows that an improve- ment in the conditions of cash crop farmers and their families was respon- sible for more than half of the improvement in poverty between the IHS and MS-3. Improvements in the lot of food crop farmers made a more modest contribution to the fall in the headcount, but these accounted for around a quarter of the improvement in other poverty indicators. Other sectors made more modest contributions, with trade, manufacturing, and government ser- vices being the more noticeable (largely due to their size). The table also re- veals the worsening poverty of those in households whose head was not working. Population shifts between sectors also help explain some of the improvement in poverty, but their contribution is less, only around 2 to 4 percent. Interaction effects were positive, implying that people moved into sectors where poverty was falling faster, such as cash crop farming. Changes in Poverty and Inequality 111 Table 4.14. Sectoral Decomposition of Changes in Poverty between IHS and MS-3 Percentage contribution to Sector P0 P1 P2 Food crop 14.1 24.2 27.7 Cash crop 50.8 52.6 52.6 Noncrop agriculture 6.0 5.3 5.2 Mining -0.5 -0.2 -0.1 Manufacturing 9.3 6.8 6.6 Public utilities 0.3 0.1 0.1 Construction 0.3 1.0 0.9 Trade 5.3 4.8 4.3 Hotels 0.5 0.4 0.4 Transport and communication 3.8 1.7 1.2 Miscellaneous services -0.3 -0.2 0.1 Government services 6.9 4.8 4.8 Not working -2.5 -6.8 -9.4 Total intrasectoral 94.0 94.3 94.4 Total intersectoral 2.9 3.3 3.7 Total interaction 3.0 2.4 1.9 Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Summary and Conclusions The data on private consumption from five recent Ugandan household sur- veys provide a picture of rising living standards in accordance with the mac- roeconomic data on growth. The finding that urban living standards have risen is unsurprising, given the many indicators of strong performance of nonagricultural sectors and the visible progress in the major towns. How- ever, the household survey data is perhaps the strongest evidence available that living standards in rural areas have also improved commensurately with the macroeconomic statistics. The growth in the household surveys was not driven by any one year being atypical. Average living standards improved between each of the five surveys. Moreover, the growth was broadly based as it was shared across the income distribution. Real consumption per capita has risen for all deciles, implying a reduction in poverty regardless of where the poverty line is set. Indeed, living standards grew most rapidly for the poorer deciles, leading to a fall in inequality. We drew an absolute poverty line for Uganda, sufficient to meet calorie needs given the typical diet of the bottom half of the population and to meet minimum nonfood requirements. The line implies that 56 percent of Ugandans were poor in the first survey in 1992. This percentage fell to 44 112 Simon Appleton percent in last survey in 1997/98, a significant and substantial reduction in poverty during a relatively short interval of five years. The reduction in poverty is explained mainly by growth, although falling inequality also contributed. Poverty reduction has been uneven across economic sectors, with those engaged in cash crop farming, manufacturing, and trade faring particularly well. The improvement in the living standards of those grow- ing cash crops accounts for more than half of the fall in poverty during the period. Although the data generally imply improvements in welfare, growth was not uniform, and a number of less favorable trends were identified. Regional disparities were exacerbated during the period, with the central region growing the most strongly and the eastern region lagging behind. At the median, living standards rose more in urban areas than in rural ar- eas. Finally, poverty worsened in the first four surveys for those households where the head was not working. The rise in living standards observed in the surveys is evidence of broadly based growth. Although there is much debate about whether growth "trickles down," such terminology is clearly inappropriate here. If anything, growth in living standards has been strongest among poorer households. Nonetheless, many questions remain concerning implications for the future and for our understanding of the recent past. The period considered is relatively short- five years-and whether the impressive reduction in poverty observed here can be sustained in the long term has yet to be determined. Note that the growth in the living standards of the very poorest was somewhat erratic, with no im- provements between 1993/94 and 1995/96. The extent of the reduction in pov- erty in the future will partly depend on whether growth can be sustained, but also on how growth is distributed. On the latter point, the experience of 1992- 98 may be a poor guide for the future, because growth at that time was driven partly by the coffee boom and partly by a process of recovery from the eco- nomic collapse of the 1970s and 1980s. The temporary nature of the coffee boom raises the issue of whether the associated rise in living standards observed will only be temporary. A merit of using consumption rather than income to measure welfare is that, ac- cording to the permanent income hypothesis, temporary windfalls will have less effect on consumption. Indeed, empirical research has tended to con- firm that export crop farmers do often save heavily out of any windfalls resulting from price booms (Bevan, Collier, and Gunning 1993). Conse- quently, consumption is unlikely to fall with the end of coffee boom and, indeed, the last survey, MS-4, provides no evidence that it has. However, whether future growth will have the same effect on poverty as that arising from the coffee boom is not clear. Much will depend on the sources of growth, such as the extent to which the poor derive income from growing sectors or can enter such sectors. One reason why growth during the pe- riod has reduced poverty so much is that before the boom, coffee farmers were as poor as the average Ugandan. Changes in Poverty and Inequality 113 More generally, the period studied may mark Uganda's transition from recovery to fresh growth. Although long-term comparisons are problematic, it appears that Uganda has not yet returned to the real per capita income levels enjoyed in the early 1970s. The process of recovery to achieve those income levels may have involved a pattern of growth that is quite different from what will arise with subsequent development. Recovery has necessi- tated the rehabilitation of traditional export crops, the restoration of the pub- lic sector, and a reversal of the retreat to subsistence. Although predicting the nature of future growth is hard, it is unlikely to be a simple continuation of the processes of recovery These considerations imply a need to continue monitoring poverty and living standards at the microeconomic level. Annex 4.1. Methodology This annex explains the main assumptions and principles that underlie the results reported in this chapter. More detailed explanations are provided in an extended version of this chapter available upon request from the author. Obtaining Consistent Estimates of Consumption Table A4.1 reports the estimates of consumption per capita as calculated in the official survey reports and after a number of adjustments.14 There were six adjustments: * The exclusion of Kasese, Kitgum, Gulu, and Bundibugyo districts. Exclu- sion of these districts was necessary because fears about safety meant that large parts of these districts were not covered in MS-4. These four districts included 6.9 percent of Uganda's population in the 1991 cen- sus (Republic of Uganda 1995). They are relatively poor, so that their omission raises mean consumption per capita by 1.8 percent in the IHS and by 2.3 percent in the MS-1. * Adjustmentfor public transportfares. The IHS omitted an item for fares on public transport. To adjust for this, a value for such an item was imputed using the regional shares in the MS-1.'5 Omission from the IHS of health expenditures for Arua district was dealt with in a 14. The figures differ very slightly from those in the official survey reports, per- haps due to subsequent cleaning of the data. 15. This is a striking example of sensitivity to questionnaire design. Both the IHS and MS had an item for "other transport expenses," but only the MS questionnaire explicitly mentioned public transport fares as an example. To adjust for the change in questionnaire design, we did not include the item as reported in the IHS, but instead assumed the item had the same share as in the MS-1 (with separate shares for rural and urban areas). Table A4.1. Adjusted Comparison of Mean Consumption Per Capita (U Sh per month) Mean consumption per capita HIS 1992/93 MS-1 1993/94 MS-2 1994/95 MS-3 1995/96 MS-4 1997/98 National As calculated in official reports 11,574 13,195 15,221 17,499 20,540 1. Excluding Gulu, Kasese, Kitgum, and Bundibugyo 11,786 13,501 15,388 17,721 20,747 2. Adjusting for public transport fares 11,981 n.a. n.a. n.a. n.a. 3. Revaluing home consumed food at market prices 12,769 14,748 16,643 18,568 21,976 4. Adjusting for regional prices 13,187 15,267 17,064 18,973 22,139 5. Adjusting for inflation (1989 prices) 5,452 5,825 6,058 6,187 6,353 6. Reweighting MS-1 5,452 5,718 6,058 6,187 6,353 Rural As calculated in official reports 9,547 10,116 12,470 14,303 17,210 1. Excluding Gulu, Kasese, Kitgum, and Bundibugyo 9,675 10,351 12,564 14,411 17,367 2. Adjusting for public transport fares 9,788 n.a. n.a. n.a. n.a. 3. Revaluing home consumed food at market prices 10,633 11,685 13,887 15,323 18,714 4. Adjusting for regional prices 11,400 12,571 14,669 16,082 19,141 5. Adjusting for inflation 4,701 4,794 5,206 5,242 5,488 6. Reweighting MS-1 4,735 4,862 5,206 5,242 5,488 Urban As calculated in official reports 25,869 34,092 34,334 37,194 42,047 1. Excluding Gulu, Kasese, Kitgum, and Bundibugyo 26,697 35,177 35,312 38,929 42,746 2. Adjusting for public transport fares 27,471 n.a. n.a. n.a. n.a. 3. Revaluing home consumed food at market prices 27,858 35,833 36,085 39,362 43,205 4. Adjusting for regional prices 25,805 33,822 33,957 37,498 41,647 5. Adjusting for inflation 10,752 12,919 12,067 12,246 11,979 6. Reweighting MS-1 10,752 11,645 12,067 12,246 11,979 n.a. Not applicable. Source: Author's calculations from household survey data provided by the Uganda Bureau of Statistics. Changes in Poverty and Inequality 115 similar manner. These adjustments together raise the mean consump- tion figure for the IHS by 1.7 percent. * Revaluing home-consumedfood at market values. Interviewers were sup- posed to make sure that respondents valued home consumption of food in farm-gate prices. Farm-gate and market prices were estimated based on the unit values in the surveys for home consumption and purchases of food, respectively.'6 Estimates were done separately by region and urban-rural location (that is, eight sets of prices were iden- tified). The revaluation increased the estimated value of home con- sumption of food by approximately 30 percent. * Adjusting for regional prices. Food prices are markedly higher in some areas of Uganda, particularly in urban areas, than they are in others. It was possible to use unit values for purchases of major food items to construct regional food price indexes for each survey. Median unit val- ues were used to make the results insensitive to outliers. Nonfood prices were assumed to be constant across the country. It is sometimes argued that nonfood prices may be higher in rural areas due to transport costs, but this is not well established. In a study of the C6te d'Ivoire, Grootaert and Kanbur (1994) found nonfood prices to be generally lower in rural areas than in urban areas. The regional price adjustment is important primarily when making intracountry (for example, urban-rural) com- parisons rather than intertemporal comparisons."7 * Adjustingfor inflation. Adjusting for inflation is probably the most important adjustment when making comparisons over time. The composite national consumer price index (CPI) was used as the price deflator and expenditures were converted into 1989 prices. Although the CPI is only collected for major urban areas, it does appear fairly reliable. During the period, there were no price controls or other dis- tortions. Furthermore, an earlier exercise for the period 1989-92 us- ing unit values from household survey data had largely corrobo- rated the CPI. 16. This was complicated by the fact that quantities could be reported in different units, including some unspecified measures such as "heaps," "bunches," and so on. Where possible only metric measures were used. For some items most units codes were nonmetric, in which case only reports with a single unit code were used to avoid having to make different units comparable. It was not necessary to convert quantities into metric units except when calculating calories per shilling for the food poverty line. For that purpose, conversion factors from Kayiso (1993) were used for nonstand- ard unit codes for the few items where output was never reported in metric units. 17. It does raise the overall national expenditures somewhat, because prices were adjusted to survey median values. Urban areas were oversampled, and this effect is not corrected for when calculating median values, so the survey prices dispropor- tionately reflect higher urban prices. 116 Simon Appleton * Reweighting MS-1. MS-1 assigned low weights to certain households that were either previously surveyed in the IHS or were in enumera- tion areas surveyed in the IHS. Subsequent monitoring surveys did not weight such households differently from others. Consequently, MS-1 was reweighted to make its population multipliers comparable to those in MS-2 and MS-3. The main impact of this adjustment was to remove a rather implausible deterioration in urban living standards between MS-1 and MS-2. Setting the Poverty Line The MS-1 data are used to calculate an absolute poverty line. This is derived from a food poverty line, showing the cost of meeting calorie requirements given the typical diets of poor Ugandans, and an estimate of the cost of meet- ing nonfood requirements. CALORIE REQUIREMENTS. Lipton and Ravallion (1995) identify the energy re- quirements set by the World Health Organization (WHO 1985) as the most widely used official estimates. Consequently, we adopted these guidelines for Uganda. WHO calorie requirements vary by body size, age, sex, daily activities, pregnancy, and lactation. We followed the principles laid out by WHO in adjusting for these factors. Our calculations were based on the as- sumption that adults are engaged in farming, and we estimated energy re- quirements based on the time use data provided in the IHS. The calculations were involved; however, our results yielded similar multiples of basal meta- bolic rates to those given by WHO in their illustrative examples of a subsis- tence farmer and a rural woman in a developing country (WHO 1985, tables 10 and 14). The estimated calorie requirements by age and sex are presented in table A4.2. We first define the poverty line according to the needs of a man aged 18 to 30. This poverty line can then be compared with household con- sumption per adult equivalent, where the adult equivalence scales measure needs relative to a man aged 18 to 30. THE FOOD POVERTY LINE. Many combinations of foods (food baskets) could meet the requirement of 3,000 calories. We focus on the food basket of the poorest 50 percent of Ugandans, ranked by consumption per adult equiva- lent. Previous work using the IHS data defined a poverty line based on the consumption pattems of the bottom 50 percent and found that more than half of Ugandans lived below this line (World Bank 1996). To calculate the food poverty line, we first use the MS-1 data to estimate the mean quantities of 28 major food items (see table 4.2) consumed by the poorest 50 percent. These mean quantities constitute a reference food basket: the typical food basket of the poor. We then estimated how many calories were generated by the reference food basket. We did this calculation using the calorific values of East African foods as reported by West (1987) (see table 4.2). For some foods, Changes in Poverty and Inequality 117 Table A4.2. Daily Calorific Requirements and Equivalence Scales Male Female Calorie Equivalence Calorie Equivalence Age requirement scale requirement scale 0 755 0.25 700 0.23 1 1,200 0.40 1,140 0.38 2 1,410 0.47 1,310 0.44 3 1,560 0.52 1,440 0.48 4 1,690 0.56 1,540 0.51 5 1,810 0.60 1,630 0.54 6 1,900 0.63 1,700 0.57 7 1,990 0.66 1,770 0.59 8 2,070 0.69 1,830 0.61 9 2,150 0.72 1,880 0.63 10 2,190 0.73 2,015 0.67 11 2,340 0.78 2,130 0.71 12 2,440 0.81 2,225 0.74 13 2,560 0.85 2,295 0.77 14 2,735 0.91 2,370 0.79 15 2,875 0.98 2,385 0.88 16 2,990 1.00 2,425 0.89 17 3,090 1.02 2,435 0.89 18-29 3,025 1.00 2,350 0.87 30-39 2,960 0.99 2,325 0.87 40-59 2,960 0.99 2,295 0.86 60+ 2,290 0.86 1,830 0.77 Note: Equivalence scales for children aged 14 and under are obtained by dividing calorific requirements by 3,000. Equivalence scales for adults are given by 0.42 + 0.58 x (calorie requirements/3,000). Source: Calorie requirements are author's calculations from the IHS based on guidelines from WHO (1985). part of the food weight was inedible or lost in preparation. Estimates of the ratio of the food retained for consumption are given in table 4.2. Multiplying the mean quantities of foods consumed by their calorific value and retention rates, we estimated that the poorest 50 percent of Ugandans consumed around 1,373 calories per day per person (not per adult equivalent). Consequently, the typical diet of poor Ugandans would have to be scaled upward by a fac- tor of 2.18 to generate 3,000 calories per person per day. Scaling up the refer- ence food basket by this factor gave us the food basket that was costed to identify the food poverty line. The total cost of the food basket, which repre- sents our food poverty line, is U Sh 11,463 per month (in the average prices of the MS-1 survey; these MS-1 prices must be deflated by 2.63 to be converted to the 1989 prices used in reporting most real expenditures in this chapter). 118 Simon Appleton NONFOOD REQUIREMENTS. We followed Ravallion and Bidani (1994) in identi- fying nonfood requirements, NF, as the nonfood expenditure of those whose expenditure is just equal to the food poverty line, zf. The rationale for this is that, because at this level of welfare the poor have sacrificed some of their need for calories, the nonfood expenditures they have chosen to give priority to should also be regarded as meeting essential needs. Different locations were allowed different nonfood requirements. On average, the model predicts a mean food share of 0.566 for households whose total consumption is just sufficient to meet their calorie requirements (see the second column in table 4.3). This gives a national poverty line of U Sh 16,443 per adult equivalent per month (MS-1 prices). Taking a purchasing power parity exchange rate of U Sh 369 to the U.S. dollar, this is equivalent to US$44.56 per adult equivalent a month. (At the official exchange rate of U Sh 1,195 per U.S. dollar, it amounts to US$13.76 a month.) In the case of Uganda, the line is equivalent to US$34 per capita per month, and hence is comparable to the US$1 a day poverty line sometimes used for international poverty comparisons by the World Bank. Rather than use a single "all Uganda" poverty line, the lines were allowed to differ by location because estimated nonfood requirements vary (the third column of table 4.3). Predicted food shares were much lower in urban areas than in rural areas, for example, 0.49 in the central urban compared with 0.68 in western rural. Consequently, the western rural had the lowest poverty line, U Sh 15,189 (MS-1 prices) per adult equivalent per month, while the central urban had the highest, U Sh 17,314.18 These regional differences in poverty lines are relatively modest. However, note that a single food basket was used for all regions and was valued in constant prices. Because food prices are much higher in urban areas, the difference between urban and rural poverty lines is much greater when valuing in nominal terms (and not at constant prices).'9 In nominal terms, the poverty line for central urban was 106 percent higher than that for western rural. The derived poverty lines are based on the calorie requirements of a Ugan- dan man aged 18 to 30. To use the lines to assess poverty with households of different demographic composition, we need a set of equivalence scales to measure the needs of different age and sex groups. We used relative calorie 18. That westem rural should have the lowest poverty line raises some doubts about the appropriateness of working with a national food basket. One reason why the food share may be predicted to be higher in western rural (and hence the poverty line lower) is that it is more expensive to obtain sufficient calories using matooke, a favored staple in the western region. 19. The food poverty lines in nominal terms (fourth column of table 4.3) are not equal to the food poverty lines in national prices scaled by our estimated regional food price index. This is because the food price index was based on the consumption patterns of the whole population, whereas the poverty line is based on the consump- tion patterns of the poorest half of the population. Changes in Poverty and Inequality 119 requirements to measure relative food needs. All adults are assumed to have equal nonfood needs regardless of sex or age. For calculating equivalence scales for adults, we assumed that 58 percent (the mean food share) of the scale is equal to calorie requirements divided by 3,000. The remaining 42 percent of the scale was assumed to be the same for all adults. Small children can more reasonably be said to have lower nonfood requirements. Rather arbitrarily, we assumed that children's nonfood requirements are lower than those of men by the same proportion as their relative calorie requirements. The resulting equivalence scales are reported in table A4.2. References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Appleton, Simon. 1996. "Problems of Measuring Changes in Poverty over Time: The Case of Uganda 1989-1992." Institute of Development Studies Bulletin 27(1): 43-55. Appleton, Simon, I. Chessa, and J. Hoddinott. 1999. "Are Women the Fairer Sex? Looking for Gender Differences in Gender Bias in Uganda." Uni- versity of Oxford, Centre for the Study of African Economies, U.K. Processed. Bevan, David, Paul Collier, and Jan Gunning. 1993. "Trade Shocks in Devel- oping Countries: Consequences and Policy Responses." European Eco- nomic Review 37(2-3): 557-65. Datt, Gaurav, and Martin Ravallion. 1992. "Growth and Redistribution Com- ponents of Changes in Poverty Measures: A Decomposition with Ap- plication to Brazil and India in the 1980s." Journal of Development Eco- nomics 38(2): 275-95. Foster, James, Joel Greer, and Erik Thorbecke. 1984. "A Class of Decompos- able Poverty Measures." Econometrica 52(May): 761-66. Grootaert, Christian, and Ravi Kanbur. 1994. "A New Regional Price Index for C6te d'Ivoire Using Data from the International Comparisons Project." Journal of African Economies 3(1): 114-41. Kakwani, Nanak. 1993. "Statistical Inference in the Measurement of Poverty." Review of Economics and Statistics 75(4): 632-39. Kayiso, P. K. 1993. "Final Report on Conversion Factors and Regional Price Indi- ces." Ministry of Finance and Economic Planning and Program to Allevi- ate Poverty and the Social Cost of Adjustment, Kampala. Processed. Kikafunda, Joyce, Louise Serunjogi, and Michael Migadde. 1992. "Final Re- port on Establishment of a Nutrition Based Absolute Poverty Line 120 Simon Appleton for Uganda." Ministry of Finance and Economic Planning and Pro- gram to Alleviate Poverty and the Social Cost of Adjustment, Kampala. Processed. Lipton, Michael, and Martin Ravallion. 1995. "Poverty and Policy." In Jere Behrman and T. N. Srinivasan, eds., The Handbook of Development Eco- nomics, vol. III. Amsterdam Elsevier. Ravallion, Martin, and Benu Bidani. 1994. "How Robust Is a Poverty Line?" World Bank Economic Review 8(1): 75-102. Ravallion, Martin, and Monica Huppi. 1991. "Measuring Changes in Pov- erty: A Methodological Case Study of Indonesia during an Adjust- ment Period." World Bank Economic Review 5(1): 57-82. Republic of Uganda. 1994. "Background to the Budget 1994-1995." Ministry of Finance and Economic Planning, Kampala. _.__ 1995. "The 1991 Population and Housing Census Analytic Report," vol. 1, "Demographic Characteristics." Ministry of Finance and Eco- nomic Planning, Entebbe. _.___ 1996. "Background to the Budget 1995-1996." Ministry of Finance and Economic Planning, Kampala. . 1997a. "Background to the Budget 1997-1998." Ministry of Planning and Economic Development, Kampala. . 1997b. "Poverty trends in Uganda, 1989-1995." Discussion Paper no. 1. Ministry of Finance and Economic Planning, Coordination of Pov- erty Eradication Project and Statistics Department, Kampala. . 1999. "Uganda Participatory Poverty Assessment: A Summary of Key Findings and Policy Messages." Ministry of Finance and Economic Planning, Kampala. _.__ 2000. "Background to the Budget 2000-2001." Ministry of Planning and Economic Development, Kampala. UNDP (United Nations Development Programme). 1997. Uganda Human Development Report. Kampala. West, Clive. 1987. "Food Composition Table." Wageningen Agricultural Uni- versity, Department of Human Nutrition, De Drejien, Netherlands. Processed. WHO (World Health Organization). 1985. "Energy and Protein Require- ments." WHO Technical Report Series no. 724. Geneva. World Bank. 1994. World Development Report: Infrastructurefor Development. New York: Oxford University Press. Changes in Poverty and Inequality 121 . 1995. "The Social Impact of Adjustment Operations: An Overview" Report no. 14381. Operations Evaluations Department, Washington, D.C. . 1996. Uganda: The Challenge of Growth and Poverty Reduction. A Coun- try Study. Washington, D.C. 5 Rural Households: Incomes, Productivity, and Nonfarm Enterprises Klaus Deininger and John Okidi During the past decade, Uganda's economy has shown remarkable growth, which has translated into a substantial reduction in poverty. For growth to be sustainable and to reduce poverty in a sustainable fashion, it will be critical to increase agricultural productivity and rural nonfarm employment. This is be- cause about 80 percent of the labor force is concentrated in agriculture, but the sector receives less than half of the total income. In addition, more than two- thirds of the earned income of the poorest decile comes from agriculture. En- abling the poor to accumulate additional human and physical capital and in- creasing the returns to assets they already own through technical progress, increased diversification, market integration, commercialization, and growth of rural nonfarm enterprises will, therefore, be key elements of any strategy aimed at equitable growth and broadly based poverty reduction. The purpose of this chapter is to assess the extent of progress toward these goals and to explore obstacles that need to be overcome. Data from three different house- hold and community surveys were used, including the 1992/93 integrated household survey, the 1993/94 monitoring survey, and the first round of the 1999/2000 national household survey (see appendix A at the end of the book).' Use of the 1999/2000 Uganda national household survey would have not been possible without the excellent performance of the Uganda Bureau of Statistics survey team under Jackson Kanyerezi and James Muwonge, the careful data editing under Tom Emwanu, and the contribution of Bart Minten in questionnaire design and enu- merator training. The authors are deeply indebted to all of them. 1. Although enough observations (about 4,800 households) are available to make inferences that are statistically representative at the regional level, note that all the means discussed in this chapter refer to sample rather than population averages, be- cause final weights are not yet available for the latter survey. 123 124 Klaus Deininger and John Okidi This chapter first reviews major changes that occurred in the rural sector between 1992 and 1999. It then analyzes the determinants of changes in house- hold income using a panel of approximately 1,000 households that were in- terviewed in both 1992/93 and 1999/2000. Finally, the chapter explores pro- duction, input demand, and the establishment of nonfarm enterprises using information from 1992/93 and 1993/94 surveys. A Panorama of Rural Uganda To gain a better understanding of the rural environment, this section intro- duces the historical evolution of Uganda's rural sector and then describes changes in output structure, technology, operation of factor markets, and ac- cess to infrastructure and other services that occurred between 1992 and 1999. The purpose of presenting such a summary is to provide a descriptive over- view of the some of the issues the rural population faces; the extent to which these conditions have changed; and the scope for further improvements in providing rural households with access to technology, services, and infrastruc- ture. Indeed, while the descriptive evidence provides a clear indication that there has been a change for the better, for example, as regards asset ownership, it also indicates that significant interregional and rural-urban differences per- sist, especially in access to infrastructure and technology. Historical Background During 1971-85 Uganda's rural sector suffered from a combination of ill- founded nationalization of assets, problems related to civil strife, and agri- cultural price disincentives. These social and economic problems were due to implicit and explicit taxation of export crops through monopoly market- ing boards, the associated inefficiencies in input and output markets, and overvalued exchange rates. The combined effect of these factors was to dis- courage many rural producers from risking exposure to markets and make them shift to food crop production and subsistence farming. For example, cotton production declined from a peak of 87,000 tons in the early 1970s to about 2,000 tons in the mid-1980s, as producing for the market was no longer profitable. A similar decline occurred in tea and, although less dramatic, in coffee. Price disincentives, withdrawal of financial intermediaries, lack of infrastructure maintenance, and deterioration in the delivery of public goods all led to the successive decapitalization of the rural economy, erosion of in- ternational competitiveness, and a secular decline in productivity. To reverse these trends, since the late 1980s the government has attempted to reduce biases against rural producers. Coffee marketing and exports were liberalized and direct export taxation was abolished (though reintroduced temporarily during the 1994-95 coffee boom). Similar measures were taken in the cotton sector, although progress has been slower. Agricultural output grew at an annual rate of 4 to 4.5 percent in real terms during the last 10 Rural Households: Incomes, Productivity, and Nonfarm Enterprises 125 years. However, given the low level from which the sector started, this per- formance is less impressive than one may think. In reality, agricultural growth has been well below the average growth rate of the economy as a whole (6 to 7 percent), suggesting that a variety of structural impediments have thus far limited the supply response of the rural sector (see Belshaw, Lawrence, and Hubbard 1999 for a critical review). Nevertheless, agricultural growth has played an important role in re- ducing poverty. As shown by Appleton (chapter 4 in this volume), the inci- dence of poverty decreased nationally from 56 percent in 1992 to 44 percent in 1997. A decomposition analysis indicates that this was mostly due to growth rather than to redistribution. Agricultural production for the mar- ket was strongly correlated with the reduction in poverty: sectoral decom- position shows that cash (export) crop farming households account for half of the poverty reduction achieved between 1992 and 1997. This suggests that greater agricultural commercialization could play an important role in lifting the large majority of poor food crop and subsistence producers out of poverty. The importance of rural sector growth for poverty reduction is illustrated by the fact that in 1999, agriculture accounted for more than two-thirds of households' earned income, and land accounted for about half the value of the total asset endowment even of the poorest decile in the population. Any measures that raise agricultural income and the returns to land would there- fore yield significant and immediate benefits for the rural poor. It is against this background that the next section discusses output structure, the opera- tion of factor markets, and access to services in more detail. Structure of Output and Technology This section uses community survey evidence to highlight the changes in the relative importance of main commodities grown by Ugandan households, their yields, and the number of producers between 1992 and 1999.2 What emerges is a pattern whereby, with the exception of cotton and a number of fruits, traditional agricultural production in the north appears to have been relatively stagnant, and in many cases, characterized by declining yields. Other regions, especially the west, emerge as more dynamic, having diversi- fied into vegetables, while at the same time expanding in traditional com- modities such as maize, beans, millet, and cassava. Evidence also suggests a major role for the transfer and adoption of improved technology. Technology helps both to arrest diverging trends across regions (as in maize and beans, tomatoes, and cabbage), and to halt declining yields, often through disease, as in the case of coffee and matooke (plantain). 2. The available comrnmunity survey data include more than 500 communities across Uganda. A community typically corresponds to a village. 126 Klaus Deininger and John Okidi Building on the evidence on commodities, this study also examined spe- cific factors associated with the use of technology and the functioning of fac- tor markets. During the period under study, livestock ownership expanded considerably, creating opportunities for a sustained increase in the use of mechanical technology as well as the establishment of integrated systems of production and organic manuring. Even though evidence points to increased ownership and use of ox plows, their share remains extremely low. This sug- gests that the scarcity of complementary factors of production, such as labor, capital, and land, may constrain further expansion. The data also point to extremely low levels of organic fertilizer use to improve soil fertility (about 3 percent of farms use inorganic fertilizer and about 6 percent use manure). At the plot level, the data indicate a strong correlation between the adoption of high-yielding varieties (HYV) and fertilizer use (the simple correlation coef- ficient is 0.18). CHANGES IN ourur AND YIELDS. Changes in the output mix and in yields of main commodities are important indicators of producer response to shifts in incentives and opportunities in marketing and technology. They reveal whether or not the expected diversification is actually occurring. At the same time, they provide an indication of the adequacy of existing technology. This section uses community-level information on 14 main commodities grown in Uganda to make such inferences.3 It begins with staple crops and proceeds to export and nontraditional commodities, exploring the share of communities where the specific crop is grown;4 how the number of producers has changed since 1992; whether yields increased or decreased during 1992-99; and the main reason reported by communities for changes in yields (table A5.1).5 Maize, the main staple for the most of the population, is grown in 75 percent of communities countrywide, with 61 percent of villages having vir- tually everybody grow maize. Maize cultivation expanded significantly be- tween 1992 and 1999: 36 percent of communities reported an increase in the number of producers and only 11 percent reported a decrease in the number of producers. In 36 percent of communities in the east and 30 percent of com- munities in the central region, maize yields increased, attributable to improved management practices. By comparison, maize yields dropped in 60 percent of communities in the north and 43 percent of communities in the east, with the drops being attributed mainly to weather-related factors. In the north, this appears to have led to a significant move out of maize cultivation. Data 3. The survey contains information on 20 commodities. 4. The categories are "by all or many", "by some" (up to one half), and "by none" of the producers. 5. Although the original answer for both the number of producers and yields was given on a scale from one to five, these are collapsed into two categories to irn- prove the readability of table A5.1. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 127 indicate that in 37 percent of communities the share of maize producers de- creased, in marked contrast to the rest of the country. It is important to exam- ine the extent to which this reflects the region's comparative advantage, that is, whether there is scope for producers to substitute for maize with other commodities that are more profitable under local conditions. If such a substi- tution is not profitable, adaptive research to expand varieties and/or tech- niques used to grow maize would be needed. As beans are often grown together with maize, trends observed in beans were similar, including their almost universal importance. Only 20 percent of communities reported that nobody plants beans, while in 64 percent of the villages everybody grows them. All regions except the north showed marked increases in the number of communities where beans were grown. Seventy percent of the communities in the north reported a decline in the yields. At the same time, in the west yields declined in 40 percent of communities but increased in 30 percent of villages, suggesting that there may be gaps in tech- nology that could be easily bridged. In view of its drought resistance, millet is most important in the north, where almost 70 percent of communities report that everybody grows millet, followed by the west (60 percent), the east (33 percent), and the center (8 percent). Only 15 percent of communities in the north reported that nobody grows millet. It is therefore surprising that about 20 percent of communities in the north and east reported a decline in the number of millet producers, compared with 43 percent of communities in the west reporting an increase. This change in the relative importance of production appears to be caused mainly by changes in yields, which were reported to have declined in almost two-thirds of northern and one-third of eastern communities, but increased in about one-third of western communities. In view of the crop's drought resistance, weather was almost universally mentioned as the main underly- ing factor for yield decreases, a fact that would warrant attention. Sorghum is most important in the west, followed by the east and the north, while it is almost nonexistent in the central region. The number of producers shows a moderate increase in the west, together with a moderate to significant decrease in the east and north. Declining yields experienced in 31 percent of northern communities (as compared with 21 percent of western ones) appear to be the main reason for the reduced emphasis on this commodity in the north. Groundnuts are of major importance in the north, the west, and the east, but less important in the center. In contrast to the commodities discussed ear- lier, groundnuts show a marked pattern of declining yields nationwide: about one-third of communities in the east and west and two-thirds in the north reported declining yields while only about 15 percent report yield increases. Matooke was produced in about 50 percent of communities. The main pro- duction areas are in the western region, where the crop was grown in 82 percent of communities (almost universally in 55 percent and by about half in 26 percent). While of moderate importance in the central and eastern re- gion, matooke was grown only in about 10 percent of communities in the north. 128 Klaus Deininger and John Okidi Production expanded in the western and central regions, but contracted in the east and remained virtually constant in the north. Almost 60 percent of yield decreases in communities in the west and 25 percent of yield decreases in communities in central region were attributed to diseases or the weather. While the commercial market for cassava is limited, it provides an impor- tant source of calories for home consumption. Indeed, cassava is universally grown in almost half of Uganda's villages and by some producers in another quarter of the country's villages. Producers in the central region and the north appear to have shifted out of cassava, whereas the number of producers in the west and the east has increased over time. Yield declines appear to have been most marked in the north (where two-thirds of communities reported a decline) and the central region (45 percent), but were relatively equally bal- anced with yield increases in the remainder of the regions. Cultivation of coffee, Uganda's main eamer of foreign exchange, domi- nates in the western, central, and eastern regions, but is nonexistent in the north. Although the profitability of coffee was high, the geographic expan- sion of coffee growing was limited to the east, where 16 percent of the com- munities increased production and 12 percent of communities decreased pro- duction. While changes observed in the west were moderate, 30 percent of communities in the central region reported a decline in the number of coffee producers (as compared with 7 percent that reported an increase). The major reason appears to have been disease. Given the macroeconomic importance of coffee and the forward and backward linkages in the economy, efforts to reduce vulnerability to diseases, and where possible to expand cultivation, have showed a high payoff. Cotton, Uganda's other main cash crop, is important mainly in the north and the east, where about two-thirds and one-third of communities, respec- tively, reported cotton cultivation. The sector has been characterized by a long history of neglect, which over time has resulted in significantly reduced output. The fact that the number of cotton producers has increased in half of the communities in the north and a quarter of the communities in the east indicates that the dislocations associated with the past have given way to a more sustained path of consolidation and renewed growth. The expansion of the area growing cotton appears to have been accompanied by technology- driven yield increases in most communities, which is particularly encourag- ing. Although data on total output and profits will be needed before more definite conclusions can be drawn, the signs are hopeful. Few national statistics are available on the importance of fruits and vegetables, two categories of products that are often considered indispens- able for moving Uganda's agricultural sector up the value added chain. While a survey of communities cannot substitute for a more detailed as- sessment of Uganda's comparative advantage, potential markets, and op- portunities for expansion into agroprocessing, the new community-level data suggest that the focus on traditional crops that characterizes Uganda's Rural Households: Incomes, Productivity, and Nonfarm Enterprises 129 agricultural service is likely to miss out on an important element of region- specific diversification. For example, tomatoes, which were being grown in about 40 percent of communities nationwide, have achieved a level of geographical coverage greater than that of cotton and almost equal to that of coffee. The west clearly dominated, with 10 percent of communities where virtually everybody grew tomatoes, while 57 percent of communities in the west, 47 percent in the east, 22 percent in the center, and 16 percent in the north reported at least some tomato producers. The number of producers increased in 15 percent of vil- lages and decreased in 4 percent of villages nationwide. Yields showed a divergent pattem whereby a significant increase in the central and, to a lesser extent, the eastern region was counterbalanced by a marked drop in yields in the west. The adoption of the improved management techniques that are reported to have caused the yield increase in the central region could be trans- ferred to the west. Improved management techniques could smooth the path for future expansion of production. Although grown in a slightly more limited number of communities than tomato, cabbage is another high-value product that has recently attained sig- nificance. While the western and eastern regions still dominated in cabbage production, the center appears to be catching up and, in a pattern that ap- pears to be similar to the one observed for tomatoes, these regions registered considerable increases in yields (which were universal for all the communi- ties where the crop is produced). Compared with this increase in yields, many communities in the west experienced a disease-related decrease in yields. Tree crops that are relatively drought resistant might provide an opportu- nity for expansion of production in the north that, according to the 1999/2000 community survey data, seems to have done rather badly in terms of overall agricultural performance, with the exception of cotton. Production of man- goes is clearly dominated by the north, with near-universal coverage in 43 percent of communities and some coverage in another 25 percent of communi- ties. Only the western region, with 20 percent near-universal coverage and some coverage in 10 percent of communities, approaches this level of cover- age. Currently, mangoes do not seem to provide a basis for sustainable expan- sion. Contrary to conditions in the west, where the number of producers in- creased in 14 percent of communities, the number of producers decreased in other regions. The situation is similar with respect to oranges, which were grown universally in 20 percent and to some extent in 37 percent of northern commu- nities. Disease-related declines in yields that were observed in more than half of the communities (that is, virtually everywhere the crop was grown) point toward a significant deficit in terms of technology. As such technology should be easily available from other countries, more detailed examination of the rea- sons underlying its limited current adoption, as well as the scope for better cultural practices, would be important. An examination of these issues might open up opportunities for nontraditional agricultural growth in the north. 130 Klaus Deininger and John Okidi Passion fruit is another recently introduced high-value crop on which few nationally aggregate production estimates are available.6 Evidence at the community level suggests that the crop was almost universally grown in about 9 percent of communities in the west, and was of some importance in 16 percent of communities nationwide. The number of producers increased in 11 percent of communities in the west and 9 percent in the north. While yield increases were reported from 11 percent of communities in the north attributed mainly to improved labor use, 19 percent of western communities experienced a weather-related yield decline. USE OF TECHNOLOGY. In addition to being an important investment item in traditional agricultural societies, livestock ownership can affect agricultural performance by increasing producers' ability to use animal traction and me- chanical technology to expand the area cultivated, perform necessary activi- ties in a more timely manner, and through provision of manure maintain soil fertility and make use of higher-yielding varieties. The latter is relevant, be- cause at least part of the decline in yield observed in some of the commodi- ties may have been caused by lack of investment in soil improvement through either organic or inorganic fertilizer. Moreover, enabling producers to ex- pand their cultivated area beyond the current average farm size of 1.5 to 2 hectares per household will require a shift from hand-hoe technology to ani- mal traction. This shift is important, as some regions of Uganda still have potential for further expansion of cultivated areas.7 The data point to an increase in the ownership of livestock in the 1990s (table A5.2, panel 1). During 1992-99, the number of households owning live- stock increased from 11 to 20 percent for cows, 4 to 7 percent for bulls, and 1 to 2 percent for oxen. The increase was distributed equally across regions, suggesting a broad pattern toward higher levels of investment in agricul- tural technology. The value of livestock owned increased by 36 percent (from U Sh 0.74 million to U Sh 1,004 million per household), which is a substantial investment given that a high share of households did not own livestock at all in 1992. Although overall levels of plow ownership are still low, and the rate of expansion was much slower than in the case of livestock, 4 percent of producers, compared with 2.5 percent in 1992, were reported to own plows. The eastern and northern regions reported that approximately 7 percent of 6. Given the focus of most conventional production surveys on traditional com- modities, it is unlikely that reliable information on the economic importance of any recently introduced high-value crops exists at the moment. While they were at least included as separate categories in the 1999/2000 national household survey, it is not clear whether the training of enumerators was sufficient to make them probe for such nontraditional crops in each case. 7. Estimates put the potential for increasing the cultivated area from 5 million hectares in 1992 to as much as 18 million hectares (World Bank 1996). Rural Households: Incomes, Productivity, and Nonfarm Enterprises 131 producers owned plows, the highest levels of plow ownership among the regions. Note, however, that 70 percent of communities nationwide (48 per- cent in the north) reported that nobody uses ox plows. Given the relative land abundance and the relatively small areas cultivated in the north (which suggests scope for greater use of animal traction to increase the area culti- vated), it is particularly surprising that plow use seems to be slightly lower there than in the east, where about 15 percent of communities reported "many" users of ox plows. In addition, even though ox plow use increased in about 20 percent of eastern and 35 percent of northern communities, its use de- clined in others. Compared with ox plows, tractor use decreased rapidly in most of the communities (22 percent showed a decrease and only 4 percent showed an increase). The decreased tractor use probably reflects a legacy of unsustainable mechanization in earlier years. As illustrated in panel 4 of table A5.2, the share of food crop area planted in HYV tripled during the period, albeit from a very low level. Growth oc- curred fairly uniformly across regions, with the level of HYV use being high- est in the eastern region. Community data suggest that, in addition to an increased number of producers who used these varieties within specific vil- lages, the number of HYV also spread geographically. However, in 43 per- cent of communities nationwide (70 percent in the north), there was still no use of HYV, and only in about 7 percent of communities were HYV used by half or more of the population. In line with the limited spread of HYV, the use of fertilizer (a strong comple- ment to HYV) was low, with an average of 3 percent reporting use, based upon a reported use of 5 percent in the north to 2 percent in the more fertile west. More producers used pesticides than fertilizer (7 percent nationally, ranging from 11 percent in the center to 3 percent in the west). Also, the ap- plication of manure to improve soil fertility was slightly higher than that of fertilizer, with 6 percent nationally reporting manure use. The large interre- gional variation (from 1 percent of producers using manure in the north to 13 percent in the center region), despite fairly uniform levels of livestock own- ership, suggests that further examination is warranted with regard to the determinants of the use of investments to enhance soil fertility. Factor Markets Ability to access credit is important to finance the expansion of productive activities, to obtain working capital, and to insure against risk. The data show a large increase in the share of producers who have access to credit, from 8 to 16 percent between 1992 and 1999. While this information suggests that the past contraction of the credit system has been largely reversed, it does not imply that further improvements, both on the supply and the demand side, could not yield significant economic benefits. The majority of credit was used for production rather than consumption. About one-third of producers in the sample had not obtained credit, either because the bank was too far away 132 Klaus Deininger and John Okidi or because the producers lacked collateral. Due to a combination of highly covariate risks and high levels of poverty, the scope for informal credit in the northern region appears to be particularly limited, implying that formal in- stitutions are much more important in the north. Land rights and land markets link credit, productive efficiency, and pov- erty. Obviously, in view of the findings, the scope for secure, formal land rights to help producers access credit appears to be important. Moreover, if the non- farm economy becomes more vibrant, the scope for realizing efficiency gains from better functioning of land rental markets will increase significantly. In- deed, the greater importance of off-farm employment could underlie the rapid expansion of land rental markets observed between 1992 and 1999. CREDIT MARKETS. Data for 1999 suggest that households' access to credit improved considerably since 1992. In 1999 about 16 percent of households nationwide had access to credit, ranging from 24 percent in the west and 6 percent in the north (table A5.3, panel 1). Comparing this with the 9 percent of households who had an outstanding loan in 1992, the data suggest that access to credit at the household level has expanded considerably.8 The fact that the number of households having access to credit now is almost equal to those who ever had access to credit confirms this conclusion.9 To determine whether producers are credit constrained, that is, whether unsatisfied demand for credit exists under present conditions, a closer look was taken at the reasons given for nonuse of credit. Table A5.3, panel 1 shows that 42 percent did not apply because they did not need credit, and 19 percent did not apply because they did not know how to apply. Taking these two groups together still leaves 40 percent who appeared to have creditworthy projects, but did not apply.'" Only 6 percent failed to apply because interest rates were too high, suggesting that the cost of credit was no longer the most important factor limiting access to and use of credit. By contrast, 22 percent did not apply because they lacked security (even though the majority owned 8. As informal credit was explicitly included in 1992/93, there should be little bias due to differences in the survey methodology, except possibly differences in the extent of enumerator training. 9. In the 1992 survey, only 9 percent of households nationwide (4 percent in the northem region) had an outstanding loan during the survey period, with sharp dif- ferences in mean loan sizes between urban (U Sh 242,000) and rural (U Sh 66,000) areas. Moreover, even though about half the number of loans was made in rural ar- eas, these areas received only about one-tenth of the total available credit, most of which was concentrated in the central region, which is close to urban centers. From a sectoral perspective, loans were heavily concentrated in trade (44 percent of loans as well as amounts), services (19 percent), and livestock farming (15 percent). Crop farm- ing (12 percent) ranked much lower. 10. Although households that do not know where to apply may be credit con- strained, they are included in the category of nonconstrained producers to err on the conservative side. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 133 land and all owned other assets). An additional 12 percent failed to apply because the bank was too far away. This suggests that about one-third of the producers who did not use credit would, at a given cost, appear to be able to benefit from an increased ability to use existing assets as collateral and from expansion of financial infrastructure. Indeed, less than half of the communities nationwide (44 percent) had access to formal credit, with considerable regional differences reported (table A5.3, panel 2). The availability of formal credit was relatively high in the west (65 percent of communities) and very low in the north, where only 20 percent of communities had access to formal credit. The government's Entandikwa scheme continues to be the most widely available source of for- mal credit and, therefore, of loans, followed by banks and cooperatives." In the north, there was not a single community with a bank branch. The second half of panel 2 illustrates sources that were available in the past, but have since closed. Clearly, many communities where cooperatives have in the past provided loans are no longer doing so. In 13 percent of communities, Entandikwa schemes closed down as well. The majority (55 percent) of those who obtained credit received it from relatives or community funds, followed by cooperatives and government sources (21 percent), nongovernmental or- ganizations (16 percent), banks (5 percent), and other businesses (3 percent). As mentioned earlier, the pattern in the north differs markedly from that in other regions: relatives were considerably less important than elsewhere. The survey indicates (table A5.3, panel 3) that a large share of loans (45 percent) was used to establish nonagricultural enterprises, followed by ex- penditures on education and health (24 percent), purchase of inputs (15 per- cent), agricultural investments in land and livestock (9 percent), and house- hold goods (7 percent). Note the regional differences, especially between the north and other regions. In the north, the emphasis on nonconsumptive use of credit was even more pronounced than in the other regions. LAND RIGI-rs AND LAND MARK. Land rights and land markets are important for a number of reasons. First, land to which secure property rights (as nor- mally documented through a formal title document) exist can serve as collat- eral for formal credit. Second, land markets are important to enhance agricul- tural productivity and household welfare by shifting land toward its most productive use, either through sales or through rental.'2 Finally, secure land rights are normally a precondition for households to be willing to undertake 11. Entandikwa is a government soft loan scheme targeted at the poor with the primary objective of providing start-up capital for household business enterprises. The credit program was started in the mid-1990s as a revolving fund to facilitate and move households out of poverty. The program has suffered from a low recovery rate for several reasons, including people's view of the fund as a government handout. 12. The difference between sales and rental markets to land is explained, for ex- ample, in Deininger and Binswanger (1999) and Deininger and Feder (2000). 134 Klaus Deininger and John Okidi the investments necessary for sustainable increases in land productivity and/ or to maintain soil fertility.'3 Deininger (2000) demonstrates the importance of land rights for investment to enhance soil fertility, for land values, and for land market participation, and shows that more secure land ownership increases the probability of applying manure (but not fertilizer), the value of the land, and farmers' propensity to rent out land. Building on this finding, this section focuses on the extent to which land rental and land sales markets function and on the aggregate incidence of land conflicts. The land rental market helps equalize land access. In 1999, the operation of rental markets helped to reduce the Gini coefficient from 0.57 for owned land, a figure that puts Uganda in the middle league of countries interna- tionally, to about 0.50 for operated land.14 In addition to improving access to land, land rental markets are also likely to make a contribution to higher allocative efficiency. Panel 4 of table A5.3 indicates that the participation in land rental markets in the central region was high, with 25 percent of pro- ducers having reported to have rented in land and 12 percent having rented out land in 1999. Note that much of the activity in land rental markets was of recent origin. Even considering only households that cultivated land in 1992, participation in rental markets more than doubled between the two peri- ods.'5 The share of households renting land increased from 10 percent in 1992 to 24 percent in 1999, with the greatest increase observed in the east and the west. Similarly, the share of households renting out land increased from 5 to 12 percent in the central region, with the highest absolute increase in the east. Complementing this with information on the number of communities where land rental was practiced, the only region where a significant increase in this figure was observed was the north (from 9 to 25 percent). Information on land sales was only available at the community level (table A5.3, panel 5). This information shows that land prices differed markedly across regions, with the west (U Sh 526,840 per acre) being the highest and the north (U Sh 56,860 per acre) being the lowest. Not surprisingly, land sales transac- tions were rare in the north where they were reported in only 13 percent of the 13. There is some controversy as to the importance of land title in the African context in general (see Besley 1995; Brasselle, Frederic, and Platteau 1997; Platteau 1996), and for Uganda in particular (Baland and others 1999). See Deininger (2000) for a more elaborate discussion and econometric evidence on the importance of land rights in Uganda. 14. The Gini coefficient is a widely used measure of inequality that varies be- tween a value of one (for perfect inequality) to zero (for perfect equality). Land Gini coefficients are in the 0.8 to 0.9 range in Latin America and in the 0.4 to 0.5 range in Asian countries. 15. Doing so avoids the need to count households that did not exist in 1992, but which obtained land either through rental or through a pre-inheritance transfer while parents were still alive between 1992 and 1999. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 135 communities, compared with 64 percent in the central region, 63 percent in the west, and 58 percent in the east. A number of commnunities (11 percent) re- corded more than 5 land transactions per year. Also, the activity of land rental markets increased in about 40 percent of the communities during 1992-99. More detailed investigation at the household level will be needed to make infer- ences on the impact of land sales on efficiency and welfare. Land conflicts were reported by 52 percent of communities (table A5.3, panel 6). While land conflicts were virtually absent in the north (where they were reported only in 16 percent of communities, they were of considerable importance in the west (70 percent), the east (58 percent), and the central region (51 percent). In addition to having the highest incidence of land con- flicts, the west also appears to be characterized by a considerable increase in land conflicts: in 21 percent of communities, land conflicts increased signifi- cantly, and in 15 percent they increased somewhat. One of the distinguishing features of the central and western regions is that most land is held under mailo tenure, which indicates that there is considerable scope for improving tenure security on such lands.'6 Infrastructure, Services, and Social Capital Regression estimates reported by Larson and Deininger (in chapter 6 in this volume) suggest that when transaction costs were reduced, the extent to which producers participated in the market was affected by the infrastructure and other public services. Examination of the 1992/93 and 1999/2000 survey data indicates that, both at the household and the village level, the extent of changes in access to extension and infrastructure has been modest. The ease of linking to infrastructure is illustrated in table A5.4, panel 1, which gives the average time taken (in minutes), using the most common means of transport from a community in each of the regions to different infrastructure items in 1999. The average household had to spend 25 minutes to get to the next feeder road, 75 minutes to reach a tarred road, an hour to reach a bus or a truck that could transport agricultural produce, and 48 minutes to reach a taxi. Access to other services also required considerable amounts of time: to reach a hospital took 56 minutes, a factory employing more than 10 people took 63 minutes, and a post office or telephone took 70 and 75 minutes, respectively The table also illustrates the high level of regional variation, and that changes in access to infrastructure have still been quite limited. Households in the north generally had to spend about double the time of the national average: 65 minutes to reach the next feeder road and more than 2 hours to reach the next tarred road, truck, telephone post office, and hospital. According to the group of village leaders interviewed in the community survey, infrastructure access improved 16. Mailo is a form of freehold tenure that was awarded to local kings and no- tables by the British when they colonized the country in 1900 (Brett 1973). 136 Klaus Deininger and John Okidi in a limited number of communities between 1992 and 1999, with 7 percent at the national level reporting improved access. Moreover, improvements in in- frastructure appear to have been concentrated in the east and central regions (12 and 10 percent, respectively). Only 1 percent of communities in the north experienced improvements in infrastructure during the period. One of the explicit goals of agricultural extension is to help farmers cope with the challenges posed by an environment where fast-changing crop pests and diseases pose a consistent threat to production and welfare. In this con- text, it is encouraging to note that coverage with extension services increased from 11 percent of farmers in 1992 to 17 percent in 1999 (table A5.4, panel 2). This increase appears at the household level to have been highest in the north, resulting in relatively equal regional coverage in 1999. In the west, the pri- vate sector was of far greater importance than public extension agents. Ac- cording to community-level information, extension workers were the main source of information in 31 percent of communities in the east, 21 percent in the west, and 17 percent in the central region, but in none of the communities in the north. In virtually all the northern communities (96 percent) the radio was the main source of information on agricultural practices (table A5.4, panel 2). Also, in all regions households relied more on the radio than on extension workers for information on technology. One reason for this may be that, de- spite the apparent expansion of extension services' coverage, the vast major- ity of producers meet the extension agent only once a year. Only 5 percent of producers nationally, according to the survey, have had contact with an ex- tension worker more than twice a year, and this percentage has remained virtually constant throughout the period. At the village level, about 64 percent reported that the community was not at all reached by extension services (table A5.4 panel 3). The majority of pro- ducers were reached in about 21 percent of communities in the west and 1 percent in the north. Neither community nor individual data indicate any gen- der bias in such access.'7 A regionally distinct pattern of expansion and reduc- tion of extension access is apparent: extension access increased in 31 percent of communities in the west and 27 percent in the center, whereas the east and north seem to have been characterized by large-scale withdrawal. Access to extension services decreased in 31 percent of eastern and 20 percent of north- ern communities."8 All these observations may be useful, together with the importance of having access to timely information at different stages in the production process and strategies to complement attention to traditional ex- tension with mass media and private sector sources. For example, in 41 17. Data from the 1999/2000 national household survey indicate that access to extension information was virtually equal between male and female producers. 18. Note that this information, which is given for the same community at two points in time, does not suffer from limitations regarding statistical representativeness. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 137 percent of communities in the west, the private sector had overtaken both the radio and the public service as the primary information source. Access to veterinary services appears to be better than access to extension services (table A5.4, panel 4). Of the 82 percent of communities that reported cattle ownership, 66 percent had veterinary services available (the lowest coverage was observed in the north, with 48 percent). The public sector still provided the majority of these services (70 to 80 percent). Coverage with artificial insemination was low (12 percent of communities) and confined to the central (23 percent) and eastern (19 percent) regions. Issues of governance, violence, and social capital affect economic activity in low-income communities where the scope for formal contract enforcement is limited and, as a consequence, many economic exchanges rely on trust and reciprocity, often within informal kinship networks. While there were few indicators on governance, the data point to a marked increase in the inci- dence of civil strife, which affected about 8 percent of households in 1992 and 13 percent in 1999 (table A5.4, panel 5). The pattern of increase was re- gionally uneven; the largest increase (from 8 to 18 percent) was noted in the west. Compared with other regions, civil strife in the north remained con- stant, affecting 10 percent of households in both periods. Similarly, the num- ber of households affected by property thefts increased from 13 to 20 per- cent, while physical attacks remained almost constant, increasing in the aggregate from 7 to 9 percent from 1992-99. To construct an indicator of social capital endowments, households' re- action to exogenous shocks during the last seven years was evaluated.19 This indicator is defined as the percentage of households which, having experienced a shock, received help or gifts from community members. Re- sults indicated that the distribution was fairly equal across the country, with between 30 and 40 percent of households receiving help to cope with shocks and with interregional and intertemporal changes being relatively minor (table A5.4, panel 5). Intertemporal Changes in Household Income While the foregoing analysis provides an interesting account of changes in the productive and social environment that can give useful insights for gov- ernment policy, it does not establish a clear link between households' pro- ductive capacity and their overall well-being. This section aims at providing such a link by analyzing the determinants of growth in incomes at the house- hold level during 1992 and 1999. The main findings are that growth in Uganda 19. The shocks considered include an illness of one month or longer (56 percent), abandonment or separation (9 percent), loss of permanent job (6 percent), and loss of productive assets (22 percent). 138 Klaus Deininger and John Okidi has been propoor, that in a liberalized environment the opportunities pro- vided by households' endowments of physical capital and their access to electricity and financial services were of great importance, and that unob- served region-specific effects still had a large impact. The Panel Data and Descriptive Evidence To make inferences about the factors that have contributed to higher rates of income growth, over and above mere cross-sectional correlation, regression analysis was used for the 953 panel households for which information was available from both the 1992/93 and the 1999/2000 surveys.20 While using data from the same households enabled making inferences on growth, in large household surveys that contain a panel element, attrition may be high and generally follows a systematic pattern (Deaton 1997). Indeed, a probit regression for attrition (not reported) indicates that mobility and thus attri- tion was much higher for households located in urban areas (7 percent), that had access to electricity (6 percent), whose head was younger (each addi- tional year of age increases the marginal probability of staying in the sample by 0.5 percent), who had fewer children below the age of 14 (each child in- creases the marginal probability of staying in the sample by 1.8 percent), and more people above 60 (each older person decreases the probability of staying by 3.9 percent). Attrition rates were also slightly higher (3 and 5 percent, respectively) in the east and north. Table 5.1 summarizes income sources of panel households and their evo- lution over time. In 1992 and 1999 households received about 72 percent of their income from own-agricultural enterprises. By contrast, the share of in- come from agricultural wages declined considerably, from 9 to 4 percent, whereas income from nonagriculture, both in its wage and its nonwage com- ponent, increased. Across regions, the most marked change observed was a drop in the importance of agricultural self-employment income from 81 to 72 percent in the north, accompanied by an increase of nonagricultural enter- prise income from 5 to 13 percent. Similar increases in nonfarm income were observed for the remainder of the regions, although agricultural enterprise income in these regions remained more stable. Figure 5.1 depicts the cumulative distribution of earned income in both years for panel households. The distinct shift of the distribution to the right indicates an unambiguous improvement in income levels (consistent with second-order stochastic dominance). Thus, even though one cannot exclude the possibility that some households saw their income drop during the period, 20. Income instead of expenditure is used here to be able to decompose income by source (agriculture and nonagriculture). Given that income and expenditure dis- tributions are relatively similar in both years, it is unlikely that use of expenditure would lead to radically different results. Table 5.1. Income Sources for Panel Households by Region, 1992 and 1999 (percent) National Central Eastern Northern Western Category 1992 1999 1992 1999 1992 1999 1992 1999 1992 1999 , Own agricultural enterprise 71.3 72.0 71.6 72.6 65.4 66.4 80.5 71.6 72.3 75.9 Agricultural wages 9.0 4.1 9.1 4.3 8.0 4.0 9.8 4.7 9.4 3.6 Nonagricultural enterprise 10.1 12.7 10.7 12.9 14.8 16.1 5.0 12.6 7.7 10.1 Nonagricultural wages 9.7 11.2 8.7 10.3 11.8 13.5 4.7 11.2 10.6 10.4 Number of observations 911 274 233 102 302 Note: Only earned income is considered. Remittances, rental income, and so on, are therefore not included. Source: Authors' calculations based on the 1999/2000 national household survey and the 1992 integrated household survey. 140 Klaus Deininger and John Okidi Figure 5.1. Cumulative Distribution of Income, 1992 and 1999 r- 1.00 0.75 - 0 1992 0 c 0.50 - 0.25 - 0 10 11 12 13 14 Log of annual adult equivalent income (U Sh) Source: Authors' construction from the 1992 and 1999 household survey data. income levels in the aggregate showed a marked increase. The mean annual increase in household income, which was used as the dependent variable in the regressions reported later, was about 7 percent, indicating a considerable increase in overall household welfare. Analysis of Income Growth To identify initial conditions that are associated with higher subsequent income growth, and in particular, whether household- or location-specific characteris- tics are quantitatively more important, we regressed the annualized rate of in- come growth at the household level on initial household characteristics, com- munity characteristics, and a set of regional or provincial dummy variables.21 Results from the regressions are summarized in table A5.5 where, to im- prove readability, the dependent variable is the growth rate of income in percentage terms.22 A cursory look at the table reveals that initial household 21. As misreporting of income would imply that outliers could introduce consid- erable error into the dependent variable, we report results from a least absolute de- viation (LAD) estimator rather than from ordinary least squares (OLS). The former estimation technique gives lower weight to outliers, thereby reducing the possibility that extreme observations will have an unduly strong impact on the results. Results from LAD are very similar to those obtained by OLS (the latter are not reported here, but are available from the authors). 22. To illustrate the quantitative impact of certain independent variables in subse- quent discussion, their values are assumed to shift from the 25th to the 75th percentile. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 141 characteristics are important. If the household head has one additional year of education, the increase in annual income growth is estimated to be be- tween 0.55 and 0.63 percentage points. Shifting a household from zero years of education (the 25th percentile) to seven years of education (the 75th per- centile) would increase annual income growth by 3.9 to 4.4 percentage points. One could expect households with younger heads to be able to adjust more swiftly to changing economic circumstances and show higher levels of in- come growth; bridging the interquartile range (25 and 48 years) would be expected to increase growth by between 3.2 and 2.6 percentage points. Con- trary to the opinion (and the evidence from simple cross-sectional correla- tions) that a higher initial number of household members is associated with lower levels of per capita income, households with more members (initially) saw higher subsequent income growth because of higher levels of family labor or the potential for consumption smoothing through informal family networks. The magnitude of the estimated coefficient is significant; a shift from 2.38 to 4.48 adult equivalents would be expected to increase income growth by 1.5 to 1.7 percentage points. Including households' initial asset endowments and income levels in the regression demonstrates that, despite the positive correlation between in- come and assets (0.13), the two variables have very different effects in the long term. While there has been divergence in assets, households appear to have converged strongly in income. In other words, households with higher levels of initial assets experienced higher levels of income growth, whereas households with high levels of initial income experienced lower subsequent income growth. This would imply that, during the period under review, the character of the growth process gave households with low initial levels of income opportunities to catch up, although possession of physical assets and-more important from a quantitative point of view-human capital, greatly improved their ability to do so.23 To illustrate the magnitude of the associated effects, note that the difference in assets between the 25th and the 75th percentile of the asset distribution would have affected income growth by less than a percentage point. By contrast, an equivalent shift in the income distribution would have had a dramatic impact of about 7 percentage points on subsequent income changes. To examine whether, as is often asserted, gender bias posed structural obstacles for female households, two dummies were included, one if a house- hold were headed by a female in 1992, and one for widowed households. Contrary to a popular perception and in line with findings in the recent lit- erature (for example, Appleton 1996), there seemed to be little evidence of bias against female households. To the contrary, female headship emerged as a positive, although not statistically significant, characteristic, confirming the notion that women did enjoy opportunities in the trade and service sectors 23. Indeed, a dummy for households that were below the poverty line in 1992 is highly significant and positive. 142 Klaus Deininger and John Okidi (Kwagala 1999). Widowed headship was negative as expected, but not sig- nificant at conventional levels of confidence. As a proxy for access to infrastructure at the household level, a dummy variable was included. That variable equaled one if the household were con- nected to electricity in 1992. The magnitude of the coefficient is quite large, although it was significant at the 10 percent level only in one of the equations. The three location-specific characteristics included are the distance to public transport in 1992, whether the community had access to a bank in 1992, and a rural dummy. The rural dummy was positive (although insig- nificant), suggesting that policy-induced biases against rural areas were re- duced in 1992-99. Distance to transport was negative (although insignifi- cant). Access to banks emerged not only as the most significant, but also as quantitatively important (2.2 percent) in the regression with regional dum- mies. The significance of the coefficient decreased once district dummies are included. One would expect this result, because within districts there was much less variation in access to banks, so that part of the impact was ab- sorbed in the district-specific intercept. Household characteristics, in particular education, played an important role in income growth. Within a sound macroeconomic framework that pro- vides incentives to the private sector, raising the population's levels of edu- cational attainment appears as an important means of raising income growth, helping households overcome structural disadvantages, and reducing pov- erty. At the same time, given the significance of initial asset levels, any mea- sures that would improve the asset position of the poor, for example, by strengthening property rights to resources they already own, could have a significant impact on poverty. While the analysis suggests that observed location-specific characteris- tics are less important than household characteristics, we also noted the mag- nitude and statistical significance of the regional dummies for both the east- ern and northern regions. In these two regions annual income growth would be expected to be 4.2 to 4.3 percent lower than elsewhere in the country. This points to the presence of unobserved factors that have a profound impact on subsequent growth. Exploring such factors in more detail would be an inter- esting topic for future analysis. To summarize, in line with the analysis of consumption poverty (chap- ter 4 in this volume), growth appears to have been overwhelmingly propoor during 1992-99. Households who had lower initial income saw consider- able increases in their income. At the same time, the eastern and the north- ern regions appear to be characterized by structural barriers, for example, climatic endowments and access to technology, not directly related to ob- servable community attributes. This suggests not only a need for further analysis of the nature of these differences, but also a more comprehensive and integrated approach to promoting growth in these regions that links improved technology and nonfarm employment. The importance of finan- cial infrastructure suggests that mechanisms at the household level (for Rural Households: Incomes, Productivity, and Nonfarm Enterprises 143 example, expanding the range of assets that can be used as collateral) and at the community level could be important. Agricultural Productivity and Nonfarm Enterprises More detailed analysis would be necessary to test the extent to which the factors identified earlier, that is, households' education and physical assets, access to financial infrastructure, and technology, are relevant to the growth of agricultural productivity. Such analysis is not yet possible, because com- plete data on agricultural production will only be available when the two rounds of the 1999/2000 survey of crops will have been completed. How- ever, data from earlier household surveys can be used to examine both deter- minants of agricultural productivity and the start-up of rural nonfarm enter- prises. The rationale for including the latter is that, as a large and growing literature has demonstrated, complementing agricultural with nonfarm in- come offers many advantages and a potential for sustained growth.24 At the same time, credit and output market imperfections and households' endow- ments are likely to have similar effects on the scope for agricultural invest- ment and for diversification into off-farm activities. Stylized Facts At least three stylized facts characterize rural areas in Uganda. First, informa- tional imperfections give rise to high levels of credit rationing. Second, trans- action costs drive a wedge between buying and selling prices for different commodities, thereby generating a wide margin within which it is economi- cally rational for producers to remain self-sufficient. Third, households' en- dowments of human and physical capital are important not only from an effi- ciency point of view, but also for their ability to access markets. Credit market imperfections have implications for the use of recurrent inputs in agriculture as well as for investment in nonfarm activities. Even if 24. The ability to complement agricultural incomes with nonagricultural enter- prise activity is important for households to improve their ability to smooth consump- tion; reduce their exposure to risk and vulnerability (Reardon and Taylor 1996); and facilitate more efficient use of family, and especially female, labor during agricultural slack periods (Lanjouw and Lanjouw 1997). Rural nonfarm activity has been shown to be an important determinant of regional economic growth and households' ability to escape poverty in Asian countries such as China, India, and Thailand (Hayami 1998), and in South Africa and Zambia (Hazell and Hojjati 1995). Better linkages between the farm and nonfarm economy are believed to be important for broader development and have been argued to be of particular relevance in a predominantly agriculture- based economy such as Uganda (Bigsten and Kayizzi-Mugerwa 1995). Income from nonagricultural sources can also serve to generate funds for agricultural investment, especially where access to credit is limited (Reardon and Taylor 1996). 144 Klaus Deininger and John Okidi starting up a nonfarm enterprise (or the use of productivity-enhancing pur- chased inputs) would allow a household to increase the returns to all factors of production, doing so normally requires a minimum amount of liquidity or access to credit. Farmers who are not credit constrained will have a level of purchases of productive inputs or investment in nonfarm enterprises that will be closer to the optimum than those who are credit constrained. Credit constraints are affected not only by asset ownership, but also by proximity to financial services. Hence, in addition to households' levels of wealth (to be used either directly or as a collateral) and education, lack of access to finan- cial infrastructure could reduce input use and investment in nonfarm enter- prises below the socially optimal level.25 Transaction costs in output markets that arise, among others things, from distance to infrastructure, drive a wedge between purchase and sales prices of agricultural commodities. That difference in prices makes it rational for producers to remain self-sufficient, implying that price incentives will not affect their behavior and the shadow price of different factors may deviate significantly from what a commodity would command in the market.26 Households' asset endowments will, in line with the foregoing discussion on credit constrains and assets, and when markets for land, labor, or capital are imperfect, directly affect the level of agricultural input use and longer-term investment. For example, if, through the ability to use them as collateral, own- ership of assets is an inportant determinant of credit access, ownership of even nonproductive assets should affect the intensity of use of purchased in- puts for credit constrained (but not credit unconstrained) farmers. To explore these issues empirically, we used three approaches here. First, to provide insight concerning the optimality of input use, a production func- tion was estimated that included traditional inputs (own and hired labor, seeds, fertilizer, and other inputs) plus other productivity-enhancing factors such as education and agricultural experience. Coefficients from this pro- duction function were used to indicate not only the impact of household 25. Credit unconstrained farmers will equate the marginal value product of each of the inputs (hired labor, fertilizer, and home-produced seeds) to its market price. Consequently, fertilizer will be applied optimally, that is, exactly to the point where its marginal return equals the market price. Transaction costs in output markets in- crease the amount of seeds consumed and/or used as an input to agricultural pro- duction over and above the optimum without such transaction costs, irrespective of whether or not a farmer is credit constrained. 26. Output as well as market and price risk could explain deviations from profit- maximizing input quantities as well. However, it is difficult to construct a model that would explain the pattem of overapplication of one and underapplication of another input. If output or market price risk is a concern, farmers should store their wealth in a less risky asset than putting seeds into the ground. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 145 characteristics on agricultural productivity but, more interestingly, to make inferences on the degree to which input is optimal.27 Second, to identify whether credit constraints affected input use demand functions for different types of inputs (labor, seeds, and a fertilizer and pur- chased seed combination) were estimated. This allowed us to test to what de- gree these variables were affected by endowments rather than policy-related factors. If credit market imperfections are an important determinant of demand for purchased inputs (by increasing shadow prices for capital), exogenous capi- tal constraints should enter demand functions for purchased, but not for home- produced, inputs. Similarly, it is possible to identify the impact of availability of government services and other facilities on input demand. Finally, a probit equation for start-up of nonagricultural enterprises was estimated. Clearly, households' endowments with physical and human capi- tal, as well as community characteristics, would be expected to affect such investment. In addition, if, as hypothesized, the availability of financial in- frastructure affects households' ability to obtain credit for investment and working capital, households located closer to, say, banks, should be more likely to invest in and to continue to operate nonfarm enterprises. Agricultural Production and Productivity The estimation of the production function used data from 528 panel house- holds that were interviewed in the 1992/93 integrated household survey and the 1993/94 monitoring survey (see Deininger and Okidi 2000 for more detailed description of the data). Descriptive statistics were similar to the ones discussed earlier, and thus are not reported separately. Mean annual output value for the sample was about U Sh 190,000, ranging from U Sh 221,000 in the central region to U Sh 147,000 in the northern region. Farm household size averaged 4.9 people who cultivated an average of two hect- ares of land. About 17 percent of sample households used hired labor, with 27. If input use is correlated with unobservable characteristics of the household (for example, managerial ability) or the farm (for example, soil quality), coefficient estimates from a cross-section are likely to be biased. A priori determination of the sign of the bias will be difficult, in view of the multitude of potential unobservable or omitted variables. For example, fertilizer use is likely to be higher on low-quality soils, leading to an underestimation of the impact of fertilizer. At the same time, if better managers apply more fertilizer, possibly because local input traders who have at least some knowledge about producers' managerial ability are willing to approve higher lines of credit for them, the coefficient of fertilizer would capture such unob- served ability and therefore be biased upward. Panel data methods can be used to overcome this problem under the condition that unobservable variables are time- invariant (Mundlak 1978), an assumption that is likely to be satisfied in this case. 146 Klaus Deininger and John Okidi total household expenditure on hired labor averaging U Sh 15,800. While virtually all producers in the sample used seeds, only 7 percent used fertil- izer, pesticides, or other purchased inputs-a figure that was even lower (2 to 3 percent) in the western and northern regions. The mean value of nonland farm and other household assets together amounted to about U Sh 26,000 per producer. About a quarter of the farm households were female-headed and 11 percent were headed by widows. Eleven percent of households lived in communities that received advice from extension workers. Heads of sample households were on average 42 years old and had attended school for about 4 years. Mean experience, computed as the age of the agricultural enterprise, amounts to almost 20 years, implying that the average agricul- tural producer started an independent farming enterprise at the age of 23. Producers were also asked whether they had any difficulty in accessing sufficient credit to run their enterprise, a question that was answered posi- tively by 54 percent of the whole sample. All households who answered this question affirmatively were classified as credit constrained, irrespec- tive of whether or not they had actually received credit. Classifying house- holds this way suggests that credit constraints have acquired increased importance: while only 50 percent of producers had been credit constrained in 1992, 60 percent had problems in accessing credit in 1993. Results from the pooled ordinary least squares (OLS) regression and the fixed and random effects panel estimations of the production function are reported in table A5.6.28 The random-effects specification is the most appro- priate.29-3 Therefore, the subsequent discussion focuses on the coefficients from this estimator. With an elasticity of 0.36 and 0.28, respectively, labor and land were the obvious main inputs into agricultural production. For producers using hired labor, the point estimate for the production elasticity of hired labor at the mean was 0.25, which was significantly lower than the production 28. To accommodate the fact that more than 90 percent of the farmers in the sample did not use purchased inputs and more than 80 percent did not hire labor, a specifica- tion was adopted where a zero-one dummy variable for fertilizer and hired labor use was included as well as the product of these dummy variables and the observed input of fertilizer and hired labor. 29. While the coefficients obtained from the fixed-effects regression are consis- tent, they may be inefficient due to the failure to take account of variation within individual observations. The random effect estimator, which takes this variation into account, would be preferable if there were no correlation between the fixed effects and the error term. A Hausman test fails to reject the hypothesis of equality between the coefficients from fixed and random-effects estimation, suggesting that the random- effects specification is the most appropriate 30. The test statistic is distributed according to a X2 distribution with 20 degrees of freedom, and the value of 25.02 is below the critical values for the 5 percent (31.41) and the 10 percent (28.41) level. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 147 elasticity for family labor. Indeed, equality of the coefficients on own and hired labor could be rejected at any conventional level of significance, sug- gesting that, in line with the literature, supervision constraints limit sub- stitutability between family and hired labor (see, for example, Frisvold 1994). Farm assets were shown to have a production elasticity of slightly above 6 percent, which, given that they entered the production function in value terms, was equivalent to their economic return. Farming experience made a clearly positive contribution to productivity, the magnitude of which increased rapidly up to about 5 to 8 years and flattened off subse- quently, reaching its maximum at 24.5 years. Household education is relevant for production outcomes. One addi- tional year of education by the household head is estimated to increase productivity by 5 percent, in addition to a positive return on farming ex- perience. The lack of significance of the squared term suggested nondecreasing returns to education in agricultural production over the range observed in the sample. The point estimate of the coefficient is large, suggesting that having universal primary education (seven years com- pleted) for the population of farm operators would increase production by 15 percent at the margin.31 This is consistent with the weaker and more indirect evidence for payoffs to education found by Appleton and Balihuta (1996) based on 1992 data. It contrasts, however, to the result by Bigsten and Kayizzi-Mugerwa (1995) who found that in the preliberalization pe- riod, returns to education were negligible. Assuming that their result can be taken to be representative for the whole of Uganda,32 this would pro- vide evidence that economic liberalization, in particular the elimination of monopoly marketing in the agricultural sector had, by 1992/93, created an environment where returns to education increased. While a community's access to roads does not affect agricultural productivity, it may affect the quantities of input used through its impact on prices. Compared with the importance of education, a surprising finding is that community-level ac- cess to extension services (a variable that is available only at the commu- nity level), although positive, remained insignificant, and that for 1992- 93, productivity seems actually to have decreased. Examining the degree to which different trends-at the national or the regional level-have emerged in the interim would be of great interest. 31. Re-estimation of the same model with only the level of education entered does not change the magnitude of the coefficient. 32. While they only had a small sample (200 households), they surveyed one of the most dynamic and technologically advanced agricultural districts. As the received wisdom in the literature is that education is of value in dynamic environments char- acterized by economic and technological change, it is very likely that failure to find positive retums to education in this area implies the lack of such retums in other districts as well. 148 Klaus Deininger and John Okidi Efficiency of Input Use and Determinants of Factor Demand In addition to obtaining information on the determinants of productivity in general, a main objective of estimating the production function was to examine the degree to which the use of inputs has been economically optimal. Indeed, coefficients from the random effects estimation point toward underutilization of purchased inputs. Shifting farmers who do not use fertilizer to the mean level of fertilizer consumption (about U Sh 3,900) observed in the sample would increase output by almost 50 percent (U Sh 8,900), thereby providing a more than 100 percent return on the required outlay. Similarly high returns are found for a marginal increase in fertilizer use among producers who already apply fertilizer. For example, applying U Sh 1,000 more of fertilizer would increase output by U Sh 2,200, again, a more than 100 percent return. Even with provi- sions for transport costs, the use of ferftlizer would appear to be an attractive investment, with rates of return significantly above the cost of credit.-" The ob- served underuse of fertilizer may therefore point toward the existence of credit market imperfections. While farmers use too little fertilizer, regression results suggest that they apply more than the optimum amount of home-produced seeds.3 Although there is clear evidence of inefficient input use, the production function by itself did not indicate the degree to which imperfections in financial or input markets may be responsible for this inefficient use. This question provided a motivation for testing whether credit constraint was a factor underlying the apparently suboptimal factor use. To do so, we needed information on whether or not households were credit constrained, something that cannot be directly observed. To make inferences on their credit worthiness, we used producers' responses to the question that asked whether they had difficulty accessing credit to run their enterprise. Data show that only about half of the households who failed to obtain credit were credit constrained in this sense, while many farmers who obtained credit had difficulty in getting the credit they wanted, and thus were actually credit constrained.35 33. Noting that even the most remote producers are less than 600 kilometers from Kampala, transport costs would at most increase fertilizer prices by 20 percent (as- suming a fertilizer price of US$300 and a transportation cost of US$0.1 per ton- kilometer). 34. If anything, the coefficient on this variable is biased downward, because in the case of perennials, producers obtain output even without having applied seed in the current production cycle (and there is no measure of the production stock ap- plied). 35. The cross-tabulation of constrained producers with actual loan recipients (for the whole sample) is as follows: Constrained Unconstrained Total Received no loan 3,931 5,159 9,090 Received a loan 181 652 833 Total 4,112 5,811 9,923 Rural Households: Incomes, Productivity, and Nonfarm Enterprises 149 Given the small number of households who actually used fertilizer and the fact that only a few changed from nonusers during the short pe- riod observed, there was too little variation over time to make inferences on input demand from the panel estimation. We therefore pooled obser- vations for 1992 and 1993 using maximum likelihood and OLS estimators for fertilizer, hired labor, and seed use, all of them normalized per hectare of land used. We also included the total value of assets rather than indi- vidual components. The second column in table A5.7 illustrates results for the use of pur- chased inputs. As hypothesized, capital constraints significantly reduced the propensity to use fertilizer. By contrast, the possession of assets and land, which can be used as collateral in credit markets, increased the prob- ability of fertilizer use. The positive sign of family size suggests that fertil- izer and family labor complement each other. While neither experience nor age had a significant effect, fertilizer use increased more than propor- tionately with higher levels of household education. Also, the coefficient of access to extension in the fertilizer demand equation was positive and significant. This suggests that extension helps increase the intensity of fer- tilizer use, thereby bringing the producer closer to the profit-maximizing optimum. Distance to infrastructure was, as predicted, negative and sig- nificant, which reflects, in part, a price effect. Regional dummies indicate that in all regions the propensity to use fertilizer was much lower than in the central region. As all households in the sample used seeds, we estimated an OLS equation of the quantity of seed used. The results, listed in the fourth column of table A5.7, suggest that, contrary to what was found for fertilizer, capital constraints increased farmers' propensity to use seeds. The area of land owned had a sig- nificant and negative impact, whereas more family labor and higher levels of assets increased the level of seeds applied per hectare, possibly by allowing more intensive cultivation. In contrast to fertilizer use, which decreased sig- nificantly over time, farmers actually increased the rate of application of home- produced seeds. Together with the lack of significance for virtually any other variable (education, extension, distance to infrastructure) except regional dum- mies, this implies that farmers without sufficient access to working capital tried to substitute home-produced inputs for purchased inputs. The use of hired labor was determined mainly by household characteris- tics such as the amount of land owned, the gender of the household head, the amount of assets owned, and the household head's level of education. All these variables have a strong, positive impact on labor use. The impact of asset ownership (human as well as physical capital) suggests that well- educated and well-endowed households established nonagricultural enter- prises. Finding out why these households choose to adjust through the labor rather than through the land rental market is important. As theory suggests that the latter would provide superior incentives that would increase pro- ductive efficiency, identifying the obstacles to proper functioning of the land rental market could answer this question. 150 Klaus Deininger and John Okidi Nonagricultural Enterprise Start-Ups During 1988-92 a considerable number of new enterprises sprang up, most in the trade sector. Almost 50 percent of all households, and almost one-third in rural areas, started a nonagricultural enterprise during this period. The sectoral composition of new enterprises established differs between regions. New farm enterprises dominate in the national aggregate and in the eastern and northern regions. The opposite is true for the central and western re- gions, where nonfarm enterprises exceeded farming enterprise start-ups. Of the nonfarm businesses that were established during the period, most were in trade (26 percent of households), followed by manufacturing, hotels, and other services. Even within the trade sector, regional variation was pro- nounced (34 percent of households started a trade enterprise in central re- gion, but only 13 percent in the northern region).-6 Empirical results (table A5.8) suggest that new trade enterprises did not require large physical assets, but that enterprise start-ups were critically de- pendent on a minimum level of education. As coefficients in table A5.8 are marginal probabilities (at the mean of all other variables), they can be di- rectly interpreted. For example, having a bank in the community increased the probability of a household diversifying into trade by 5.6 percent; similar to directly connecting to infrastructure a household living at the mean dis- tance (33 kilometers) from a road in the north. A household whose head has completed primary education was about 5 percent more likely to establish a trading enterprise than one who did not attend school. There was no gender bias against starting a trading activity. Recent immigration had a significant and negative impact (by about 2.5 percentage points), supporting the conjec- ture that trade requires longer presence in a community to build trust and acquire information on the individuals that are likely to be involved in trans- actions so as to assess their creditworthiness accurately. The fact that high levels of initial assets were insignificant may indicate that with relatively low barriers to entry (with the exception of education), the returns from trad- ing activity were generally low and once individuals had acquired sufficient levels of wealth, it paid for them to diversify into other areas. To the degree that trade enterprises can serve as a point of entry into the off-farm economy 36. Prior to a discussion of the results, two technical issues need to be addressed. First, to control for agroclimatic and other endowments that affect placement of infrastructure, we include mean community income as one of the independent vari- ables. Second, to avoid reverse causality whereby the establishment of enterprises after 1987 but before 1992/3 has led to increased wealth-rather than the other way-it is necessary to include the household's initial capital stock. Unfortunately, we have only asset levels at the beginning for 1992. As results from re-estimating the same equations for enterprises started up only in 1992 (in which case our asset measure clearly reflects initial conditions) were virtually identical, the five-year re- gressions are reported here. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 151 and a springboard to accumulate resources and experience to make the tran- sition to other stages, the results suggested that education, financial services, and road access are important determinants of off-farm diversification. The number of enterprises established in other sectors was not only much lower, but they were apparently less dependent on infrastructure access and completely independent of educational levels. The establishment of crop farm enterprises in a relatively land-abundant economy such as Uganda seems to be even more strongly correlated with life cycle phenomena. While house- holds starting nonagricultural enterprises do so at a more advanced age, the opposite was true for household heads establishing a farming enterprise. The strongly negative coefficient on age, together with almost universal in- volvement of households in the rural farm economy, could indicate that farm enterprises serve as an important first stepping stone into the nonfarm economy. The positive coefficient on recent in-migration is likely to capture households who migrated in search of land or were displaced through war. Livestock and crop enterprises shared a location in relatively distant areas. There are, however, a number of marked differences. The significance of as- set ownership and age of the household head suggests that new livestock enterprises generally occurred at a later stage in the life cycle, a finding that is supported by the bias against female-headed households and the positive significance of household size (reflecting the need for additional household members to care for livestock). Both these results supported the conclusion that, although financial services and education improved the efficiency of the traditional farming sector, they were not an essential precondition for establishing livestock or crop enterprises. Even though they were based on earlier data, the results from the analy- sis of agricultural productivity and of nonagricultural enterprise start-ups were quite consistent with those from the earlier discussion of income growth between 1992 and 1999. There was no indication of a conflict between growth in agriculture and the nonagricultural rural sector. To the contrary, the main factors contributing to agricultural productivity (education, financial infra- structure, and asset ownership) were also key to improving economic per- formance in the nonfarm sector. Contrary to earlier studies, which found negligible returns to education in the prereform period, we find that educa- tion has become an important determinant of both agricultural and nonagri- cultural activity. This suggests that Uganda's broad macroeconomic and sectoral reforms succeeded in restoring incentives and in increasing returns to private factors of production. It also implies that by focusing on broad improvements in educational achievement, the government has identified- and is aiming to improve-one of the key constraints to future economic growth and poverty reduction in rural areas. The fact that community income levels emerged as the quantitatively most important determinant of nonfarm enterprise start-ups provides fur- ther indication for complementarity between farm and nonfarm sectors. The rural sector's ability to respond to the changed economic environment 152 Klaus Deininger and John Okidi and incentive framework has been constrained by imperfections in factor markets, especially access to productive infrastructure. Limited progress in improving productivity indicates that in addition to improving levels of human capital and the functioning of factor markets, greater efforts are required to improve the availability and awareness of improved technol- ogy through research and extension. Uganda has thus far seen an organic evolution of the off-farm sector based on agricultural income growth. This evolution is in marked contrast to coun- tries where unequal initial asset distribution (for example, education and land) has led to an unequal distribution of off-farm income, thereby causing further polarization of the income distribution (Feldman and Leones 1998; Lanjouw 1998). To maintain this relationship that, thus far, seems to have prevented increases in overall income inequality, it will be essential to ensure a regional balance in policies aimed at promoting education, infrastructure, and agricul- tural productivity to ensure broad access to economic opportunities. Failure to do so will not only cut the tight link between growth and poverty reduction that has been characteristic for Uganda thus far, but also threaten the sustainability of economic growth in a more fundamental way. Conclusions Noting the critical importance of rural income growth for overall poverty reduction in Uganda, this chapter combined descriptive evidence and econo- metric analysis to highlight the accomplishments of the past and to outline the challenges that Uganda faces in the future. The community- and household-level data suggest accomplishments in a number of areas, namely: * During 1992-99 levels of per capita income grew significantly with- out deterioration in income distribution. Households with low income levels in 1992, but with human and physical capital assets, were able to benefit the most from overall growth. * Cotton output has recovered and shows strong signs of growth, espe- cially in the northern region. Similarly, nontraditional crops (toma- toes, cabbage, and fruit) are grown more widely and could provide a basis for diversification and sustained income growth in rural areas. Half of the communities (villages) in the north reported that both the number of cotton producers and cotton yields increased between 1992 and 1999. In the eastern region 25 percent of communities reported increases in the number of growers and yields. * The extent of livestock ownership has increased significantly (the num- ber of owners more than doubled) as has investment in rural areas. The use of HYV has also increased considerably, albeit both from very low levels. * The functioning of rural factor markets has improved and the number of land rental transactions and the share of producers with access to Rural Households: Incomes, Productivity, and Nonfarm Enterprises 153 credit have increased strongly. More interesting, most of the credit obtained is used for productive investment, suggesting that produc- ers are aware of and make use of new economic opportunities. * The share of producers who have access to extension services and pri- mary school enrollments have increased greatly in rural areas. Given the importance of education for agricultural productivity and the start- up of nonfarm enterprises, this could provide the basis for more knowledge-based development in the future. * Although agriculture remains the mainstay of the rural economy, ru- ral households have used opportunities to diversify into off-farm in- come generation and establishment of nonagricultural enterprises. Analysis of the determinants of nonfarm enterprise start-ups illus- trates the crucial role of education and access to financial markets. At the same time, despite the indisputable successes, there is little room for complacency: * With the exception of cotton, the north has seen little agricultural di- versification and growth. In the rest of the country, output remains variable, mainly due to crop diseases. Differential performance by communities even within the same region suggests that better access to existing technology and information could offer large scope for in- creasing productivity. * Despite continued efforts, extension service coverage remains limited, and 64 percent of communities reported not having access to an ex- tension worker. Similarly, about a quarter of producers reported not having used credit due to the nonavailability of a bank. * Land conflicts exist in about half of the communities. Unless cost- effective ways are found to implement recent land legislation, these conflicts and other tensions could easily threaten social stability and rapid development. The analysis in this chapter demonstrates that the government's strong fo- cus on rural areas has resulted in a propoor growth pattern and has provided the preconditions for a revival of the rural sector. However, it also demon- strates that technology, financial services, and infrastructure will be needed to improve levels of human capital and structurally transform the rural sector. In all these respects Uganda has the potential to teach valuable lessons to other African countries that have recently embarked on programs of liberalization. 154 Klaus Deininger and John Okidi Annex 5.1. Tables of Estimation Results Table A5.1. Changes in Extent of Production, Number of Producers, and Yields of Main Commodities, 1992-99 Current production Changes since 1992 (percent) Crop crop is grown by Number ofproducers Yields Reasonfor and region 50%-100% 0%-50% 0% Increased Decreased Increased Decreased yield change Matooke Central 33.3 23.0 43.7 20.7 14.9 14.9 25.3 Disease Eastern 10.6 31.3 58.1 8.9 21.2 7.8 18.4 Disease Northern 2.0 8.2 89.8 4.1 3.1 2.0 5.1 n.a. Western 55.4 26.4 18.2 31.1 20.9 13.5 58.8 Weather National 25.8 24.0 50.2 16.4 16.6 9.6 28.7 Weather Maize Central 54.0 5.7 40.2 34.5 2.3 29.9 13.8 Fallow Eastern 60.3 10.1 29.6 40.2 2.8 35.8 9.5 Other Northern 48.0 26.5 25.5 14.3 36.7 12.2 60.2 Weather Western 73.6 14.2 12.2 47.3 8.8 28.4 43.2 Weather National 60.7 13.7 25.6 36.3 10.9 28.1 29.7 Weather Beans Central 51.7 11.5 36.8 31.0 8.0 26.4 20.7 Fallow Eastern 50.8 30.2 19.0 41.3 7.3 25.1 13.4 Other Northern 73.5 11.2 15.3 11.2 21.4 11.2 70.4 Weather Western 81.8 4.7 13.5 51.4 6.8 29.1 39.2 Weather National 64.3 16.0 19.7 36.7 10.0 23.8 33.0 Weather (table continues onfollowing page) Table A5.1 continued Current production Changes since 1992 (percent) Crop crop is grown by Number of producers Yields Reason for and region 50%-100% 0%-50% 0% Increased Decreased Increased Decreased yield change Sorghum Central 2.3 16.1 81.6 5.7 2.3 3.4 3.4 n.a. Eastern 31.3 24.6 44.1 7.8 16.2 13.4 16.2 n.a. Northern 27.6 18.4 54.1 4.1 18.4 4.1 30.6 Weather Western 33.1 35.1 31.8 16.2 8.8 14.2 20.9 Weather National 26.2 25.0 48.8 9.2 12.1 10.2 18.2 Weather Q Millet Central 8.0 11.5 80.5 8.0 3.4 5.7 3.4 n.a. Eastern 33.0 29.1 38.0 9.5 20.1 8.9 31.3 Weather Northern 68.4 16.3 15.3 7.1 22.4 8.2 61.2 Weather Western 59.5 18.2 22.3 42.6 8.1 31.1 20.3 Weather National 43.2 20.5 36.3 18.4 14.3 14.6 29.1 Weather Groundnut Central 20.7 26.4 52.9 20.7 12.6 16.1 23.0 Animals Eastern 31.8 39.1 29.1 17.9 31.3 16.2 34.1 Disease Northern 37.8 31.6 30.6 14.3 36.7 8.2 66.3 Weather Western 37.2 36.5 26.4 31.1 10.8 14.2 33.8 Weather National 32.6 34.8 32.6 21.5 23.2 14.1 38.3 Weather (table continues on following page) Table A5.1 continued Current production Changes since 1992 (percent) Crop crop is grown by Number of producers Yields Reasonfor and region 50%-100% 0%-50% 0% Increased Decreased Increased Decreased yield change Cassava Central 32.2 26.4 41.4 12.6 35.6 6.9 44.8 Disease Eastern 36.3 28.5 35.2 33.5 26.3 29.1 26.8 Disease Northern 59.2 25.5 15.3 15.3 27.6 15.3 65.3 Weather Westem 61.5 14.2 24.3 45.3 6.1 25.7 31.8 Disease National 47.3 23.4 29.3 29.9 22.3 21.7 38.7 Disease Coffee Central 19.5 25.3 55.2 6.9 29.9 2.3 39.1 Disease Eastem 21.2 19.6 59.2 16.2 12.3 12.8 16.8 Disease Northern 0.0 1.0 99.0 1.0 0.0 0.0 0.0 n.a. Westem 27.0 27.0 45.9 18.2 19.6 6.1 32.4 Disease National 18.6 19.1 62.3 12.3 15.0 6.6 21.9 Disease Cotton Central 1.1 6.9 92.0 4.6 2.3 3.4 4.6 n.a. Eastem 9.5 28.5 62.0 24.6 8.9 21.8 8.4 Technology Northern 30.6 35.7 33.7 50.0 11.2 48.0 16.3 Other Westem 2.0 3.4 94.6 3.4 3.4 0.7 1.4 n.a. National 10.0 18.9 71.1 19.9 6.6 17.6 7.2 Other (table continues on following page) Table A5.1 continued Current production Changes since 1992 (percent) Crop crop is grown by Number of producers Yields Reasonfor and region 50%-100% 0%-50% 0% Increased Decreased Increased Decreased yield change Tomato Central 5.7 16.1 78.2 14.9 2.3 17.2 2.3 Input and labor use Easterm 3.9 43.6 52.5 16.2 5.6 10.1 7.3 Disease Northern 2.0 14.3 83.7 10.2 0.0 5.1 8.2 Weather Western 10.1 47.3 42.6 17.6 5.4 1.4 20.9 Disease National 5.7 34.4 60.0 15.2 3.9 7.8 10.5 Disease X Cabbage Central 2.3 8.0 89.7 9.2 1.1 10.3 0.0 n.a. Easterm 3.9 35.2 60.9 2.8 5.6 2.8 6.7 Disease Northern 1.0 9.2 89.8 2.0 0.0 0.0 3.1 n.a. Western 4.7 48.0 47.3 11.5 7.4 0.0 22.3 Disease National 3.3 29.3 67.4 6.3 4.3 2.7 9.4 Disease Mango Central 3.4 4.6 92.0 0.0 4.6 2.3 1.1 n.a. Eastern 1.1 8.9 89.9 0.0 0.0 0.0 0.0 n.a. Northern 42.9 25.5 31.6 2.0 5.1 3.1 36.7 Weather Western 19.6 10.1 70.3 14.2 0.7 1.4 3.4 n.a. National 14.8 11.7 73.4 4.5 2.0 1.4 8.2 Weather (table continues onfollowing page) Table A5.1 continued Current production Changes since 1992 (percent) Crop crop is grown by Number of producers Yields Reasonfor and region 50%-100% 0%-50% 0% Increased Decreased Increased Decreased yield change Orange Central 2.3 3.4 94.3 0.0 0.0 0.0 2.3 n.a. Eastern 1.1 27.4 71.5 2.2 6.7 0.6 3.4 n.a. Northern 20.4 36.7 42.9 2.0 1.0 2.0 53.1 Disease Western 2.0 6.8 91.2 3.4 3.4 0.0 6.8 n.a. t National 5.3 19.1 75.6 2.1 3.5 0.6 13.7 Disease Passion fruit Central 1.1 2.3 96.6 2.3 0.0 0.0 1.1 n.a. Eastern 0.0 23.5 76.5 1.1 4.5 0.6 1.1 Reduced fallow Northern 1.0 15.3 83.7 9.2 0.0 11.2 1.0 Labor use Western 8.8 16.9 74.3 10.8 1.4 2.0 9.5 Weather National 2.9 16.4 80.7 5.7 2.0 2.9 3.5 Weather n.a. Not applicable. Source: Authors' calculations based on key informant interviews in 512 commurnities; integrated household survey 1992/93 and the national household survey 1999/2000. Table A5.2. Changes in Technology and Input Use, 1992-99 Panel 1. Ownership of livestock and mechanical equipment (percent of household level) Value of livestock owned Owning cows in Owning bulls in Owning oxen in Owning plow in (million U Sh) Region 1999 1992 1999 1992 1999 1992 1999 1992 1999 1992 Central 15.7 7.8 4.1 2.2 0.2 0.1 0.9 0.7 1,182.03 956.08 East 22.1 11.9 8.7 4.4 4.4 2.1 7.2 4.1 612.46 354.20 North 18.3 8.9 8.9 4.7 2.7 1.6 6.7 4.3 844.10 586.88 West 22.4 13.7 7.9 4.4 0.2 0.1 2.0 1.6 1,444.22 1,130.03 z Total 19.8 10.7 7.2 3.8 1.8 0.9 4.0 2.5 1,004.42 738.36 Panel 2. Use of ox plows, community level (percent) Used by Increased Did not Decreased Region 100% 75% 50% 25% 0 >25% 0-25% change 0-25% > 25% Central 0.0 0.0 0.0 3.4 5.0 0.0 3.5 87.9 1.7 6.9 East 2.6 14.5 6.6 26.3 50.0 11.2 9.9 51.3 5.3 22.4 North 0.0 0.0 0.0 52.3 47.7 1.2 34.5 41.7 2.4 20.2 West 0.0 0.0 0.7 3.7 95.6 3.7 1.5 81.3 0.8 12.7 Total 0.9 5.1 2.6 21.3 70.1 5.4 11.2 63.8 2.8 16.8 (table continues onfollowing page) Table A5.2 continued Panel 3. Use of tractor, community level (percent) Used by Increased Did not Decreased Region 100% 75% 50% 25% 0 >25% 0-25% change 0-25% > 25% Central 0.0 0.0 1.6 9.8 88.5 1.7 1.7 73.3 10.0 13.3 East 0.0 2.0 1.3 22.4 74.3 0.0 4.6 64.5 9.2 21.7 North 0.0 0.0 0.0 6.2 93.8 0.0 0.0 81.3 1.3 17.5 West 3.0 2.2 0.7 5.9 88.2 3.0 3.0 79.9 3.0 11.2 Total 0.9 1.4 0.9 12.4 84.4 1.2 2.8 73.7 5.9 16.4 Panel 4. Use of hybrid seeds,fertilizer, and pesticides (percent) Household-level data Community-level data Area planted to HYV in Use of Fertilizer used by Pesticides used by Region 1999 1992 Manure Fertilizer Pesticides 0%-50% 0.0% >50% 0%-50% 0.0% Central 5.0 1.1 13.3 3.4 11.4 21.3 78.7 4.9 23.0 62.3 East 9.1 3.7 1.9 3.0 9.6 15.2 84.8 13.3 35.4 45.6 North 3.2 0.9 0.6 4.6 3.3 8.5 91.5 1.1 2.1 91.5 West 1.2 0.3 6.4 1.6 3.1 10.3 89.7 8.1 22.1 67.6 Total 4.8 1.6 5.8 3.0 7.1 13.1 86.9 8.0 22.7 64.1 (table continues onfollowing page) Table A5.2 continued Panel 5. Use of hybrid seeds and changes in such use, community level (percent) Hybrid seed used by Increased Did not Decreased Region 100% 75% 50% 25% 0%/0 >25% 0%-25% change 0-25% > 25% S Central 0.0 1.6 0.0 59.0 39.3 18.0 26.2 42.6 3.3 9.8 East 0.0 9.2 6.6 55.3 29.0 10.5 19.1 44.1 11.2 15.1 North 1.2 2.4 2.4 23.5 70.6 2.4 10.7 66.7 2.4 17.9 West 1.5 1.5 2.3 51.1 43.6 6.8 33.1 44.4 6.8 9.0 Total 0.7 4.4 3.5 48.3 43.2 8.8 22.8 48.4 7.0 13.0 HYV High-yielding varieties. Source: Authors' calculations from the 1999/2000 national household survey. Table A5.3. Changes in Functioning of Credit and Land Markets, 1992-99 Panel 1. Credit use, household level (percent) Household had a loan Reason for not applying Region In past In 1999 No need Don't know No bank No security High rates Central 15.4 11.8 38.0 18.4 9.6 27.1 6.8 East 20.2 20.0 37.2 17.3 12.3 28.5 4.7 North 6.1 6.3 49.3 21.4 14.1 11.3 4.0 West 24.0 23.7 47.1 18.6 10.8 16.2 7.4 Total 17.6 16.5 42.1 18.7 11.5 21.9 5.9 Panel 2. Credit availability (n 432) (percent) Source offormal credit Total Registered Local Institution closed down Region availabilitya Bank Entandikwab cooperative cooperative Total Bank Entandikwa Cooperative Central 49.2 23.0 32.8 4.9 10.0 32.8 6.4 14.8 24.4 East 38.2 13.2 25.7 5.3 9.2 36.9 1.5 19.7 23.9 North 20.0 0.0 10.6 7.1 4.7 10.6 1.2 3.5 7.9 West 64.7 21.1 34.6 13.5 19.6 28.0 4.8 11.3 29.9 Total 44.3 14.4 26.5 8.1 11.6 28.3 3.0 13.2 21.8 (table continues onfollowing page) Table A5.3 continued Panel 3. Credit use, household level (percent) Source of loan Purpose of loan Cooperative! Household Educationl Region Bank government NGO Business Relatives goods health Inputs Enterprisec Land d Central 6.1 14.7 27.8 3.5 48.0 5.1 16.0 12.0 54.3 12.6 East 1.7 13.7 16.7 2.3 65.7 7.0 29.6 6.6 49.4 7.4 North 8.2 35.3 29.4 3.5 23.5 1.3 6.3 40.5 50.6 1.3 West 7.0 26.7 5.5 4.2 56.7 10.6 28.2 15.9 34.8 10.6 Total 5.2 20.6 16.2 3.4 54.7 7.4 23.9 14.5 45.1 9.1 Panel 4. Land use and land rental markets, household level (percent) Cultivated land (in acres) Renting in land Renting out land Region 1999 1992 1999 1992 1999 1992 Central 2.53 1.91 25.3 9.7 12.3 5.4 East 2.26 1.54 30.4 7.5 13.4 5.5 North 1.77 1.29 10.6 3.1 8.3 2.7 West 2.28 1.58 23.2 6.0 5.7 1.3 Total 2.25 1.60 24.0 7.1 10.1 3.8 (table continues on following page) Table A5.3 continued Panel 5. Land sale and rental markets, community level (percent) Land price U Sh Landfor rental thousandsl Number of annual land sales Land sales market activity availablee Region acre 0 1-2 3-5 > 5 Increased Constant Decreased 1999 1992 Central 466.38 35.7 16.1 30.4 17.9 37.7 32.1 30.2 53.2 53.2 East 459.89 42.0 18.2 28.0 11.9 28.1 48.2 23.7 66.3 66.3 North 56.86 86.9 4.8 7.1 1.2 6.6 93.4 0.0 24.7 9.3 West 526.84 37.0 20.7 28.9 13.3 42.3 46.9 10.8 63.0 54.1 Total 398.23 48.6 16.0 24.4 11.0 29.9 54.3 15.8 55.2 49.5 Panel 6. Prevalence of land conflicts, community level (percent) Number of land conflicts Increased Did not Decreased Region 0 1-5 > 5 >25% 0%-25% change 0-25% > 25% Central 48.3 45.0 6.7 4.9 11.5 52.5 13.1 18.0 East 41.9 49.3 8.8 4.0 12.7 64.0 12.7 6.7 North 84.3 13.3 2.4 0.0 6.0 94.0 0.0 0.0 West 30.8 56.7 12.5 21.3 14.7 56.6 2.2 5.1 Total 48.2 43.6 8.3 8.8 11.9 65.8 7.0 6.5 NGO Nongovernmental organization. a. As conmmunities may have more than one source of formal credit, the percentages in this column will not necessarily be equal to sums of subsequent columns. b. Entandikwa is a govermnent credit scheme. c. Establishment of enterprise. d. Purchase of land or livestock. e. Land rental is practiced in the community. Source: Authors' calculations based on 1999/2000 national household survey. Table A5.4. Changes in Access to Infrastructure, Services, and Governance, 1992-99 Panel 1. Access to infrastructure and otherfacilities (community level) Time taken in minutes to Feeder Tarred Region road road Bus Taxi Truck Post Telephone Hospital Factory Changes a Central 9 42 49 35 47 45 46 46 39 10.3% East 18 63 52 43 47 55 61 48 58 12.2% North 66 123 87 78 115 131 139 119 107 1.0% West 20 80 55 42 52 62 66 41 66 3.4% Total 26 75 59 48 61 70 75 57 63 7.2% Panel 2. Access to extension advice and main source of information on technology (percent) Household level Community level 1-2 contacts a year > 2 contacts a year Main source of information on technology Extension Private Region 1998 1992 1998 1992 Radio agent sector Farmers Other Central 18.5 13.3 10.5 10.8 64.8 16.7 3.7 13.0 1.9 East 15.9 8.9 2.9 3.2 38.4 30.5 7.3 6.0 17.9 North 15.5 8.1 1.1 1.4 96.3 0.0 3.7 0.0 0.0 West 16.9 11.9 5.9 5.6 23.3 20.9 41.1 6.2 8.5 Total 16.8 10.7 5.3 5.5 48.6 19.7 16.6 5.8 9.4 (table continues onfollowing page) Table A5.4 continued Panel 3. Farmers' access to extension worker, community level (percent) Access to extension worker by region Increased Did not Decreased Region 100% 75% 50% 25% 0% >25% 0%-25% change 0-25% > 25% Central 0.0 8.2 3.3 21.3 67.2 6.7 20.0 65.0 6.7 1.7 East 2.0 10.6 4.6 28.5 54.3 4.7 9.4 55.0 7.4 23.5 North 0.0 1.2 1.2 7.3 90.2 1.3 1.3 77.5 1.3 18.8 West 9.6 11.9 1.5 18.5 58.5 9.7 21.6 56.7 3.0 9.0 Total 3.7 8.9 2.8 20.3 64.3 5.9 13.2 61.2 4.7 14.9 Panel 4. Cattle ownership and access to veterinary services, community level (percent) Veterinary services are Region Holding cattle Available Public Private Do artifical insemination Central 90.3 67.7 53.2 19.4 22.6 East 87.7 77.2 66.0 14.2 18.5 North 80.6 48.4 35.5 9.7 1.1 West 73.8 63.8 55.3 13.5 5.7 Total 82.3 65.9 54.8 13.8 11.6 (table continues on following page) Table A5.4 continued Panel 5. Physical violence, household level Percentage of households sufferingfrom Theft in Civil strife in Physical attack in Social capital c (percent) Region 1999 1992 1999 1992 1999 1992 1992 1998 c Central 17.5 10.6 8.0 3.6 7.8 6.6 33.0 33.0 East 22.1 13.5 15.2 10.1 10.5 8.6 40.1 40.1 North 21.1 16.8 10.3 10.4 9.3 7.4 39.7 35.3 West 19.1 12.9 18.4 7.8 8.3 6.9 31.1 26.8 Total 19.9 13.1 13.3 7.7 8.9 7.4 35.7 34.1 a. Whether or not changes in infrastructure access occurred between 1992 and 1999 b. Veterinaries that practice artificial insemination (Al) c. As explained in the text, social capital is defined as the share of households that, after experiencing a shock, received assistance from their communities. Source: Authors' calculations based on the 1999/2000 national household survey. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 169 Table A5.5. Determinants of Household-Level Income Growth, 1992-99 Robust regression estimates Item Coefficient t Coefficient t Mean Initial household characteristics Head's education in 1992 0.5522 4.11 0.6304 4.75 4.23 Head's age in 1992 -0.1395 -4.36 -0.1136 -3.55 36.76 Household members (adjusted equality) in 1992 0.7222 2.36 0.8511 2.81 3.55 Assets in 1992 (U Sh 1,000) 0.0004 5.48 0.0001 4.66 3489.96 Income in 1992 (U Sh 1,000) -0.0122 -28.39 -0.0113 -26.33 833.03 Female headed in 1992 0.6900 0.54 1.4081 1.10 0.24 Widowed in 1992 -2.1066 -1.19 -1.1334 -0.64 0.11 Electricity available in 1992 2.6040 1.40 3.4671a 1.81 0.10 Initial community characteristics District to public transport in 1992 -0.0253 -0.92 -0.0028 -0.10 11.90 Bank within 10 kms in 1992 2.2289 2.15 1.4444 1.32 0.38 Rural 2.4523 1.21 1.5507 0.75 0.91 Regional dummies Eastem region -4.6649 -3.65 0.25 Northern region -4.2278 -2.62 0.12 Western region 0.5006 0.42 0.34 Intercept 15.3028 5.20 18.6727 3.85 R2 adjustment/F 76.620 25.790 Number of observations 953 953 Note: Numbers in bold indicate that the estimated coefficient is statistically significant at the conventional level. a. Coefficient is significant at 10 percent. Source: Authors' calculations based on 1999/2000 national household survey and the 1992 integrated household survey. Table A5.6. Results from the Agricultural Production Function Estimation OLS pooled Fixed effects Random effects Item Coefficient t Coefficient t Coefficient z Family labor 0.3324 18.10 0.2228 2.04 0.3599 6.13 Hired labor dummy -1.3746 -7.50 -1.6366 -2.35 -1.7190 -3.17 Hired labor (log) 0.1663 8.73 0.1814 2.51 0.2003 3.57 Seed (log) 0.0970 18.43 0.1028 4.92 0.0949 6.00 Fertilizer dummy -1.4709 -7.67 -1.7179 -2.24 -1.9080 -3.30 Fertilizer (log) 0.1996 8.60 0.2418 2.65 0.2773 4.04 Other inputs (log) 0.0242 8.29 0.0296 2.59 0.0219 2.51 Land (log) 0.5140 18.76 0.1985 1.89 0.2842 3.74 Farm assets (log) 0.0564 8.99 0.0368 1.56 0.0633 3.61 Nonfarm assets (log) 0.0016 0.77 0.0043 0.52 0.0027 0.43 Female head dummy -0.1291 -5.34 -0.1663 -0.77 -0.1437 -1.77 Experience (log) 0.0152 7.76 0.1960 0.71 0.4904 2.53 Experience (log) squared -0.0002 -6.84 -0.0333 -0.63 -0.0759 -2.01 Age of head 0.0002 0.06 0.0280 1.11 0.0154 1.39 Age of head squared 0.0000 -0.43 -0.0003 -1.14 -0.0001 -1.29 Head's years of education 0.0382 5.40 -0.0297 -0.61 0.0497 2.13 Head's education years squared -0.0028 -4.76 0.0042 1.08 -0.0016 -0.79 Access to extension 0.0119 0.37 0.1844 1.67 0.1295 1.53 Tune dummy -0.1358 -5.62 -0.0346 -0.48 -0.1315 -2.17 Road distance -0.0004 -2.36 -0.0004 -0.60 0.0003 0.71 Rural dummy 0.1045 3.76 0.0597 0.70 Western dummy -0.0182 -0.65 -0.1486 -1.73 Eastern dummy -0.3660 -12.61 -0.3312 -3.61 (table continues on following page) Table A5.6 continued OLS pooled Fixed effects Random effects Item Coefficient t Coefficient t Coefficient z Northern dummy -0.5927 -18.80 -0.5925 -5.53 9 Constant 7.3323 70.02 7.1287 9.96 6.4980 18.47 Number of observations 8,651 1,046 1,046 R2 adjustment 0.3101 0.2505 0.3224 Hausman test 23.38 Note: Numbers in bold indicate that the estimated coefficient is statistically significant at the conventional level. Source: Authors' calculations based on the 1992 integrated household survey and the 1993/94 first monitoring survey (MS-i1). Table A5.7. Demand Functions for Fertilizer, Seeds, and Hired Labor Fertilizer (probit) Seed (OLS) Hired labor (Tobit) Item dF/dx z Coefficient t Coefficient t Capital constraints -2.037 -4.60 14.466 3.03 0.501 1.24 Land (log) 2.569 4.76 -58.894 -9.17 6.880 13.71 Household members (log) 0.839 2.05 26.888 6.06 0.390 1.03 Assets (log) 0.620 4.52 4.810 3.26 0.647 5.24 Female head dummy -0.338 -0.62 11.244 1.92 2.021 4.06 Experience -0.004 -0.10 0.370 0.80 -0.058 -1.46 Experience squared 0.000 -0.24 -0.008 -1.38 0.001 0.95 Age of head -0.105 -1.20 -0.008 -0.01 -0.018 -0.23 Age of head squared 0.001 0.70 -0.006 -0.67 0.000 -0.21 Head's education years -0.083 -0.55 2.855 1.67 0.602 4.30 Head's education years squared 0.036 3.30 -0.103 -0.74 0.011 1.08 Access to extension 2.288 3.31 2.478 0.32 1.083 1.64 Road distance -0.028 -4.15 -0.058 -1.29 -0.005 -1.36 Time dummy -3.203 -6.29 92.811 16.01 -2.611 -5.31 Rural dummy -1.154 -1.82 13.629 2.04 -3.999 -7.69 Western dummy -5.953 -10.94 56.745 8.47 -4.747 -8.38 Eastern dummy -1.812 -3.62 67.607 9.99 -2.485 -4.56 Northern dummy -4.128 -7.47 58.777 8.14 -5.001 -8.32 Constant 8.446 35.01 -19.096 -9.09 Log likelihood/R2 adjustment -1,844.491 0.083 -7,686.263 Note: Numbers in bold indicate that the estimated coefficient is statistically significant at the conventional level. Source: Authors' calculations based on the 1992 integrated household survey and the 1993/94 first monitoring survey. Table A5.8. Probit Estimates for the Probability of Enterprise Startups in 1987/88-1992/93 Nonagricultural enterprises Agricultural enterprises Trade enterprise Hotel enterprise Farm enterprise Livestock enterprise Item dF/dx z dF/dx z dF/dx z dF/dx z Total nonland assets (log) -0.7214 -3.96 0.5859 5.94 -0.4939 -1.80 1.3077 9.00 Bank in community 5.5962 4.77 2.2709 4.23 -4.0147 -2.61 -3.6549 -5.86 Moneylender in community 0.0894 0.06 -0.5469 -0.91 2.7562 1.34 -1.3260 -1.37 Household members (log) -0.1771 -1.15 -0.1558 -2.12 -0.0458 -0.22 0.3311 4.36 In-migration during last 5 years -2.4989 -2.45 0.9387 1.92 2.8544 2.01 -0.3367 -0.61 Female head dummy 1.2599 1.20 0.5406 1.12 -0.5825 -0.41 -1.8864 -3.36 Age of head 0.1738 1.09 0.2109 2.77 -1.4591 -6.66 0.2096 2.35 Age of head squared -0.0001 -0.07 -0.0021 -2.62 0.0116 5.13 -0.0019 -2.02 Head's education years 1.1136 3.58 -0.1984 -1.40 -0.0956 -0.22 -0.0060 -0.04 Head's education years squared -0.0549 -2.50 0.0021 0.21 -0.0324 -0.98 0.0165 1.35 Road distance -1.5513 -4.68 0.3818 2.58 2.5508 5.63 1.2528 6.67 Mean community income (U Sh/month) 13.4616 11.99 1.9381 3.88 -18.2777 -11.84 -1.0467 -1.74 Rural dummy -11.1854 -8.27 -0.4913 -0.83 20.8621 11.96 2.0507 2.85 Northern dummy -8.9709 -6.53 0.8875 1.11 0.8978 0.46 1.4919 1.97 Westem dummy 2.6348 2.07 9.0618 10.93 -2.5607 -1.42 -4.9195 -7.36 Eastem dummy -5.1673 -4.47 3.9793 5.56 3.4024 2.04 -0.5099 -0.82 Log-likelihood -3,711.17 -1,772.56 -4,626.67 -2,200.63 Pseudo R2 0.235 0.128 0.234 0.125 Note: Numbers in bold indicate that the estimated coefficient is statistically significant at the conventional level. Year dummnies included but not reported. Source: Authors' calculations based on the 1992 integrated household survey and the 1993/94 first monitoring survey. 174 Klaus Deininger and John Okidi References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Appleton, Simon. 1996. "Women-Headed Households and Household Wel- fare: An Empirical Deconstruction for Uganda." World Development 24(12): 1811-27. Appleton, Simon, and Arsene Balihuta. 1996. "Education and Agricultural Productivity: Evidence from Uganda." Journal of International Develop- ment 8(3): 415-44. Baland, Jean-Marie, Gaspart Frederic, Frank Place, and Jean-Phillippe Platteau. 1999. "Poverty, Tenure Security and Access to Land in Cen- tral Uganda: The Role of Market and Non-Market Processes." Research Series no. 216, pp. 1-39. Notre-Dame de la Paix, Faculty of Economics and Social Sciences, Namur, Belgium. Belshaw, Deryke, Peter Lawrence, and Michael Hubbard. 1999. "Agricultural Tradables and Economic Recovery in Uganda: The Limitations of Struc- tural Adjustment in Practice." World Development 27(4): 673-90. Besley, Timothy. 1995. "Property Rights and Investment Incentives: Theory and Evidence from Ghana." Journal of Political Economy 103(5): 903-37. Bigsten, Ame, and Steve Kayizzi-Mugerwa. 1995. "Rural Sector Responses to Economic Crisis in Uganda." Journal of International Development 7(2): 181-209. Brasselle, Anne-Sophie, Gaspart Fr6deric, and Jean-Phillippe Platteau. 1997. "Land Tenure Security and Investment Incentives: Some Further Puz- zling Evidence from Burkina Faso." Research Series no. 201, pp. 1-53. Notre-Dame de la Paix, Faculty of Economics and Social Sciences, Namur, Belgium. Brett, E. A. 1973. Colonialism and Underdevelopment in East Africa: The Politics of Economic Change, 1919-1939. London: Heinemann. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore and London: The Johns Hopkins University Press. Deininger, Klaus. 2000. "Is Land Tenure in Africa Really Unimportant for Investment?" Development Research Group, World Bank, Washing- ton, D.C. Processed. Deininger, Klaus, and Hans Binswanger. 1999. "The Evolution of the World Bank's Land Policy." World Bank Research Observer 14(2):47-76. Rural Households: Incomes, Productivity, and Nonfarm Enterprises 175 Deininger, Klaus, and Gershon Feder. 2000."Land Institutions and Land Markets." In B. Gardner and G. Raussser, eds., Handbook of Agricul- tural Economics. Amsterdam: Elsevier-North Holland. Deininger, Klaus, and John Okidi. 2000. "Market Participation, Agricultural Productivity, and Nonagricultural Enterprise Start-Ups." Development Research Group, World Bank, Washington, D.C. Processed. Feldman, S., and J. P. Leones. 1998. "Nonfarm Activity and Rural Household Income: Evidence from Philippine Microdata." Economic Development and Cultural Change 46(4): 789-806. Frisvold, George B. 1994. "Does Supervision Matter? Some Hypotheses Tests Using Indian Farm Level Data." Journal of Development Economics 43(2): 460-71. Hayami, Yujiro, ed. 1998. Towards the Rural-Based Development of Commerce and Industry. Selected Experiencesfrom East Asia. WBI Learning Resource Series. Washington D.C.: World Bank. Hazell, Peter B. R., and Behjat Hojjati. 1995. "Farm/Nonfarm Growth Link- ages in Zambia." Journal of African Economies 4: 406-35. Kwagala, Betty. 1999. "Integrating Women's Reproductive Roles with Pro- ductive Activities in Commerce: The Case of Businesswomen in Kampala, Uganda." Urban Studies 36(9): 1535-50. Lanjouw, Peter. 1998. "Rural Nonagricultural Employment and Poverty in Ecuador." Economic Development and Cultural Change 48(1): 91-122. Lanjouw Peter, and Jean 0. Lanjouw. 1997. "Rural Nonfarm Employment. A Survey." Policy Research Working Paper no. 1463. World Bank, De- velopment Research Group, Washington, D.C. Mundlak, Yair. 1978. "On the Pooling of Time Series and Cross-Section Data Sources." Econometrica 46(1): 69-85. Platteau, Jean-Phillippe. 1996. "The Evolutionary Theory of Land Rights as Applied to Sub-Saharan Africa: A Critical Assessment." Development and Change 27: 29-86. Reardon, Thomas, and Edward J. Taylor. 1996. "Agroclimatic Shock, Income Inequality, and Poverty: Evidence from Burkina Faso." World Develop- ment 24(5): 901-14. World Bank. 1996. Uganda: The Challenge of Growth and Poverty Reduction. Washington, D.C. 6 Crop Markets and Household Participation Donald Larson and Klaus Deininger During years of violence, political instability, and economic collapse, Ugan- dans retreated into self-sufficient agriculture. Even today, farming for home consumption remains a primary activity for most rural households. Crop mar- kets are therefore a key link between subsistence-oriented households and the recovering formal economy. Evidence from other countries indicates that some- times these markets do not work well, and high transaction costs-often re- lated to poor information and uncertainty about property rights and contract performance-can limit the benefits of participating in formal markets. This chapter presents evidence from survey data that suggest that crop markets in Uganda effectively convey prices from district to local markets, but that transaction costs differ significantly among crops. The community and household survey data also show that the variance in crop market characteris- tics and quality measures largely explains differences in prices. Current mar- ket conditions encourage households willing to produce export crops rather than food crops, since both transaction costs and uncertainty are lower for export crops (see Martin 1962 for a historical perspective). Given that the sur- vey includes community-level information about several crop markets, a dis- tinction can be made between factors that explain differences in average price levels between communities and those in crop markets. These results also suggest that transaction costs could be reduced by the development of private market institutions that foster product stan- dardization and help gather and disseminate market information. The scope for direct public interventions, however, is limited. In contrast to export crops, transaction costs are likely to be high in food crops because of a lack of specialization in Ugandan agriculture and labor markets more gener- ally, which gives rise to uncertain demand and thin markets. Some evi- dence exists, however, of diversification within sectors. A continued change 177 178 Donald Larson and Klaus Deininger in the composition of the economy along with growth in incomes can cre- ate greater opportunities for households to participate in formal domestic crop markets and export markets. Market Participation in the Early 1990s Uganda remained overwhelmingly rural in the early 1990s with only 12 per- cent of the population living in urban areas. Seventy-seven percent of Ugan- dans were farmers, and 94 percent of all rural households engaged in farming (table 6.1). Households engaged in other income-generating activities and many received remittances, so that agriculture generated just less than 60 percent of household income. Although sometimes on communal lands, farms were mostly family operated, and households provided most of the labor. Only 17 percent of households hired labor to work on crop farms, and only 13 percent of livestock farmers hired outside labor in 1992. Few households (4 percent) used fertilizers and few (10 percent) purchased seeds (Okidi 1999). On average, about 20 percent of the 1992/93 survey farm output was mar- keted. However, farmers who did market their crop tended to farm on a larger scale than those who did not. As a result, the share of the crop marketed was greater than the share a typical farmer marketed. Generally, rural households were largely self-reliant and consumed most of what they produced. Accord- ing to the survey, one-third were subsistence farmers, while half marketed less than 10 percent of their output. Of the farmers who grew cotton or coffee, the primary cash crops in Uganda, most marketed 25 percent of their output or more (figure 6.1). Farmers who grew cotton were also more likely to market other crops (table 6.2). Still, few of either type marketed more than 25 percent of their crops other than coffee or cotton (figure 6.2). The share of the crop marketed also varied significantly from district to district. Farmers in Luwero and Bundibugyo marketed 30 percent or more of their crop, while farmers in Kotido, Moroto, and Soroti marketed less than 10 percent. Nevertheless, the volumes sold were sufficient to give rise to local and district markets throughout Uganda. For example, respondents in all but four of the more than 800 communities surveyed noted that district markets were accessible. In only 39 of the communities had accessibility been limited by seasonal rains. While crop markets were generally available, results from the household surveys suggest that the volumes of particular crops reaching the general market varied greatly. Moreover, household surveys indicate significant dif- ferences in the number of farmers engaged in marketing each type of crop. Table 6.3 reports the share of production marketed by crop. The shares range from a low of 6 percent for sweet potatoes to 92 percent for coffee. The table also reports the amount the average household markets. For export crops, household averages for cotton and coffee match marketwide averages, but for several crops-for example, cassava and sim-sim-the two averages are significantly different, indicating that a smaller share of the farmers growing Table 6.1. Percentage of Ugandan Households Engaged in Farming, 1992 Rural Urban Urban and rural Livestock Livestock Livestock Location Agriculture Cropfarm farm Agriculture Cropfarm farm Agriculture Cropfarm farm W All Uganda 94.0 93.0 14.8 39.7 37.9 5.7 76.6 75.3 11.9 Central 91.2 88.9 14.4 33.4 29.9 6.7 70.0 67.2 11.6 Eastern 95.4 95.0 16.7 46.9 46.0 4.7 80.3 79.8 13.0 Northern 96.9 96.4 22.9 52.8 51.4 8.9 84.0 83.3 18.8 Western 92.7 92.0 6.1 30.0 28.8 2.8 73.5 72.7 5.1 Source: Okidi (1999). 180 Donald Larson and Klaus Deininger Figure 6.1. Marketed Share, Cumulative Frequency by Share of Crop Marketed, 1992/93 Marketed share by farm type 1.2 - ~1.0 0.8 - 0.6 - U 0.2- ,. ,.' 0 10 20 30 40 50 60 70 80 90 100 Marketed share of production (percent) All farms - - ---- Cotton and coffee farmers No coffee or cotton produced Source: 1992/93 integrated household survey these crops supplied crop markets. Together, the results suggest that some markets are thin and that these markets are supplied by a concentrated num- ber of households. Average margins-the difference between district and local market prices expressed as a percentage of local prices-showed a great deal of heteroge- neity among crops. Moreover, many are high in absolute terms (figure 6.3). In addition to the low volumes and concentration of suppliers already men- tioned, a number of additional reasons exist that may generate differences in margins among crops. These reasons are discussed in the next section. A Market Model for Community Trade Generally, crop markets in Uganda, as in most developing countries, are cash based. Performance risk is high and contract enforcement is weak. Informa- tion systems are informal for many crops and often network based. Assur- ances of all types-for quality and performance-tend to be based on repu- tation and personal trust. Property rights are insecure. This contrasts with most crop markets in industrial countries where transactions are impersonal and formal institutions guarantee performance and deliver information Crop Markets and Household Participation 181 Table 6.2. Number of Farms by Share of Output Marketed Crops other than Share All crops coffee or cotton marketed Farms with Other Farms with Other (percent) Allfarms cashcrops farms cashcrops farms 0 2,309 58 2,251 477 2,251 0-5 617 118 499 145 499 5-10 554 148 406 118 406 10-15 436 132 304 113 304 15-20 377 113 264 82 264 20-25 343 96 247 67 247 25-30 294 96 198 54 198 30-35 227 83 144 45 144 35-40 226 65 161 52 161 40-45 219 77 142 47 142 45-50 201 68 133 41 133 50-55 178 59 119 26 119 55-60 149 50 99 26 99 60-65 131 49 82 29 82 65-70 114 44 70 22 70 70-75 108 37 71 18 71 75-80 92 31 61 19 61 80-85 68 26 42 16 42 85-90 51 27 24 5 24 90-95 46 18 28 2 28 95-100 48 24 24 5 24 Total 6,788 1,419 5,369 1,409 5,369 Source: Author's calculations from the 1992/93 integrated household survey. (Fafchamps and Minten 1999 provide a good description of the differences between formal and informal markets). Differences in how markets are orga- nized are expected to affect outcomes as well. Transaction costs are generally lower in formal, impersonal markets, and the development of the institu- tions that promote such markets is considered key to long-term economic progress (see, for example, North 1989). All other things being equal, lower transaction costs result in lower marketing margins between communities that generate opportunities for trade, specialization, and productivity gains. In practice, formal and informal and personal and impersonal markets often coexist (see Kranton 1996, for example). It is likely that common, ex- isting marketing practices, despite limits, best solve the special circum- stances associated with each market. Circumstances can differ among crops, such that crop markets work differently in the same country, or even in the same community. In turn, the way the market works affects transaction 182 Donald Larson and Klaus Deininger Figure 6.2. Marketing of Crops Other Than Cotton and Coffee, 1992 2,500 - 2,000 - Q 1,500- 0 -S 1,000- 500- 0- 0 10 20 30 40 50 60 70 80 90 100 Marketed share of production (percent) Farms without cash crops Farms with cash crops Source: 1992/93 integrated household survey. costs, marketing margins, and household choices, including the decision even to enter markets at all. For crops largely produced and consumed lo- cally-but not exported cash crops-household decisions will affect the demand characteristics of the crop market, for example, the liquidity and volatility of the local market and the quality of goods demanded. Conse- quently, household choices and market organization are linked. One way to measure how well markets work is to see if transfer costs ex- plain differences in a given crop price among different villages. More formally, the spatial arbitrage model stipulates that when trade takes place between two locations, the observed crop price difference can be explained by the full cost of transferring the crop between the two locations. If the difference becomes larger, then traders will move enough of the crop between the two locations to bring prices back in line. Consequently, price changes in one location will be reflected in the other. When transfer costs are significant however, some price differences will be too small to cover the cost of transferring the crop. Trade will stop and prices can drift independently over some limited range. For esti- mation purposes the spatial arbitrage model can be expressed as: (6.1) E,[P,i + Kti(w; c, s)] + X,ij = E,(P,j), where PI and Pi are the contemporaneous price of a commodity in markets i and j; where Ki' is the associated transfer cost, which is a function of the variable Crop Markets and Household Participation 183 Table 6.3. Share of Crop Marketed, by Crop (percent) Crop Average Typical household Sweet potatoes 6 6 Millet 12 11 Sorghum 13 13 Cowpeas 16 7 Peas 17 11 Matooke 21 11 Other pulses, nuts, seeds 23 15 Irish potatoes 24 15 Beans 26 15 Maize cob 28 9 Bananas 34 39 Maize grain 35 22 Soybeans 40 24 Sim-sim 42 25 Groundnuts 49 16 Cabbage 62 36 Onions 64 33 Rice 66 49 Cassava 76 44 Cotton 76 74 Tomatoes 78 41 Coffee 92 91 Source: Authors' calculations from the 1992/93 integrated household survey. input prices w and fixed capital stock levels c, conditional on a set of state vari- ables s; and where X is a slack variable that equals zero when trade takes place between markets i and ] (Larson 2000). Variable input costs include labor, fuel, and working capital. Capital stocks reflect past public or private investment. Examples of capital include roads and communication infrastructure, but may also include difficult to observe human or social capital investments, such as past investments in reputation or in informal information systems. State variables are not subject to choice, but nonetheless affect the way markets are organized; that is, state variables affect choices by market participants about how they will conduct trade. For example, a lack of contract enforcement may encourage exchanges based on personal trust, differences in end-market con- ditions among crops may give rise to varying levels of risk and uncertainty, differences in security may lead to differences in risk by location, or govem- ment policy may encourage smuggling some goods into the country. The arbitrage model is a short-term measure of markets that asks whether markets convey appropriate price signals and incentives, while taking as given the condition of roads, the availability of information, the nature of demand, and the capacity of institutions. In the longer term investments can be made, 184 Donald Larson and Klaus Deininger Figure 6.3. Average Gross Margin between District and Local Market Prices, 1992 Average margins Cotton Other pulses, nuts, seeds Sim-sim Irish potatoes Millet Groundnuts Coffee Sorghum Maize cob Peas Cowpeas U Sweet potatoes Maize grain Cassava Onions Soybeans Bananas Tomatoes Rice Matooke Beans Cabbage e 1 = I I I 1 II 0 20 40 60 80 100 120 140 160 Share of local price (percent) Source: Authors' calculations based on 1992/93 integrated household survey. demographic and income developments can change local demand, and in- stitutions can evolve. Evidence from the 1992/93 surveys is helpful in ad- dressing the larger issue of market development. Crop Markets in 1992 The 1992/93 integrated household survey is especially valuable for research- ers interested in spatial arbitrage. The questionnaire provides information on price pairs-local and district-for crops in each community. While trans- Crop Markets and Household Participation 185 fer costs were not directly surveyed, community members were asked about factors likely to influence the cost of transfer, such as the distance to the dis- trict market and issues related to transportation and communications infra- structure, including the distance to all-weather roads, public transportation, rail stations, or a public telephone. Community members were also asked to choose the best description of credit access conditions and common marketing practices from a list. The choices were discrete, for example, the credit access conditions included no access, nearby access, access within 5 kilometers, and access within 10 kilo- meters. Together, these discrete and continuous variables can be used to ex- amine factors affecting marketing margins and differences in margins among communities and among crops. As discussed earlier, variations in community and marketing characteris- tics can be used to proxy the transfer costs given in equation 6.2: (6.2) Pic = bPpp d + z + I b.zmc + vc where the price of crop i in community c is a function of the price of the same crop in the district market d that survey respondents designate as most rel- evant, a set of cost and state variables z = [w c s], and a random error v. The matrix z includes continuous and dummy variables associated with I com- munity characteristics and m crop characteristics, where the continuous vari- ables are expressed in natural logs. Table 6.4 reports regression results based on 3,326 price observations from 666 communities, and 31 crops reported in the 1992/93 survey.' A summary of the main findings follows. PRICES. When interspatial arbitrage occurs, local prices are expected to move together with district prices. The regression results suggest that local and district markets are linked. All other attributes being equal, the regres- sion results indicate that a 10 percent increase in the district price for any crop would lead, on average, to a 9.4 percent increase in the local price. In the narrow sense, markets do indeed work in Uganda. Prices in local markets relate to markets in district markets. INFRASTRUCTURE. The survey measures types of infrastructure likely to af- fect transportation and transaction costs-distances to all-weather roads, public transportation, rail stations, and a public telephone-in kilometers, and local prices are expected to decline as these distances increase. 1. The model was estimated using the Mixed Model Procedure from SAS (1992). About 140 communities were excluded because of missing values. In addition, not all communities produced all crops. Estimates of community prices predicted from in- strumental variables were used in the reported results, although ordinary least square estimates were similar. 186 Donald Larson and Klaus Deininger Table 6.4. Model Results for 1992/93 Survey Continuous variables Estimate t-statistic District price 0.936 47.48a Distance to district market -0.003 -0.75 Phone -0.016 -2.52b All-weather road 0.004 1.51 Public transport 0.000 0.07 Testforfixed effects X2 statistic District 305.91a Crop 68.36a Unit 75.83a Credit availability 3.06c Inventories available 48.93a Marketing scale 0.39 Low district price 210.18a a. Significant at the 99 percent confidence level. b. Significant at the 95 percent confidence level. c. Significant at the 90 percent confidence level. Source: Authors' calculations. Transportation systems depend not only on the quality of the road that runs near a community, but also on the quality of the roads to which it connects. For example, in Kisoro District most communities are near all-weather roads, while, on average, traders must cover nearly 24 kilometers of seasonal roads to reach communities in Mukono District. A district effect is included in the regression because the efficiency of local infrastructure is affected by trans- portation and information systems and to compensate for missing informa- tion on regional input prices. Only the district effects and telephone variables were significant among these variables. According to the regression results, decreasing the distance to a telephone by 10 percent would lead to a 1.6 percent increase in local prices. Individually, district effects are not expected to have a specific sign, but taken together, the district effects are statistically important in explain- ing observed price differences. The effects of the other variables, including distance and road quality, are not significant. As with prices, these results are robust under a number of alternative assumptions, which we discuss later. CROP EFFECTS. The relationship between transport costs and infrastructure is likely to differ among crops. For example, poorly maintained roads might greatly affect the cost of matooke (plantain) transport because of bruising. Such differences are important when infrastructure and distance are used to proxy transfer costs, because they imply differences in parameter values. More- over, just as farmers choose among different production technologies based Crop Markets and Household Participation 187 on their own needs and the risks and benefits associated with the various approaches, market participants will decide on strategies based partly on the physical and market characteristics of the crop. For example, quality might be relatively more difficult to gauge for some crops, or the crop might spoil more readily. The demand and information characteristics of crops markets will likely differ as well, especially between export and food crops. Conse- quently, we include crop dummies in the estimation, which proved signifi- cant (table 6.4). PHYSICAL UNITS OF MEASURE. There is no standard unit of measurement for most commodities traded in rural Uganda. Coffee and cotton are generally measured in kilograms, but many communities trade other crops in local measurement units, for example, cans of various sizes. Altogether, the 1992 community survey data include 22 units of measure with each of the 31 crops measured in more than one unit. Fortunately, respondents quoted local and district prices in similar units. The Ugandan Bureau of Statistics provides a table of detailed conversion rates for the units so that, potentially, prices could be converted into stan- dard units. Early experimentation with the data showed that the conversion process itself introduced additional, and sometimes large, increases in price variability. In part, these differences may reflect true transformation costs, that is, the cost of sorting, grading, packaging, and so forth. Furthermore, units may indicate quality and information differences that are lost during conversion. For example, a heap (a unit of measure in the survey) is an inex- act measure and probably indicates a sack weighing less than 50 kilograms. For estimation purposes, prices were converted to logs, but fixed unit effects were included to account for quality of product and information effects (see Deaton 1997 for a good discussion of quality in the context of consumer de- mand). These proved highly significant. C(REDr. Working capital is required to trade crops, and often the volume of a trader's business is limited by lack of credit.2 Access to credit is likely to differ according to the crop. The role of credit and its availability may also contribute to differences in transfer costs among crops. For example, credit may be more readily available for export crops than for food crops, because cotton ginneries and coffee hullers and exporters will often finance their purchasing agents, sometimes by passing through offshore financing from buyers. Furthermore, for storable crops, the cost of storage that, in turn, is affected by access to credit, will affect price variability. This is significant, because most communities re- port that formal bank credit is unavailable (505 out of 666 communities, or 76 2. See Jones' (1972) account of the crop trade in Senegal; or more recently, Barrett's (1997) account of traders in Madagascar; or Baulch, Jaim, and Zohir's (1998) account of grain markets in Bangladesh. 188 Donald Larson and Klaus Deininger percent). Working capital is required to arbitrage markets, so access to credit is expected to be an important determinant of local prices. The regression results are mixed. Local prices increase by about 4 percent when credit is available; however, the result is statistically significant at a somewhat lower (91 percent) level of significance. STATE VARIABLE 1: INVENTORIES. When commodities are storable, owners not only decide whether to sell their crop locally or in another town, but they also decide whether to make that decision today or tomorrow. More specifically, when crops can be stored, prices must meet an intertemporal condition as well as the spatial arbitrage condition (Larson 1994). Using a simulation model, Williams and Wright (1991) show that spatial and tem- poral arbitrage work together to reduce price volatility and decrease the volume of trade needed to bring spatial markets into equilibrium. When stock-outs occur, the intertemporal condition need not hold. For crops that are not continuously produced, a lack of inventories may also signal that the crop is unavailable for trade. By drawing on the household survey, we generated a dummy variable to identify observations in which community members held stocks of a particular commodity. We included the variable in the regression, which proved significant (table 6.4). STATE VARIABLE 2: FAILURE TO COVER TRANSFER COSTS. When transfer costs are high relative to the price of the commodity, there is a large range of positive district prices over which trade will not occur. If such a condition prevailed in a large number of observations, the regression results could be biased. To address this possibility, we sorted the observations by crop and ranked by district price level. A dummy variable was included in the regression and set to one when an observation contained a district price among the bottom 20 percent of district crop prices. This dummy variable also proved significant. STATE VARIABLE 3: COMMUNT MARKETING ORGANIZATIONS. Often farmers will organize formal or informal associations for marketing purposes to reduce transportation costs, to strengthen negotiating positions, or to take advan- tage of other economies of scale. Survey participants were asked to choose among four options that describe how crops are organized for market: by individual farmers, by small groups, by large groups, or by formal coopera- tives. We included these responses in the regression as fixed effects (market- ing scale variable). The model results suggest, however, that marketing crops collectively has little influence on the prices observed in local markets. Explaining Differences among Communities The questionnaire covers both communities and commodities and a regres- sion of the type reported earlier attempts to explain the variation in prices due Crop Markets and Household Participation 189 to both, making it possible to decompose the variation into parts (Mundlak 1978). By averaging across commodities in any given community, it is possible to examine the determinants of community differences and explain why, on average, prices are higher in one community than another. Table 6.5 reports these between community regression results. As expected, distance and road quality are highly significant in explaining this portion of the total sample varia- tion. Nevertheless, the effects remain quantitatively small. Together with the results from the earlier regression, these results sug- gest that investments in infrastructure are important in explaining why crops generally receive a higher price in communities that have good access to trans- portation. The results also suggest that much of the price spread among com- munities is to be found in crop differences and that infrastructure does little to explain this portion of price variation. This point is also brought out by the crop model results discussed next. Crop Models Using the 1992/93 survey, the model was estimated by crop.3 Table 6.6 pro- vides selected results from the crop models that are largely consistent with the aggregate models. The price parameters are consistently significant and quantitatively close to one. For the most part, transport and communication variables were not statistically different from zero. Together with the earlier results, this suggests that the differences among crops-both the physical nature of the crop and the associated market characteristics-largely deter- mine differences among crop marketing margins. Table 6.5. Explaining Differences in Average Community Price Levels Variable Estimate t-score Price 0.958 350.81a Distance To market -0.005 _2.93a To phone -0.003 _1.45b To all-weather road -0.004 -4.17a To public transport -0.003 -2.83a Note: R2 equals 0.975. a. Significant at the 95 percent confidence level. b. Significant at the 90 percent confidence level. Source: Author's calculations based on 1992/93 integrated survey 3. In this version of the model unit effects were dropped to save on degrees of freedom. Table 6.6. Community Prices: Crop Model Estimation Results Distance to Distance to District price Distance to market Distance to phone all- weather road public transport Distance to rail stop Probability Probability Probability Probability Probability Probability Crop Estimate > tl Estimate >| t Estimate > |t Estimate > I t I Estimate > |t I Estimate > I t I Bananas 0.923 0.000 -0.056 0.148 -0.183 0.032 0.011 0.693 -0.158 0.258 0.104 0.479 Beans 0.992 0.000 -0.018 0.104 -0.056 0.014 -0.016 0.064 -0.004 0.744 -0.039 0.033 Cabbage 1.104 0.000 0.006 0.878 -0.208 0.208 0.004 0.896 0.027 0.659 -0.391 0.050 Cassava 0.925 0.000 0.002 0.854 -0.053 0.017 0.026 0.006 0.000 1.000 0.048 0.028 Coffee 0.931 0.000 -0.007 0.410 0.017 0.264 0.008 0.211 0.002 0.779 -0.012 0.361 Cotton 0.969 0.000 -0.015 0.372 0.017 0.577 0.000 0.978 0.004 0.670 -0.016 0.394 Cowpeas 1.064 0.000 -0.163 0.057 0.260 0.231 0.062 0.048 -0.011 0.797 -0.263 0.251 Groundnuts 0.995 0.000 -0.017 0.222 -0.024 0.318 -0.007 0.439 -0.015 0.159 0.070 0.010 Irish potatoes 0.934 0.000 -0.069 0.051 0.019 0.711 0.019 0.425 0.010 0.789 0.002 0.961 Maize cob 0.978 0.000 0.025 0.319 0.153 0.007 0.039 0.029 -0.005 0.784 -0.098 0.180 Maize gramn 0.944 0.000 -0.003 0.851 0.028 0.293 0.008 0.429 -0.013 0.265 0.012 0.657 Matooke 0.875 0.000 0.004 0.749 0.007 0.673 0.004 0.649 0.004 0.797 0.005 0.717 Millet 0.967 0.000 0.003 0.841 0.033 0.196 -0.013 0.100 0.011 0.248 -0.001 0.939 t Onions 1.110 0.000 0.005 0.943 -0.080 0.595 0.011 0.771 -0.011 0.878 0.287 0.076 m Other pulses, nuts, seeds 0.875 0.000 -0.211 0.343 0.222 0.016 -0.017 0.276 0.079 0.002 0.007 0.934 Peas 1.168 0.000 -0.078 0.311 -0.055 0.830 -0.091 0.153 0.147 0.074 -0.051 0.431 Rice 0.881 0.000 -0.008 0.752 -0.038 0.215 0.019 0.230 0.002 0.919 0.018 0.628 Sim-sim 0.950 0.000 -0.014 0.442 -0.036 0.487 -0.014 0.190 -0.013 0.266 0.021 0.510 >< Sorghum 0.949 0.000 -0.002 0.944 0.042 0.358 -0.019 0.089 -0.002 0.887 0.018 0.481 ą Soybeans 0.974 0.000 0.008 0.740 -0.033 0.428 0.031 0.126 0.028 0.215 -0.007 0.849 Sweet potatoes 0.865 0.000 -0.011 0.461 -0.003 0.892 -0.003 0.783 0.001 0.932 -0.015 0.429 Tomatoes 1.169 0.000 -0.013 0.697 -0.044 0.682 0.038 0.116 -0.039 0.414 0.017 0.712 . Source: Reported estimation results based on the 1992/93 integrated household survey. The Determinants of Market Participation Transfer costs in commodity markets set bounds on potential gains from trade for poor rural households. Most analysts recognize that rural households in many poor countries and regions face imperfect markets.4 Apparent from the analysis above, differences prevail among commodity markets. More- over, since transfer costs are also partly determined by distance to rmarket, credit markets, and other community characteristics, the set of relative prices and the characteristics of the markets faced by households will differ by lo- cation. In turn, the capacity of households to produce, take on risk, and par- ticipate in commodity markets will differ from household to householcd. Con- sequently, some households may choose to participate in commodity output markets while others do not. Moreover, participation in some comrnodity markets may be more attractive to households than participation irn other markets. As already discussed, survey data show that, in general, few rural households participate significantly in crop markets. This section examines the determinants of the decision to participate. Formally, we assumed that conditional on the state variables, households formulate a decision price, t1 j, which when compared with market prices, trig- gers their participation in markets; that is, Yh Ž 0 O, for Ph,; 2 th j(H, I, E, V), where y is the share of crop i production marketed by household h; and where H, I, E, and V are state variables representing household, commodity, enterprise, and community characteristics (de Janvry, Fafchamps, and Sadoulet 1991). Because the decision price is unobserved, the following relationship is estimated: (6.3) Yh,i = s(Phj; (Hh, Ij, Eh, Vh). Table 6.7 provides estimates of household participation in Ugandan crop markets based on nearly 11,700 observations from the 1992/93 survey. The subsequent sections discuss the implications of the key results. Prices Relative prices are important when farmers decide how much of their crop to market. On average, across all crops, a 10 percent increase in price will riesult in a 2 percent increase in the amount of the crop sold. The source of the price increase can be a general rise in district-level prices or a decrease in transfer costs. The latter point is significant, given the relatively large margins associ- ated with many of the food crops. Furthermore, the 0.2 participation elasticity is in addition to household supply elasticities, that is, households can produce 4. For example, Ellis (1993, p. 13) provides the following definition: "Peasants are households which derive their livelihoods mainly from agriculture, utilisE mainly family labour in farm production, and are characterised by partial engagement in input and output markets which are often imperfect or incomplete." 191 192 Donald Larson and Klaus Deininger Table 6.7. Market Participation: Household Tobit Results Variable Estimate X2 statistic Probability > X2 Price 0.20 23.27 0.00 Market characteristics Crop effects (joint significance) n.a. 2,255.82 0.00 Region effects (joint significance) n.a. 195.49 0.00 Household characteristics Size of household -0.01 0.14 0.71 Male head of household 0.05 4.37 0.04 Education of household head (joint significance) n.a. 8.20 0.04 None -1.13 2.52 0.11 Primary -1.08 2.31 0.13 Secondary -1.11 2.43 0.12 Sources of capital From savings/family -0.18 1.66 0.20 From government/NGO 0.08 0.12 0.73 Money lender 0.03 0.06 0.81 Formal credit -0.46 3.11 0.08 Enterprise characteristics Number of paid workers 0.01 4.98 0.03 Number of unpaid workers -0.10 37.22 0.00 Investment capital 0.04 26.13 0.00 Community characteristics Distance to market 0.00 2.48 0.12 Distance to phone -0.03 15.91 0.00 Distance to all-weather road 0.00 2.25 0.13 Common method of sale Sold by cooperative -0.23 0.46 0.50 Sold by individual farmer 0.06 0.15 0.70 Sold by large group 0.09 0.28 0.60 n.a. Not applicable. Note: Continuous variables not already expressed as shares were converted to natural logs prior to estimation. Source: Estimates based on 1992/93 integrated household surveys. for either home consumption or for market. Consequently, all other things be- ing equal, a permanent reduction in transfer costs would result in both an in- crease in output and an increase in the share of output marketed. Market Characteristics Crop markets have different information and storage characteristics that influ- ence household marketing decisions in the same way they influence markets Crop Markets and Household Participation 193 between communities. In the regression, crop effects and regional effects were significant when taken as a group. Table 6.8 reports the estimated crop effects individually. Among the crops, farmers were more likely to export a greater share of traditional export crops. Household Characteristics Among the household characteristics included in the regression, the gender of the household head proved significant: male heads of households were more likely to participate in crop markets. The size of the family and differences in Table 6.8. Estimated Fixed Commodity Effects Commodity Estimated effect xI Probability > xI Coffee 1.33 6.13 0.01 Other fruit 1.22 4.70 0.03 Cotton 1.21 5.09 0.02 Tomatoes 0.97 3.14 0.08 Tobacco 0.95 2.94 0.09 Sugar cane 0.90 2.19 0.14 Rice 0.87 2.61 0.11 Onions 0.85 2.41 0.12 Other cash crop 0.84 1.22 0.27 Other grains 0.81 1.92 0.17 Cabbage 0.72 1.66 0.20 Oranges 0.67 1.11 0.29 Pineapples 0.65 1.08 0.30 Soybeans 0.50 0.88 0.35 Maize grain 0.38 0.50 0.48 Bananas 0.37 0.46 0.50 Sim-sim 0.37 0.47 0.49 Peas 0.24 0.19 0.66 Maize cob 0.14 0.07 0.79 Groundnuts 0.14 0.07 0.80 Beans 0.06 0.01 0.91 Sorghum 0.03 0.00 0.95 Millet 0.02 0.00 0.98 Matooke 0.00 0.00 1.00 Irish potatoes -0.01 0.00 0.99 Other vegetables -0.02 0.00 0.97 Other pulses, nuts, seeds -0.10 0.03 0.86 Cassava -0.17 0.10 0.75 Cowpeas -0.24 0.19 0.66 Sweet potatoes -0.31 0.32 0.57 Note: Fixed effect for yams excluded. Source: Authors' fixed-effect parameter estimates. 194 Donald Larson and Klaus Deininger educational attainment were not important in explaining market participa- tion, nor were differences in the sources of credit important. The coefficient on formal credit sources was significant, although not of the expected sign. This may well be coincidental, because less than 2 percent of the sample included households with credit obtained from formal sources. Enterprise Characteristics Past investment reflected in current investment capital levels was significant in explaining differences in market participation rates. The number of paid employees was associated with greater market participation, that is, enter- prises with commercial outputs also participated in formal labor markets. The opposite was also true, and farms that relied on unpaid-usually family- labor were also less likely to market their output. Community Characteristics The effects of community characteristics, which are important to trade among communities, are likely to be reflected in farm-gate prices. They may also play a role in the type of information households receive and the range of opportunities available to households. Among the community characteris- tics included in the regression, only access to telephones proved both quan- titatively and statistically significant. The Effects of Price Changes on Household Welfare Crop models provide a way to quantify the effects of exogenous changes in district prices on community prices. However, the quantitative significance of such changes on welfare is not clear, because the relationship between local prices, district prices, and community market characteristics differ among crops; market characteristics differ among communities; and the composition of pro- duction differs among farms. Fortunately, the 1992/93 household survey con- tains information on the composition of household production and also pro- vides a way of linking households with communities. Linking the crop pricing models with the household composition data provides a way of measuring the welfare effects of changes in policy and public investment. Annex 6.1 ex- plains how the welfare measure was calculated. The model results discussed earlier suggest that changes in district prices are reflected in local prices (table 6.4). In turn, tradable crop prices are linked to international prices. However, in 1992/93 Uganda households were only par- tially linked to formal crop markets, because most households marketed only a small share of their output. Cash crops were the exception. Table 6.9 shows the effects of a simulated 10 percent increase in cash crop prices. The change is significant, but mostly for a concentrated group of households. This is because Crop Markets and Household Participation 195 Table 6.9. Welfare Effect of a 10-Percent Increase in Cash Crop Prices (percent) Range Share of households Average change 0-1 84.8 0.0 1-2 4.6 1.5 2-3 3.1 2.4 3-4 2.4 3.5 4-5 1.3 4.4 5-6 1.2 5.5 6-7 0.8 6.5 7-8 0.7 7.5 8-9 0.4 8.5 9-10 0.3 9.4 10-11 0.2 10.5 11-12 0.2 11.6 12-13 0.2 12.4 Total 100.0 0.6 Source: Authors' simulation. Ugandan households choose production diversification as a survival and risk management strategy. As might be expected, a price increase also has regional effects. Coffee appears to dominate the results, with coffee-growing districts like Kasese showing greater changes under the simulation.5 Because of the composition of production, gains from temporary price movements are also small and unlikely to contribute in a significant way to the accumulation of household assets. In contrast, reductions in transfer costs will improve the household terms of trade in a permanent way. Results from the previous section suggest that, all other things being equal, farmers would consequently increase their market participation. Moreover, changes in trans- fer costs may lead to reduced risk levels as well. As a result, the true devel- opment impact of improvements in crop markets are likely to come from changes in household production and market participation made in response to new opportunities. Are Crop Markets Developing? This section considers subsequent evidence that suggests that crop markets are developing in Uganda. In 1992 crop markets worked in the sense that 5. Note that an export tax-sometimes proposed to stabilize exchange rates dur- ing a boom-would have exactly the opposite results. 196 Donald Larson and Klaus Deininger relevant price signals were relayed among markets. Transfer costs, however, were high and differed significantly among crops. Most households engaged in farming and consumed most of what they produced. Rural households did not specialize in other sector activities nor did they specialize in produc- ing particular crops that would give rise to growing food crop markets. Clearly there was scope for transaction costs to fall, and for crop markets, especially food crop markets, to deepen. Increased Participation in the 1990s Subsequent household surveys show small but steady changes in income levels and sources of income, which together suggest some growth in the domestic demand for food crops (see book appendix A for surveys). Changes in the survey instruments make some comparisons difficult; nevertheless, it appears that between 1992 and 1995 real income levels grew as did income from wages (table 6.10). At the same time, the share of household income from farming grew between 1992/93 and 1994/95. Moreover, the percentage of heads of households reporting agriculture as their primary occupation grew between 1992/93 and 1995/96 (table 6.11). Using the same household data, Okidi (1999) reports some changes in the composition of household production, including a greater role for coffee, but it is difficult to tell if this result is due to changing prices only. A change in the way the survey was conducted further obfuscates any changes in the composition of household production. Nevertheless, there is some evidence from the early surveys to suggest that households became significantly more integrated into crop mar- kets during this period. In 1999/2000, a revised household survey included questions about mar- ket participation, not only for the period coinciding with the survey, but for previous years as well. Partial and preliminary data from the 1999/2000 sur- vey provides strong evidence that farmer participation is increasing. Table 6.12 reports results from a probit regression that predicts participa- tion in crop markets based on the household characteristics of the respondents. The regression suggests that the availability of storage facilities and of market information are important determinants of market participation. The use of high-yielding seed varieties (proxy for the technical sophistication of the farmer) was not significant, but having at least one visit by an extension officer was significant. In addition, separately included time effects suggest that market participation increased significantly between the early surveys of 1992/93 and 1994/95 and the subsequent surveys in 1995/96 and 1999/2000. This general result-that farmers report increasing participation in for- mal markets over time-remains at odds with participation as measured by the share of income from formal markets. Nonetheless, the results suggest at least a perception of increased use of formal markets. Moreover, the positive effect of extension on participation reinforces the earlier finding that infor- mation is an important component of market development. Table 6.10. Household Income by Source, 1992-96 Share of incomefrom Total household income agriculture Share of incomefrom (U Sh 1,000) (percent) all wages (percent) Location 1992/93 1994/95 1995/96 1992/93 1994/95 1992/93 1994/95 1995/96 All Uganda 54.2 69.5 74.6 49.4 51.7 12.0 13.5 15.7 Central 68.9 104.2 103.7 42.7 40.4 18.7 19.8 20.5 Eastern 49.9 51.5 64.2 49.4 54.7 9.7 10.8 15.4 Northem 42.8 46.7 50.1 52.1 61.8 5.6 7.7 9.3 Western 50.3 56.2 67.4 55.4 57.3 11.4 11.4 14.6 Source: Okidi (1999). Table 6.11. Main Occupation of Household Head, 1992 and 1995 (percent) Self-employed Employed Agriculture Nonagriculture Agriculture Nonagriculture Location 1992 1995 1992 1995 1992 1995 1992 1995 All Uganda 57.3 63.2 20.1 16.0 3.2 3.1 19.5 17.8 Central 50.8 57.4 17.6 18.7 5.0 5.0 26.6 18.9 Eastern 60.9 63.9 15.2 16.3 1.6 2.1 22.3 17.8 Northern 59.9 69.1 33.7 11.6 1.1 1.2 5.3 18.1 Western 58.7 64.3 16.3 15.9 4.4 3.4 20.7 16.4 Source: Okidi (1999). Crop Markets and Household Participation 199 Table 6.12. Explaining Positive Responses to "Did You Sell Your Output?" Variable Estimate z Probability > z Storage facility 0.47 3.75 0.00 HYV, 1-20% -0.31 -1.15 0.25 HIYV > 20% -0.15 -0.81 0.42 Number of extension visits One 0.62 2.88 0.00 Two to three 0.54 2.50 0.01 More than three 0.51 2.21 0.03 Received market information Somewhat useful 0.64 5.13 0.00 Useful 0.37 2.28 0.02 Year effect 1994 0.10 0.79 0.43 1996 0.25 2.05 0.04 1999 0.26 2.17 0.03 HYV High-yielding varieties. Source: Estimates based on preliminary data from 1999/2000 national household survey. Crop Market Resultsfrom Subsequent Surveys The community surveys of 1992/93, 1993/94, and 1995/96 asked different questions about crop prices. Only the 1992/93 survey asked about district prices. The 1993/94 survey inquired about local prices and prices in the most common market, and the 1995/96 questionnaire asked about farm-gate prices and most common market prices. Moreover, information on infrastructure, credit access, and marketing practices was more limited in the 1993/94 and 1995/96 surveys. Consequently, putting together a panel as detailed as the model applied to the 1992/93 data is not possible. It is feasible to use the 1993/94 and 1995/96 data together and separately in a limited way. Using the 1993/94 survey, observed nearest market prices were regressed on most common market, distance to market, infrastructure measures, and fixed effects for crop, unit, and region. Distance to market was assumed to equal the distance to the most important produce market minus the distance to the local market. When the distance was zero, the observation was dropped on the assumption that the local market was also the most im- portant one. Because of the way the question was asked, one cannot unam- biguously associate the distance variable with a market pair. Similarly, the 1995/96 survey can be used by regressing farm-gate prices on producer prices in the most common market, transport infrastructure, and fixed effects. The distance between farm and community must, unfortunately, be ignored. 200 Donald Larson and Klaus Deininger Despite ambiguities about distance, the regression results are quite similar to the results from the better-designed 1992/93 survey (table 6.13) and broadly support earlier findings. Prices in related markets largely explain that local prices and quality effects are important. Crop and regional effects turn up sig- nificant for the 1995/96 survey, but not for the 1993/94 survey. Transportation infrastructure effects are neither large nor significant. Unfortunately, commu- nication or other types of information measures were not available. Evidence on Transfer Costs As discussed earlier, even when markets are integrated and arbitrage condi- tions hold, high transfer margins encourage self-sufficiency and limit the benefits of trade. Nevertheless, markets are an amorphous blend of informa- tion, business practices, institutions, credit, and transport costs. Thus an ad- ditional question of interest then is how marketing margins have changed over time given the many changes to the Ugandan economy. One cannot make direct comparisons of margins over time because of changes to the surveys. A related question, however, can be addressed: Have average spreads between observed market prices and district averages de- clined over time? This is done by regressing local most common crop prices, available in both the 1993/94 and 1995/96 surveys, against transportation variables and crop, unit, and regional dummies together with a year dummy. Table 6.13. Results from the 1993 and 1995 Surveys 1993/94 survey 1995/96 survey Variable Estimate t-score Estimate t-score Price 0.952 37.840 0.921 59.610 Distance to market 0.000 -0.170 -0.004 -1.060 Distance to tarred road 0.004 0.690 0.003 1.020 Distance to public transport -0.007 -1.510 0.002 0.490 Distance to rail stop -0.032 -1.010 -0.004 -0.870 Testsforfixed effects v2 statistic Probability > XI X2 statistic Probability > XI Crop 23.300 0.669 99.860 0.0001 Unit of measure 36.060 0.003 41.800 0.0001 Region 22.220 0.507 88.190 0.0001 Source: Based on 1992/93 and 1995/96 household surveys. Crop Markets and Household Participation 201 Together, the crop, unit, and regional dummies take on the value of the regional average price for a particular crop in a particular unit of measure. The year dummy takes on a value of one for 1993/94 observations and is otherwise zero. The results are given in table 6.14 and answer yes to this question, that is, average spreads have fallen with all other things being equal. In fact, the regression results imply that the spread between the two surveys fell by ap- proximately 32 percent, suggesting that opportunity for trade increased for most communities. Evidence on Changing Domestic Demand In rural communities where most households are largely self-sufficient in food, there is scant and inconsistent demand for local food crops. When trans- portation costs are high and the information supply is low, households can- not depend on markets for food crops to generate income. In contrast, an export market always exists for coffee, although the price may be uncertain. As already noted, improving communications and access to information can lower uncertainty and transaction costs. Changes in the composition of the economy that lead to a consistent demand for local food crops can do the same. Evidence from national accounts suggests this has been happening to some extent in recent years. Nevertheless, the number of households engaged in agriculture has not fallen, and rural incomes remain highly dependent on agriculture. As discussed in chapter 2, a significant portion of growth has come from the formal nonagricultural sectors of the economy. To the extent Table 6.14. Modeling District Spreads Variable Estimate Probability> It I Distance to market 0.00005 0.993 Distance to tarred road -0.01178 0.024 Distance to public transport 0.00739 0.107 Distance to rail stop -0.00603 0.511 Testsforfixed effects X2 statistic Probability > X2 Crop 952.81 0.0001 Unit of measure 6,166.28 0.0001 Region 167.15 0.0001 Year 88.49 0.0001 Test that year effects are the same Estimate Probability> It I 1995 effect + 1993 effect -0.38688 0.0001 Source: Authors' calculations. 202 Donald Larson and Klaus Deininger that the changing composition of the economy continues to reflect changing specialization by households, consistent demand for local food crops will create incentives for greater participation by farming households in formal markets for food crops. Conclusions and Policy Implications Information from community and household surveys conducted in 1992/ 93, 1993/94, and 1995/96 and partial survey results from 1999/2000, to- gether with modeling work, suggest that household participation in crop markets is limited. The consequences are less severe for export crops that eventually make their way into intemational markets. For food crops, how- ever, the lack of domestic participation most likely results in illiquid mar- kets. Gross margins between local and district markets for most food crops are large, both absolutely and relative to cash crops. This fact alone would encourage self-sufficiency in food crops and trade in cash crops. Large mar- gins and high transfer costs also imply a large range of positive prices that fail to cover transfer costs, increasing the risks of going to market when prices are uncertain. Units of measure, which probably indicate quality dif- ferences, are important determinants of local prices relative to district prices, suggesting that demand markets are further fragmented by uncertainty about quality. Together these factors indicate that marketing food crops is riskier than marketing export crops, encouraging households to remain both self-sufficient and diversified in their production. The mechanism is self- enforcing, because a lack of specialization also limits the demand for local food crops and the depth of local markets. Despite high margins, there are no indications of widespread market fail- ure. Results from a spatial arbitrage model suggest that district and local markets are integrated to the extent that local prices largely reflect district prices once transfer costs and commodity quality characteristics are taken into account. This sign of market integration is significant and fundamental, because the most basic economic measures of income, poverty, and welfare are premised on the notion that prices are comparable across time and space. Although differences among the surveys make comparisons difficult, evidence indicates that crop markets are improving and that transfer mar- gins fell between 1992 and 1995. Nevertheless, in 1995 most rural house- holds were still engaged in self-reliant farming, and their participation in crop markets remained limited. Working markets were not sufficient to coax many poor households out of self-sufficiency. In addition, while in- vestments in infrastructure that are statistically significant can explain average crop price differences among communities, the quantitative im- pact of public investment in infrastructure-in the short run-appears to be small. Access to telecommunications and credit seems to be significant for crop markets, but again, the estimated short-run benefits are small. Crop Markets and Household Participation 203 Nonetheless, the benefits of public investments in infrastructure and ac- cess to credit probably extend beyond crop markets, and policies toward investments in infrastructure must include a balanced consideration of the effects of public investments in infrastructure. Taken together, the survey and modeling results suggest that the prob- lems in crop markets up to 1995 were symptomatic of limited specialization in the Ugandan economy and labor markets. More recent sectoral income data suggest that this is changing and the economy is becoming more di- verse. Consequently, policies that enable households to accumulate savings and human and physical capital and, thereby, take on the additional risks of specialization, are likely to result in improved crop market performance. In tandem, government support for the public and private institutions that strengthen crop markets by reducing price, quality, and transaction uncer- tainty can speed the development of crop markets. Annex 6.1. Calculating Household Welfare From the 1992/93 integrated household survey, crop production shares were calculated for each of the nearly 6,000 households in the sample and crop weights were calculated, so that: (A6.1) w,l = v,11 v Vh, where h denotes households, c denotes crops, and v represents a value of production calculated from the household survey. Price changes, dpc, were simulated under different scenarios. The two series were then combined to provide a Laspayres measure of welfare change: (A6.2) ALh = _ dpcw h. Because households would reevaluate their optimization strategy in light of a relative price change, the Laspayres measure gives a lower-bound estimate. References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Barrett, C. B. 1997. "Food Marketing Liberalization and Trader Entry: Evi- dence from Madagascar." World Development 25(5): 763-77. Baulch, Bob, W. M. H. Jaim, and Sajjad Zohir. 1998. "The Spatial Integration and Pricing Efficiency of the Private Sector Grain Trade in Bangladesh." Briefing paper. Institute of Development Studies, Brighton, United Kingdom. 204 Donald Larson and Klaus Deininger Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore, Maryland: The Johns Hopkins University Press. de Janvry, Marcel Fafchamps, and Elisabeth Sadoulet. 1991. "Peasant House- hold Behaviour with Missing Markets: Some Paradoxes Explained." Economic Journal 101(409):1400-17. Ellis, Frank. 1993. Peasant Economics: Farm Households and Agrarian Develop- ment. Cambridge, U.K.: Cambridge University Press. Fafchamps, Marcel, and Bart Minten. 1999. "Property Rights in a Flea Market Economy." Working Paper no. WPS/99.25. University of Oxford, Cen- tre for the Study of African Economies, United Kingdom. Jones, William. 1972. Marketing Staple Food Crops in Tropical Africa. Ithaca, New York: Comell University Press. Kranton, Rachel. 1996. "Reciprocal Exchange: A Self-sustaining System." American Economic Review 86(4): 830-51. Larson, Donald F. 1994. "Copper and the Negative Price of Storage." Policy Research Working Paper no. 1282. World Bank, Development Research Group, Washington, D.C. . 2000. "Measuring Market Development: Crop Markets in Uganda." World Bank, Development Research Group, Washington, D.C. Pro- cessed. Martin, Anne. 1962. The Marketing of Minor Crops in Uganda: A Factual Study. London: Her Majesty's Stationery Office. Mundlak, Yair. 1978. "On the Pooling of Time Series and Cross-section Data." Econometrica 46(1): 69-85. North, Douglas. 1989. "Institutions and Economic Growth: An Historical In- troduction." World Development 17(9): 1319-32. Okidi, John A. 1999. "Regional Growth Disparities and Household Economic Performance in Uganda." Research Series no. 17. Economic Policy Research Centre, Kampala. Processed. SAS Institute, Inc. 1992. "SAS Technical Report P-229, SAS/STAT Software: Changes and Enhancement, Release 6.07." Cary, North Carolina. Williams, Jeffrey C., and Brian Wright. 1991. Storage and Commodity Markets. Cambridge, U.K.: Cambridge University Press. Part III Firm Responses and Constraints u 7 Confronting Competition: Investment, Profit, and Risk Ritva Reinikka and Jakob Svensson This chapter examines the investment performance of the private firm sector. The ensuing analysis is based mostly on the Uganda enterprise survey carried out in 1998 (for survey details, see appendix B at the end of the book). Household enterprises, the focus of chapters 5 and 6, are also part of the private sector, but they are typically very small (micro) enterprises. This chapter focuses on larger firms, that is, firms with five or more employees. As Collier and Reinikka argued in chapter 2, given the capital flight and the depletion of capital stock during the 1970s and 1980s, sustained growth and reduction in poverty beyond recovery re- quire a strong private investment response. Collier and Reinikka's mac- roeconomic data show that investment as a share of gross domestic prod- uct increased in the 1990s, but is still well below, say, that of East Asian countries. This chapter analyzes the microeconomic or firm-level evidence of private investment. How important is investment generally for economic growth? Invest- ment or physical capital accumulation has long played a central role in the The findings reported in this chapter are based on data from the 1998 Uganda enterprise survey, which was carried out by the Uganda Manufacturers Association Consultancy and Information Service on behalf of the Ugandan Private Sector Founda- tion and the World Bank, and was managed by William Kalema and Frances Nzonsi. The survey design benefited from the Regional Program on Enterprise Development and contributions from Andrew Stone. Alex Bilson-Darku, and Mimi Klutstein-Meyer assisted in data analysis. Useful comments were received from participants at the an- nual seminar on the Ugandan economy, organized by the Economic Policy Research Centre (Kampala) in May 1999, as well as from Catherine Pattillo and Francis Teal. 207 208 Ritva Reinikka and Jakob Svensson literature on economic growth and development. Few economic ideas are as intuitive as the notion that increasing investment is a good way to raise output and income. Recent empirical research also supports this view-the rate of investment is robustly and positively correlated with the rate of eco- nomic growth in cross-country, long-run growth regressions. Early research on growth and investment took a rather mechanical ap- proach: growth was constrained by a lack of investment that, in turn, was constrained by a lack of finance (see Easterly 1997). Consequently, if financ- ing was made available, it was argued, physical capital investment and, ulti- mately, growth would follow.1 This chapter contends that both investment and growth, as well as inno- vation and technical change, are driven by the prevailing policies and eco- nomic, social, and legal institutions. While some of these polices, particu- larly macroeconomic policies, can be measured directly, the effect and efficiency of other policy areas are much more difficult to assess. By studying the determinants of private investment, it is possible to study a larger set of institutional and policy issues that affect firms. The basic idea in the initial wave of the so-called endogenous growth theory is that growth differences could be sustained indefinitely because the return to capital would not diminish as economies develop (Lucas 1988; Rebelo 1991; Romer 1986). Unlike the growth theory of the 1960s, recent re- search reflects closer attention to the relationship between theory and data. Indeed, a large empirical literature developed in the 1990s in which virtually every possible variable has been used to explain this divergence in growth over time within the cross-country framework (Barro 1991; see Barro and Sala-i-Martin 1995 for a review). Most of this work explains cross-country differences in growth, but a few studies also attempt to explain Africa's poor performance (Easterly and Levine 1997; Sachs and Warner 1995, 1996; see Collier and Gunning 1999 for a review). While the explanatory power of many of the proposed variables has been shown to depend on specification, sample, or measurement, some variables appear to be robustly correlated with growth (see Levine and Renelt 1992 for a critical review). These variables include investment rate (DeLong and Summers 1991; Mankiw, Romer, and Weil 1992), level of initial income, human capital stock, openness to trade, financial depth, and fiscal stance. The African growth "tragedy" has been explained by addi- tional factors, including high volatility (high incidence of shocks originating from external terms of trade, climate, or policy), deficient public infrastruc- ture, and ethnic fragmentation. 1. Recent research based on data for a cross-section of countries during 1970-97 shows that public investment has not been correlated with growth in Africa (Devarajan, Easterly, and Pack 1999). Similarly, private investment has not been correlated with growth, unless Botswana is included in the sample. This result is not surprising given the poor policy and institutional environment in most of these countries during most of the sample period. Confronting Competition: Investment, Profit, and Risk 209 This chapter has two objectives. First, using new microeconomic data from Uganda, it examines the extent to which liberalization and the profound macroeconomic and structural reforms implemented in the late 1980s and the 1990s translate into higher private investment. Second, while at present households are important economic agents in agriculture and other sectors, sustainability of economic growth depends on the growth of firms, because households seldom achieve significant economies of scale necessary for sus- tained growth. Using quantitative and qualitative survey data, the chapter analyzes factors that constrain investment and the growth of Ugandan firms. Investment Response Firm surveys have proven a useful tool to explore private sector responses to macroeconomic reforms and to increase our understanding of microeconomic constraints to investment. Such surveys can also help policymakers prioritize policies and interventions to improve the business environment. In Africa, the Regional Program on Enterprise Development, initiated by the World Bank, has over time produced valuable quantitative data on manufacturing firms for Burundi, Cameroon, C6te d'Ivoire, Ghana, Kenya, Tanzania, Zambia, and Zimbabwe (Biggs and Srivastava 1996). The 1998 Uganda enterprise survey benefited from the Regional Program on Enterprise Development model and, hence, is comparable to the other Af- rican surveys. However, the Uganda survey is somewhat more limited in its scope (it excludes detailed labor and finance questions), but covers a wider range of sectors-in addition to manufacturing it includes firms in commercial agriculture, construction, and tourism (for details, see appen- dix B at the end of the book). In addition, it includes a wider range of ques- tions on infrastructure, taxation, and corruption. Investment Data Before analyzing the regression results, it is useful to examine the Ugandan investment data and compare them with similar data for four other African countries: Cameroon, Ghana, Kenya, and Zimbabwe. The survey provides quantitative data on employment, capital stock, investment, sales, and value added for 192 Ugandan firms during 1995-97. Because changes are used in some of the variables, one year of observations (1995) is lost. Thus, data per- mitting, each firm has two observations, making the total number of obser- vations 367. Initial inspection of the data resulted in discarding 14 of these observations as outliers, leaving a sample size of 353.2 2. Observations with reported value added to capital above 1,000 percent or below -100 percent are dropped. A closer inspection of the data revealed thatmisreported or erroneous recording of capital stock data was the source of these extreme values. 210 Ritva Reinikka and Jakob Svensson As shown in table A7.1, about half of the Ugandan firms made an invest- ment in machinery and equipment in both 1996 and 1997. This is similar to the African country average. For individual countries where comparable in- formation exists, the percentage of Ugandan firms that invested is somewhat higher than in Cameroon, Ghana, and Kenya, but lower than in Zimbabwe (Bigsten and others 1999). While large firms are more likely to invest (77 per- cent of large and 45 percent of small firms in Uganda), they invest less rela- tive to their capital stock than smaller firms. For the Ugandan firms that in- vested, the value of investment relative to the capital stock (investment rate) was, on average, 11 percent for large firms and 30 percent for small firms. For all Ugandan firms, both those that did and did not invest, the investment rate was 13 percent in 1996 and 11 percent for 1997. Again, this pattern is quite similar to the African comparator country average. With respect to in- dividual comparator countries, the investment rate for the firms that invested in Uganda is lower than that in Cameroon and Ghana, about the same as in Kenya, and higher than in Zimbabwe. Averages, however, can be misleading when the underlying distribution is skewed. At the median firm, the Ugandan investment rate is very low: less than 1 percent for all firms and 4.7 percent for those firms that invest. The picture is similar in the four comparator countries, that is, median invest- ment rates for all firms range from zero in Cameroon and Kenya, to less than 1 percent in Ghana, to 3 percent in Zimbabwe.3 As shown in table 7.1, there are obvious differences between firms that invest and those that do not invest. Investing firms, on average, have higher profits, tend to experience positive changes in demand and value added, are larger in terms of value added and employment, and are somewhat more recently established. Uganda and Ghana are the only countries that experi- ence a positive change in value added (and gross sales for Uganda) at the median, reflecting a growing economy and relatively good economic poli- cies. For Ugandan firms that invest, the sales-to-capital stock ratio increased by 42 percent, on average (9 percent at the median), while for firms that did not invest, the change in sales was negative (O at the median). Another notable characteristic of African firms is the very high mean and median profit rates, that is, profit as a share of the installed capital stock is high. These gross profits are calculated as the firm's value added less wages and interest payments. Compared with the rest of the world, the high profit-to-capital ratios are likely to be driven by the low level of installed machinery and equipment. For the four comparator countries, Bigsten and 3. The firm survey data seem to be generally consistent with the trend depicted by Uganda's macroeconomic data. As shown in chapter 2, private investment was relatively stable during the survey period of 1995-97, while the overall share of in- vestment in machinery and equipment in gross domestic product fell somewhat after the 1994-95 coffee boom. Confronting Competition: Investment, Profit, and Risk 211 Table 7.1. Summary Statistics for Ugandan Firms, Pooled Data, 1996-97 Firms that Firms that do Variable invest not invest Allfirms Profit rate 0.914 0.565 0.747 [0.306] [0.177] [0.256] Change in sales 0.418 -0.023 0.207 [0.090] [0.001] [0.028] Change in value added 0.214 0.012 0.117 [0.027] [-0.001] [0.007] Value added 1.39 0.890 1.149 [0.501] [0.330] [0.414] Size (employment) 150 51 103 [50] [19] [28] Age (years) 12 14 13 [9] [11] [10] Investment rate 0.234 n.a. 0.122 [0.047] n.a. [0.002] n.a. Not applicable. Note: Mean values, with median values in brackets. There were 184 observations with positive investment and 169 with zero investment. Variables are expressed as a ratio of lagged capital stock, except for size and age. Source: Authors' calculations based on the 1998 enterprise survey. others (1999) report an average profit rate of 198 percent and a median of 40 percent for all firms. While the Ugandan investment rates do not differ much from the African average, average profit rates are clearly lower. They are also lower than in any individual comparator country. Indeed, profit rates in Uganda, both at the median and the mean, are only about half of those re- ported for the pooled African sample: for those Ugandan firms that invested, the mean profit rate was 91 percent (31 percent at the median), while for all firms the mean was 75 percent (26 percent at the median). Flexible Accelerator Model of Investment To what extent is investment across Ugandan firms driven by changes in de- mand as suggested by the flexible accelerator model of investment? Are firms in general constrained by liquidity? Do age and size matter? Are there any clear geographical or sectoral differences in investment behavior? To answer these questions a simple flexible accelerator model is estimated (see annex 7.2). In this model, fluctuations in demand are assumed to motivate investment. Given the weaknesses of the financial sector in African economies, a model is adopted in which firms do not have access to credit and simply allocate cur- rent profits to investment (for details see Tybout 1983). A similar approach has 212 Ritva Reinikka and Jakob Svensson been applied to four other African countries, namely, Cameroon, Ghana, Kenya, and Zimbabwe (Bigsten and others 1999). By replicating their specification, this section explores whether Uganda, with its better macroeconomic record, differs from the other countries in terms of firms' investment response. As in the case of the comparator countries, data on investment in machinery and equipment are used. The flexible accelerator model of investment for a profit maximizing firm i, which is liquidity constrained, can be written as follows (see annex 7.2 for details): (7.1) I.(t) = ax. + aQAQ,(t) + ct, (t) + ca,I(t -1) + ax X+ dt+ Ps, where Ii (t) is the level of investment for firm i at time t, coi is the constant for firm i, AQ, denotes the change in sales, Tit is the level of profits, X. denotes firm-specific characteristics (age, size), d, is a time dummy, and 8i is the error term. To avoid the heteroskedasticity problem with respect to size in the esti- mation, the variables are expressed in rates, that is, scaled by the inverse of capital stock at the end of the previous period, K(t - 1). The empirical model set out in equation (7.1) treats investment as a con- tinuous variable. However, capital investment is typically lumpy, which constrains the firm's investment behavior. In a given year the firm may not be able to invest the desired amount and, therefore, chooses not to invest at all. In other words, the observable data on firms' investment rates are inci- dentally truncated and, thus, equation (7.1) is estimated in two stages.4 The two-stage procedure involves, first, the estimation of a probit model of the decision to invest and, second, an estimation of the investment rate equa- tion for the firms that invested, accounting for the selection of firms with only positive investment. Regression Results This section explores how well the flexible accelerator model, as expressed in equation (7.1), can explain Ugandan firms' decisions to invest and the amount. Table 7.2 reports the basic results, including the two-stage estimation and the Tobit regression. Apart from the variables defined above, each regression in- cludes industrial category and location-specific dummies. Column 1 shows the result of the first-stage probit model concerning the decision to invest. At the 90 percent confidence level, both the accelerator (change in sales) and the liquidity constraint (profit) are found to be important in the decision to invest. Thus, according to the prediction of the accelerator model, Ugandan firms in- deed invest to meet increases in demand, provided that they have sufficient 4. Heckman's (1979) two-step procedure. If the factors that determine the deci- sion to invest and the amount of investment are the same, the correct specification is the Tobit model. Confronting Competition: Investment, Profit, and Risk 213 Table 7.2. Investment Regressions for All Ugandan Firms, 1995-97 (2) Ordinary (1) Probit least squares (3) Tobit Variable regression regression regression Constant -1.15a 0.992 _0.430b (0.470) (0.525) (0.232) Change in sales-to-capital stock 0.164b -0.055 0.032 (0.073) (0.042) (0.028) Profit rate 0.090C 0.076b 0.100a (0.054) (0.035) (0.024) Age (log) -0.250a -0.028 -0.147a (0.092) (0.054) (0.045) Size (log) 0.372a -0.120 0.087' (0.064) (0.075) (0.030) Time dummy 0.060 -0.082 -0.005 (0.144) (0.084) (0.072) District dummies significant No No No Industrial category dummies significant Yes No Yes Agroprocessing 0.844a n.a. 0.258c (0.288) n.a. (0.137) Tourism 0.644b n.a. 0.281c (0.320) n.a. (0.158) Predictability 0.70 n.a. n.a. R 2 n.a. 0.15 n.a. Observations 353 184 353 n.a. Not applicable. Note: The dependent variable in regression (1) takes the value one if the firm invested and zero otherwise. Standard errors (in parenthesis) adjusted for heteroskedasticity (White 1980). Regressions (2) and (3) were adjusted for selectivity. (The inverse Mills ratio is not reported.) a. Significant at the 1 percent level. b. Significant at the 5 percent level. c. Significant at the 10 percent level. Source: Authors' calculations based on the 1998 enterprise survey. funds-that is, adequate profits-to do so. If they do not have adequate prof- its, they cannot invest, even if the demand for their product is increasing.5 Age and size also enter significantly into the decision to invest. Bigsten and others (1999) argue that size may proxy the likelihood that indivisibilities in investment constrain capital accumulation (the constraint is less likely to bind for large firms), and that older firms are likely to have better access to bank finance. The Ugandan data support the first of these assumptions (size is positively correlated with the probability to invest), but rejects the second 5. The results are very similar when using the lagged profit-to-capital ratio in- stead of the profit-to-lagged-capital ratio. 214 Ritva Reinikka and Jakob Svensson (age enters significantly, but with a negative sign). A possible explanation for the latter result is that older firms in the sample were first established in an environment with a very different incentive system. While many establish- ments in the 1996 census update began operating during the first half of the 1990s (37 percent), many of the older firms were endowed with a capital stock that, because of drastic changes in the policy environment, is no longer viable (for example, equipment to produce an import-substituting good). These firms are therefore less willing to invest. Two industrial category dum- mies are also significant. Holding changes in demand and profit constant, firms in agroprocessing and tourism are more likely to invest. Column 2 in table 7.2 reports the second-stage regression, which exam- ines the amount of investment for those firms that invested in machinery and equipment.6 Now only profit enters significantly. Thus, while demand changes play a role in determining whether or not to invest, profit is the only binding constraint for the level of investment. The results suggest that most (but not all) firms can generate funds for some investment if demand is increasing, but they cannot realize their desired investment level if cur- rent profits are not sufficient. Interestingly, neither age nor size nor any of the sector and location dummies enter significantly. Indivisibilities and sec- tor-specific factors are important for the decision to invest, but they do not influence the actual investment level. This interpretation is supported by the Tobit regression reported in column 3. The profit rate is highly signifi- cant, but the accelerator is insignificant at the conventionally accepted sig- nificance levels.78 6. The flexible accelerator model was also applied to investment data on build- ings and land, for which valuation is much more difficult. In the probit model, only size and some district and industrial category dummies are significant (at the 10 per- cent level) for the decision to invest, while none of the variables are significant in the second-stage regression. 7. When using a dynamic specification of the model, that is, including a lagged dependent variable, all qualitative results continue to hold. The main difference is that the size of the coefficient on the profit term is reduced from 0.100 to 0.059 in the Tobit model. Lagged investment is insignificant in all three specifications: decision to invest, investment level regression, and Tobit model. Given the lack of significance, and because around a dozen observations are lost by including the lagged dependent variable, the restricted model (reported in table 7.2) appears preferable. 8. Another potential objection to the reported results could be that they may be driven by unobservable firm-specific factors. To test this, a second-stage regression with fixed effects is run, using deviations from means. The results imply a lower, but highly significant, coefficient on the profit term (0.034 with a t-value of 4.80). How- ever, a test of the hypothesis that the fixed effects were all equal across firms indi- cated that the fixed effect specification was not efficient. In other words, the fixed effects are picking up important cross-firm differences in profits and demand, reduc- ing the explanatory power of these variables in the regression. Confronting Competition: Investment, Profit, and Risk 215 In table A7.2 the sample is partitioned into small firms (100 employees or less) and large firms (more than 100 employees). The results reveal some interesting trends. First, for the decision to invest (columns 1 and 3), for small firms only the profit term is significantly positive, while for large firms the important explanatory variable is changes in demand. Second, the second- stage regressions (columns 2 and 4) show a similar pattern for small firms, while neither profit nor the accelerator is significant for large firms. Third, as before, only the age of the firm appears significant and negative, for the large firms.9"l0 The results suggest that firms, particularly small ones, are liquidity constrained in the sense that they cannot invest (or can invest only small amounts) when demand is increasing if they do not have sufficient funds available. However, given the reported high profit-to-capital ratio in Uganda (and in the four comparator countries), it is hard to argue that the liquidity constraint is binding in most cases. Comparing the results from the Ugandan firm survey with the evidence from other African countries is interesting." Regarding the decision to in- vest and using the same model specification, the Ugandan coefficient for profit is found to be somewhat larger. In the level of investment, the esti- mated coefficient for profit in Uganda is also larger (0.076 versus 0.03 else- where). This holds for all firms and when the firms are divided into two groups according to size. While Bigsten and others (1999) find no robust correlation between the accelerator and investment in the other African countries, this study finds some evidence that demand plays a role in in- vestment for large Ugandan firms. Age and size of the firm behave simi- larly in Uganda as elsewhere. Compared with the rest of the world, the estimated coefficient on profit (and accelerator) is small in Uganda, even though it is larger than in the African comparator countries (see, for ex- ample, Athey and Laumas 1994; Bigsten and others 1999; Bond and others 1997; Tybout 1983). 9. The lack of clear results for large firms in the second-stage regression may be driven by the small sample size. By estimating a Tobit regression, degrees of freedom can be saved. 10. While in both the 1994 and 1998 firm surveys interest rates were ranked as one of the leading constraints by firms of all sizes, firms' perceptions varied con- siderably regarding access to finance. As in the quantitative analysis, the percep- tions of larger enterprises seem to be different from those of smaller ones. For large enterprises that had not borrowed money recently, the leading reason after "high interest rates" was "no need to borrow." Nor did collateral requirements prevent large firms from borrowing; the smaller the firm, the more collateral proved a prob- lem. Liquidity constraints may be binding for start-ups, however. Firms reported that about 70 percent of their private investment was financed by profits and per- sonal savings. 11. As Bigsten and others (1999) do not report marginal effects, the results are compared at each stage. 216 Ritva Reinikka and Jakob Svensson Constraints to Investment So far this chapter has examined the determinants of private investment by different types of firms in the single country context. In general, the Ugan- dan results are strikingly similar to those obtained from several other Afri- can countries. This section takes the viewpoint of a typical, or average, Ugan- dan firm and examines differences across countries. In particular, it attempts to explain the observation that firms' profit rates are lower in Uganda, while investment rates are similar. Profit Rates Table 7.3 reports a series of regressions of profit rates on size and foreign ownership, using data from both the Ugandan firm survey and the four other surveys described in Bigsten and others (1999). Colurmn 1 illustrates Table 7.3. Profit Rate Regressions, Pooled Data for Cameroon, Ghana, Kenya, Zimbabwe, and Uganda (1) Profit (2) Profit (3) Profit (4) Profit (5) Profit Variable rate rate rate rate rate Constant 3.46a 3.99a 3.41a 2.02a 1.81a (0.444) (0.510) (0.631) (0.172) (0.221) Foreign 0.933c 0.801c 0.856c -0.014 -0.007 (0.493) (0.480) (0.481) (0.105) (0.105) Size (log) -0.631a -0.623a -0.523a -0.267a -0.238a (0.128) (0.109) (0.104) (0.035) (0.037) Uganda n.a. -1.23a -1.03a _0.559a -0.447a n.a. (0.194) (0.373) (0.090) (0.152) Cameroon n.a. n.a. -0.557 n.a. -0.005 n.a. n.a. (0.476) n.a. (0.211) Zimbabwe n.a. n.a. -0.345 n.a. -0.018 n.a. n.a. (0.363) n.a. (0.151) Ghana n.a. n.a. 1.51a n.a. 0.452b n.a. n.a. (0.691) n.a. (0.212) R2 0.05 0.07 0.09 0.09 0.10 Observations 1,287 1,287 1,287 1,058 1,058 n.a. Not applicable. Note: The dependent variable is the profit rate (profit-to-capital ratio); foreign is a binary variable taking the value one if the firm is foreign owned, zero otherwise. Standard errors (in parenthesis) adjusted for heteroskedasticity (White 1980). Regressions (4) and (5) exdude outliers. a. Significant at the 1 percent level. b. Significant at the 5 percent level. c. Significant at the 10 percent level. Source: Authors' calculations based on the 1998 enterprise survey; Bigsten and others (1999). Confronting Competition: Investment, Profit, and Risk 217 the result when pooling all variables (altogether, 1,287 observations). As evident from the table, the size of the firm (logarithm of total employment) is significantly negatively correlated with the profit rate (profit-to-capital ratio). Foreign ownership is positively related to profit (although the dummy variable enters only marginally significant at the 10 percent level). In col- umn 2 a dummy for Ugandan firms is added. The dummy enters with a large, negative coefficient and is highly significant. Thus, controlling for size and ownership, Ugandan firms, on average, have significantly lower profits than firms in the four comparator countries. Significant differences are apparent across the four comparator countries. When adding (individually) country controls for the four comparators (col- umn 1), the country dummies for Cameroon, Kenya, and Zimbabwe differ insignificantly from zero, while the Ghana dummy is significantly positive. As reported in column 3, the result is similar if all country controls are in- cluded (one has to be dropped to estimate the regression). The Uganda dummy is significantly negative, while the Cameroon and Zimbabwe (and Kenya if we replace Zimbabwe with Kenya) controls are insignificant, and Ghana is significantly positive. There are at least two possible objections to the pooled results in col- umns 1-3. First, while the Uganda sample includes both manufacturing firms and firms in commercial agriculture, tourism, and construction, the sample of firms in the comparator countries only includes manufacturing firms, including agroprocessing firms. To control for this possibility, all Ugandan firms in commercial agriculture, tourism, and construction are excluded. Second, in the Uganda sample a few firms with extreme value added were excluded, while the sample of firms of the comparator coun- tries include a few firms with extreme profit rates (and value added) of more than 1,000 (up to almost 8,000) percent. While these observations may not necessarily be misreported, it would be of concern if the results were driven by them. To examine this possibility, all observations with profit rates larger than 1,000 percent and lower than -100 percent were dropped. The new results are depicted in columns 4 and 5. As evident, the results are very similar qualitatively to those reported earlier. The Uganda dummy remains negative and highly significant, but with a smaller coefficient (in absolute terms). On average, controlling for size and ownership, the Ugandan firms' profit rate is 56 percentage points lower than in other African countries. Again, some differences exist across the four comparators. Repeating the procedure described, the country dummies for Cameroon, Kenya, and Zimbabwe again differ insignificantly from zero, while the Ghana dummy is significantly positive. As shown in column 5, including all country controls simultaneously yields a similar result. The Uganda dummy is significantly negative, while the Cameroon and Kenya (and Zim- babwe if we replace Kenya with Zimbabwe) controls are insignificant and Ghana is significantly positive. Finally, the coefficient on size is now only 218 Ritva Reinikka and Jakob Svensson one-third of that reported in column 1, suggesting that a few extreme obser- vations significantly affect the absolute value of the coefficient.12 Conceptual Framework How can Ugandan investment rates be similar to those in other African coun- tries when Uganda's profit rates are lower? This section constructs a simple conceptual framework suggesting one possible answer. Consider a two-period model of a representative firm. A risk-neutral manager decides on the firm's level of investment in period one to maximize the present value of its cash flow cl + Pc2, where f8 = 1 / (1 + 0) is the discount factor. One can think of 0 as capturing expectations about the future. The assumption is that the firm can borrow in period one. The interest on the borrowed amount b is r. To avoid extreme solutions, we assume that r> 0, implying that the firm will only borrow to finance investment. The budget constraint in period one is then: (7.2) cl + i < 7t, + b, where nit is the initial profit available to the firm and i is the level of invest- ment. The return to investment (or gross profit) is captured by the concave and strictly positive revenue function nt2(i:x), where x is a vector of variables that affect the profit but which the firm cannot control (degree of competi- tion, quality of infrastructure, and so on). The budget constraint in period two can be expressed as follows: (7.3) C2= iT2(i:X) - (1 + r)b. The model is easily solved by maximizing the firm's cash flow subject to the budget constraints. Provided that the firm has sufficient internal funds, it will not borrow. Then the first-order condition that defines the optimal level of investment i* can be written as follows:13 (7.4) f 2(j*) - (1 + 0) = O. The first term in equation (7.4) is the marginal return (MR) curve. The second term is the discounted opportunity cost. The equilibrium is illustrated in the middle graph in figure 7.1. This simple model has a number of interesting implications. First, a policy change that, other things being equal, reduces profits (for example, increased competition from abroad resulting from trade liberalization) shifts the MR curve 12. Indeed, when dropping all firms with profit rates larger than 300 percent, no significant statistical relationship exists between size and profit. The relationship be- tween profit rates and size for Ugandan manufacturing firms is also significantly negative (coefficient = 0.17). 13. If the firm does not have sufficient internal funds, that is, Olt'2(11) - 1 > 0, it will borrow. The first order condition then becomes it'2(7t1 + b) - (1 + r) = 0. Confronting Competition: Investment, Profit, and Risk 219 Figure 7.1. Investment and Profit in Uganda and Other African Countries Profit ~~~~~~~~~~X07th (i;X) Rutg (i;x) io Investment Marginal return (1 + ()Oth (1 + )UgRoth __________ MRuga io Investment Profit rate irIKUga ------------- IKOth t/K/Uga 2/Kuga |----Ko g io Investment MR Marginal return. o Discount rate. i Level of investment. Source: Authors. 220 Ritva Reinikka and Jakob Svensson inward, leading to a lower level of investment for a given r and 6 for the exist- ing firms.'4 Second, a lower discount rate 0 (for example, better economic po- lices are expected in the future) would shift the horizontal curve down, lead- ing to a higher investment level as future income becomes more valuable. Comparing Uganda with other African countries, the model offers one potential explanation as to how investment rates can be similar while profit rates are lower. Increased competition has reduced profits and would, every- thing else being equal, have reduced investment rates as well. However, less uncertainty about future policies, resulting in a lower 0, counterbalances the negative effect of tougher competition on the level of capital accumulation. In equilibrium (figure 7.1), investment remains the same while profits and profit rates are lower. While it would be interesting to test this simple model statistically using the Ugandan survey data, endogeneity problems and a lack of suitable in- struments effectively prevent this. Instead, the conceptual framework can be used for a diagnostic discussion of the factors that are likely to affect the MR curve and the discount rate (0) of an average Ugandan firm. The analysis poses two hypothetical questions: Why is the Ugandan MR curve likely to be to the left of that of other African countries? Why is the discount rate of Ugan- dan firms likely to be smaller than elsewhere in Africa? The diagnostics are based on both quantitative and qualitative survey data from Uganda and focus on firms' perceptions of constraints to investment, competitive envi- ronment, costs beyond firms' control (infrastructure, corruption), risk, and policy credibility. Note, however, that similar data are not available for the comparator countries. Hence, the diagnostics presented in the rest of the chap- ter are tentative at best. Firms' Perceptions of Constraints This section examines qualitative data on constraints to investment. Rankings of constraints reported by firms give us a general idea of likely factors affecting both the marginal return to investment and the discount rate. In the 1998 survey Ugandan enterprises identified price and quality of utility services (electricity, telephones, water, and so on), high taxes, and interest rates as "major" (four on a scale of one to five) constraints to in- vestment (figure 7.2). Corruption, access to finance, tax administration, and the cost of raw materials formed a second tier of leading constraints. Fi- nally, the group of "moderate" (three on a scale of one to five) constraints included the problems of local competition, lack of demand, lack of busi- ness support services, crime and security, lack of skilled labor, and uncer- tainty about government policies. The largest variance in responses between firms occurred in access to finance and raw materials. 14. In this context we disregard the fact that increased competition may have other effects, such as raising productivity, which would shift the MR curve outward. Confronting Competition: Investment, Profit, and Risk 221 Figure 7.2. Ranking of Constraints to Investment, 1998 High utility prices High taxes Poor utility services Interest rates Corruption Tax administration Access to finance Crime and security Uncertainty about government policies Lack of skilled labor Exchange rate Cost of raw materials and supplies Insufficient demand Inflation Lack of business support services Government's debt burden Competition from local firms Other regulations Import and export regulations Competition from imports Access to raw materials and supplies Access to land Unclear property rights Politicial instability 1 2 3 4 5 No Minor Moderate Major Severe obstacle obstacle obstacle obstacle obstacle Source: Authors' calculations based on the 1998 enterprise survey. 222 Ritva Reinikka and Jakob Svensson A similar survey carried out in 1994 provides an interesting dynamic com- parison (figure 7.3).15 In the earlier survey, only high taxes were ranked a "major" constraint, while together with availability of inputs, lack of demand, and economic policy uncertainty, cost and access to finance and infrastruc- ture formed a second tier of "moderate" constraints. In that survey infra- structure included both the quality and the price of utility services. In addi- tion to a general elevation of constraints in their perceived severity, the major differences between 1994 and 1998 are the top rating of utility prices when offered in the 1998 survey as a separate constraint choice; the identification Figure 7.3. Ranking of Constraints to Future Operations and Growth in 1994 High taxes Cost of finance Access to finance Infrastructure Availability of inputs Demand Economic policy uncertainty Other regulation Inflation Policy uncertainty Labor force Business support services Trade regulation Exchange rate level fluctuations _ ~ ~ ~ ~ ~ ~~I I I I 1 2 3 4 5 No Minor Moderate Major Severe obstacle obstacle obstacle obstacle obstacle Source: World Bank (1994). 15. The 1994 survey differed slightly in its formulation of constraints, offered fewer choices of constraints to rank, and included firms from more subsectors of the economy. Confronting Competition: Investment, Profit, and Risk 223 of corruption as a leading constraint when offered in the 1998 survey, the recognition of labor force skills as a moderate constraint, and the new evalu- ation of the lack of business services as a moderate constraint. A closer look at the constraints by firm category shows little difference between the relative rankings in 1998 by small and large firms. For large firms, however, constraints were generally more binding, as reflected in higher perception scores. For foreign firms (and construction industry firms), cor- ruption was the second constraint in severity (see figure AI0. 1 in chapter 10). For Kampala-based firms, access to utility services was less binding than for other locations, while commercial farms and construction companies were less concerned about high taxes than the other firms. Competitive Environment When asked whether competition for their principal product had changed during the past three years, 88 percent of firms said it had increased, 10 percent reported unchanged competition, and only 2 percent said it had decreased. Similarly, the number of new firms exceeded those that had ex- ited. The firm-level evidence of increased competition accords with the lib- eralization of the economy and the continued start-up of new firms. Fair- ness constitutes another feature of competition. In 1994 a perception of unfairness existed in tax and regulatory administration. In 1998 this per- ception remained, with tax evasion as a leading constraint in relation to unfair competition. Firms in commercial agriculture reported the lowest incidence of unfair competition. However, the numerical constraint scores for competitors evading taxes, undercutting fair prices, or smuggling have all declined. Hence, while the overall level of competition has increased, firms' perception is that it has become slightly fairer since 1994. Lower profits are thus consistent with the observation of increased com- petition and the pressure it places on firms to reduce costs. Many of the re- ported cost constraints, such as utility prices, cost of imported inputs, and interest rates, are outside firms' direct control. One can therefore infer from the perception data that increased competition may not have been matched by corresponding improvements in physical and other support systems, par- ticularly those in the public domain. This makes it difficult for firms to re- spond to the challenge of increased competition brought about by external liberalization by cutting costs. Costs beyond Firms' Control As noted earlier, the Ugandan firm survey of 1998 points to at least three categories of costs that are beyond the firm's control, but that nonetheless tend to lower their profits. First, transport and other import-related costs add about 50 percent, on average, to the cost of imported capital goods and inputs compared with their cost in the country of origin. Second, infrastruc- ture services are highly deficient and costly, which also affects profits and 224 Ritva Reinikka and Jakob Svensson tends to shift the MR curve to the left. The 1998 survey confirmed that the cost of utilities is the most binding constraint to all types of Ugandan firms. Reliability and adequacy of electric power supply remain the leading infra- structure constraints to Ugandan enterprises, the only "major" constraints in the evaluation of respondents. Responses suggest that the electric power sup- ply has worsened in the last few years as demand has increased. Given the poor quality of infrastructure services, investment in productive capacity often requires an additional investment in complementary capital by the firm, such as electric power generators (see Reinikka and Svensson 1999). Third, cor- ruption is another factor that adversely affects returns to investment and, hence, shifts the MR curve inwards. As Svensson notes in chapter 10, the Ugandan survey data show that the larger, more profitable, more export- oriented the firm, the higher the incidence and the amount of bribe payments. Risk The relatively high profit rates in African firms point to a high cost of capital and high risk. The latter affects the discount rate 0 and tends to shift the horizontal line in figure 7.1 upward. The Ugandan firm survey reveals at least three types of risks that can adversely affect firms' expectations of fu- ture returns. First, erratic transport and other infrastructure services create a high risk in terms of unexpected delays (and related extra costs) in produc- tion, imports, and exports. For example, in 1998 it took an average of 30 days for imported inputs to arrive from their original destination in the port (typi- cally, Mombasa), another 30 days from the port to Ugandan customs, and an extra 9 days to the firm. While these figures are ex post averages, there is considerable variance among firms. In electric power supply, firms report that 87 operating days are lost annually due to power cuts. Although vari- ance between firms is smaller with respect to power shortages than other infrastructure services, blackouts and brown-outs create uncertainty about the returns to investment projects, including uncertainty about future im- provement in these services (Reinikka and Svensson 1999). Second, while the past decade has shown improvement, the tax administra- tion is still plagued by arbitrary tax assessments and audits. When firms do not know their tax liability in advance, returns to investment become uncertain. Crime poses a third major risk for Ugandan firms. The survey shows that 54 percent of the firms experienced merchandise robbery or theft of goods and equipment in 1995-97. Thirty-seven percent of the firms had also been victims of fraud. The loss from all these incidents was equivalent to US$7,500 at the median firm during the three years. Compared with corruption, for example, the incidence of crime seems to be relatively random. There is no evidence that the incidence of robbery or fraud, or the size of the loss from them, are corre- lated with profit, sales, or other cost- and revenue-related data from the firms. No evidence supports that certain sectors, foreign-owned firms, or those en- gaged in trade more often experience crime. The only characteristic of firms Confronting Competition: Investment, Profit, and Risk 225 that seems to matter is size (proxied by employment) and location. Larger firms are more often exposed to crime, and Kampala firms encounter an approxi- mate 20 percent increase in the probability of robbery or theft, independent of the size of the firm. In the sample, the probability that the average [median] firm in Kampala with 120 [35] employees had suffered from robbery and/or theft during the past three years is around 70 [63] percent. Not surprisingly, larger firms and firms located in Kampala spend significantly more on secu- rity. The annual cost of security for the median firm is equivalent to US$1,800, which equals the median firms' reported corruption payment per year. The data reveal that a 1 percent increase in employment (that is, firm size) corre- sponds with a 1.5 percent increase in security spending. Finally, noncommercial risk (captured by "political instability" in the over- all ranking of constraints) does not seem to concern many firms already in operation. According to a foreign investor survey, however, these risks were more of a concern for potential investors (World Bank 1999). Policy Credibility At the time of the firm survey in 1998, the private sector in Uganda seemed fairly confident that good macroeconomic management would continue both in the short and medium term, that is, one and three years from the time of the interview. This optimism was spread across all five sectors. On average, firms expected the exchange rate to remain about the same for the short term as at the time when the survey was carried out. Foreign-owned firms antici- pated a slightly higher depreciation, however. In the medium term a slight depreciation was expected (less than 10 percent). These results indicate that firms did not expect any major exchange rate volatility either in the short or medium term. Subsequent depreciation has been more substantial than the firms' expectations in 1998. Inflation forecasts were also relatively favorable. More than half of the firms expected that the country's single-digit average annual inflation-which had been maintained consistently since 1992/93- would continue both in the short and medium term. Two-thirds of the enterprises expected the trade regime to be further lib- eralized, and almost all firms expected the privatization program to continue. Indeed, at the time of the survey privatization appeared to be the most cred- ible of all the government's economic reforms. As discussed in chapter 2, while a large number of productive enterprises have been privatized in re- cent years, privatization of a few high-profile enterprises subsequently failed and corruption investigations were initiated. As a result, the privatization program was partially halted in 1998/99. Firms were less optimistic about the financial sector reform and its impact on future interest rates. About half the respondents expected interest rates to be lower in three years' time. However, close to 40 percent of firms did not believe that the banking sector could be reformed in the medium term and expected even higher interest rates. Concerning access to bank financing, four 226 Ritva Reinikka and Jakob Svensson out of every five respondents expected the situation to remain the same or to improve. In 1999 the Ugandan financial sector saw a number of bank closures, so firms might have appeared even more pessimistic about the financial sector had the survey been conducted in 1999. While this may be a temporary set- back and even a sign of more effective banking supervision, it is likely to have a negative effect on investor confidence, at least in the short term. Firms seemed to believe in continued growth in 1998: more than two- thirds of firms anticipated that their production would increase during the next three years. However, regarding expected future tax rates, they showed some pessimism: more than half anticipated that tax rates would be increased, and only 25 percent believed that rates would decrease."6 Conclusions and Policy Recommendations This chapter shows that investment rates in Uganda are relatively similar to those in other African countries. On average, the investment rate is slightly more than 10 percent, while at the median firm it is only about 1 percent. Such low investment rates in response to economic reform pose a serious policy problem. Unlike other African comparators, most firms in Uganda (and Ghana) experienced a positive change in their value added and gross sales. Investment by small firms seemed to be partly constrained by liquid- ity, while large firms, on average, could have chosen to invest more from retained earnings. As shown elsewhere, poor electricity supply substan- tially hinders private investment (Reinikka and Svensson 1999). Further- more, Ugandan profits are considerably lower than profit rates elsewhere in Africa. These results are consistent with the view that during the latter half of the 1990s, Ugandan firms displayed more confidence in the economy than their counterparts in many other African countries. Thus, for a given profit rate Ugandan firms invest more. At the same time increased competition, due to far-reaching economic liberalization, has pressured firms to cut costs. Many of the costs, such as utility prices, transport costs, and interest rates, are not in the firms' control, however. As there has been no matching improvement in infrastructure services or the financial sector, firms have failed to fully meet the challenge of increased competition. Thus, profits have been squeezed. 16. When asked an open-ended question about the best investment opportunity in the Ugandan economy in the medium term, firms listed a large variety of eco- nomic activities. Agriculture (horticulture, fruit, flowers, fishing, cattle, and so on) and agroprocessing were the most popular choices. Tourism and manufacturing (the latter mainly for the local market) were also frequently mentioned as good opportu- nities. A few firms considered trading (rather than production) as the most profitable activity, but the share of these firms was relatively small in the total survey. Confronting Competition: Investment, Profit, and Risk 227 The survey identified a number of cost factors to explain the observed low level of investment in Africa in general and the lower profits in Uganda in particular. First, capital goods are more expensive, largely due to higher transport costs and inefficiencies in transit transport and ports. Second, apart from investing in productive assets, firms often need to purchase comple- mentary capital, such as power generators, to stay in operation. Third, cor- ruption is a problem for most firms, particularly for those that invest more and employ more workers, are active in the formal sector, and are trade ori- ented. Risk factors likely to increase the discount factor firms apply to the future cash flow from investment and make longer-term investment less at- tractive include erratic infrastructure services, arbitrary tax administration, and crime. At the same time, macroeconomic policy credibility and investor confidence improved considerably in Uganda in the 1990s, and the risk of economic policy reversal is perceived to be relatively small. This in turn re- duces the discount factor of firms. The survey findings suggest four key policy priorities. First, the electric power sector urgently needs an effective reform program, combined with privatization and new investment in large-scale hydropower capacity. This is key to growth of the firm sector. Without a major improvement in the power supply, the sustainability of current growth rates is uncertain. Other utilities also need to improve their services. Second, while the government has com- mitted in its most recent budgets not to raise tax rates, tax administration needs improvement. One way could be to initiate a trust-building effort through establishment of a systematic mechanism of consultation between the tax collector and taxpayers, as well as proper appeals procedures. Third, a concerted effort to reduce corruption and improve contract enforcement is required. Such efforts are likely to take time, and it is initially important to choose measures that have a strong signaling effect. A recent household sur- vey found that the judiciary and police are one of the most corrupt institu- tions (Republic of Uganda 1998). Tackling corruption in these institutions, as well as in tax administration, should lead to less crime and reduced security costs, both of which are now a serious problem for firms. Finally, a more efficient transport route to the coast is needed, both in terms of improving the infrastructure and reducing red tape. The international donor commu- nity in Uganda could play a role in this effort, as Uganda alone will likely find it difficult to effect major changes in transit transport when part of the problem lies with the neighboring countries. 228 Ritva Reinikka and Jakob Svensson Annex 7.1. Data and Estimation Results Table A7.1. Investment in Machinery and Equipment by African Firms (mean) Country and Proportion of Investment-capital Investment-capital category firms investing stockfor allfirms stock iffirms invest Cameroon 1993-94 0.125 0.059 0.479 1994-95 0.347 0.132 0.382 Ghana 1992 0.363 0.090 0.428 1993 0.536 0.136 0.254 Kenya 1993 0.357 0.072 0.202 1994 0.459 0.127 0.277 Uganda 1996 0.506 0.134 0.263 1997 0.529 0.111 0.208 Large firms 0.765 0.083 0.109 Small firms 0.445 0.133 0.300 Zimbabwe 1993 0.621 0.069 0.111 1994 0.738 0.142 0.193 Comparator average All firms 0.535 0.128 0.239 Large firms 0.738 0.113 0.152 Small firms 0.458 0.134 0.291 Note: Large firms have more than 100 employees, while small firms have 100 or less employees. Source: Bigsten and others (1999); authors' calculations based on the 1998 enterprise survey. Table A7.2. Investment Regressions for Small and Large Ugandan Firms (2) Ordinary (4) Ordinary (1) Probit least squares (3) Probit least squares (5) Tobit (6) Tobit Variable (smallfirms) (smallfirms) (largefirms) (largefirms) (smallfirms) (largefirms) Constant 1.14b -0.005 4.95b 0.468' -0.727 0.169 (0.582) (0.216) (2.33) (0.227) (0.365) (0.219) Change in sales-to-capital stock 0.102 0.040 0.94a 0.006 0.010 0.048c (0.076) (0.038) (0.342) (0.026) (0.036) (0.028) Profit rate 0.143b 0.109, -0.12 0.036 0.145, 0.011 (0.065) (0.051) (0.139) (0.026) (0.034) (0.017) Age (log) -0.306a n.a. -0.062 {0.065' -0.193, -0.064b (0.104) n.a. (0.333) (0.031) (0.064) (0.028) NJ Size (log) 0.395a n.a. -0.828b n.a. 0.154b 0.014 (0 (0.105) n.a. (0.399) n.a. (0.064) (0.036) rime dummy 0.019 -0.098 0.524 0.026 -0.042 0.066 (0.160) (0.111) (0.411) (0.049) (0.099) (0.046) District dununies significant No Yes No No No Yes Mbale n.a. n.a. n.a. n.a. n.a. 0.305b n.a. n.a. n.a. n.a. n.a. (0.141) Kampala n.a. 0.218c n.a. n.a. n.a. n.a. n.a. (0.114) n.a. n.a. n.a. n.a. Mukono n.a. 0.389c n.a. n.a. n.a. n.a. n.a. (0.232) n.a. n.a. n.a. n.a. Industrial category dummies significant Yes No Yes No No No Agroprocessing 0.708b n.a. 2.06b n.a. n.a. n.a. (0.350) n.a. (0.814) n.a. n.a. n.a. (table continues onfollowing page) Table A7.2 continued (2) Ordinary (4) Ordinary (1) Probit least squares (3) Probit least squares (5) Tobit (6) Tobit Variable (smallfirms) (smallfirms) (largefirms) (largefirms) (smallfirms) (largefirms) Predictability 0.67 n.a. 0.80 n.a. n.a. n.a. , R2 n.a. 0.16 n.a. 0.27 n.a. n.a. O Observations 278 126 75 58 278 75 n.a. Not applicable. Note: The dependent variable in regression (1) takes the value one if the firm invested and zero otherwise. Standard errors (in parenthesis) are adjusted for heteroskedasticity (White 1980). Regressions (2), (4), (5), and (6) are adjusted for selectivity. The inverse Mills ratio is not reported. The tourism dummy had to be dropped from regression 3 because all large firms in this sector invested. a. Significant at the 1 percent level. b. Significant at the 5 percent level. c. Significant at the 10 percent level. Source: Authors' calculations based on the 1998 enterprise survey. Confronting Competition: Investment, Profit, and Risk 231 Annex 7.2. Derivation of the Investment Equation Let the cost of instantaneous net investment be given by C(I), where I is net investment and C is a cost function with C(O) = 0, and C', C"O for all I > 0. Let profit be a concave function of the capital stock ic = n(t, K), and assume that the firm takes product and factor prices as given. As shown by Tybout (1983), with constant relative prices, investment can be expressed as (A2.1) I(t) = P[K - K(t)], where K is the desired capital stock implicitly determined by t'(K) = rC'(O), and 3 is a composite variable (constant) of the discount rate, r, and it" and C" evaluated at K and 0, respectively. Hence, in the flexible accelerator model, investment is driven by the gap between the desired and actual capital stock, where the relative sluggishness of adjustment depends on the user cost of capital. Assume that managers expect that the future demand for their out- put will be Q*, and let K'(t) = yQ*(t), where y is determined by relative prices. In discrete time, equation (A2.1) can be written as (A2.2) I(t) = 0[yQ*(t) - K(t - 1)]. Demand expectations are assumed to be linear functions of current output. Thus, (A2.3) I(t) = P[70Q(t) - K(t - 1)]. By first-differencing equation (A2.3) and noting that I(t - 1) = K(t - 1) - K(t - 2), equation (A2.3) can be written as (A2.4) I(t) = a%AQ(t) + (1 - ,B)I(t - 1), where 1xQ _ 5 and AQ(t) = Q(t) - Q(t - 1). This is the traditional flexible accelerator model in which fluctuations in sales motivate changes in capital spending, that is, investment is driven by demand. As shown in Tybout (1983), if firms must finance all investment out of profits and retained eamings, the firms will behave according to (A2.4) when they have funds to do so. However, with currently binding shortages, they will simply allocate current profits to investment. Hence, (A2.5) I(t) = C-1[11(t)]. A general empirical model can now be formed by nesting (A2.4) and (A2.5), (A2.6) i(t) = aiO + (xQAQ.(t) + ao,it(t) + cx,I.(t - 1) + ax'X, + d, + .,, where aio is a constant for firm i, ax is a n x 1 vector of coefficients, Xi is a n x 1 vector of firm specific controls (firm age and size), dt is a time dummy, and Ei is an iid error term. To avoid heteroskedasticity problem with respect to size, I(t), AQ,(t) and itc(t) are scaled by the inverse of the end of the previ- ous period capital stock, K(t - 1). Thus, we are regressing investment rate, I#(t)/K#(t - 1), on change in output (value added) rate, AQ,(t)/K,(t - 1) and profit rate, ni(t)/Ki(t - 1). 232 Ritva Reinikka and Jakob Svensson A number of variations of (A2.6) are estimated: with fixed effects (ct..), with a common constant (ao), and with and without the lagged investment vari- able. Given the short panel, there are clear costs of estimating the more com- plex regressions. With fixed effects all firms that do not have observations for all three years are lost.'7 Similarly, including a lagged dependent variable im- plies that we lose observations for firms that started up after 1995, and fixed effects in a dynamic model with a short time dimension result in biased esti- mates (Nickell 1981) that cannot be overcome by instrument variables tech- niques (due to the short panel) as suggested by Arellano and Bond (1991). References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Arellano, Manuel, and Stephen Bond. 1991. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations." Review of Economic Studies 58(April): 277-97. Athey, M. J., and P. S. Laumas. 1994. "Internal Funds and Corporate Invest- ment in India." Journal of Development Economics 45(2): 287-303. Barro, Robert. 1991. "Economic Growth in a Cross-Section of Countries." Quarterly Journal of Economics 106(2): 407-43. Barro, Robert, and X. Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hill. Biggs, Tyler, and Pradeep Srivastava. 1996. "Structural Aspects of Manufac- turing in Sub-Saharan Africa: Findings from a Seven Country Enter- prise Survey." World Bank Discussion Paper no. 346, Africa Technical Department Series. World Bank, Washington, D.C. Bigsten, Arne, Paul Collier, Stefan Dercon, Bernard Gauthier, Jan Willem Gunning, Anders Isaksson, Abena Oduro, Remco Oostendorp, Cathy Pattilo, Mans Soderbom, Michel Sylvain, Francis Teal, and Albert Zeufack. 1999. "Investment in Africa's Manufacturing Sector: A Four- Country Panel Data Analysis." Oxford Bulletin of Economics and Statis- tics 61(4): 489-512. Bond, Stephen, Julie Ann Elston, Jacques Mairesse, and Benoit Mulkay. 1997. "Financial Factors and Investment in Belgium, France, Germany, and the UK: AComparison Using Company Panel Data." Working Paper no.5900. National Bureau of Economic Research, Cambridge, Massachusetts. 17. Note that to create a panel with, at the most, two observations for each firm, we must use data for three years since AQ(t) = Q(t) - Q(t - 1). Confronting Competition: Investment, Profit, and Risk 233 Collier, Paul, and Jan Willem Gunning. 1999. "Explaining African Economic Performance." Journal of Economic Literature 37(March): 64-111. DeLong, J. B., and L. H. Summers. 1991. "Equipment Investment and Eco- nomic Growth." Quarterly Journal of Economics 106(2): 445-502. Devarajan, Shantayanan, William Easterly, and Howard Pack. 1999. "Is In- vestment in Africa Too Low or Too High?" World Bank, Development Research Group, Washington, D.C. Processed. Easterly, William. 1997. "The Ghost of Financing Gap-How the Harrod- Domar Growth Model Still Haunts Development Economics." Policy Research Working Paper no. 1807. World Bank, Development Research Group, Washington, D.C. Easterly, William, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Division." Quarterly Journal of Economics 112 (4): 1203-50. Heckman, James. 1979. "Sample Selection Bias as a Specification Error." Econometrica 47(1): 153-61. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross- Country Growth Regressions." American Economic Review 82(4): 942-63. Lucas, R. E. 1988. "On the Mechanism of Economic Development." Journal of Monetary Economics 22(1): 3-42. Mankiw, N. G., D. Romer, and D. N. Weil. 1992. "A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107(2): 407-38. Nickell, Steven J. 1981. "Biases in Dynamic Models with Fixed Effects." Econometrica 52(6): 203-7. Rebelo, Sergio 1991. "Long-Run PolicyAnalysis and Long-Run Growth." Jour- nal of Political Economy 99(3): 500-21. Reinikka, Ritva, and Jakob Svensson. 1999. "How Inadequate Provision of Public Infrastructure and Services Affects Private Investment." Policy Research Working Paper no. 2262. World Bank, Development Research Group, Washington, D.C. Republic of Uganda. 1998. National Integrity Survey. Kampala: Inspector Gen- eral of Government. Romer, P. M. 1986. "Increasing Returns and Long-Run Growth." Journal of Political Economy 94(5): 1002-37. Sachs, J. D., and A. M. Warner. 1995. "Natural Resource Abundance and Eco- nomic Growth." Working Paper no. 5298. National Bureau of Economic Research, Cambridge, Massachusetts. 234 Ritva Reinikka and Jakob Svensson . 1996. "Sources of Slow Growth in African Economies." HIID Devel- opment Discussion Paper no. 545. Harvard Institute for International Development, Cambridge, Massachusetts. Tybout, J. R. 1983. "Credit Rationing and Investment Behavior in a Develop- ing Country." Review of Economics and Statistics 65(4): 598-607. White, Halbert. 1980. "A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity." Econometrica 48(4): 817-38. World Bank. 1994. "The Private Sector in Uganda: Results of the World Bank Enterprise Survey." World Bank, Eastern Africa Department, Wash- ington, D.C. Processed. .1999. "Survey of Foreign Investors." A Report by the Consorzio Italiano Consulenti, Bologna, Italy. Processed. 8 Productivity and Exports Bernard Gauthier Uganda has liberalized its trade and exchange rate regimes to scale back trade barriers and price distortions (see chapters 2 and 3 in this volume). Have the reforms generated a significant response from firms and substantial produc- tivity gains? Have they encouraged the development of an outward-oriented industrial sector? While external competition is perceived as favoring effi- ciency through increased productivity and a shift of resources from ineffi- cient to efficient sectors, the transition from a restrictive to an open trade regime can impose short-term adjustment costs in sectors newly exposed to external competition. The answers to these questions are important, there- fore, in understanding what can be done to speed and smooth the transition toward an efficient, outward-oriented industrial base. To address these issues we examine the impact of trade and exchange rate reforms on private enterprises. Firm-level productivity and technical efficiency measures, as well as other performance indicators, are constructed using detailed information collected in a 1998 survey of firms by the World Bank and the Ugandan Private Sector Foundation (see appendix B at the end of the book). Performance measures show whether firms have shifted re- sources by increasing or decreasing output following the change in incen- tives; whether they have become more productive in terms of labor produc- tivity, total unit cost, total factor productivity (TFP), and technical efficiency (using stochastic production frontier models); whether market shares have The author thanks Ritva Reinikka for insightful discussions. Excellent research assistance was provided by Jean Habarurema, Michel Sylvain, and Alex Darku at different stages of the project. 235 236 Bernard Gauthier shifted toward high-productivity firms; and whether an association exists between productivity gains and the ability to export. This chapter discusses the relationship between trade reform, exports, and productivity. It examines the response of enterprises to the new environ- ment of liberalized trade and exchange in terms of output and productivity growth, and documents in more detail the export response to trade reforms. The conclusion suggests three types of policies to enhance the export orien- tation of Ugandan enterprises. Trade Liberalization, Exports, and Productivity In addition to improvements in the policy environment and macroeconomic stability, the core element of the economic reform programs implemented and supported by external donors in many Sub-Saharan African countries since the late 1980s has been the implementation of trade liberalization and the de- velopment of an outward-looking development strategy. Given the small size of the domestic market in Africa and the dependency on imported intermedi- ate goods and capital, the development of an export-oriented sector has been perceived as essential for investment and development (Husain and Faruqee 1994; UNCTAD 1998; World Bank 1994). Trade liberalization and export orien- tation is seen to have a positive effect on productivity and investment. The perception is that exposure to international markets favors technology acqui- sition and market discipline, allowing firms to achieve economies of scale. By introducing competition among previously protected domestic firms, trade reforms induce changes in firms' behavior and performance. The benefits of more open trade and an export orientation are transmit- ted through several channels. Exposure to external competition encourages domestic firms to adopt newer and more efficient technology or to use the same technology with less waste or less x-inefficiency to reduce costs and compete against international firms (Nishimizu and Robinson 1984). Fur- thermore, the removal of less efficient firms, previously able to operate inef- ficiently because of protection, results in a lower average cost and higher productivity. The firms that remain in the industry must adjust by expand- ing the scale of their production, exploiting economies of scale, and reducing technical inefficiency. Because domestically produced goods cannot replace imported intermediate and capital goods in developing countries, imported inputs tend to increase knowledge and improve technical efficiency. At the same time, increased imports and exports augment the spillover of interna- tional technical knowledge (Grossman and Helpman 1991). Despite these expected positive effects, more open trade and an outward- oriented strategy may have adverse effects on domestic producers compet- ing with imports. Rodrik (1988) and Tybout (1992) emphasize that a negative transition cost could result for domestic producers in industries where econo- mies of scale existed and that contract or exit due to greater import penetra- tion in the domestic market. Productivity and Exports 237 More generally, the potential benefits of trade liberalization and export orientation have not been fully exploited in Africa, and some analysts con- tend that this explains the development of the small industrial sector: [M]anufacturing industries in Africa have not been exposed to market discipline through exports, and in addition they have failed to benefit from the scale advantages needed to compete internationally. These factors have, in turn, further restricted the development of such industries to small and sluggish do- mestic markets, perpetuating high costs and giving rise to inef- ficiencies and low levels of productivity (UNCTAD 1998, p. 196). Empirical evidence of the effects of trade liberalization and export orien- tation has been mixed. Pack (1988, p. 353) reviewed numerous empirical stud- ies on the effects of export orientation on the industrial sector, observing that "there is no clear confirmation of the hypothesis that countries with an exter- nal orientation benefit from greater growth in technical efficiency in the com- ponents sectors of manufacturing." In the particular case of Sub-Saharan Africa, some analysts doubt that more open trading conditions will produce an industrial response. Several factors could compromise this response, in- cluding inconsistent macroeconomic policies and weaknesses in institutions, infrastructure, and available human resources. Indeed, even if economic re- forms were credible and producers did respond to the new incentives, the resulting productivity gains could be offset by declines in factor accumula- tion (Elbadawi 1992; Matin 1992). Several analysts have found evidence of output growth and productivity gains following liberalization and export orientation in the African context. Harrison (1994) analyzed the changes in firm behavior in C6te d'Ivoire and found that liberalization had significant affected productivity. In Morocco, Haddad (1993) found a positive relationship between productivity and ex- ports at the firm level. She suggested that firms closest to the maximum effi- ciency level tended to have high export shares. Roberts and Tybout (1997) in their examination of four countries found that differences in productivity within an industry are typically greater in industries protected from international competition, suggesting that protec- tion nurtures inefficiency. However, in another study of four countries, in- cluding Morocco, Clerides, Lach, and Tybout (1998) found little evidence of an export efficiency effect. By contrast, Bigsten and others (forthcoming), using a four-country panel of African manufacturing firms in four sectors, found a significant efficiency gain from exporting, the gains being even larger for new entrants into the exporting market. Using firm-level data from five sectors in Uganda, this chapter exam- ines the evidence of association between productivity and exports within the framework of trade liberalization. The next section examines the re- sponse of enterprises to the new environment of trade liberalization and economic reforms. 238 Bernard Gauthier Enterprise Responses to Changing Incentives This section examines firms' responses to changes in incentives following trade and economic reforms. It first looks at whether the firms shifted re- sources toward tradable products by examining output responses between 1995 and 1997. It then examines whether firms improved their productivity by constructing indexes of productivity performance and analyzing the determinants of productivity growth to determine whether any differences can be explained by the firms' export orientation, the import intensity of in- termediates, or other firm-level characteristics. Nature of the Sample and Variables The data used in this study comprise a balanced panel data set of 139 firms.' Data were collected in a recent firm-level survey conducted in Uganda by the World Bank and the Ugandan Private Sector Foundation during 1995- 97. Table A8.1 presents basic characteristics of the sample of firms drawn from five sectors, representing a wide spectrum of private sector activities. Firms are classified into three size categories based on the number of em- ployees: small (5-20 employees), medium (21-100 employees), and large (more than 100 employees). The distribution is also broken down accord- ing to whether the firm was an exporter in 1997 and its reliance on im- ported inputs. A firm is classified as an exporter if it exports any percent- age of the value of its output. A firm is classified as imported-input intensive when it imports more than 50 percent of the value of its inputs. Firms are additionally classified by source of capital (domestic, foreign, and joint ownership firms). As illustrated in table A8.1, the agroprocessing and manufacturing sec- tors are heavily represented, with 21 and 49 percent, respectively. While the manufacturing sector is also the most important in terms of total output value (52 percent), the agroprocessing sector is the most important in terms of total employment (43 percent). Seventy-four percent of the firms in the sample are domestically owned, but represent a much smaller proportion of total out- put and employment. Foreign firms are the most important for total employment (37 percent), while the 11 percent of jointly owned firms account for more than 71 percent of total output. With respect to market orientation, exporting firms account 1. The original data set comprised 243 firms in five sectors. Restricting the sample to firms with complete time series in all variables of interest reduced the size of the data set by about one-third. Of the remaining firms, those that reported data inconsis- tent with the following criteria corresponding to twice the standard deviation were rejected as data errors or outliers: replacement value of machinery and equipment over gross output value greater than 50, growth in unit cost over the period greater than 1.5, growth in gross output greater than 500 percent, and growth in employment greater than 500 percent. Productivity and Exports 239 for almost half of total employment and nearly 40 percent of total output, despite their representation of only a quarter of the sample in terms of num- ber of firms. With respect to size distribution, large firms represent less than 20 percent of the sample, but employ 77 percent of the total work force and account for 86 percent of total output value. Table A8.2 presents the variables used in the empirical analysis of the study, defined as follows. Output corresponds to sales revenue from all out- put produced by the firm during the year. Capital is defined as the replace- ment value of machinery and equipment. Intermediate inputs include the cost of raw materials, utilities (telephone, electricity, water), and fuel. Wage is the total wage bill, including allowances, benefits, bonuses, and statutory payments. Labor is defined as total number of employees. Where applicable, data are expressed in constant 1995 prices. Output and Productivity Growth A fundamental objective of trade liberalization and the economic reform pro- gram in Uganda was to encourage the production of tradable goods. Firms were expected to increase outputs if changes in relative prices increased prof- itability. Furthermore, if markets were made more competitive by the removal of trade barriers, domestic producers were expected to increase production and reduce inefficiency in production. Were these goals achieved? To investi- gate differences in the output and productivity responses of categories of firms, several indicators of performance are constructed, including output growth and productivity indexes. Table A8.3 summarizes growth rates in real output for the categories of firms during 1995-97. It presents unweighted and weighted averages, as well as medians and interquartile ranges.2 All firm-level growth rates are cumula- tive for the period 1995-97. As shown in table A8.3, output response was relatively strong during the period, as the average firm shows an increase in real output of more than 35 percent (unweighted) between 1995 and 1997 (29 percent weighted). There is, however, a wide variation across firms, as the median output increased by 12 percent, but more than a quarter of the sample exhibited negative output growth. Breaking down the sample by sector reveals that output growth is con- centrated in the manufacturing sector as well as the agroprocessing and com- mercial agriculture sectors. Output-weighted figures in particular show an increase of close to 50 percent in the real output value in manufacturing. Export-oriented firms fared much better than nonexporting firms, with the weighted average figures exhibiting an increase of 40 percent, compared with only 15 percent for nonexporters. Unweighted average figures reveal a simi- lar picture, although the gap is smaller. 2. Firm gross output is used as the weight, which is used to account for firms' relative size within the sample in their contribution to the sample mean. 240 Bernard Gauthier Regarding categories of input use, for firms relying intensely on domes- tic inputs, the median firm recorded an increase of 13 percent in output, com- pared with only 6 percent for firms relying principally on imported inputs. However, average real output growth is greater for imported-input- intensive firms than for domestic input firms because of a greater variation in performance among the latter category of firms and, presumably, because many of them produce tradable goods. Regarding categories of ownership, foreign-owned firms did better than domestically-owned firms. Joint ownership firms also performed better than entirely domestic firms, with more than three-quarters of them recording posi- tive output growth. Larger units fared better than other size categories during the period, especially medium-size firms, as shown in the unweighted aver- age of 35 percent compared with 20 percent, and the weighted average in- crease of 36 percent compared with a drop of -3 percent for medium-size firms. Although significant growth in real output was registered mainly by larger producers in the categories of tradable goods producers, it is unclear whether this increase paralleled an increase in productivity due to a shift of resources from inefficient to efficient activities or through a more efficient use of re- sources. To assess the effects of trade liberalization on firm-level productiv- ity, four commonly used indexes of productivity are constructed: labor pro- ductivity, total unit cost, TFP, and technical efficiency. Labor productivity is measured as the logarithm of the output per em- ployee in constant 1995 prices. Total unit cost is measured as the long-term average cost of production in constant 1995 prices, and is calculated as the logarithm of the cost of capital, wages, and intermediate inputs divided by the value of gross output. The TFP represents the level of output not ex- plained by the level of inputs. It is constructed as the residual of a constant returns Cobb-Douglas production function using capital, labor, and material as inputs (see annex 8.1 for details). Finally, technical efficiency is a continu- ous index ranging from zero to one representing the degree to which firms fail to reach a best practice frontier. This index is measured using the stochas- tic frontier production function methodology (see Bauer 1990 or Schmidt and Sickles 1984 for a survey). In this model, the best practice production frontier is estimated that defines the maximum output achievable for a given set of inputs. All firm output is then compared to this frontier. Deviations from the frontier mean that the firm produces less than its technical capacity, imply- ing some degree of inefficiency (see annex 8.1 for details). Productivity Level Table 8.1 presents the unweighted average levels for 1995-97 of each of the four productivity indexes by firm category. All figures are presented in con- stant 1995 prices. Examining the differences in productivity levels among categories of firms confirms that all measures of productivity considerably favor exporting firms. On average, exporters enjoy more than 60 percent more output per employee than nonexporters, while for the average Productivity and Exports 241 Table 8.1. Productivity Levels by Firm Characteristics (unweighted averages, 1995-97) Number Labor Totalfactor Characteristic offirms productivity Unit cost productivity Efficiency By sector Commercial agriculture 16 8.54 1.94 2.75 0.27 Agroprocessing 29 20.19 4.86 1.95 0.18 Manufacturing 68 19.33 3.08 2.00 0.17 Construction 11 7.36 4.19 2.02 0.19 Tourism 15 5.44 1.37 3.25 0.23 By exporters Exporter 32 22.22 1.96 2.41 0.23 Nonexporter 107 13.91 3.60 2.15 0.18 By importer Domestic input intensive 97 8.39 3.01 2.40 0.19 Imported input intensive 42 32.99 3.73 1.77 0.19 By ownership Local 103 9.46 5.53 2.24 0.18 Foreign 21 26.25 2.27 2.10 0.17 Joint 15 44.89 2.49 2.15 0.32 By size Small 63 9.53 3.59 2.12 0.18 Medium 50 17.93 3.35 2.48 0.20 Large 26 27.02 2.10 1.91 0.22 Total 139 15.82 3.22 2.21 0.19 Note: Categories in 1995. Labor productivity, unit cost, and total factor productivity in logs. Efficiency is a 0-1 index, where 1 indicates full efficiency. Source: Author's calculations based on the 1998 enterprise survey. exporting firm the TFP index is more than 12 percent greater than for nonexporters. Total cost per unit of revenue is 46 percent less for exporters, and the index of technical efficiency is an average of 28 percent greater for exporting firms, indicating more homogeneity in the distribution of export- ers. These patterns are robust when measured by weighted averages, medi- ans, or unweighted means. Among categories of ownership, foreign-owned firms and those with joint foreign and local ownership generally enjoy higher levels of productivity. The labor productivity of local firms is almost five times less than that of joint ownership firms; the total unit cost is twice as high and the efficiency index is 44 percent lower. These patterns are similar for median and weighted average figures (table A8.4). 242 Bernard Gauthier Differences in productivity levels are also marked among sectors, because the agroprocessing and manufacturing sectors exhibit close to 2.5 times more labor productivity than the agriculture and construction sectors. Productiv- ity levels by size categories are also as expected, because smaller firms ex- hibit 65 percent less labor productivity than large firms, 70 percent higher total costs per unit of revenue, and 19 percent less technical efficiency. With respect to the TFP index, medium-size firms show a higher unweighted av- erage and large firms show a higher weighted average (31 percent more than small firms, see table A8.4), indicating that the larger segment of the large- size category exhibits greater productivity levels. Table 8.2 compares technical efficiency levels in Uganda with manufac- turing firms in four other African countries. All technical efficiency figures were computed using a random-effect model and estimated with a gener- alized least squares approach. Table 8.2 reveals that among all five African countries technical efficiency among exporting firms is consistently greater, on average, than that of domestic-oriented firms. However, efficiency among the sample of firms in Uganda is low among both exporters and nonexporters relative to the other African countries. Low technical efficiency in Uganda indicates more potential waste and x-inefficiency in produc- tion. It may also indicate untapped opportunities for productivity improve- ment through learning, possibly reflecting less homogeneity in technology within the distribution of firms. Productivity Growth An increase in output generally leads to an increase in productivity because of a reduction in idle capacity and better use of economies of scale. Furthermore, if market competition were increased through trade liberalization, Ugandan firms may have responded to the new environment by further improving their productivity. Some preliminary evidence reveals better output response and higher productivity levels, particularly from exporters. Table 8.3 shows productivity growth among categories of firms and pre- sents cumulative productivity growth rates for the four indexes by category. Table 8.2. Efficiency Levels of Exporters in Five African Countries (unweighted averages) Category Uganda Cameroon Ghana Kenya Zimbabwe Exporters 0.23 0.52 0.49 0.32 0.40 Nonexporters 0.18 0.31 0.25 0.18 0.34 All 0.19 0.38 0.27 0.22 0.37 Number of firms 139 50 93 70 94 Period 1995-97 1993-95 1991-93 1992-94 1992-94 Source: Author's calculations based on the 1998 enterprise survey; Bigsten and others (2000, table 3). Productivity and Exports 243 Table 8.3. Real Productivity Growth (cumulative percentages unweighted, 1995-97) Number Labor Total Totalfactor Characteristic offirms productivity unit cost productivity Efficiency By sector Commercial agriculture 16 -1.2 -4.6 14.5 -1.8 Agroprocessing 29 14.2 2.3 2.6 -0.2 Manufacturing 68 6.5 5.9 10.2 -4.9 Construction 11 -15.9 15.6 -11.5 0.8 Tourism 15 25.9 14.0 -0.8 1.9 By exporters Exporter 32 25.8 1.1 10.4 7.3 Nonexporter 107 2.1 6.9 4.9 -5.3 By importer Domestic input intensive 97 2.0 7.1 4.3 -1.7 Imported input intensive 42 20.5 2.0 10.5 -4.0 By ownership Local 103 5.1 8.5 3.7 -3.6 Foreign 21 18.5 -4.5 21.2 6.5 Joint 15 9.0 -0.5 2.1 -6.7 By size Small 63 6.3 8.5 8.8 -4.0 Medium 50 6.0 4.7 2.7 -3.2 Large 26 13.5 0.2 6.5 3.0 Total 139 7.5 5.6 6.2 -2.4 Note: Categories in 1995. Source: Author's calculations based on the 1998 enterprise survey. All figures are unweighted averages for 1995-97 in constant 1995 prices. An examination of the four indexes of productivity reveals a mixed overall re- sponse. While some indexes exhibit a positive trend, others have regressed. Both unweighted and weighted average figures (see table A8.5) show im- provements in labor productivity and TFP for the average firm in the sample, with the related indexes rising by 8 and 6 percent, respectively (unweighted). By contrast, total unit cost and technical efficiency exhibited a negative over- all trend, with an increase in unit cost of 6 percent and a drop in efficiency of 2 percent. Significant differences in productivity performance among catego- ries of firms explain these trends. The export sector, however, performed noticeably better than domestic market-oriented producers for all productivity indexes. During the period, la- bor productivity grew more than 10 times faster, and total unit cost grew 6 244 Bernard Gauthier times slower. Moreover, exporters recorded an (unweighted) average increase in their TFP of 10 percent-compared with 5 percent for nonexporters-and an increase of 7 percent in efficiency index, compared with a decline of 5 per- cent for nonexporters. Interestingly, in terms of labor productivity, total unit cost, and TFP, firms relying intensely on imported inputs achieved higher pro- ductivity growth than firms relying on domestic inputs. With respect to size categories, large firms performed better than smaller units, especially in terms of labor productivity, efficiency, and lower unit cost growth. With respect to ownership, foreign-owned firms fared better than their domestic counterparts for all four productivity measures (both weighted and unweighted). Table 8.4 compares technical efficiency growth in Uganda during the lib- eralization period with observed efficiency growth in four other African coun- tries also going though a process of trade liberalization. Efficiency growth in the export sector in Uganda is consistent with similar growth observed in the other four African countries, with export-oriented firms outperforming do- mestic market-oriented producers during the liberalization period. Explaining Productivity Growth A series of regressions is estimated to more rigorously examine how eco- nomic reforms produced increases in productivity. Firm-level productivity is modeled as a function of various explanatory variables, including exports, market participation, and other firm characteristics. More specifically, the following equation is used: (8.1) AAit=aIDEit_l+a +,e.t where AAit is a measure of productivity growth calculated above for firm i at time t, DEit l is a dummy of initial exports, and Xi, l is a vector of exogenous variables of firm characteristics, particularly size and sector. In a small country like Uganda, exporting firms are expected to adjust relatively easily to changes in relative prices and other external changes due to the absence of demand constraints. Indeed, export-oriented firms Table 8.4. Efficiency Growth of Exporters in Five African Countries (cumulative percentages, unweighted) Category Uganda Cameroon Ghana Kenya Zimbabwe Exporters 7.3 12.8 15.2 8.4 8.6 Nonexporters -5.3 -9.8 -3.0 2.0 1.9 Al -2.4 -2.6 -1.7 4.0 5.8 Number of observations 139 50 93 70 94 Period 1995-97 1993-95 1991-93 1992-94 1992-94 Source: Author's calculations based on the 1998 enterprise survey; Bigsten and others (2000, table 4). Productivity and Exports 245 exposed to a more competitive environment are expected to have a greater incentive to increase productivity. To assess this incentive adequately, and to identify the effect of exporting, a dummy variable takes the value of one when a firm was an exporter initially and zero when it did not export at the beginning of the period. Equation (8.1) is estimated using Huber-White correction for heteroskedasticity. Table A8.6 presents the results, which tend to confirm the evidence presented in table 8.3. Export orientation is a significant determi- nant of productivity growth according to several productivity measures. As table A8.6 shows, the coefficients of the dummy variable of initial exporters is positive and significant in both the TFP and technical efficiency regres- sions, indicating that initial exporters tend to show higher growth of TFP and efficiency compared with nonexporters during the period. As for labor productivity growth, the export dummy, while not significant, has the ex- pected positive sign, and the export dummy for total unit cost has the ex- pected negative sign, as initial exporters exhibit a lower total cost per unit of revenue over the period. These results accord with the resource shift documented earlier toward export-oriented activities and are consistent with those of Roberts and Tybout (1997) and Kraay (1999), who observed that exporting firms are more productive than their domestically-oriented counterparts. Indeed, Kraay (1999), studying a panel of Chinese firms, also observed that past exports were positively associated with higher growth in productivity measures. The results are also consistent with those of Bigsten and others (2000) in a study of a comparable group of Sub-Saharan countries. Using firm-level panel data from four Sub-Saharan African countries (Cameroon, Ghana, Kenya, and Zimbabwe), Bigsten and others (2000) examined the effects of exporting on technical efficiency over a three-year period. They showed that the effects were quite substantial, with initial exporters exhibiting 11 percent higher efficiency growth than nonexporters over the period. In Uganda the effect of exporting on technical efficiency is also positive and significant, indicating an important learning effect associated with export- ing activities among the sample firms during the period. Caution is necessary when analyzing these results because of endogeneity problems between exporting and efficiency. On the basis of the present analy- sis, it is impossible to answer the question of whether exporting leads to effi- ciency gains or if the relationship runs from efficiency to exporting. Indeed, according to several productivity measures, the correlation between export status and ex post productivity levels suggests that high productivity precedes entry into the export market. One likely explanation investigated in the recent literature (but not yet pursued with the Ugandan survey data) is that high- productivity producers can afford the cost of entering the export market (Rob- erts and Tybout 1997). Work on U.S. firms and middle-income countries has documented that high productivity levels correlate with subsequent entry (Ber- nard and Jensen 1999). Further research would be required to disentangle the 246 Bernard Gauthier direction of causality (see Bigsten and others 2000; Clerides, Lach, and Tybout 1998).3 In conclusion, export orientation is associated with significantly greater output growth during the period of liberalization in Uganda, and with higher productivity levels and growth in terms of several measures of productivity. Export Response As shown previously, trade liberalization in Uganda was accompanied by output growth among export-oriented activities as well as greater levels and growth in productivity among these firms. This section documents in more detail the export response to trade liberalization. It examines the source of export response by breaking down export growth by firm category (in- cumbent, new entrants, and quitters), as well as the determinants of the decision to export. The sample in this section comprises a balanced panel data set of 177 firms that reported data on the decision to export, percentage of exports in each year, and destination of exports.4 As can be seen from table A8.7, which presents summary statistics on the exporters, the average percentage of ex- ports to gross output value in the sample increased from 9 to 10 percent dur- ing the period. Exporting firms exported an average of 37 percent of their output in 1995, a figure that remained stable in 1997 (38 percent). When weighted by the value of output to account for relative firm size, the weighted export average increased to 15 percent of total sales value in 1997, up from 12 percent in 1995. An interesting element concerns the destination of Ugandan exports and the changes during the liberalization period. As reported in table 8.5, the most important destination was Europe, which received 60 percent of total export value in 1997, followed by East Africa and other non-European, non- African countries, both with 18 percent. Between 1995 and 1997, export values for the sample increased by 90 per- cent. As documented in table 8.5, the largest increase over the period (327 3. Clerides, Lach, and Tybout (1998) have performed a type of Granger causality test by using a full information maximum likelihood (FIML) estimator on a dynamic model of productivity and exports with serially correlated errors, as well as a gener- alized method of moments estimator on an average variable cost function. Examin- ing three middle-income countries, they have not found evidence that exporting ex- perience reduces costs, except in the apparel and leather products industries in Morocco. Bigsten and others (2000), using a comparable nonparametric FIML dy- namic model with correlated random effects, found a significant and positive effect of export history on technical efficiency among manufacturing firms in four Sub- Saharan countries. 4. Note that the sample in this section is larger by 38 firms than in the section on enterprise responses. This is due to the smaller requirement in the number of variables in each year in this section. Firm number 75 was deleted because of data-entry error. Productivity and Exports 247 Table 8.5. Nominal Exports Value and Shares by Destination, 1995-97 1995 1996 1997 USh USh USh billions billions billions Destination (current) Percent (current) Percent (current) Percent East Africa 10.76 27.4 15.92 23.5 13.56 18.2 Rest of Africa 1.21 3.1 1.71 2.5 2.74 3.7 Europe 24.05 61.4 38.50 56.8 44.66 60.0 Other countries 3.16 8.1 11.71 17.2 13.48 18.1 Total exports 39.18 100.0 67.84 100.0 74.44 100.0 Note: Number of firms is 177. Source: Author's calculations based on the 1998 enterprise survey. percent) was registered in the "other countries" category (non-European and non-African countries). This destination now represents 18 percent of total export value in the sample, compared with just 8 percent in 1995. Exports to the rest of Africa increased by 126 percent, but still represent a small fraction of total exports (3 percent). European destinations increased by 86 percent and represent the most important export destination with 60 percent of total export value. With a below average increase in exports of 26 percent during the period, regional exports to East African countries decreased in relative terms from 28 to 18 percent of total export value between 1995 and 1997. Did new firms enter the export market during the period of trade liberal- ization? If so, decisions to enter the export market would signal the credibil- ity of the trade reforms and suggest that entrepreneurs believed the future benefits of foreign sales outweighed the start-up cost of exporting. To under- stand this question in more depth, table A8.8 documents the transition pat- tern for the 177 firms providing complete export data in the full sample be- tween 1995 and 1997. Of the 177 firms, 41 (23 percent) exported in 1995, compared with 47 in 1997 (27 percent). This increase is due to nine firms entering the foreign market, while only three ceased exporting during the period, leaving a net entry of six firms. All nine of the new entrants export to Africa, three to the East African market exclusively, three to other African markets, and three to Africa and elsewhere. All three of the firms that ceased exporting during the period were active in the European market. Still, earlier export growth figures show the African market entrants and European market quitters left the overall market share of European destinations unchanged at 60 per- cent of total value, while the relative importance of African markets de- creased from 31 to 22 percent. The situation in Uganda is relatively similar to that in Cameroon follow- ing trade liberalization and devaluation. In Cameroon, between 1993 and 248 Bernard Gauthier 1995 few firms entered the export market: the net entry rate was only 5 per- cent (11 entrant and 1 quitter among a sample of 187 firms). Essentially, most of the entry and exit was in the African market among relatively small firms. None of the exporters specializing in the African market entered the Euro- pean market and few selling outside Africa began selling in African markets (see Tybout and others 1997). The pattern in Cameroon suggested that the two export markets were segmented, which appears to be the case in Uganda. Similarly, in Chad and Gabon, where trade and exchange rate reforms affected the relative profitability of different markets between 1993 and 1996, essentially no shift occurred between markets during the period. Entry and exit from the export market occurred only for exporters to Africa and among small firms. The net entry rate was negative in Gabon (3 firms exited and 1 firm entered among a sample of 80 firms), while in Chad 2 new firms started exporting within the regional free-trade area (among a sample of 54 firms) during the period. Thus, sunk costs for export market entry appeared rela- tively high (see Barba Navaretti, Faini, and Gauthier 1998). The source of the growth in exports and differences in behavior among categories of firms in Uganda can be understood more clearly through a de- composition analysis. Following Barba Navaretti, Faini, and Gauthier (1998); Sullivan, Tybout, and Roberts (1995); and Tybout and others (1997), nominal export growth is broken down by three categories of firms, incumbent ex- porters (continuous), new exporters (entrants), and quitters.5 The incumbent effect in table A8.9 is the contribution of continuous ex- porters to samplewide export growth. This is a weighted average of the growth in exports among firms that continue to sell abroad, the weights being their share in total exports. The net entry effect measures the effect of net changes in the number of exporters on growth, that is, the difference between the number of firms that enter the export market between periods 5. The following equation is used for decomposition: _____- = x {Qf1 )lQQitt- .I l\ QT{Qt X f _ ic\Q l Q.t l / e iq \Qf , / =Ql-Q,I t Ine-nt - I |Q-Q'tX\ -Q It I(ne-n1 -I it-c Qft-I + e ,t l g- Qf -l\ Qet Qt where 54W 1 denotes the share of total exports of the jth firm in year t - 1, nf refers to the number of exporting firms, Qf is output value sold in foreign markets during year t, and overbars denote period averages. The index i stands for the ith firm, e subscripts refer to firms entering the export market, q subscripts refer to firms that will quit the export market during the next period, and c subscripts to continuous exporters. Ag- gregates without these subscripts refer to the entire set of exporting firms (see Sullivan, Tybout, and Roberts 1995 for further details). Productivity and Exports 249 t - 1 and t, and the number of firms that cease exporting over the same interval. The turnover effect describes the effect on export growth of re- placing firms ceasing to export with firms entering the export market. Note that if quitters and entrants export the same value per firm, the turnover effect is zero. However, if large exporters leave foreign markets and small exporters enter them, turnover can lead to a decrease in total export value. Table A8.9 presents the results of the breakdown; it also shows the result of a similar decomposition performed on data from Cameroon and Gabon. Note that the incumbent effect, the net entry effect, and the turnover ef- fect in the table sum to nominal export growth. In addition, both the net entry effect and the turnover effect break down into their multiplicative com- ponents. For example, the net entry rate times the relative size of entrants equals the net entry effect. As observed from table A8.9, the net entry rate in Uganda was 12 percent between 1995 and 1997. However, new exporting firms exported, on average, only 35 percent as much per firm as incumbent exporters (see relative size), so the net entry effect amounted to only 4 percent of total growth in export value. Furthermore, the export value of entrants represented only 51 percent as much per firm as that of quitters over the sample period, so the replacement of exit- ing firms with entering firms tended to reduce total export value. Indeed, ac- cording to the observed pattern, some large-scale exporters dropped out of foreign markets, and the firms that replaced them exported less. Combining these entry and exit effects, virtually all the export growth in the sample in Uganda between 1995 and 1997 can be attributed to incumbent firms (95 percent). The Ugandan pattern is reminiscent of that observed in Cameroon and Gabon following trade liberalization in which no export boom was observed (table A8.9). Indeed, export growth in these countries occurred among incumbent firms and did not result from a surge of new entrants into the export market. This contrasts with export booms in Mexico, Morocco, and Columbia driven by a net entry of more than 50 percent of total growth of exports over a five-year period (Roberts and Tybout 1995). Tables 8.5 and A8.8 show that the growth in export value by existing pro- ducers in Uganda takes place in the European and other non-African coun- tries. As noted earlier, the few new producers in the export market represent regional exporters to Africa and tend to be smaller. This pattern may indicate the existence of significant start-up costs for the export market, especially to non-African countries. It thus appears that despite regional initiatives and the various trade re- forms implemented in Uganda since the late 1980s, a number of constraints on export development still exist. As reported in the Uganda survey, trade regulation is still perceived as a constraint by exporters and private busi- nesses considering the export market. Figure 8.1 shows that constraints on export increases principally relate to the cost of transportation, the lack of finance, and the quality of transportation due to poor infrastructure. For the 250 Bernard Gauthier Figure 8.1. Main Constraints to Increased Exports Cost of transportation Quality of transportation Lower profitability in exports Lack of finance 1 2 3 4 5 No Minor Moderate Major Severe obstacle Firm Size * Large F Medium E Small Source: Author's calculations based on the 1998 enterprise survey. larger exporters, transportation costs are the main issue; for the smaller ex- porters it is lack of finance. As figure 8.2 shows, the elements that prevent firms from starting to export are associated mainly with the cost of transpor- tation, lack of finance, and lack of information about export markets. Again, the larger firms tend to cite transportation costs as the main constraint, while smaller ones cite lack of finance. To further pursue the conjecture of significant start-up costs for the ex- port market, the export behavior of firms in the sample is explained using a simple model of the decision to export. This choice relates to evolution and level of total unit cost, controlling for previous export history and sector- based characteristics. Table A8.10 presents the results of two simple regres- sions performed on the Ugandan survey data and contrasts them with simi- lar regressions performed on Cameroonian firms. The dependent variable is a dummy that takes the value of one if the firm exported at the end of the period and zero otherwise. As shown in the first regression, (a), expressing the probability of export- ing in the last period as a function of cost and industry dummies, firms in Uganda with lower total unit costs are more likely to be exporters. Similar results were observed in Cameroon among a sample of 114 firms between 1992/93 and 1994/95 (Tybout and others 1997). These results imply that measures to reduce unit costs (through an increase in output price relative to intermediate price) should induce firms to enter the export market. Productivity and Exports 251 Figure 8.2. Main Constraints to New Exporters Cost of transportation Quality of transportation Lower profitability in exports Lack of finance _ ~ ~~I I II 1 2 3 4 5 No Minor Moderate Major Severe obstacle Firm Size * Large ] Medium 2 Small Source: Author's calculations based on the 1998 enterprise survey. The second regression, (b), accounts for export history and controls for initial cost and changes in average cost. Table A8.10 shows that in Uganda, as in Cameroon, unit cost at the beginning of the period and change in cost have the expected negative sign but are not statistically significant. The ini- tial exporter dummy is positive, however, as well as significant, indicating that firms that have already adapted their products and processes and estab- lished distribution channels and mechanisms to deal with custom authori- ties will be more likely to export at the end of the period. Still, there are also other firm characteristics that may remain important over time, such as loca- tion, foreign ownership status, managerial skills, and so forth, that relate to the firm's capacity to be an exporter. In short, substantial export growth occurred in Uganda during the pe- riod of trade liberalization. However, for the most part this growth can be explained by increased exports by firms already active in the export mar- ket. New entry is limited in terms of number of firms and relative impor- tance. This may suggest that, as observed in previous studies in Cameroon, Chad and Gabon, the barriers to export market entry remain high in Uganda. Indeed, only a small number of firms shifted toward the export market. The small number of entries likely reflects the existence of start-up costs. If such costs are high, firms are reluctant to redirect their operations toward foreign markets and incur costs for retooling, establishing distribution chan- nels, and researching foreign market conditions. As suggested in the small 252 Bernard Gauthier net entry effect, the reforms associated with trade liberalization may not have been enough to convince firms that incurring these costs is a wise business decision (see Barba Navaretti, Faini, and Gauthier 1998; Tybout and others 1997). Conclusions Using the detailed information collected in the 1998 survey of firms by the World Bank and the Ugandan Private Sector Foundation, this chapter shows that trade liberalization has been accompanied by significant growth in output and productivity in Uganda's private sector firms. Reallocation of resources toward the efficient export sector is apparent as export-oriented firms show almost 50 percent more growth, on average, in real output (unweighted) compared with nonexporters during the period. Furthermore, using several measures of productivity, a significant productivity gap ap- pears between exporters and firms producing exclusively for the domestic market. Exporters enjoy, on average, more than 60 percent more output per employee than nonexporters, while for the average firm the TFP index is more than 12 percent greater than for nonexporters. Total cost per unit of revenue is 46 percent less for exporters, and the index of technical efficiency is also 28 percent greater, on average, for exporting firms. In addition, ex- porters achieved significantly more productivity growth during the period compared with nonexporters, particularly in terms of TFP and efficiency growth. Whereas the export sector is growing in nominal value terms and relative to total industry sales, few new firms appear to be entering the market. Those who do so tend to be smaller than existing exporters and focus on the African market. The Ugandan pattern of export growth in the absence of an export boom is similar to that observed in Cameroon, Chad, and Gabon, where export growth following liberalization and foreign ex- change modification was due to incumbent firms rather than to a surge of new entrants into the export market. The absence of an export boom points toward the substantial role played by start-up costs in reducing firms' response to relative price changes and policy reforms. The findings suggest that trade liberalization and export ori- entation in Uganda can be enhanced by three types of policies, namely: * Policies that emphasize both increased specialization of incumbent producers in the export market and reduced barriers faced by new exporters. * Policies that identify and correct factors that prevent firms from in- vesting in new equipment, upgrading product quality, and research- ing foreign markets to the extent necessary for export market entry. * Policies that target deficiencies in public infrastructure and regulatory constraints, particularly those that add to production and transport. Productivity and Exports 253 Annex 8.1. Productivity Measures The four indexes of productivity used in this chapter are constructed as fol- lows. It is assumed that production relationships at the firm level can be char- acterized by a general function of the form Q =f(K, L, M, A), where Q is gross output, K is our measure of capital, L is labor, M is material, and A is a pro- ductivity index. Assuming a neoclassical Cobb-Douglas production function, a measure of total factor productivity is given by (A8. 1) In Aj>t = In Qijl - Kjl In 1i,, - sLj, In Lj,, - Sm,, hn htj, where i is the index of the ith firm (i = 1 ....... N) at time t in sector j, while sv, is the share of the vh input in total costs. Assuming that firms behave optimally and that factors are remunerated at the value of their marginal product, out- put elasticities could be associated with input shares. These output elastici- ties are calculated for labor and material inputs in each sectorj at each period as the current price ratios of total wages and materials to gross output value, in that sector and that year. Furthermore, assuming a constant return to scale, the capital output elasticity is measured as one minus the two other elastici- ties. (Table A8.11 presents these output elasticities by sector and by year used in the computation of the TFP index.) Growth rate in the productivity measure, which gives the variation in output not explained by input changes, is obtained through a second-order Tomqvist approximation given by (A8.2) Aln Aitn = Aln Q,- s , Aln Kif,- s LiAln Lfti- s MjAln Mit where s J is the share of the vth input in total costs in sector j, averaged over the two periods. The technical efficiency index is measured using the stochastic frontier production function methodology. In this model a production frontier is esti- mated that defines the maximum output achievable for a given set of inputs. The degree to which firms fail to reach the frontier is attributed to ineffi- ciency of production. Note that the stochastic element of the model allows some observations to lie above the frontier, which makes the model less vul- nerable to the influence of outliers than deterministic models. Assuming again a Cobb-Douglas production function, the frontier technology can be repre- sented in the following form: (A8.3) In Yij, = oxif + al In Kift + a2ln Lift + ot3ln Mi# + vi, + ui, where Ya, is the observed value of gross output of the ii firm (I = I....... N) at time t, K represents the replacement value of equipment, L the total number of employees, and M the value of intermediate inputs, in firm i in period t, and ai is a vector of technology parameters to be estimated. The compound disturbance is composed of two terms. The first, vit, is a random disturbance assumed to be distributed identically and independently 254 Bernard Gauthier across plants as N(O, a2). It represents factors such as luck, weather condi- tions, and unpredicted variation in inputs. The second, u11, is a firm-specific effect that reflects firm efficiency and management skills. Its distribution is one-sided, reflecting the fact that output must lie on or below the frontier. u,t is assumed to be independently and identically distributed across plants as the nonpositive part of a N(ji, 0y2) distribution truncated above at zero. Both v and u are assumed to be distributed independently of the exogenous vari- ables in the model. Following Aigner and Schmidt (1977), Jondrow and others (1982), and Battese and Coelli (1992), an estimate of the efficiency measure of the iIh firm at the t time period is given by (A8.4) eff, = exp(ai1). Table A8.12 presents the estimated coefficients of the production function using a random-effect estimator (generalized least square). Furthermore, la- bor productivity is measured as the logarithm of the ratio of output per em- ployee, while total unit cost is measured as the long-term average cost of production (A8.5) UCit = ln(LRC1,) - ln(Q,,), where LRCi, is the long-run cost of firm i at time t, as measured by the loga- rithm of the cost of capital, wages, and intermediate inputs, and Q1, is value of gross output of firm i at time t. Table A8.1. Distribution of Sample by Categories, 1997 Number Employment Gross output value a Category offirms Percentage of total Mean Percentage of total Mean Percentage of total By sector Commercial agriculture 16 11.5 58.6 6.2 451.4 1.7 Agroprocessing 29 20.9 225.4 43.1 5,972.7 40.9 Manufacturing 68 48.9 60.9 27.3 3,209.3 51.5 Construction 11 7.9 265.9 19.3 1,944.4 5.0 Tourism 15 10.8 42.8 4.2 233.2 0.8 By ownership Local 103 74.1 50.8 34.5 712.6 17.3 Foreign 21 15.1 263.6 36.5 2,432.3 12.1 Joint 15 10.8 294.1 29.1 19,937.9 70.6 By exporters Exporter 37 26.6 195.2 47.6 6,950.7 39.3 Nonexporter 102 73.4 78.0 52.4 1,631.1 60.7 By importer Domestic input intensive 94 67.6 111.0 68.7 1,883.1 41.8 Imported input intensive 45 32.4 105.5 31.3 5,479.6 58.2 By size Small 63 45.3 13.1 5.4 156.4 2.3 Medium 50 36.0 52.9 17.4 1,032.5 12.2 Large 26 18.7 450.4 77.1 13,927.6 85.5 Total 139 100.0 109.2 100.0 3,047.1 100.0 a. Gross output value in million Ugandan shillings. Source: Author's calculations based on the 1998 enterprise survey. 256 Bernard Gauthier Table A8.2. Summary Statistics of Variables (average, 1995-97) Sample standard Variable Sample mean deviation Minimum Maximum Output 2,491.6 9,312.2 1.0 73,933.3 Capital 2,875.9 13,082.9 0.1 99,933.3 Employment 101.1 254.6 3.0 1,866.7 Wage cost 177.3 651.3 0.2 6,154.5 Intermediate inputs 1,074.8 4,969.9 0.3 56,239.9 Foreign (%) 15.0 35.9 0.0 100.0 Share exported (%) 9.0 23.4 0.0 100.0 Note: Output, capital, wage cost, and intermediate inputs are in millions of constant 1995 Ugandan shillings. The number of firms is 139. Source: Author's calculations based on the 1998 enterprise survey. Productivity and Exports 257 Table A8.3. Real Output Growth by Firm Characteristics (cumulative percentage, 1995-97) Number Unweighted Weighted Interquartile Category offirms average average a Median range By sector Commercial agriculture 16 31.2 13.2 14.0 -3.8 to 41.5 Agroprocessing 29 26.2 11.6 11.7 -21.8 to 39.7 Manufacturing 68 27.9 49.1 9.6 -18.5 to 40.0 Construction 11 7.1 28.6 15.5 -27.8 to 32.2 Tourism 15 18.7 4.9 9.6 -24.8 to 47.5 By ownership Local 103 18.8 10.1 10.8 -23.1 to 38.3 Foreign 21 59.7 21.5 13.2 -7.7 to 88.7 Joint 15 21.4 36.4 15.4 0.4 to 43.6 By exporters Exporter 32 33.8 39.9 18.2 -7.8 to 47.6 Nonexporter 107 22.7 15.2 9.6 -21.8 to 37.9 By importer Domestic input intensive 97 21.6 20.6 13.3 -20.9 to 40.4 Imported input intensive 42 33.8 34.9 6.0 -19.2 to 38.3 By size Small 63 25.8 10.3 6.4 -23.2 to 39.1 Medium 50 19.5 -26.0 15.3 -13.2 to 41.2 Large 26 35.1 36.1 14.4 -7.6 to 47.5 Total 139 25.3 29.2 11.7 -19.2 to 40.4 a. Weighted by firms' gross output. Note: Categories in 1995. Source: Author's calculations based on the 1998 enterprise survey. 258 Bernard Gauthier Table A8.4. Productivity Levels by Firm Characteristics (weighted averages, 1995-97) Number Labor Totalfactor Characteristic offirms productivity Unit cost productivity Efficiency By sector Commercial agriculture 16 8.50 0.99 2.89 0.72 Agroprocessing 29 23.87 1.51 4.12 0.52 Manufacturing 68 42.35 2.04 1.40 0.28 Construction 11 7.00 1.02 3.29 0.19 Tourism 15 4.83 1.44 3.59 0.20 By exporters Exporter 32 29.45 1.88 3.37 0.46 Nonexporter 107 19.52 1.50 2.14 0.32 By importer Domestic input intensive 97 14.27 0.89 4.67 0.53 Imported input intensive 42 44.98 2.27 1.61 0.31 By ownership Local 103 37.16 1.76 1.87 0.16 Foreign 21 55.73 1.37 2.44 0.19 Joint 15 79.07 1.77 3.20 0.51 By size Small 63 10.28 2.27 2.26 0.21 Medium 50 20.93 1.16 2.35 0.25 Large 26 27.18 1.81 2.95 0.44 Total 139 24.23 1.71 2.83 0.40 Note: Observations are weighted by the firm gross output value. Labor productivity, unit cost, and total factor productivity in logs. Technical efficiency is a zero to one index, where one indicates full efficiency. Source: Author's calculations based on the 1998 enterprise survey. Productivity and Exports 259 Table A8.5. Real Productivity Growth (cumulative percentages weighted, 1995-97) Number Labor Total unit Totalfactor Characteristic offirms productivity cost productivity Efficiency By sector Commercial agriculture 16 3.4 -7.4 6.1 1.0 Agroprocessing 29 63.1 17.5 -19.9 -12.4 Manufacturing 68 23.3 -15.7 25.1 3.9 Construction 11 3.5 0.0 9.7 -1.9 Tourism 15 30.7 14.0 -8.2 -2.9 By exporters Exporter 32 71.4 -1.9 2.5 -1.7 Nonexporter 107 5.3 5.4 2.1 -7.5 By importer Domestic input intensive 97 42.0 7.8 -9.8 -1.0 Imported input intensive 42 25.6 -3.1 10.4 -6.4 By ownership Local 103 -8.2 11.7 2.1 -6.1 Foreign 21 6.2 -4.1 11.8 2.5 Joint 15 12.7 -0.9 0.6 -4.9 By size Small 63 2.6 9.3 3.1 -4.6 Medium 50 8.5 4.8 0.4 -11.2 Large 26 45.6 0.3 2.7 -2.8 Total 139 36.7 1.3 2.3 -4.2 Note: Observations are weighted by the firm's gross output value. Source: Author's calculations based on the 1998 enterprise survey. 260 Bernard Gauthier Table A8.6. Regression of Productivity Growth (dependent variable: AlnA) Labor Total unit Totalfactor Independent variable productivity cost productivity Efficiency Constant -0.002 0.234b -0.014b -0.109 (0.788) (2.223) (2.376) (0.838) Initial export 0.009 -0.093 0.006a 0.154a (0.959) (-1.291) (1.663) (1.687) Agricultural -0.010 -0.244a 0.014b 0.071 (0.724) (-1.828) (2.077) (0.555) Agroprocessing -0.002 -0.161 0.003 0.071 (0.135) (-1.308) (0.446) (0.462) Manufacturing -0.003 -0.149 0.Olla 0.070 (0.219) (-1.305) (1.765) (0.394) Construction -0.016 -0.045 0.007 0.129 (-1.102) (-0.225) (0.971) (0.934) Medium 0.004 -0.025 -0.001 -0.018 (0.721) (-0.297) (0.197) (0.332) Large -0.002 -0.052 0.002 -0.006 (0.187) (-0.566) (0.497) (-0.967) Sample size 278 278 278 278 R2 0.036 0.032 0.056 0.034 Note: Robust t-statistics in parentheses. Initial export is a dummy that takes the value of one if the firm exported at the beginning of the period and zero otherwise. Sector and size dummies take the value of one if the firm is in the category and zero otherwise. The service sector and the small-size dummnies are omitted. a. Significant at the 10 percent level. b. Significant at the 5 percent level. Source: Author's calculations based on the 1998 enterprise survey. Table A8.7. Summary Statistics of Exporters, 1995-97 Category 1995 1996 1997 Exporters in the sample (percent) 23.2 26.6 26.6 Number of firms 41 47 47 Export (percent) 39.2 36.9 37.9 Export/gross output value (unweighted) 9.1 9.8 10.1 Mean output Nonexporter 1.159 1.38 1.61 Exporter 4.479 5.49 6.05 Employment Nonexporter 60.4 62.7 74.1 Exporter 178.5 189.9 190.1 Note: Mean output in billions of Ugandan shillings. Number of firms is 177. Source: Author's calculations based on the 1998 enterprise survey. Table A8.8. Exporting Status, 1995 versus 1997 1997 status Other Joint East Africa Rest of Europe countries Joint Africa and 1995 status only Africa only only only Africa elsewhere Nonexporter Total s East Africa only 6 0 0 0 0 0 1 7 Rest of Africa only 0 5 0 0 0 0 0 5 Europeonly 0 0 1 2 0 0 2 5 Other countries only 0 0 0 3 0 2 0 5 Joint Africa 0 0 0 0 2 0 0 2 Joint Africa and elsewhere 0 0 0 0 0 17 0 17 Nonexporter 4 3 0 0 0 2 127 136 Total 10 8 1 5 2 21 130 177 Source: Author's calculations based on the 1998 enterprise survey. 262 Bernard Gauthier Table A8.9. Nominal Export Growth Decomposition, Selected African Countries (percentage) Category Uganda Cameroon Gabon Nominal export growth 92.9 82.5 115.2 Incumbent effect 94.9 85.4 115.4 Net entry effect 4.1 8.6 -23.7 Net entry rate 12.0 20.8 -10.0 Relative size 0.345 0.413 0.237 Turnover effect -6.14 -11.5 -1.53 Turnover rate 12.0 43.8 25.0 Size difference -0.512 -0.26 -0.06 Period 1995-97 1993-95 1993-96 Source: Author's calculations based on the 1998 enterprise survey for Uganda; Tybout and others (1997) for Cameroon; Barba Navaretti, Faini, and Gauthier (1998) for Gabon. Productivity and Exports 263 Table A8.10. Probit Models of the Decision to Export, Uganda and Cameroon, 1997 Uganda Cameroon a Independent variable (a) (b) (a) (b) Constant 1.512c 0.058 -0.433 -0.944 (0.513) (0.888) (-0.198) (-0.29) Ln (UC) initial n.a. -0.185 n.a. -0.325 n.a. (0.224) n.a. (0.336) Ln (UC) final -0.193 n.a. -0.343C n.a. (0.153) n.a. (0.115) n.a. ALn (UC) n.a. -0.349 n.a. -0.125 n.a. (0.527) n.a. (0.176) Exporter (initial) n.a. 3.277c n.a. 1.513c n.a. (0.644) n.a. (0.312) Agricultural -2.307c -2.075b n.a. n.a. (0.632) (1.153) n.a. n.a. Agroprocessing -2.045c -1.780b n.a. n.a. (0.588) (-1.018) n.a. n.a. Manufacturing -2.314c -1.263 n.a. n.a. (0.546) (0.926) n.a. n.a. Construction -8.289 -6.867 n.a. n.a. (21,124.9) (21,044.2) n.a. n.a. Wood product n.a. n.a. 0.036 -0.011 n.a. n.a. (0.307) (0.403) Textiles/apparel n.a. n.a. -0.009 -0.049 n.a. n.a. (0.301) (-0.407) Metal products n.a. n.a. -0.159 -0.19 n.a. n.a. (0.267) (0.372) Sample size 126 126 114 114 Log-likelihood function -52.250 -25.36 - - - Not available. n.a. Not applicable. Note: Standard error in parentheses. Ln (UC) is the log of unit cost, and ALn (UC) is the variation in the log of unit cost during the period. Exporter (initial) is a dummy variable that takes the value of one if the firm in an exporter in the first period and zero otherwise. Sector dummies take the value of one if the firm is part of the category and zero otherwise. The service sector dummy is omitted in Uganda, and the food sector dummy is omitted in Cameroon. Initial periods: Uganda 1995, Cameroon 1993. Final periods: Uganda 1997, Cameroon 1995. a. Cameroon: Export dummy in 1995. b. Significant at the 10 percent level. c. Significant at the 5 percent level. Source: Author's calculations based on 1998 enterprise survey for Uganda; Tybout and others (1997) for Cameroon. 264 Bernard Gauthier Table A8.11. Output Elasticities Used in Computing the TFP, 1995-97 Sector Factor 1995 1996 1997 Commercial agriculture Labor 0.212 0.206 0.198 Materials 0.244 0.228 0.221 Capital 0.543 0.566 0.581 Agroprocessing Labor 0.155 0.121 0.109 Materials 0.592 0.542 0.535 Capital 0.254 0.337 0.356 Manufacturing Labor 0.125 0.124 0.132 Materials 0.463 0.431 0.436 Capital 0.412 0.445 0.432 Construction Labor 0.193 0.165 0.163 Materials 0.388 0.379 0.326 Capital 0.419 0.456 0.511 Tourism Labor 0.179 0.153 0.173 Materials 0.240 0.197 0.218 Capital 0.581 0.650 0.610 Total Labor 0.153 0.139 0.141 Materials 0.435 0.401 0.400 Capital 0.413 0.460 0.459 Note: Output elasticities for labor and materials are calculated as the share expenditures on materials (including intermediate inputs, utilities, and fuel) and wages (including allowances, bonuses, and statutory payments), respectively The capital elasticity is calculated as one minus the other two elasticities (see text for details). Source: Author's calculations based on the 1998 enterprise survey. Table A8.12. Estimated Parameters of the Frontier Production Function Independent variable Random effect Constant 3.586a (8.755) Labor 0.151a (4.361) Capital 0.132a (5.036) Intermediate inputs 0.692a (30.728) Sample size 139 R2 0.857 Note: Standard error on parentheses. a. Significant at the 5 percent level. Source: Author's calculations based on the 1998 enterprise survey. Productivity and Exports 265 References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Aigner, Denis, and Peter Schmidt. 1977. "Formulation and Estimation of Sto- chastic Frontier Production Function Models." Journal of Econometrics 6(1): 21-37. Barba Navaretti, Giorgio, Ricardo Faini, and Bemard Gauthier. 1998. "Enter- prise Response to the Devaluation and Fiscal Reforms in Chad and Gabon." Union douani6re et economique de l'Afrique centrale (UDEAC) and World Bank, Africa Region, Washington, D.C. Processed. Battese George E., and Tim J. Coelli. 1992. "Frontier Production Functions, Technical Efficiency and Panel Data: With Application to Paddy Farm- ers in India." The Journal of Productivity Analysis 3(1-2): 149-65. Bauer, Paul W. 1990. "Recent Developments in the Econometric Estimation of Frontiers." Journal of Econometrics 46(October-November): 39-56. Bernard, Andrew B., and J. Bradford Jensen. 1999. "Exporting and Pro- ductivity." Yale School of Management, New Haven, Connecticut. Processed. Bigsten, Arne, Paul Collier, Stefan Dercon, Marcel Fafchamps, Bernard Gauthier, Jan Gunning, Jean Haraburema, Abena Oduro, Remco Oostendorp, Catherine Pattillo, Mans Soderbom, Francis Teal, and Albert Zeufack. 2000. "Exports and Firm-Level Efficiency in African Manufacturing." Working Paper Series 2000.16. Oxford University, Centre for the Study of African Economies, U.K. Forthcoming. "Are There Efficiency Gains from Exporting in African Manufacturing." In Augustin Kwasi Fosu, Saleh Nsouli, and Aristomene Varoudakis, eds., Policies to Foster Manufacturing Competi- tiveness in Sub-Saharan Africa. Paris: Organisation for Economic Co- operation and Development, Development Centre. Clerides, Sofronis, Saul Lach, and James Tybout. 1998. "Is Learning by Ex- porting Important? Micro-Dynamic Evidence from Colombia, Mexico, and Morocco." Quarterly Journal of Economics 113(3): 903-47. Elbadawi, Ibrahim A. 1992. "World Bank Adjustment Lending and Economic Performance in Sub-Saharan Africa in the 1980s: A Comparison of Early Adjusters, Late Adjusters, and Nonadjusters." Policy Research Work- ing Paper no. 1001. World Bank, Development Research Group, Wash- ington, D.C. Grossman, Gene M., and Elhanan Helpman. 1991. Innovation and Growth in the Global Economy. Cambridge, Massachusetts: MIT Press. 266 Bernard Gauthier Haddad, Mona. 1993. "How Trade Liberalization Affected Productivity in Morocco." Policy Research Working Paper no. 1096. World Bank, De- velopment Research Group, Washington, D.C. Harrison, Ann E. 1994. "Productivity, Imperfect Competition, and Trade Reform: Theory and Evidence." Journal of International Economics 36(1-2): 53-73. Husain, Ishrat, and Rashid Faruqee, eds. 1994. Adjustment in Africa: Lessons from Country Case Studies. World Bank Regional and Country Studies. Washington, D.C.: World Bank. Jondrow James, C. A. K. Lovell, I. S. Materov, and P. Schmidt. 1982. "On Esti- mation of Technical Inefficiency in Stochastic Frontier Production Model." Journal of Econometrics 19(2-3): 233-38. Kraay, Aart. 1999. "Exportations et performances economiques: etude d'un panel d'entreprises chinoises." Revue d'Economie du Developpement 7(1- 2): 183-207. Lovell, C.A.K., J. Defourny, and A.G.M. N'Gbo. 1992. "Variation in Produc- tive Efficiency in French Workers Cooperative." Journal of Productivity Analysis 3(1-2): 103-117. Matin, Kazi M. 1992. "Openness and Economic Performance in Sub-Saharan Africa: Evidence from Time-Series Cross-Country Analysis." Policy Research Working Paper no. 1025. World Bank, Development Research Group, Washington, D.C. Nishimizu, Mieko, and Sherman Robinson. 1984. "Trade Policies and Pro- ductivity in Semi-Industrialized Countries." Journal of Development Economics 16(1-2): 177-206. Pack, Howard. 1988. "Industrialization and Trade." In Hollis Chenery and T. N. Srinivasan, eds., Handbook of Development Economics. Amsterdam: North-Holland. Roberts, Mark, and James R. Tybout. 1995. "An Empirical Model of Sunk Costs and the Decision to Export." Policy Research Working Paper no. 1436. World Bank, Development Research Group, Washington, D.C. Processed. . 1997. "Producer Turnover and Productivity Growth in Developing Countries." The World Bank Research Observer 12(1): 1-18. Rodrik, Dani. 1988. "Imperfect Competition, Scale Economies and Trade Policy in Developing Countries." In Robert E. Baldwin, ed., Trade Policy Is- sues and Empirical Analyses. National Bureau of Economic Research Conference Report Series. Chicago: University of Chicago Press. Schmidt, Peter, and Robin C. Sickles. 1984. "Production Frontiers and Panel Data." Journal of Business and Economic Statistics 2(4): 367-74. Productivity and Exports 267 Sullivan, Theresa, James R. Tybout, and Mark Roberts. 1995. "What Makes Exports Boom? Evidence from Plant-Level Panel Data." World Bank, Africa Region, Washington, D.C. Processed. Tybout, James R. 1992. "Linking Trade and Productivity: New Research Di- rections." World Bank Economic Review 6(2): 189-211. Tybout, James R., Bernard Gauthier, Giorgio Barba Navaretti, and Jaime DeMelo. 1997. "Firm-Level Response to the CFA Devaluation in Cameroon." Journal of African Economies 6(1): 3-34. UNCTAD (United Nations Conference on Trade and Development). 1998. Trade and Development Report. Paris. World Bank. 1994. Adjustment in Africa: Reforms, Results and the Road Ahead. New York: Oxford University Press. Part IV Government Performance from a Beneficiary Perspective 9 A Quest for Revenue and Tax Incidence Duanjie Chen, John Matovu, and Ritva Reinikka One of the main accomplishments of the Ugandan government in the 1990s was the removal of massive implicit taxation on exports. This chapter as- sesses how tax policy evolved after export taxation was eliminated, how the government was able to meet its revenue needs in a less predatory fash- ion, and how these policies affected households and firms. Because of the past predatory taxation and prolonged conflict, government revenue was only 5 percent of gross domestic product (GDP) when Uganda began its recovery in 1986. Simultaneously, the needs for public spending on social services and infrastructure were massive to support impoverished house- holds' efforts to increase their production, consumption, and welfare, and to encourage enterprises to invest and diversify. This led policymakers to pursue a rapid increase in domestic revenue and a corresponding increase in public services. Rebuilding the government's revenue base was an es- sential feature of Uganda's economic recovery. Institution building for tax administration resulted in the semiautonomous Uganda Revenue Author- ity (URA), established in 1991, inspired by Ghana's example. Because the URA is not part of the civil service, it can offer higher pay and attract more qualified staff. Consequently, domestic revenue more than doubled in real terms during the first half of the 1990s, and by 1996 was 11.3 percent of GDP (table 9.1).1 In conjunction with large aid inflows, this allowed public expenditure to grow far more rapidly than GDP (which itself was growing rapidly) without destabilizing the economy. 1. In Uganda GDP includes the nonmonetary (subsistence) sector. Domestic rev- enue was 7.6 percent of monetary GDP in 1986 and increased to 15.5 percent of mon- etary GDP in 1999. 271 Table 9.1. Central Govemment Revenues, 1991/92-1998/99 Revenue category 1991/92 1992/93 1993/94 1994/95 1995/96 1996/97 1997/98 1998/99 U Sh millions Taxes on income and profits 23,600 40,900 53,000 77,200 82,600 102,200 124,750 170,040 Excise taxes 15,000 18,800 40,500 50,600 217,000 301,500 304,050 322,870 Petroleum products n.a. n.a. n.a. n.a. 149,900 197,500 188,270 193,210 Other n.a. n.a. n.a. n.a. 67,100 104,000 115,780 129,660 Taxes on goods and services 55,500 75,100 92,800 153,000 188,700 209,600 247,200 298,600 Value added tax n.a. n.a. n.a. n.a. n.a. 209,600 247,200 298,600 Sales tax 43,400 62,900 75,300 128,700 162,300 n.a. n.a. n.a. Commercial transaction levy 5,400 9,600 15,300 22,300 25,600 n.a. n.a. n.a. Other 6,700 2,600 2,200 2,000 800 n.a. n.a. n.a. Taxes on international trade 78,600 124,230 152,500 205,500 100,500 74,800 78,400 96,530 Import duties 76,600 124,230 152,500 176,700 75,900 72,300 78,050 96,480 Export duties (coffee) 2,000 0 0 28,800 24,600 2,500 350 50 Total tax revenue 172,700 259,030 338,800 486,300 588,800 688,100 754,400 888,040 Total nontax revenue (fees and licenses) 13,295 22,404 25,063 40,400 38,400 43,300 47,060 62,700 Total revenue 185,995 281,434 363,863 526,700 627,200 731,400 801,460 950,740 (table continues onfollowing page) Table 9.1 continued Revenue category 1991/92 1992/93 1993/94 1994/95 1995/96 1996/97 1997/98 1998/99 CPI, annual average (1991 = 100) 195 253 270 287 308 332 352 351 Real domestic revenue 95,415 111,075 134,725 183,795 203,612 220,098 227,974 270,866 GDP at factor cost 2,588,800 3,625,938 4,069,439 4,922,397 5,565,388 6,022,953 7,104,303 7,887,246 GDP at market prices 2,745,491 3,870,388 4,400,270 5,367,456 6,122,089 6,663,235 7,791,426 8,647,425 Monetary GDP at factor cost 1,794,145 2,481,870 2,890,811 3,619,057 4,213,995 4,717,950 5,467,267 6,119,562 Percentage share of , total domestic revenue Taxes on income and profits 12.7 14.5 14.6 14.7 13.2 14.0 15.6 17.9 Excise taxes 8.1 6.7 11.1 9.6 34.6 41.2 37.9 34.0 Petroleum products n.a. n.a. n.a. n.a. 23.9 27.0 23.5 20.3 Other n.a. n.a. n.a. n.a. 10.7 14.2 14.4 13.6 Taxes on goods and services 29.8 26.7 25.5 29.0 30.1 28.7 30.8 31.4 Value added tax n.a. n.a. n.a. n.a. n.a. 28.7 30.8 31.4 Sales tax 23.3 22.3 20.7 24.4 25.9 n.a. n.a. n.a. Commercial transaction levy 2.9 3.4 4.2 4.2 4.1 n.a. n.a. n.a. Other 3.6 0.9 0.6 0.4 0.1 n.a. n.a. n.a. (table continues onfollowing page) Table 9.1 continued Revenue category 1991/92 1992/93 1993/94 1994/95 1995/96 1996/97 1997/98 1998/99 Taxes on international trade 42.3 44.1 41.9 39.0 16.0 10.2 9.8 10.2 Import duties 41.2 44.1 41.9 33.5 12.1 9.9 9.7 10.1 Export duties (coffee) 1.1 0.0 0.0 5.5 3.9 0.3 0.1 0.0 Total tax revenue 92.9 92.0 93.1 92.3 93.9 94.1 94.1 93.4 Total nontax revenue (fees and licenses) 7.1 8.0 6.9 7.7 6.1 5.9 5.9 6.6 Total revenue 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Real change in total revenue -4.5 16.4 21.3 36.4 10.8 8.1 3.6 18.8 Total revenue as share of GDP at factor cost 7.2 7.8 8.9 10.7 11.3 12.1 11.3 12.1 Total revenue as share of GDP at market prices 6.8 7.3 8.3 9.8 10.2 11.0 10.3 11.0 Total revenue as share of monetary GDP at factor cost 10.4 11.3 12.6 14.6 14.9 15.5 14.7 15.5 n.a. Not applicable. CPI Consumer price index. Source: Ministry of Finance, Planning, and Economic Development data. A Questfor Revenue and Tax Incidence 275 The policy of rapidly increasing public revenue presented a tradeoff for the economic liberalization program. In particular, it curtailed the scope for trade reform. The coffee export tax was abolished early on, but tariffs and other import taxes were retained, initially at a high level because of the quest for revenue. Even by 1996, import taxes (including petroleum) still accounted for more than half of total revenue. As argued in chapter 2, the Ugandan government initially did not recognize the close relationship between export taxes and import taxes, specifically that import taxes are ultimately borne by export producers, in particular, coffee farmers. Hence, the switch from ex- port taxation to import taxation probably achieved less than expected in terms of export orientation and diversification. For most of the 1990s the government had an explicit target of increasing revenue by one percentage point of GDP each year. This target was not backed, however, by a concrete strategy and administrative measures to encourage such growth. Over time the government increasingly relied on ad hoc in- creases in tax rates-particularly fuel taxes-to achieve the revenue target, without specific knowledge of the supply-side effects. While import tariffs protected the producers oriented toward the local market, high tax rates gen- erally encouraged seeking and granting of firm-specific exemptions and tax holidays, adversely affecting competition. The revenue target was met in the early 1990s, but since 1996 the recovery has nearly stalled. The rapid increase in domestic revenue may not have been the best strat- egy. The corresponding expansion in government expenditure may not have paid off well at the margin in terms of service delivery (see chapter 11 in this volume), while the cost of taxation was probably high in terms of bureau- cratic control and opportunities for corruption (see chapter 10 in this vol- ume). Thus resource misallocation and foregone private investment likely undermined growth and the prospects for increasing public revenue in a more sustainable manner. As discussed in chapters 2 and 3, Ugandan policymakers have demon- strated a remarkable ability to readjust economic policy when necessary. During the second half of the 1990s, import tariffs were reduced consider- ably and several tax reforms were implemented, including the value added tax (VAT) and income taxation. Consequently, the tax system is gradually being transformed from high tax rates and selective incentives toward lower rates and more standard provisions across the board. The challenges now are to build a tax culture of compliance and administration that the public per- ceives as fair and efficient. This chapter takes first a closer look at the policy of rapidly increasing revenue and recent tax reforms. It then uses household survey evidence (see appendix A at the end of the book for details) to iden- tify which taxes are progressive and whether tax reforms helped the poor or left them worse off. Similarly, it examines business taxation to answer three questions: What is the actual tax burden on firms' capital investment and the overall cost of production across various industries? How does this burden compare to that in Kenya and Tanzania, which compete for the same foreign 276 Duanjie Chen, John Matovu, and Ritva Reinikka investment? How does poor compliance and tax administration affect tax incidence on the enterprise sector? Revenue Trends and Tax Reforms In the 1990s, import duties levied on both consumer goods and raw materi- als accounted for the largest share of central government revenues, reflecting a policy switch from export to import taxation (table 9.1). The share of im- port duties reached its highest level of 44 percent in 1992/93. Including sales tax on imported goods, import taxation has been well over half of total rev- enues. Subsequent import liberalization produced a more uniform tariff tax structure and reduced the level and dispersion of tariffs (from 10 to 350 per- cent in 1992/93 to 0 to 15 percent today for nonregional trade). Petroleum is the most important taxable item; since 1991/92 it has contributed more than 30 percent of total revenues, including import duty and excise tax. The ad valorem rate for fuels ranges from 100 percent to more than 200 percent for paraffin, diesel, and petroleum products, with an estimated weighted aver- age rate of 174 percent (table A9.6). A high fuel tax is common in many in- dustrial economies, particularly for environmental reasons. What makes it problematic in Uganda is that Kenya and Tanzania's much lower fuel tax rates result in substantial smuggling (table A9.9). A tax on coffee exports was reintroduced as a stabilization measure dur- ing the coffee boom in 1994/95 (see chapter 3 in this volume). A year later, international coffee prices declined to less than the price threshold set for the coffee tax. The tax rate of 32 percent above the threshold level was re- vised to 25 percent in 1995/96, and later abolished. In principle, a coffee tax could shift forward to buyers of the export good, or it could shift to its producers in the form of a reduced producer price. Elastic supply of and inelastic demand for the taxed commodity usually shift the tax to consum- ers, but with inelastic supply, elastic demand, and a competitive producer market, the tax burden falls on producers. Small developing countries that produce primary commodities are usually price-takers internationally. Hence, even though the coffee stabilization tax was not targeted at coffee farmers, it did indirectly tax them. The second largest source of revenue is the VAT, which followed the sales tax on goods and the commercial transaction levy on services. Introducing the VAT was regarded as crucial to the government's longer-term strategy to broaden the tax base, improve compliance, and increase revenue collection. By 1991/92, taxes on goods and services constituted 30 percent of total rev- enues, and have remained at that level since then. The introduction of the VAT in 1996 was accompanied by a small temporary decrease in its share of revenue compared with the sales tax, which has since been recovered. While sales taxes varied considerably by commodity (table A9.1), the VAT has a single rate of 17 percent. To take into account equity concerns, some goods considered to have substantial budget shares in the consumption basket of A Questfor Revenue and Tax Incidence 277 the lowest income earners were exempted, including unprocessed foodstuffs, social services, passenger transport services, and fuels.2 Before liberalization, the government instituted a range of tax incentives in the early 1990s to compensate firms that undertook major investment projects for prevailing distortions. The 1991 Investment Code included project- based licensing of large investments. A typical license entitled its holder to a full or partial income tax holiday and duty exemptions on imported inputs (see chapter 2 in this volume). As distortions were later reduced by the eco- nomic liberalization program, the government implemented an income tax reform in 1997 to streamline the tax incentive system. (Duty-free treatment of imported capital goods for all firms had been introduced in 1995.) The objective of the income tax reform was to broaden the tax base, increase ad- ministrative simplicity, and encourage long-term investment and technol- ogy transfer. Table A9.6 summarizes key features of the pre- and post-1997 business taxation, including capital taxes, indirect taxes applicable to inputs, and payroll taxes.3 Finally, the graduated personal tax (GPT) is a major source of locally raised revenue, although its revenue yield is limited. The legal base of the GPT is income, but in practice it is an income tax, a wealth tax, or a poll tax, depending on the district and subject of taxation (Bahl 1997). Because the threshold is about half the lower threshold of the central government in- come tax on individuals, the tax is regressive over much of the income scale.4 The GPT is assessed and collected as a presumptive tax on almost all tax subjects in rural areas and the self-employed in urban areas. In its present form the GPT represents an expensive form of tax administration and, be- cause presumptive assessment can imply subjectivity and arbitrariness, is often perceived as unfair. 2. Excise taxes are of two types: those levied on imports and those levied on locally produced goods that are often considered as luxury items. The share of nonpetroleum excise duties in total revenues was only 8.1 percent in 1991/92, and increased to about 14 percent in 1998/99. Income taxes are composed of the corporate (profit) tax and pay-as-you-earn personal income tax. The overall contribution of in- come and profit taxes to revenue remained relatively modest, averaging about 14 percent throughout the 1990s. Only since 1997/98 has this contribution increased noticeably, likely reflecting the income tax reform of 1997. 3. Investment analysis in chapter 7 relates the probability of a firm to invest and its investment level to several variables, including firm characteristics, changes in demand, and profits. Using the same flexible accelerator model of investment, tax exemptions are added in the regression. They enter negatively but are insignificant. Hence, despite their important role as policy instrument, tax exemptions do not seem to explain either the probability that a firm invests or the level of investment of Ugan- dan firms in 1996-97. 4. In 1997 the GPT had 36 rate brackets, with a specific rate up to a maximum payment of U Sh 80,000 at the annual income level of U Sh 820,000. 278 Duanjie Chen, John Matovu, and Ritva Reinikka Method and Data for Tax Incidence Analysis In examining the impact of tax reforms on households, particularly whether tax reforms have made the poor better or worse off, this chapter applies the welfare dominance analysis (see annex 9.1). The method is based on the con- centration curves that measure the fraction of total expenditure on a com- modity ascribed to different income groups when ordered according to the level of income or consumption expenditure.5 In this analysis, concentration curves plot households from poorest to wealthiest on the horizontal axis against the cumulative proportion of taxes paid by households (figures A9.1 and A9.2). In this method, the more a concentration curve moves away from the 45-degree straight line, the more progressive is the tax. For one tax to dominate the other, the difference in their concentration curves must be non- negative over the whole range of incomes. When the Gini coefficient for a given tax is greater than the Gini coefficient of per capita expenditure, then the tax is considered to be progressive. Similarly, the comparison of Gini co- efficients before and after the tax reforms indicate how the progressivity of taxation has changed over time (tables A9.2 and A9.3). Several assumptions are made. The factors that produce the incomes are assumed to pay the associated direct taxes, while households that consume the taxed items are assumed to pay the indirect taxes. Thus smokers pay taxes on tobacco and households that use paraffin pay taxes on paraffin. Im- port duties are more difficult to capture from a household survey, because of the lack of differentiation between domestically produced and imported con- sumer goods. The prices of all goods for which imports compose a large share of the market are assumed to go up by the amount of the tariff when it is levied. Finally, most of the analysis relies on statutory tax rates rather than on any estimates of taxes actually paid. The data on household expenditures are obtained from the 1992/93 inte- grated household survey (see appendix A at the end of the book). Table A9.1 shows the various consumer goods and their corresponding tax rates before and after reforms. Most import duties were reduced considerably by 1995. This table also shows the various rates of the 1992 sales tax, which was re- placed by a uniform 17 percent VAT rate in 1996 (some goods are zero-rated). Uganda uses a different import tariff regime for regional and external trade. The different tax rates on various products depending on the import source 5. The concentration curve is similar to the Lorenz curve, which is a graphical presentation of inequality. For the Lorenz curve household expenditures are arranged in ascending order, and the cumulative share of total expenditures is plotted against the cumulative share of population. For complete equality the Lorenz curve would be a straight line; it becomes more curved when inequality rises. The Gini coefficient is the ratio of the area between the straight line and the Lorenz curve to the total area under the straight line. A Questfor Revenue and Tax Incidence 279 could be identified, but identifying the origin of the imported product con- sumed by a particular household would be difficult. Therefore, no attempt is made to calculate import duties based on the countries of import, but the external regime of import duties is used. For the tax incidence on firms, the marginal effective tax rate (METR) on investment and production costs is chosen as the quantitative indicator. The key assumption underlying the METR concept is that a profit-maximizing firm invests (or produces) as long as the after tax marginal revenue from its investment (production) exceeds the marginal cost. While the marginal rev- enue is not easily observable in practice, data on the marginal cost can be obtained. For example, when estimating the METR on capital, the marginal cost is the sum of the financing cost of investment and the economic depre- ciation rate, adjusted for all relevant taxes and tax allowances. Hence, the marginal effective tax rate measures the impact of a tax system on an incre- mental unit of capital investment or business activity (see annex 9.2).6 The METR incorporates the effects of both statutory tax rates and related tax incentives (such as tax depreciation, tax credit, tax deductibility, and tax holidays) as well as various industry-specific and economywide factors in- teracting with these taxes (including financial costs, inflation, and capital struc- ture). Because of this interaction, the effective tax rate can vary by industry or tax jurisdictions under the same tax regime. The difference in the METR across various investors or sectors quantifies the tax bias at the margin and, other things being equal, indicates how tax policy is likely to affect invest- ment decisions. In a low-income country like Uganda where the tax administration is rela- tively weak, the actual tax incidence is likely to differ from the formal tax structure. While the analysis can be extended to compare the impact of the formal tax structure across industries or jurisdictions, obtaining adequate information about actual administrative practices and detailed industrial parameters is more difficult. Thus the issue is not so much whether the METR method can handle the real world, but how well analysts understand the real world and are able to quantify the differences between the formal tax struc- ture and actual tax collection. Although the analysis presented in this chap- ter is based on Uganda's formal tax system, it uses actual firm-level data for key nontax parameters (see appendix B at the end of the book). The capital structure by industry was obtained from the URA taxpayer database, while the cost structure by industry was estimated from 1992 input-output tables 6. For example, if the gross-of-tax rate of return to capital is 15 percent and the net-of-tax rate of return is 12 percent, the marginal effective tax rate on capital is 25 percent if the after tax return is used as the denominator, or 20 percent if the before tax return is the denominator. This study uses the former convention, as it is more convenient when calculating the METR on the cost of production. 280 Duanjie Chen, John Matovu, and Ritva Reinikka (Republic of Uganda 1995). Firm survey evidence is also used to explore the effect of compliance and tax administration on the METR. Tax Incidence on Households This section presents the tax incidence analysis on .households both before and after tax reforms. It first explores the extent to which the overall tax sys- tem is progressive or regressive. To determine this, all the taxes paid by the household are aggregated and the (extended) Gini coefficients of the aggre- gate taxes are compared with those of total household expenditure. The re- sults presented in table A9.2 indicate that, by and large, the tax structure was progressive before reforms. Concentration curves for the main tax categories shown in figure A9.1 confirm this. Most individual tax categories were also found to be progressive before reforms, with the exception of the excise tax and the graduated personal tax. In particular, the excise tax on paraffin- which is heavily consumed by the poor-was highly regressive. By attach- ing a higher weight to commodities consumed by the poor (parameter v in tables A9.2 and A9.3), excise taxes on paraffin became even more regressive. Pay-as-you-earn was the most progressive tax. This tax is levied on formal sector employees and hence tends to be concentrated among the better-off. Because the minimum threshold to be liable for this tax is relatively high, it exempts the lowest income groups. Import taxes were the second most pro- gressive category, followed by the sales tax. The incidence analysis shows that after reforms, the overall tax system remained progressive (table A9.3). The results are consistent with findings of similar studies on Ghana and Madagascar (Younger 1996; Younger and oth- ers 1999). The Gini coefficients before and after reforms confirm that substi- tution of the VAT for the sales tax does not necessarily worsen the welfare of the poor (tables A9.2 and A9.3).7 The pay-as-you-earn tax remains the most progressive tax after reforms. However, determining conclusively which of the other taxes (VAT, import tax, and excise tax) dominates after the reforms is not possible. Some important changes in the Gini coefficients occurred after the tax re- forms (table A9.3). The coefficient of the aggregate excise taxes shows that these taxes were made more progressive by the reform. Import duties, however, be- came more regressive. The Gini coefficient of the coffee stabilization tax is con- siderably below that of other taxes, implying that an export tax on a primary commodity can be highly regressive. The burden lies heavily on rural produc- ers as exporters shift the tax to them. As confirmed by the test statistic (table A9.5), the coffee stabilization tax dominates all other taxes. 7. While the results show that the VAT is a progressive tax, it is inconclusive from the welfare dominance test whether the VAT is much more progressive than a sales tax (tables A9.4-A9.5). A Questfor Revenue and Tax Incidence 281 To raise public revenue, petroleum products have been heavily relied on, as their demand is considered inelastic. Applying the statutory tax rates di- rectly to petroleum consumption shows that petroleum taxes (apart from those on paraffin) are very progressive. This incidence analysis, however, ignores the indirect or intermediate effects of petroleum taxation on the other sectors. These indirect price effects of petroleum taxes can be obtained from the input-output table and assigned to the corresponding commodities in the household survey (Republic of Uganda 1995). Two types of taxes are con- sidered. First, import duties levied on petroleum products are imputed on all other sectors. Second, the excise tax has a strong effect on prices in the transport sector. When these effects are taken into account, petroleum taxes are no longer as progressive as in the initial analysis.8 Marginal Effective Tax Rate for Firms This section covers estimates of the METR on capital and cost of production for Ugandan firms operating in commercial agriculture, agroprocessing, manufacturing, construction, transportation, communication, and tourism. The METR on capital includes four types of assets (buildings, machinery, inventories, and land), two different tax regimes (the pre- and post-1997 tax system), and three tax codes (regular taxable, tax holiday, and small firms). Various policy options are also simulated.9 The METR estimation on the cost of production includes three key inputs: capital, labor, and fuel. The estimation of the METR is not only sensitive to tax policy, but also to the choice of nontax parameters, such as macroeconomic indicators and industry-specific parameters. These include inflation rate, interest rate, debt- to-assets ratio, economic depreciation rate, capital structure, and cost struc- ture (table A9.7). While inflation and the interest rate are usually the same for all industries within an economy, the other parameters vary by sector. De- preciable assets used by different industries can have a different useful life and replacement cost, which results in a different economic depreciation rate. Capital structures also vary by industry. Compared with tourism, for example, the capital structure in manufacturing is more intensive in machinery and inventories and less intensive in buildings.10 8. While the excise tax on gasoline-without taking into account indirect effects on other sectors-gives progressive Gini coefficients of 0.899 to 0.992 (for v = 2, 4, 6, 8, 10), taking into account indirect effects on other sectors yields less progressive results of 0.436 to 0.726 for the same values of v. The higher the value of the parameter v, the higher weight is attached to goods consumed by the poor. 9. Overall, the discussion focuses on large and medium-size firms (firms that have more than 20 employees): small firms are discussed as a special case. 10. Chen and Reinikka (1999) provide a discussion on nontax parameters as well as sensitivity analyses for the base case assumptions. 282 Duanjie Chen, John Matovu, and Ritva Reinikka METR on Capital Capital investment generally involves two categories of capital, depreciable and nondepreciable assets. These categories can be further divided into build- ings and machinery (depreciable), and inventory and land (nondepreciable). Capital taxes in Uganda are summarized in box 9.1. As mentioned earlier, Box 9.1. Capital Taxes Capital taxes include company income tax (and related tax allowances), per- sonal income taxes on investment income, presumptive tax on small businesses, municipal property taxes, and import duties applicable to capital goods. The company income tax is 30 percent. Firms are allowed to carry over their operat- ing losses indefinitely, except for those firms that enjoy a tax holiday. Two types of deductions from the company income tax are allowed: the initial investment allowance and the annual depreciation allowance. Investment in machinery and plant is strongly encouraged through tax incentives; such investment is entitled to both the initial allowance and the annual depreciation allowance available to all taxable firms. The initial allowance for investment in machinery and plant (except for vehicles) is 50 percent in five main industrial locations-Kampala, Entebbe, Namanve, Jinja, and Njeru-and 75 percent elsewhere in Uganda. The annual depreciation rate is 40, 35, 30, and 20 percent for the four different classes of machinery and plant, respectively. For industrial buildings, there is no initial allowance, and the annual depreciation rate is much smaller (5 percent) than for machinery. However, expenditures on acquiring farm structures are entitled to a higher annual depreciation allowance of 20 percent. Before the 1997 tax reform, the annual depreciation rate for structures was 4 percent, while ma- chinery and plant were divided into three classes, with the annual deprecia- tion rate at 50, 40, and 20 percent, respectively. The classification of machinery was also changed significantly in 1997. Before the income tax reform, a holder of the certificate of investment incen- tives was exempted from company income tax, withholding tax, and tax on dividends for a certain period, depending on the total value of the investment. New tax holidays were repealed in 1997, and interest and dividends are both taxed at 15 percent. A presumptive tax on small businesses was introduced in 1997, while previously most small firms had no tax obligations. Instead of pay- ing a regular income tax, a small firm with annual turnover below U Sh 50 mil- lion is subject to a presumptive tax up to 1 percent of its gross tuniover, unless it opts to file the regular income tax return. This tax is final and no deductions for capital expenditure or other business expenses are allowed. Finally, municipali- ties impose a property tax on immovable property or buildings, but not on va- cant land. For example, in Kampala the property tax rate is 10 percent on the ratable value, which is obtained by deducting maintenance cost from the gross value, or the rent one may expect to receive from the property. A Questfor Revenue and Tax Incidence 283 capital investment by asset type varies by industry. Consequently, even if a certain type of asset incurs the same METR, the different capital structure by industry will result in a different aggregate METR on capital across indus- tries. The cost structure by input varies also by industry. ASSET TYPE. The base case is the 1997 regular taxable firm. As table 9.2 shows, machinery is the lowest taxed asset in Uganda. This is mainly because of the generous initial allowance of 50 percent, along with the annual depreciation allowance that begins the first year. In fact, the METR on machinery is nega- tive in several industries, which indicates a tax subsidy." The transportation sector, however, incurs a relatively high METR of 17 percent on machinery, mainly because vehicles are not eligible for the initial allowance. Inventories are the highest taxed asset, with an METR of 45 percent. This is mainly because of the first-in-first-out (FIFO) accounting method used by most Ugandan firms, combined with a positive inflation rate. Build- ings, except those used by commercial agriculture, are taxed the second highest (an METR of more than 40 percent), mainly because of the local property tax on buildings, combined with less generous tax depreciation allowances. Because of a more generous depreciation allowance for farm works, buildings used in commercial agriculture bear a low tax burden (an METR of 12 percent. Structures used by the construction industry incur a higher METR than other sectors, mainly because of a higher economic de- preciation rate. Finally, nonfarm land is also subject to the local property tax, resulting in a relatively high METR (42 percent), while farmland incurs a significantly lower METR (28 percent). As shown in table 9.2, while nondepreciable assets such as inventories and land are taxed at the same level across industries, depreciable assets, such as buildings and machinery, are taxed unevenly. This is because depre- ciable assets used by different industries have different useful lives and dif- ferent tax depreciation allowances. For a given depreciable asset, the wider the gap between the economic and tax depreciation rate, the higher the METR. INDUSTRIES. The aggregate METR for each industry is simply a propor- tional difference between the weighted average of the before tax and after tax rate of return by asset, based on the industry-specific capital structure. Obviously, the larger the share of the assets that are highly taxed, the higher the industry's aggregate METR. As shown in table 9.2, tourism incurs the highest METR (39 percent) in the base case. This is mainly a result of its high capital weight in buildings, the second highest taxed asset. Manufac- turing incurs the second highest METR (33 percent), mainly because the 11. As a firm is taxed as a whole rather than by asset type or at the margin, this tax subsidy on machinery can be thought of as reducing the tax on income generated by other type of investment. Table 9.2. Marginal Effective Tax Rate on Capital for Ugandan Firns (percent) Commercial Category agriculture Agroprocessing Manufacturing Construction Transportation Communications Tourism Regular taxable case, 1997 (interindustry dispersion: 3.9) Buildings 11.7 43.4 44.3 48.4 42.7 44.9 43.0 Machinery -0.3 0.6 1.4 -0.4 16.6 -1.1 2.9 Inventory 45.2 45.2 45.2 45.2 45.2 45.2 45.2 M Land 27.5 41.7 41.7 41.7 41.7 41.7 41.7 a Aggregate 26.2 23.2 32.9 23.5 20.9 31.0 39.2 Regular taxable case, pre-1997 (interindustry dispersion: 3.4) Buildings 30.4 46.5 47.6 51.9 45.9 48.2 46.2 Machinery 20.4 29.9 32.9 21.0 25.5 32.6 30.4 Inventory 45.2 45.2 45.2 45.2 45.2 45.2 45.2 Land 27.5 41.7 41.7 41.7 41.7 41.7 41.7 Aggregate 32.3 38.1 42.5 34.1 28.7 42.8 43.9 Difference from the 1997 regular taxable case 6.1 14.9 9.6 10.6 7.8 11.9 4.7 (table continues on following page) Table 9.2 continued Commercial Category agriculture Agroprocessing Manufacturing Construction Transportation Communications Tourism Tax-holiday case, pre-1997 (interindustry dispersion: 2.4) t Buildings 15.8 28.5 29.0 31.1 28.2 32.1 28.4 Machinery 26.7 25.3 26.6 25.6 27.6 31.5 28.7 Inventory 15.3 12.8 12.8 12.8 12.8 15.3 12.8 Land 8.0 18.2 18.2 18.2 18.2 20.0 18.2 Aggregate 15.0 23.6 22.3 21.0 26.9 30.2 26.3 Difference from the 1997 regular taxable case -11.2 0.4 -10.6 -2.5 6.0 -0.7 -12.9 Source: Authors' calculations based on data provided by the Ministry of Finance, Planning, and Economic Development and the URA. 286 Duanjie Chen, John Matovu, and Ritva Reinikka sector invests about two-thirds of its total capital in the two highest taxed assets, inventories and buildings. In contrast, transportation enjoys the lowest METR on capital of all sec- tors (21 percent). The primary reason is its heavy capital weight in machin- ery, particularly vehicles which have a relatively high annual depreciation allowance (30 percent). For the same reason, agroprocessing and construc- tion incur a relatively low METR (23 and 24 percent, respectively). The METR on capital for commercial agriculture and the communications industry are in the middle (with METRs of 26 and 31 percent, respectively). Agriculture has a high capital share in inventories, while communications has a high capital share in buildings. SmALL FImvis. Small firms do not pay regular income taxes unless they opt to do so, but are instead levied a presumptive tax up to 1 percent of their gross turnover. Here, small firms refers to firms qualifying for and choosing to pay the presumptive tax. Because the presumptive tax is imposed on the gross re- ceipts without any adjustments, small firms are neither entitled to the generous initial allowance for investment in machinery nor subject to any restrictions on writing off business expenditure. Consequently, the METR for small firms is lower than for large and medium-size regular taxable firms on all other assets but machinery (table A9.8). However, unless engaged in commercial agricul- ture, small firms still pay municipal property taxes. Therefore, buildings and land are taxed higher than investment in machinery and inventory by small firms. As depreciable assets wear off at a different pace from industry to indus- try, buildings and machinery incur a different METR across industries, despite being subject to the same presumptive tax rate and having no differentiated sector-specific tax allowances. Compared with the base case (regular taxable firm) by industry, small firms are taxed significantly less as measured by the aggregate METR on capital. The gap ranges from 15 percentage points in agroprocessing to more than 24 in manufacturing. Furthermore, the interindus- try dispersion is smaller than in the base case of the regular taxable firm. Impact of Tax Reform This section examines the impact of income tax reforms on regular taxable firms and compares them with the firms that had been granted tax holidays. REGULAR TAXABLE FIRMs. As shown in table 9.2, the tax burden incurred by large and medium-size regular taxable firms was significantly reduced follow- ing the 1997 income tax reform. The difference in the aggregate METR between the two systems is 5 to 15 percentage points. The most striking change is the difference in the METR on machinery, varying from 9 percentage points for transportation to 34 percentage points for the communications sector. This is mainly because of the generous initial allowance for investment in machinery and equipment available to all tax paying firms under the new system. The other contributor is the zero-rated import duty for imported machinery. A Quest for Revenue and Tax Incidence 287 Following the reform, the METR on buildings declined about three per- centage points, mainly because the annual depreciation allowance increased from 4 to 5 percent. The wider gap (about 19 percentage points) for commer- cial agriculture reflects a higher annual allowance for farm works. The METR for inventory and land did not change. REGULAR TAXABLE VERSUS TAX HOLIDAY FIRM. Corporate tax holidays were abol- ished in 1997 and replaced mainly by an initial investment allowance for ma- chinery. Consequently, the METR on machinery was reduced approximately 25 percentage points across industries, except in transportation. This indicates that, given the generous allowances, profitable firms that invest heavily in machinery can benefit from opting out from the tax holiday status. For all other assets, however, the METR was lower under the tax holiday regime.'2 Those investing heavily in machinery gained most from the tax reform, reflecting the policymakers' desire to provide incentives for acquiring new technologies. The most evident example is the transportation industry, where the advantage measured by the METR for regular taxable firms over their tax holiday counterparts is 6 percentage points. However, the METR for the regular taxable firms in the tourism sector is 13 percentage points more than their tax holiday counterparts, because of the high capital share in structures. Similarly, commercial agriculture and manufacturing incur a higher METR (11 percentage points) under the new system, as these industries invest more in nondepreciable assets, particularly inventories for which the tax holiday regime was more advantageous."' METR on Cost Production The METR on cost of production is used to evaluate the impact of all busi- ness taxes-including capital, payroll, and indirect taxes-on overall busi- ness activities. It is estimated as an integration of the METR on various in- puts, using the augmented Cobb-Douglas production function (see annex 9.2). As fuel tax is an important revenue source in Uganda, motor fuel-along with capital and labor-is included as an input for production."4 As shown in 12. As can be seen from table 9.2, interindustry tax distortion increased following the tax reform (see annex 9.2 for definition). Further analysis shows that the main contributor was the difference in the METR between commercial agriculture and all other sectors. As farm works are entitled to a fast write-off and properties used for commercial agriculture are exempt from municipal property tax, buildings and land are taxed much less than in the other sectors. 13. Chen and Reinikka (1999) provide policy simulations and sensitivity analy- ses for nontax parameters, including choice of accounting method, initial allowance for buildings, municipal property tax on small firms, inflation rate, debt-to-assets ratio, and economic depreciation rate. 14. The combined fuel tax rate for Uganda is the ad valorem rate on the total cost, insurance, and freight destination warehouse cost, including all handling charges. The 288 Duanjie Chen, John Matovu, and Ritva Reinikka table 9.3, the cost structure varies across industries. Capital accounts for the largest share, which probably reflects the low labor costs in Uganda. Further- more, as agroprocessing requires a higher share of transportation services than commercial agriculture, the share of fuel in its total cost is 9 percent, while it is only 1 percent in commercial agriculture. Table 9.3 summarizes the METR on each of the three inputs as well as on the overall cost of production by industry. The METR on capital uses the base case (regular taxable firm under the 1997 tax system), the METR on labor is the statutory payroll tax rate of 10 percent, and the METR on fuel is estimated at 174 percent (table A9.6).'5 As the METR on fuel is significantly higher than on capital, industries that use more fuel incur a higher METR on production cost than on capital alone. Agroprocessing and transportation, with the low- est METR on capital, fall in that category. In other words, the high fuel tax may actually negate some of the benefits of the tax reform-which strongly encourages investment in machinery and equipment in agroprocessing and the transportation sector-as these two sectors spend the most on fuel. In contrast, all other industries incur a lower METR on production cost than on capital, mainly because of the low METR on labor and the small share of fuel in the total cost. As concerns the cost of production, tourism and manufactur- ing are still the highest taxed industries in Uganda, while construction re- places transportation as the lowest taxed industry. Cross-Border Comparison for Foreign Firms This section compares the impact of taxation on foreign direct investment in Kenya, Tanzania, and Uganda. It attempts to determine which of the three countries could best attract foreign investors, if tax cost were the only decid- ing factor. Manufacturing and tourism are the focus, as these are key areas for foreign direct investment in Eastern Africa. For simplicity, tax provisions and economic parameters for foreign firms are based on the United Kingdom's tax system, which accounts for the largest share (about 25 percent) of the total actual foreign investment in Uganda. Cross-Border Comparison of METR on Capital Tax rates and provisions for Kenya and Tanzania are summarized in table A9.9. To focus the cross-country comparison exclusively on the burden of taxation, Uganda's nontax parameters and capital structure are also applied weighted average rate is based on data provided by the URA on fuel sales by product in 1997. 15. As the payroll tax in Uganda is imposed on the total payroll without ceilings, the statutory payroll tax rate can be seen as the marginal rate. Ignoring the shift effect assumes that the employer's share of payroll tax is fully borne by the employer. Table 9.3. Marginal Effective Tax Rate on Cost of Production for Ugandan Firns (percent) Commercial Factor of production agriculture Agroprocessing Manufacturing Construction Transportation Communications Tourism METRO Capital 26.2 23.2 32.9 23.5 20.9 31.0 39.2 Labor 10.0 10.0 10.0 10.0 10.0 10.0 10.0 Fuel 174.0 174.0 174.0 174.0 174.0 174.0 174.0 ° Aggregate 25.8 26.7 30.6 21.1 24.3 25.5 34.2 Cost structure" Capital 89.2 52.2 66.8 50.2 67.1 60.4 68.8 Labor 9.5 38.8 28.2 45.6 26.4 36.7 27.2 Fuel 1.3 9.0 5.0 4.2 6.4 2.9 4.0 a. METR by input and on overall cost of production. b. Input share in total cost of production (excluding other inputs). Source: Authors' calculations based on data provided by the Ministry of Finance, Planning, and Economic Development and the Uganda Revenue Authority; Republic of Uganda (1995) for cost structure. 290 Duanjie Chen, John Matovu, and Ritva Reinikka to Kenya and Tanzania."6 With these assumptions, Uganda has a tax disad- vantage compared with Kenya in both manufacturing and tourism, mainly because of Kenya's preferential tax treatment targeted to these two sectors (table 9.4). In tourism, Uganda is also less competitive than Tanzania in terms of taxation, mainly because of its local property tax on buildings, which ac- counts for 71 percent of capital in the tourism sector. Various factors contribute to this outcome, including the following: * Kenya and Tanzania have no property tax on structures. Consequently, even without considering the initial investment allowances available in Kenya, buildings are taxed significantly less in Kenya and Tanza- nia. 17 A slightly more generous tax depreciation rate for buildings in the tourism sector (6 percent versus 5 percent) also contributes to a lower METR on buildings in Tanzania. * Kenya provides an initial investment allowance of 60 percent for both buildings and machinery for manufacturing and tourism. Therefore, despite Kenya's slightly higher corporate income tax rate, buildings are taxed much more lightly than in Uganda and Tanzania. Table 9.4. Marginal Effective Tax Rate on Capital for Foreign Firms (percent) Factor of production Uganda Kenya Tanzania Manufacturing Buildings 38.9 1.9 25.6 Machinery -3.9 12.3 31.0 Inventory 59.0 69.0 61.9 Land 32.8 27.5 39.0 Aggregate 33.8 28.8 40.0 Tourism Buildings 36.5 0.9 15.9 Machinery -1.8 10.0 28.5 Inventory 59.0 69.0 61.9 Land 32.8 27.5 39.0 Aggregate 32.6 7.5 21.9 Note: Uganda's nontax parameters are applied to Kenya and Tanzania. Source: Authors' calculations based on data provided by the Uganda Revenue Authority; Ministry of Finance, Planning, and Economic Development; Bureau of Statistics; and the World Bank. 16. A simulation using country-specific parameters is carried out in Chen and Reinikka (1999). 17. Should buildings also be exempted from the municipal property tax in Uganda, Uganda could gain a tax advantage over Kenya and Tanzania in manufacturing and over Tanzania in tourism. A Quest for Revenue and Tax Incidence 291 * A nonzero import duty on most machinery imported to Kenya and Tanzania contributes significantly to higher METRs on machinery in these two countries. * The different property tax rates on land affect the METR on land: Tan- zania, at 39 percent, has the highest METR on land, followed by Uganda with 33 percent and Kenya with 28 percent. Cross-Border Comparison of Cost of Production Again, to isolate the impact of taxation, Uganda's nontax parameters, in- cluding the cost structure, are applied to Kenya and Tanzania. As before, the METR on labor is the average payroll tax payable by employers, and the METR on fuel is the effective average tax rate on motor fuels. As shown in table 9.5, Kenya has the lowest METR on labor, followed by Tanzania (0.1 and 4 percent, respectively, compared with 10 percent in Uganda). Tanzania has the lowest METR on fuel, followed by Kenya (26 and 64 percent, respec- tively, compared with 174 percent in Uganda).'" As a result, measured on cost of production, Uganda becomes the highest taxed country in both manu- facturing and tourism. Tanzania's tax competitiveness in tourism becomes Table 9.5. Marginal Effective Tax Rate on Cost of Production for Foreign Firms (percent) Factor of production Uganda Kenya Tanzania Manufacturing Capital 33.8 28.8 40.0 Labor 10.0 0.1 4.0 Fuel 174.0 62.0 25.4 Overall 30.7 21.5 28.8 Tourism Capital 32.6 7.5 21.9 Labor 10.0 0.1 4.0 Fuel 174.0 62 25.4 Overall 29.9 7.2 17.0 Note: Uganda's nontax parameters are applied to Kenya and Tanzania. Source: Authors' calculations based on data provided by the Uganda Revenue Authority; Ministry of Finance, Planning, and Economic Development; Bureau of Statistics; the Uganda Investment Authority; and the World Bank. 18. For Kenya and Tanzania, the fuel tax rate by product is estimated based on the tax and the price per liter, while Uganda's shares of various products in total sales were used as weights to estimate the combined fuel tax rate. 292 Duanjie Chen, John Matovu, and Ritva Reinikka more evident, while its manufacturing sector now has a lower tax burden than its counterpart in Uganda. Kenya has an even greater tax advantage over Uganda in both sectors. Compliance and Tax Administration A typical tax incidence analysis assesses the tax structure without dealing with administrative realities. Administration, however, can create major distortions in even a well-designed tax system if it is not managed efficiently and fairly. This section examines key features of taxpayer compliance and tax adminis- tration, based on firm survey evidence (see appendix B at the end of the book). The purpose is to isolate factors likely to change the true tax burden on firms from what the formal system and the METR analysis indicate. Tax Compliance Taxpayer compliance depends on economic incentives embedded in the tax structure and the effectiveness in detecting and penalizing noncompliance (see Das-Gupta and Mookherjee 1998). According to the 1998 firm survey, a third of Ugandan firms were in a tax loss position in 1997, that is, they neither paid the corporate income tax nor had a tax holiday (table A9.10). While this esti- mate may appear high, this ratio is not out of line with international experi- ence. For example, Canadian statistics show that an average of more than 40 percent of active nonfinancial firms are in a tax loss position. Twenty-six per- cent of Ugandan firms did not pay the VAT in 1997, possibly because many smaller firms are not registered for the VAT. Commercial agriculture has the largest share of non-VAT paying firms. This is broadly consistent with the de- sign of the VAT system (for instance, foods are zero-rated in general). Eight percent of Ugandan firms with five or more employees do not pay taxes. VVhether or not firms are content with their own level of taxes, their own- ers clearly feel disadvantaged when they see their competitors escaping taxa- tion. In the 1994 survey of Ugandan firms, respondents identified competi- tors' evasion of taxes as a major constraint (World Bank 1994). Some 60 percent of firms reported that they faced unfair competition. Furthermore, firms esti- mated the informal economy (part of the economy evading taxes, duties, or laws and regulations) to be as high as 43 percent. In 1998 this perception remained, with tax evasion considered the leading constraint from unfair competition. However, the numerical constraint scores for competitors smug- gling or evading taxes have declined. Despite some improvement in perceptions, the legacy of a predatory state, coupled with limited improvement in service delivery, continues to adversely affect tax compliance in Uganda. In the 1998 survey, firms in manufacturing-the second highest taxed sector measured by the METR- estimated that half of their competitors gain an advantage through tax eva- sion. In construction and agroprocessing, the reported share was about 40 A Questfor Revenue and Tax Incidence 293 percent. In tourism, the highest taxed sector as measured by the METR, firms reported that a third of their competitors evade taxes. In commercial agriculture, however, where the share of tax paying firms is the lowest, only 5 percent of competitors were perceived to evade taxes. Tax Administration A prominent feature of the Ugandan tax administration is frequent tax au- dits, which are either desk or field operations or a mixture of both. Predeter- mined criteria do not exist for conducting an audit, but factors such as com- pliance record, quality of returns submitted, and size of firm are considered important. Sixty-eight percent of all firms were audited either for the corpo- rate income tax, VAT, or both during 1995-97. Forty-one percent of firms re- ported that they were audited for the corporate income tax, while as many as 60 percent of all firms were audited for the VAT. The latter is equivalent to three-quarters of the VAT-paying firms. In the international comparison, Uganda's audit figures are extremely high. For example, in Canada all large corporations (about 1,000) are audited, and the remaining 13,000 or so corpo- rations face audit rates of 5 percent or less. The high auditing frequency indi- cates a serious lack of voluntary compliance and a low level of mutual trust between the tax authority and the taxpayer. The URA routinely "assesses" tax returns submitted by taxpayers. These assessments are typically desk reviews of self-declarations and supporting documents. The tax officer may accept the taxpayer's declaration as is, or "assess" an additional tax to be paid. A tax audit may also be involved that may lead to a demand for additional taxes to be paid as an "assessment." As shown in table A9.10, as many as 51 percent of Ugandan firms disagreed with the URA on their assessment during 1995-97. Sixty-eight percent of these cases were resolved through negotiation between the firm and URA officers, while 10 percent appealed to a third party. None of the disputes was taken to court. The rest remained unsettled at the time of the survey At the end, roughly a third of the resolved disputes ended with a result closer to the taxpayer's own assessment, a third were closer to the URA's assessment, and the rest were between the two assessments. Depreciation allowances appear to be one of the main causes for disputes in the corporate income tax assessments. The firm survey also indicates that most tax holiday firms have little or no involvement with the tax authority, which may be an additional incentive for initially acquiring tax holiday status. The Impact on the METR The firm survey reveals important differences between the formal tax sys- tem and actual practice, which can affect the METR results presented here. First, among the firms that were audited, at least every third firm had to pay additional taxes, while every fourth firm incurred additional costs, such as 294 Duanjie Chen, John Matovu, and Ritva Reinikka bribes (see chapter 10 in this volume). All firms whose tax assessment dif- fered by 100 percent or more "always" (five on the scale of one to five) had to pay bribes to URA officials, while on average, all survey firms reported that bribes were "seldom" required (two on the scale of one to five). Bribes may affect the effective tax burden in two ways. On the one hand, despite being a cost, bribes can reduce the tax burden (measured by the METR) if they pro- vide an opportunity for tax evasion. On the other hand, the extra costs may increase the tax burden when used, say, to avoid a lengthy appeal and settle- ment process (which in itself would increase the burden, but is not captured by the METR based on the formal tax system). Second, as the VAT is a consumption tax and therefore should not affect capital investment and taxable business activities, the METR model gener- ally ignores it. However, if the input tax credit under the VAT system is not refunded quickly or not at all, then VAT can place an additional tax burden on the business sector.'9 As the VAT was introduced in Uganda only in 1996, implementation problems can be expected to arise. In 1998 the main com- plaint from the business sector concerned refunding the input VAT credit. As table A9.10 shows, 81 percent of firms purchase inputs from VAT-registered suppliers but only 56 percent of these firms claim input tax credits. Whether this results from the VAT credit and liability offset procedure is not clear.20 Another potential reason is that firms with excess input tax credits decline to claim for refunds, for example, because of higher compliance costs. This could be tempting for firms that can pass on the input VAT cost to consumers, but less for the firms that have to absorb the cost themselves. In the former case, the VAT would cascade and increase tax revenue in the short term, but at the cost of consumer welfare in the long run. In the latter case, firms may incur a profit loss that can affect the corporate income tax revenue in turn. Fifty-two percent of the firms that claimed an input tax refund received their expected amount in 1998. However, a significant portion (18 percent) of firms that claimed the input tax credit did not receive any refund, while the rest (40 percent) received a partial refund. Furthermore, the waiting period for even a partial refund of the input VAT credit can be lengthy. Of the firms that received at least a partial refund, more than half waited more than six weeks, while 10 percent waited more than six months. The lengthy process for input VAT refund is likely to curb compliance as well as in- crease the cost of doing business. It ties up a considerable portion of 19. When the input tax credit is not refunded, the VAT could be modeled as a sales tax on capital or any other taxable input. In the case where the refund period is abnormally long and no interest is paid by the revenue authority, the interest cost could be modeled as an increment on the cost of financing. 20. When offset procedures are being used, apparently no supporting documenta- tion is required and the approval is granted after a desk review, subject to a later audit. However, such a loose arrangement can cause major difficulties at the audit stage. A Questfor Revenue and Tax Incidence 295 working capital that has a high opportunity cost, considering a bank lend- ing rate of more than 20 percent. Hence, two types of opposing factors emerge from the survey evidence that could alter the METR results. First, tax evasion would reduce the actual METRs compared with the formal tax system. Because compliance is firm specific and tax administration also tends to treat firms differently, this im- pact is not the same across industries, or even within a particular sector. Sec- ond, delays in the VAT refunds and payment of bribes could have the oppo- site effect of increasing the tax burden compared with the formal tax system. The net effect is ambiguous. Similarly, the impact of frequent tax audits and assessments on the METR is also ambiguous, depending on whether these contribute to enforcement of the formal rules or cause an extra cost to firms. Conclusions The National Resistance Movement government rescinded predatory implicit and explicit taxation on exports in the early 1990s, one of the major accom- plishment in Uganda's recovery. However, whether the rapid increase in do- mestic revenue was a good strategy is less clear. The corresponding expansion in government expenditure may not, at the margin, have had a high payoff in terms of service delivery, while the cost of taxation was high because of bu- reaucratic control, resource misallocation, and foregone household consump- tion and private investment. These effects were likely to undermine growth, and hence the prospects for sustainable increases in public revenue. Ugandan policymakers, however, have been able to readjust their eco- nomic policy when necessary. During the second half of the 1990s, once it became obvious that import taxes were an implicit tax on exports, these taxes were considerably reduced and several other tax reforms were introduced, including the VAT and income taxation. Consequently, the Ugandan tax sys- tem is gradually being transformed from high tax rates and selective incen- tives and exemptions toward lower rates and more standard provisions. Household survey analysis reveals that tax reforms implemented in the 1990s were generally propoor. First, given the zero rating of goods consumed by the poor, replacing the sales tax with the VAT did not lead to the poor being worse off. Second, import taxation remained progressive after tax re- forms, but less so. In aggregate, excise taxes became more progressive. Third, increased taxation on paraffin is highly regressive, while taxes on other pe- troleum products are progressive. Fourth, given the liberalized market, ex- port taxes on coffee used during commodity booms tend to hurt the poor. The METR analysis demonstrates that-even when the country's level of public revenue is low at the macroeconomic level-rapidly increasing taxation may constrain private investment at the microeconomic level, for two reasons. First, the formal enterprise sector in these economies typically represents a small share of output, but a high proportion of the effective tax base. Second, access to credit is limited and interest rates are high, 296 Duanjie Chen, John Matovu, and Ritva Reinikka particularly for smaller firms, and hence most private investment is financed by profits and personal savings. Consequently, taxation reduces both the expected revenue from a given investment project and the availability of investment finance. From the perspective of foreign investors, Uganda appears to have higher taxes than neighboring countries, particularly Kenya. Raising nominal tax rates is therefore no longer a feasible policy option for Uganda. At the microeconomic level, the Kenyan tax system appears to place the lowest bur- den on firms investing in manufacturing and tourism. However, at the mac- roeconomic level, Kenya's share of tax revenue in GDP is the highest of the three countries. Uganda's tax disadvantage results mainly from a property tax on buildings, which does not exist in Kenya and Tanzania, and its signifi- cantly higher fuel taxation. A strong case exists for harmonization of fuel taxes within the region. To level the playing field, discretionary corporate tax holidays were abol- ished in 1997 in Uganda and replaced by an initial investment allowance for machinery for all firms. Consequently, the METR on machinery was signifi- cantly reduced. The analysis indicates that profitable firms that invest heavily in machinery clearly benefited from this policy change. However, for all other assets the METR was lower under the tax holiday regime. The METR estimates reflect the formal tax structure. Tax administration, if not fair and efficient, can distort the best intentions of policymakers and produce a very different outcome in terms of the actual tax burden firms face. Using firm survey evidence, several factors that can alter the METR results were identified. First, widespread tax evasion and firm-specific ex- emptions-which show up strongly in the 1997 data despite efforts to curb them in prior years-are likely to reduce the METRs. Second, delays in VAT refunds and payment of bribes are likely to have the opposite effect of in- creasing the METR compared with the formal tax system. However, the net effect is ambiguous. Tax administration is an important area to be tackled in Uganda in the future. In particular, efforts to combat corruption and mechanisms to resolve grievances between the business sector and the tax authority are critical. These efforts require regular dialogue with the private sector to build trust, and tax education and training for both taxpayers and administration staff. Annex 9.1. Household Incidence Analysis and the Concept of Welfare Dominance The theoretical model used in this chapter for the household incidence analysis of tax reforms relies on the work of Yitzhaki and Slemrod (1991). In this model, for any social welfare function favoring equitable distribution of income, a marginal reduction in taxes on, say, good x5 and a marginal increase in taxes on, say, good x,, that keep the tax revenues constant, will improve social wel- fare if the x5's concentration curve is below x 's curve everywhere. A Questfor Revenue and Tax Incidence 297 Formally, let the social welfare function be given by (A9.1) D = I gh)h (Yh' Pl' ... I Pd) where uh is the indirect utility of household h, Yh is the income of household h, p, are commodity prices (with i = 1 ,..., n) and 'h is the social weight for each households h's indirect utility. Suppose the government considers a tax reform involving only two commodities, x5 and x,. It considers marginally increasing the tax on commodity x, and marginally decreasing that on com- modity x5 to leave total revenue constant (a revenue neutral change). If we denote with x,h the consumption of commodity i by household h and with X. the total consumption of commodity i by all households, then the tax reform keeps total tax revenues R unchanged, with (A9.2) R = EtkX, kkk where k are the taxed commodities and k is the tax rate on commodity k. Under these assumptions, it can be shown that the welfare of a household h is not worsened by the proposed tax reform if and only if (A9.3) X-- aS' X I > °' where ca, is defined by Wildasin (1984) and Mayshar (1988) as the marginal social cost of raising one dollar of revenue by taxing the t-th commodity. This may be generalized to consider all households h with h=1, ... ,m, (A9.4) " X - aSt X > °. The expression (A9.4) can be seen as the difference between the height of the relative concentration curve of commodity x5 and the height of the rela- tive concentration curve of commodity xt multiplied by a constant. These concentration curves are similar to the familiar Lorenz curve, but instead of total income, they consider the fraction of total expenditure on a commodity attributable to different income groups. Consequently, for any additive so- cial welfare function, a tax change increases social welfare if and only if the concentration curve of commodity x5 is not as high as the concentration curve of commodity x1 (multiplied by a constant) along the entire income distribu- tion. This method is generally referred to as welfare dominance. The dominance test may often be inconclusive because of the require- ment that each concentration curve must be above the other everywhere along the income distribution. In this case, conclusions can be only drawn by speci- fying the weights attached to each household in the social welfare function. Yitzhaki (1983), for example, provides a framework for analyzing welfare 298 Duanjie Chen, John Matovu, and Ritva Reinikka dominance by using extended Gini coefficients. These allow for adjustments in the social weights given to various households and provide a clearer no- tion of how alternative social welfare functions differ with tax regimes. Spe- cifically, the extended Gini coefficient is a weighted integral of the area be- tween the Lorenz curve and the 45-degree line as a fraction of 0.5 (which is the total area under the 45-degree line) and is given by (A9.5) G(v) =- vCOV[e(l - F(e))v -1], where v is a parameter that affects the weighing of the points on the concen- tration curve, F(e) is cumulative tax payment, and e measures the household's tax payment. When v = 2, G(2) yields the traditional Gini coefficient, while higher values would give more weight to commodities consumed by the poorest households. Using equation (A9.5), it can then be shown that a rev- enue neutral decrease in the tax on commodity x5, financed by an increase in commodity x,, decreases the extended Gini index, if (A9.6) f [DS(F) - asOt(DF)](I - F)v-2 dF > 0, where 4D1(F) for [i = s,t] is the concentration curve. Both concepts of welfare dominance and of extended Gini coefficient are used in this chapter to exam- ine the welfare implications of tax reforms. In practice, all welfare dominance techniques tend to be difficult, as con- centration curves tend to cross each other, especially toward the end of the distribution. A solution to this problem was developed by Davidson and Duclos (1997), who proposed a set of variance estimators to test the hypoth- esis that two concentration curves are statistically different from one another. Annex 9.2. Marginal Effective Tax Rate The METR on capital calculated in this study is the effective corporate tax rate on capital, while the METR on cost of production is an integration of the METRs on all inputs, using the augmented Cobb-Douglas production func- tion. The METR is estimated for both domestic and foreign firms. Unless otherwise specified, all estimates are based on the 1997 tax regime and recent economic indicators. The METR calculation is based on the assumption that profit-maximizing firms base their investment or business decisions on the foreseeable incremental net revenue at the present value. Taxes reduce the profits accruing to the firm, while tax allowances mitigate such a reduction. Because of the interaction between statutory tax provisions and actual eco- nomic and industrial conditions, the effective tax rate can vary by industry under the same tax regime. Furthermore, for a cross-jurisdiction compari- son, the effect of taxation can be singled out by applying the same set of economic and industrial conditions to different tax regimes. A Questfor Revenue and Tax Incidence 299 The method used to estimate the METR has been extensively documented, by Broadway, Bruce, and Mintz (1984); Chen and Mintz (1993); McKenzie, Mintz, and Scharf (1992), Mintz (1990); and others. Other useful references include Dunn and Pellechio (1990) and Shah (1995). METR on Capital As described, the METR on a given type of real capital investment is defined as the proportional difference between the gross-of-tax rate of return (rG) and the net-of-tax rate of return (rN) required by financial investors. The gross-of- tax rate of return (rG) is the marginal revenue product, or user cost of capital, net of economic depreciation. The net-of-tax rate of return (rN) is the weighted average of the return to debt and equity securities held by the financial in- vestor. Thus, the effective tax rate (t) is defined as (A9.7) t = (rG - rN)/rG or t = (rG - rN)/rN. The latter definition is used in this chapter. Real Cost of Financing For domestic firms, the real cost of financing (rf) is defined by (A9.8) rf = i(l - U) + (1- 13)p - t where P is the debt-to-assets ratio, i is the cost of debt, U is the statutory corporate income tax rate, p is the cost of equity, and Xt is the inflation rate. While interest costs are deductible for income tax purposes, the cost of eq- uity is not. That is, the cost of financing for a domestic firm is the weighted average cost of financing, net of inflation rate. For foreign firms, the real cost of financing (rf) is defined by (A9.9) rf' = [3'I'(1 - U') + (1 - 13') p'](l- 'y)/(1 - x) + y[i(1 - U) - it + '] - it', where ' is the debt-to-assets ratio in the home country, I' is the cost of debt in the home country, U' is the statutory corporate income tax rate in the home country, p' is the cost of equity in the home country, y is the ratio of debt raised in the host country to total investment fund, x is the weighted average withholding tax rate in the host country, i is the cost of debt in the host country, U is the statutory corporate income tax rate in the host coun- try, it' is the inflation rate in the home country, and it is the inflation rate in the host country. According to the equation (A9.9), the cost of financing to a foreign firm is the weighted average of the cost of its investment funds taken from the home country and debt raised in the host country. The former is the weighted aver- age cost of financing at home net of withholding tax payable in the host coun- try, and the latter is the cost of debt in the host country adjusted for income 300 Duanjie Chen, John Matovu, and Ritva Reinikka tax deductibility and the difference in inflation rates between the home and the host country. Net-of-Tax Rate of Return on Capital For domestic financial investors, the net-of-tax rate of return on capital is defined by the formula (A9.10) rN = pi + (1 - ,B)p - it. This is the rate of return on capital required by the financial investor or the supplier of investment funds. For foreign investors, the formula is (A9.11) rN = [T1'(1 - U') + (1 - ,B') p' - n'](1 - y) + y(i - i). This is the net-of-tax rate of return on capital required by fund suppliers, including foreign financial investors in the host country. Applying equations (A9.10) and (A9.11) to equation (A9.7), respectively, yields the effective cor- porate tax rate on capital for domestic and foreign firms. Gross-of-Tax Rate of Return on Capital DEPRECIABLE ASSETS. For domestic firms, the formula is (A9.12)rG = (1 + tm)(rf + 8)(1 -k)[1 -A + t(1 - U)/(a + rf+ ir)]/[(1 - )(1 - tp-tg)] -8, where tm is the tax on transfer of property or transaction tax (for example, import duty) on capital goods where applicable, 8 is the economic deprecia- tion rate, k is the investment tax credit rate, A is the present tax value of the accumulated capital cost allowance, r is the capital tax rate, a is the tax de- preciation rate, tp is the property tax rate, and tg is the gross receipts tax rate or presumptive tax. For international firms, the formula is (A9.13) rG= (1 + tm)(rf' + 8)(1 -k)[1 -A + x(1 - U)/(a + rf' + it)]/ [(1 - U)(1 - tp - tg)] - 6. INVENTORY. For domestic firms, the formula is (A9.14) rG = (1 + tm)(rf + Uir4)/[(1 - U)(1 - tg)] + r, where tm is the sales tax on inventory where applicable, and 4 is one for the FIFO accounting method and zero for the last-in-first-out (LIFO). For inter- national firms, the formula is the same except that the financing cost should be the one relevant to the international firms, that is, rf should be replaced by rf'. LAND. For domestic firms, the formula is (A9.15) rG = rf(l + tm)(I + r)(1 - U)/(rf + ir)/[(1 - U)(1 - tp - tg)]. A Questfor Revenue and Tax Incidence 301 For international firms, the formula is the same except that the financing cost should be the one relevant to the international investors, that is, rf should be replaced by rf'. Aggregation The effective tax rate for a given industry is the proportional difference be- tween the weighted average of the before tax rate of return by asset type and the after tax rate of retum, which is the same across asset type within the industry. That is, the marginal effective tax rate t. for industry i is calculated as (A9.16) t, = (Y rIGwI.I--rIN) / ri, where j denotes asset type (such as investments in buildings, machinery, in- ventories, and land) and wI. denotes the weight of asset type j in industry i. The above equations are general formats of the formulas used in this chapter. Because of the variance among different sectors or jurisdictions, some vari- ables can be zero for some sectors or jurisdictions. For example, none of the three countries in this study have capital-based taxes, and hence X = 0 in equation (A9.12) - (A9.15). METR Dispersion METR dispersion, or the weighted standard deviation, is used to measure the tax distortion. There are three measures of dispersions: overall, interin- dustry, and interassets dispersion. Only interindustry dispersion is estimated in this chapter. Let w;, w;, and w.. denote the capital weights for the ii indus- try and the ji type of asset, respectively. The interindustry METR dispersion 1, is calculated as the weighted standard deviation: (A9.17) a, = 1IWI{lIwII(tII- tI)2} /2. The expression tI is the average effective tax rate for the asset j across in- dustries, and t.. is the effective tax rate for the jh asset type in the it industry. ii METR in Other Inputs and Cost of Production METR ON LABOR. This chapter assumes that only payroll taxes paid by em- ployers are effective labor taxes borne by employers. Another assumption is that the marginal unit of labor input is an average worker. Therefore, the METR on labor is the total payroll taxes paid by employers on average labor costs. Because payroll taxes in Tanzania and Uganda are imposed on total payrolls, the statutory tax rate itself can be seen as the effective tax rate on labor. In Kenya, the ceiling of taxable payroll is K Sh 80 per month, well below the monthly payroll. As a result, the METR on labor in Kenya is esti- mated to be as low as 0.1 percent. According to the International Labour 302 Duanjie Chen, John Matovu, and Ritva Reinikka Organisation (1997), the average monthly payroll in Kenya was K Sh 3,324 for the manufacturing industry and tourism (1991 figure). METR ON OTHER INPuTS. The METR on other inputs for production is the transaction taxes firms have to pay on these inputs. Motor fuel is the only other input included apart from capital and labor. The average transaction tax rate (the fuel tax rate) is used as the METR. METR ON COST OF PRODUCTION. By using the augmented Cobb-Douglas pro- duction function, the METR on cost of production T can be estimated as (A9.18) T = fl(1 + Q)a - 1. In the formula, i indicates an input (capital, labor, and fuel), t, is the METR on each input i, and a, is the share of total cost for input i. The detailed deri- vation may be found in McKenzie, Mintz, and Scharf (1992). Annex 9.3. Figures and Tables for Household Incidence and METR Figure A9.1. Concentration Curves for Main Taxes before Reform 1.0 Qr. 0.8 - 0.6 0.4- ~0.2 - 0- 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Cumulative household expenditure 45 degree -------- Graduated personal tax Total expenditures -------- Excise tax - - - Sales tax - - - Importtax Income tax Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. A Questfor Revenue and Tax Incidence 303 Figure A9.2. Concentration Curves for Main Taxes after Reform 1.0 0.8- 0.6- >0.4 - 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Cumulative household expenditure 45 degree ........ Value added tax Total expenditures -------- Excise tax Income tax -- - Import tax Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. 304 Duanjie Chen, John Matovu, and Ritva Reinikka Table A9.1. Consumption Goods and Corresponding Tax Rates (percent) 1992 import Sales 1995 import Consumption good duties tax/CTL duties a VAT Matooke, potatoes, maize, cassava 0 0 0 0 Rice 30 30 20;4 0 Bread, macaroni, spaghetti 30 30 20;6 17 Meat, poultry and fish 0 0 0 0 Milk fresh (liquid) 0 10 0 0 Milk (powdered) 30 10 10;2 0 Other dairy products 30 10 0 0 Butter 10 20 20;6 17 Ghee 10 20 10;2 17 Hydrogenated oil 10 30 10;2 17 Margarine 10 30 20;4 17 Refined cooking oil and other oils 10 20 10;2 17 Fruits, beans, lentils and nuts 0 0 0 0 Sugar (Uganda) 0 10 0 17 Sugar (imported) 20 10 20;6 17 Cocoa 30 30 20;6 17 Salt 10 10 10;2 17 Soda (all brands) 0 40 20;12 17 Passion fruit/orange juice 0 20 20;6 17 Other nonalcoholic drinks 0 40 20;12 17 Beer 350 70 20;12 17 Uganda waragi-refined 0 50 20;12 17 Other alcoholic beverages 100 50 20;12 17 Cigarettes (all) 100 50 30;12 17 Expenditure in restaurants and cafes 0 10 0 0 Matches 40 20 20;6 17 Soap, detergents, toothpaste 30 50 20;6 17 Cosmetics 50 70 20;6 17 Shaving equipment, insecticide, and shoe polish 30 30 20;6 17 Clothing, footwear, household furnishings 30 10 20;4 17 Bags 30 30 20;6 17 Tapes and records 30 50 20;6 17 Rent (including imputed) 0 0 0 0 Water charges 0 10 0 17 Electricity 0 10 0 17 Paraffin 0 30 0 0 (table continues on following page) A Questfor Revenue and Tax Incidence 305 Table A9.1 continued 1992 import Sales 1995 import Commodity duties tax/CTL duties a VAT Other fuel and power 20 30 0 0 Plastic utensils 30 10 20;4 17 Enamel and metal utensils 20 10 10;2 17 Porcelain, glass, chinaware 30 30 20;4 17 Cutlery and kitchen tools 30 30 10;2 17 Bulbs, switches, plugs, cables 20 30 10;2 17 Tires, tubes, and other parts and tools for transport 20 30 10;2 17 Petrol, diesel, oil, greases 75 30 0 0 Stamps, aerogrammes, telephones 0 10 20;4 17 Expenditures on sports and theatres 0 10 0 0 Hotels and other touring 0 10 0 0 Beds, sofas, chairs, other furniture 0 30 20;6 17 Carpets, mats, decoration articles 30 30 20;6 17 Personal cars and vehicles 30 20 20;6 17 Bicycles 20 10 20;6 17 Television sets 10 30 20;6 17 Cassette players and musical systems 40 20 20;6 17 Video decks, cameras, musical instruments 30 50 20;6 17 Jewelry 50 10 20;6 17 Watches 30 30 20;6 17 CTL Commercial transaction levy. a. Figures after the semicolon represent the import duties charged on commodities under the preferential trade area and/or customs union. Source: Authors' calculations based on the 1992/93 integrated household survey and data from the Ministry of Finance, Planning, and Economic Development and the Uganda Revenue Authority. Table A9.2. Extended Gini Coefficients of Taxes before Reforms, 1992 Alcoholic Beverage drinks Tobacco Excise Import Graduated Petroleum Paraffin excise excise excise Aggregate Total V taxes duties Sales tax PAYE tax tax tax tax taxes taxes taxes taxes paid expenditures 2 0.452 0.540 0.521 0.904 0.303 0.889 0.334 0.746 0.649 0.515 0.557 0.426 4 0.620 0.717 0.699 0.971 0.489 0.981 0.522 0.905 0.786 0.705 0.722 0.626 6 0.685 0.775 0.759 0.986 0.569 0.989 0.602 0.934 0.833 0.769 0.777 0.699 8 0.723 0.805 0.791 0.991 0.616 0.991 0.649 0.945 0.858 0.805 0.807 0.739 10 0.748 0.825 0.811 0.992 0.646 0.992 0.681 0.949 0.875 0.828 0.825 0.764 PAYE Pay-as-you-earn. Note: The higher the value of the parameter v, the higher weight is attached to goods consumed by the poor. Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. Table A9.3. Extended Gini Coefficients of Taxes after Reforms, 1995-96 Alcoholic Coffee Beverage drinks Tobacco Excise Import stabilization Petroleum Paraffin excise excise excise Aggregate Total V taxes duties VAT PAYE tax tax tax tax taxes taxes taxes taxes paid expenditures 2 0.537 0.504 0.525 0.904 0.209 0.889 0.334 0.746 0.690 0.515 0.538 0.426 ' 4 0.692 0.691 0.712 0.971 0.407 0.981 0.522 0.905 0.820 0.705 0.712 0.626 6 0.746 0.753 0.772 0.986 0.488 0.989 0.602 0.934 0.861 0.769 0.769 0.699 8 0.777 0.787 0.803 0.991 0.529 0.991 0.649 0.945 0.883 0.805 0.801 0.739 10 0.798 0.808 0.823 0.992 0.555 0.992 0.681 0.949 0.897 0.828 0.820 0.764 PAYE Pay-as-you-earn. VAT Value-added tax. Note: The higher the value of the parameter v, the higher weight is attached to goods consumed by the poor. Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. Table A9.4. Summary of Welfare Dominance Test Statistics, 1992 Alcoholic drinks Beverage Expenditure Import Graduated excise excise Tobacco Petroleum Tax Paraffin (total) duties Sales tax Excise tax PAYE tax taxes duties excise excise Paraffin 0 1 1 1 1 1 0 1 1 1 1 Expenditure (total) 0 0 1 1 -1 1 0 1 1 1 1 Imports 0 0 0 0 0 1 0 1 1 0 1 Sales tax 0 0 1 0 0 1 0 1 1 0 1 Excise tax 1 -1 1 1 0 1 0 1 1 0 1 PAYE 0 0 0 0 0 0 0 0 0 0 0 Graduated tax 1 1 1 1 1 1 0 1 1 1 1 Alcoholic excise 0 0 0 0 0 1 0 0 0 0 1 Beverage excise 0 0 0 0 0 1 0 0 0 0 1 Tobacco excise 0 0 0 0 0 1 0 0 1 0 1 Petroleum tax 0 0 0 0 0 0 0 0 0 0 0 PAYE Pay-as-you-earn. Note: 1 in rows implies that tax is dominated (or more regressive), 0 implies that the tax dominates other taxes, -1 represents that the tax is neither dominant nor dominated (indecisive). Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. Table A9.5. Summary of Welfare Dominance Test Statistics, 1995-96 Alcoholic drinks Beverage Expenditure Import Coffee excise excise Tobacco Petroleum Tax Paraffin (total) duties VAT Excise tax PAYE tax duties duties excise excise Paraffin 0 1 1 1 1 1 0 1 1 1 1 Expenditure (total) 0 0 1 1 1 1 0 1 1 1 1 Imports 0 0 0 1 -1 1 0 1 1 0 1 VAT 0 0 0 0 -1 1 0 1 1 0 1 u Excise tax 0 0 -1 -1 0 1 0 1 1 0 1 t PAYE 0 0 0 0 0 0 0 0 0 0 0 Coffee tax 1 1 1 1 1 1 0 1 1 1 1 Alcoholic excise 0 0 0 0 0 1 0 0 0 0 1 Beverage excise 0 0 0 0 0 1 0 0 0 0 1 Tobacco excise 0 0 0 0 0 1 0 0 1 0 1 Petroleum tax 0 0 0 0 0 0 0 0 0 0 0 PAYE Pay-as-you-earn. VAT Value-added tax. Note: 1 in rows implies that tax is dominated (or more regressive), 0 implies that the tax dominates other taxes, -1 represents that the tax is neither dominant nor dominated. Source: Authors' calculations based on the 1992/93 integrated household survey and data provided by the Ministry of Finance, Planning, and Economic Development. 310 Duanjie Chen, John Matovu, and Ritva Reinikka Table A9.6. Business Taxes in Uganda (percent) Taxes and allowances 1997 system Pre-1997 system Capital taxes Company income tax 30 30 (resident) 35 (nonresident) Tax holidays n.a 3-6 years Investment allowance Structures n.a "Approved business" only Machinery 50-75 "Approved business" only Tax depreciation rate Industrial buildingsa 5 4 Machineryb Class 1 40 50 Class 2 35 40 Class 3 30 20 Class 4 20 n.a Inventory accounting FIFO/LIFO FIFO/LIFO Loss carry-over Forward Forward indefinitely indefinitely Personal tax on investment income Withholding tax on interests 15 15 Withholding tax on dividends 15 15 Presumptive tax on small firms 1 on tumover n.a Property tax 10 10 Indirect taxes on business Import duty on capital goods n.a 5+ Average fuel tax 174 [Not calculated] Payroll taxes 10 10 n.a Not applicable. a. Straight line method. b. Declining balance. The classification of machinery and equipment for tax depreciation allowance varies significantly between the 1997 and pre-1997 systems. For example, computers belonged to Class 3 under the pre-1997 tax system but Class 1 under the 1997 system (see Chen and Reinikka 1999, footnote 16, for details). Source: Ministry of Finance, Planning, and Economic Development and Uganda Revenue Authority data. Table A9.7. Nontax Parameters for Uganda (percent) Commercial Category agriculture Agroprocessing Manufacturing Construction Transportation Communications Tourism Expected inflation rate 4.9 4.9 4.9 4.9 4.9 4.9 4.9 Expected interest rate 21.4 21.4 21.4 21.4 21.4 21.4 21.4 Debt-to-assets ratio 25.0 25.0 25.0 25.0 25.0 25.0 25.0 Economic depreciation rate, Buildings 4.1 3.7 4.0 5.3 3.5 4.2 3.6 Machinery 14.2 16.5 18.7 18.9 22.7 21.2 23.9 Tax depreciation allowanceI c Buildings 5.0 5.0 5.0 5.0 5.0 5.0 5.0 Machinery 30.0 30.0 30.0 35.0 30.0 39.0 30.0 Capital structure by asset type Buildings 10.6 28.7 33.5 10.2 9.9 57.2 71.1 Machinery 20.0 47.8 26.9 47.9 83.8 29.6 9.0 Inventory 33.4 17.7 33.7 37.3 3.2 2.1 1.2 Land 36.0 5.8 5.9 4.6 3.1 11.1 18.7 Cost structure by inputfor production Capital 96.6 54.1 72.9 66.0 74.0 60.6 69.3 Labor 3.0 37.2 23.0 31.1 20.9 36.5 26.7 Motor fuel 0.4 8.7 4.1 2.8 5.1 2.9 4.0 a. Based on Canadian data. b. As a reference. Source: Ministry of Finance, Planning, and Economic Development; Uganda Revenue Authority; and World Bank. Table A9.8. Marginal Effective Tax Rate on Capital for Small Ugandan Firms (percent) Commercial Asset agriculture Agroprocessing Manufacturing Construction Transportation Communications Tourism Buildings 4.5 18.9 19.2 20.4 18.7 19.4 18.8 , Machinery 2.0 2.1 2.3 2.3 2.5 2.4 2.6 j Inventory 3.5 3.5 3.5 3.5 3.5 3.5 3.5 Land 1.0 12.4 12.4 12.4 12.4 12.4 12.4 Aggregate 2.4 7.8 8.9 5.1 4.4 13.2 15.9 Difference from the 1997 regular taxable case -23.8 -15.4 -23.9 -18.5 -16.5 -17.7 -23.3 Interindustry dispersion: 2.2 Source: Authors' calculations based on data provided by the Ministry of Finance, Planning, and Economic Development and the Uganda Revenue Authority. A Quest for Revenue and Tax Incidence 313 Table A9.9. Business Tax Provisions Applicable to Manufacturing and Tourism in Uganda, Kenya, and Tanzania, 1998 (percent) Taxes and allowances Uganda Kenya Tanzania Capital taxes Corporate income tax 30 32.5 30 Investment allowance Buildings n.a. 60 n.a. Machinery 50-75 60 n.a. Tax depreciation rate Buildings Manufacturing 5 SL 2.5 SL 4 SL Tourism 5 SL 2.5 SL 6 SL Machinery Manufacturing 30 DB 14.2 DB 14.2 SL Tourism 31 DB 22.3 DB 19.5 SL Inventory accounting FIFO/LIFO FIFO/LIFO FIFO/LIFO Loss carryover Forward Forward Forward indefinitely indefinitely indefinitely Withholding tax on dividends 15 7.5 15 Property tax Structure 10 n.a. n.a. Land 10 8 11.5-12.5 Payroll tax 10 5 up to 4 80K Sh/mo Indirect taxes Import duty on capital goods 0 5 0-5 Import duty on raw materials 7 15 10-20 Taxes on fuel (average) 174 64 26 n.a. Not applicable. SL Straight line method. DB Based on declining balance. FIFO First-in-first-out. LIFO Last-in-first-out. Source: Ministry of Finance, Planning, and Economic Development; Uganda Revenue Authority; and World Bank data. Table A9.10. Summary of Firm Survey Results on Tax Administration (percent) Commercial Firm category agriculture Agroprocessing Manufacturing Construction Tourism Total Tax-payingfirms, 1997 Corporate income 29 46 41 80 54 46 VAT 19 80 80 96 79 74 Paid no taxes 7 11 8 4 7 8 Tax holidayfirms 1995 13 51 36 5 33 32 1996 15 50 37 13 26 31 ^ 1997 18 48 42 12 26 35 Disagreed with assessment 37 51 51 64 57 51 Resolution Negotiations 73 65 71 69 63 68 Court 0 0 0 0 0 0 Appeal 9 4 12 6 12 10 Unresolved cases 18 31 17 25 25 22 Total 100 100 100 100 100 100 Firms audited 41 69 71 80 71 68 Corporate income 26 46 35 72 43 41 VAT 30 59 66 68 64 60 (table continues onfollowing page) Table A9.1O continued Commercial Firm category agriculture Agroprocessing Manufacturing Construction Tourism Total Audit resulted in Additional taxes 20 31 30 30 50 32 Other costs 20 29 25 10 30 24 Firms with inputs VAT credit 87 79 76 100 76 81 Filed for refund 22 54 49 60 22 45 Received expected refund 4 11 29 20 18 19 Received less or equal 50 7 19 10 20 0 12 Received more than 50 7 12 2 12 0 6 Received no refund 4 12 8 8 4 8 Waiting periodfor VAT refund Up to 1 week 0 6 21 26 35 18 2-5 weeks 0 21 37 31 35 30 6-13 weeks 78 31 29 26 15 30 14-26 weeks 22 27 10 0 0 12 Over 26 weeks 0 15 3 17 15 10 Total 100 100 100 100 100 100 VAT Value-added tax. Note: Figures are a percentage of the total number of responses in each question. Source: Authors' calculations based on the 1998 enterprises survey 316 Duanjie Chen, John Matovu, and Ritva Reinikka References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Bahl, Roy. 1997. "Issues in Local Taxation in Uganda." World Bank, Eastern Africa Department, Washington, D.C. Processed. Broadway, Robin, Neil Bruce, and Jack M. Mintz. 1984. "Taxation, Inflation, and the Effective Marginal Tax Rate in Canada." Canadian Journal of Economics 27(1): 286-99. Chen, Duanjie, and Jack M. Mintz. 1993. "Taxation of Capital in Canada: An Inter-Industry and Inter-Provincial Comparison." In Allan Maslove, ed., Business Taxation in Ontario. Toronto: University of Toronto Press. Chen, Duanjie, and Ritva Reinikka. 1999. "Business Taxation in a Low- Revenue Economy. A Study on Uganda in Comparison with Neigh- boring Countries." Africa Region Working Paper no. 3. World Bank, Washington, D.C. Processed. Das-Gupta, Arindam, and Dilip Mookherjee. 1998. Incentives and Institutional Reform in Tax Enforcement: An Analysis of Developing Country Experi- ence. Oxford, U.K.: Oxford University Press. Davidson, Russell, and Jean-Yves Duclos. 1997. "Statistical Inference for the Measurement of the Incidence of Taxes and Transfers." Econometrica 65(6): 1453-65. Dunn, David, and Anthony Pellechio. 1990. Analyzing Taxes on Business In- come with the Marginal Effective Tax Rate Model. Discussion Paper no. 79. Washington, D.C.: World Bank. International Labour Organisation. 1997. Yearbook ofLabour Statistics. Geneva. Mayshar, Joram. 1988. "Note on Measuring the Marginal (Welfare) Cost of Taxation." Working Paper no. 175 (anuary). Hebrew University of Jerusalem, Department of Economics, Jerusalem. Processed. McKenzie, Kenneth, Jack M. Mintz, and Kim Scharf. 1992. "Measuring Effec- tive Taxes in the Presence of Multiple Inputs: A Production Based Approach." International Tax and Public Finance 4(3): 337-57. Mintz, Jack M. 1990. Corporate Holidays and Investment. World Bank Eco- nomic Review 4(1): 81-102. Republic of Uganda. 1995. "Input/Output Tables for Uganda (1989 & 1992)." Ministry of Finance and Economic Planning, Statistics Department, Entebbe. Shah, Anwar, ed. 1995. Fiscal Incentivesfor Investment and Innovation. Oxford, U.K.: Oxford University Press. A Quest for Revenue and Tax Incidence 317 Wildasin, David E. 1984. "On Public Goods Provision with Distortionary Taxes." Economic Inquiry 22(2): 227-43. World Bank. 1994. "The Private Sector in Uganda: Results of the World Bank Enterprise Survey." World Bank, Eastern Africa Department, Wash- ington, D.C. Processed. Yitzhaki, Shlomo. 1983. "On an Extension of the Gini Inequality Index." In- ternational Economic Review 24(3): 617-28. Yitzhaki, Shlomo, and Joel Slemrod. 1991. "Welfare Dominance: An Applica- tion to Commodity Taxation." American Economic Review 81(3): 480-96. Younger, Stephen D. 1996. "Estimating Tax Incidence in Ghana: An Exercise Using Household Data." In David E. Sahn, ed., Economic Reform and the Poor in Africa. Oxford, U.K.: Clarendon Press. Younger, Stephen D., David E. Sahn, Steven Haggblade, and Paul A. Dorosh. 1999. "Tax Incidence in Madagascar: An Analysis Using Household Data." World Bank Economic Review 13(2): 303-31. 10 The Cost of Doing Business: Firms' Experience with Corruption Jakob Svensson Firms in Uganda perceive corruption as one of the most serious impediments to conducting business. Despite this, little is known about the incidence and cost of corruption in the private sector, nor about its effect on firm perfor- mance, because until recently it was considered impossible to measure cor- ruption systematically. However, with appropriate survey methods and in- terview techniques, quantitative data on corruption can be collected. This chapter exploits such data from a recent survey of private enterprises in Uganda (see appendix B at the end of the book).' This chapter presents three main findings. First, firms typically must pay bribes when dealing with public officials whose actions directly affect the firms' business operations, and more than three-quarters of firms must pay bribes. Second, the amount paid could partly be explained by certain charac- teristics of the firm, such as profitability, which suggests that the amount paid in bribes depends on how much a firm can afford. Third, firms that pay higher bribes, on average, apparently do not receive more beneficial govern- ment favors in return. The econometric work suggests that the relationship between ability to pay and amount paid is not driven by reverse causation, and that bribery slows firm growth, far more than taxation does. Further- more, the time required to receive a public service is apparently not affected by the amount of bribes paid. These findings shed light on a hotly contested issue: the consequences of corruption on firm growth and performance. At a conceptual level this has been debated for several decades (for an excellent review, see Bardhan 1997). On the one hand, corruption is considered similar to a tax, with the primary 1. See Republic of Uganda (1998) for results of a household survey on corruption. 319 320 Jakob Svensson difference that the payment does not end up as public revenues.2 This "tax effect" reduces both the return to private capital (because part of output will be extracted in bribes) and the amount of internally generated funds or re- tained profits firms can use for capital investment. To the extent that corrup- tion also deprives the government of revenue required to provide produc- tive public goods, corruption is likely to slow growth more than taxation. In addition, the uncertainty and secrecy that necessarily accompany bribe pay- ments are likely to compound this difference (see Shleifer and Vishny 1993). On the other hand, proponents of the "grease argument" claim that in an economy plagued by bureaucratic delays, bribery allows firms to avoid tax and regulatory burdens and get things done faster. We find no support for the grease argument, but robust evidence that higher corruption is associ- ated with lower firm growth. The quantitative data are consistent with the firm managers' perceptions of corruption. Figure A10.1 displays the top five constraints (of 24 constraints listed in the questionnaire) as perceived by firm managers. In the sample of all firms, corruption (based on average values) is ranked as the fifth most serious constraint to business operations. Median values show that manag- ers perceive six areas, including corruption, as a major problem. When restricting the sample to subgroups of the sample population, cor- ruption is an even bigger problem. Figure A10.1 displays the top five con- straints for large firms (those with more than 100 employees), foreign-owned firms (majority foreign owned), and exporting firms. For both large and for- eign-owned firms, corruption is perceived as the second most important con- straint. Exporting firms' perceptions are very similar. Taken together, the results support the claim that corruption has a large adverse effect on firms. Of course, some firms may benefit from corrup- tion, possibly a great deal. Some firms may choose to compete based on costly preferential bureaucratic access-by devoting resources to obtain valuable licenses, preferential market access, control of privatized compa- nies, and so forth-instead of focusing on improving productivity. In cer- tain areas and for some firms, bribes may substitute for other costs, such as taxes. What this type of econometric work identifies is what is true on aver- age, or in general. On average, the Ugandan data reveal that corruption is a heavy burden on firms. The next section describes the data collection effort in detail, followed by a discussion of the general pattern of bribe payments with respect to incidence, level, and effect on firm growth. Next, three typical (or average) firms from subgroups of the sample are considered: one trying to obtain connection to public services, one involved in trade, and one paying taxes. Conclusions follow. 2. See Johnson, Kaufmann, and Shleifer (1997) on the public finance aspect of corruption and Bardhan (1997), Tanzi (1998), and Wei (1999) for reviews of ex- isting literature. The Cost of Doing Business: Firms' Experience with Corruption 321 The Data Can reliable data on corruption be collected? For a long time, the common view has been that given the secretive nature of corrupt activities, it is virtu- ally impossible to collect reliable quantitative information on corruption. However, Kaufmann (1997) forcefully argues that this presumption is incor- rect. With appropriate survey methods and interview techniques, firm man- agers are willing to discuss corruption with remarkable candor.3 The empirical strategy used to collect information on bribe payments across firms in Uganda featured the following six components: * An industry association (Ugandan Manufacturers' Association) car- ried out the survey. In Uganda, as in many other countries, people have a deep-rooted distrust of the public sector. To avoid suspicion of the overall objective of the data collection effort, the survey was done by a body in which most firms had confidence. * Corruption-related questions and the entire survey were carefully pi- loted and built on existing surveys on regulatory constraints. * Survey experts trained the enumerators. * Questions on corruption were phrased indirectly to avoid implicating the respondent of wrongdoing. * The corruption-related questions were asked at the end of the inter- view, by which time the enumerator presumably had established cred- ibility and trust. * To enhance the reliability of the corruption data, multiple questions on corruption were asked in different sections of the questionnaire. (Consistent findings across different measures significantly increase the reliability of the data.) The survey instrument had roughly 500 entries, with a handful of them related to corruption. The data collection effort was also aided by the fact that the issue of cor- ruption has largely been desensitized in Uganda. The past few years have seen several awareness-raising campaigns on the consequences of corrup- tion, and the media regularly and freely report on corruption cases (see Ruzindana, Langseth, and Gakwandi 1998; World Bank 1998). Incidence, Level, and Effects of Corruption The survey provides bribery data for 176 firms of 243 sampled. Of the 67 firms that did not respond to the main corruption questions, about one-third declined to answer other sensitive questions (for example, about costs and 3. The Ugandan enterprise survey (see appendix B), carried out during January- June 1998, was initiated by the World Bank and the Ugandan Private Sector Foun- dation. Its primary goal was to collect data on constraints facing private enterprises in Uganda. 322 Jakob Svensson sales). As a group, the approximately 40 firms that did not answer questions about corruption in particular did not differ significantly in size, profits, and location from the firms that did reply to corruption-related questions. Thus, no evidence suggests that the sample of 176 firms is not representative. Incidence Of the 176 firms that answered the question on bribe payment, 19 percent (33 firms) reported that they did not have to pay bribes, while 81 percent (143 firms) reported that they did. Table 10.1 shows noticeable differences between the two groups of firms. Nonbribing firms have characteristics suggesting they operate in sectors with little or no contact with the public sector, that is, in the informal sector. They receive significantly fewer public services, proxied by infrastructure services. They are less involved in foreign trade, proxied by share Table 10.1. Sample Characteristics Firms that reported Firms that reported Characteristic zero bribe payments positive bribe payments Infrastructure service provision 3.24 3.70b Export share 0.15 0.33a Pay tax index 2.58 3.04 Pay tax index (non-tax exempted only) 2.50 3.28a Time spent dealing with taxes and regulations (log) 1.93 2.49a Cost of accountant, and so forth (log) 3.30 4.74a Cost of security (log) 7.17 7.48 Incidence of robbery and theft 0.52 0.58 Size (log) 3.61 3.88 Note: Average values. Variable definition: infrastructure service = index (0-5) of availability of public services (electricity, water, telephones, waste disposal, paved roads), 1 if available, 0 otherwise, index is the sum of the binary availability variables for the five services; export = share of sales exported (1997); pay tax = index (0-6), sum of six binary (O = no, 1 = yes) variables reflecting types of taxes the firm pays (import duty, import commission, withholding tax, excise tax, VAT, corporate income tax) (1997); time spent dealing with taxes and regulations = percentage of senior management's time spent each month dealing with government regulations (1997); cost of accountant = monthly cost of accountant, lawyer, agent, specialized service provider to deal with regulation and taxes in US$ (1997); cost of security = annual cost of security in US$ (1997); incidence of robbery and theft = binary variable taking the value 1 if the firm was a victim of robbery, and/or theft during 1995-97, 0 otherwise; size = total employment (1997). a. Rejection of the null hypothesis that the two means are equal at the 5 percent level. b. Rejection of the null hypothesis that the two means are equal at the 10 percent level. Source: Author's calculations based on the 1998 enterprise survey. The Cost of Doing Business: Firms' Experience with Corruption 323 of output exported. They pay fewer types of taxes, particularly when control- ling for tax exemptions. These findings suggest that firms typically must pay bribes when dealing with public officials whose actions could seriously affect business operations. This interpretation is further supported by the finding that firms reporting positive bribe payments spend significantly more time dealing with government regulations, and spend more money on accountants and specialized service providers to deal with regulations and taxes. The results support the bureaucratic extortion model presented in Svensson (2000a) (see also Bliss and Di Tella 1997; Svensson 2000b). An inte- gral assumption in this model is that public servants have discretionary power within the given regulatory system to customize the nature and amount of harassment on firms to extract bribes. Svensson shows that the extent to which this can be done depends on how tightly the civil servants can control the firm's business decisions and influence the firm's cash flow. These indirect "control rights" stem from the existing regulatory system and the discretion bureaucrats have in implementing, executing, and enforcing rules and ben- efits that affect the firm, such as business regulation, licensing requirements, permissions, taxes, exemptions, and provision of public goods and services. The last two rows but one in table 10.1 show that the cost of security and incidence of robbery and theft is similar for the two groups. In fact, the cost of security per worker is higher for the nonbribing firms. Thus, while being in the informal sector where civil servants have few control rights over the firm's business operations insulates the firm from public corruption, it does not pro- tect the firm from other sources of discretionary redistribution, such as theft. The average firm in the nonbribing group has fewer employees-mostly be- cause of the existence of a few large firms in the nonbribing group-and the difference is significant if three outliers (large firms) are dropped from the nonbribing sample. Dropping these firms results in a significant difference between the two groups; larger firms are more likely to have to pay bribes. Level The evidence suggests that bribe payments constitute a heavy burden on firms. For the firms that reported positive bribes, the average amount of cor- rupt payments was about US$8,280, with a median payment of US$1,820.4 These are large amounts, corresponding, on average, to US$88 per worker, or roughly 7.9 percent of total costs (1 percent in the median). Including firms reporting zero bribe payments, the average payment is US$6,730 with a me- dian payment of US$450, or 6.4 and 0.5 percent, respectively. Approximately 50 percent of the firms reporting positive bribe payments pay more annually in graft than for security (including guards, equipment, 4. Using an exchange rate of US$1 = U Sh 1,100. 324 Jakob Svensson and so forth). Table 10.2 compares the size of reported graft with other cost items: wages, interest payments and cost of fuel. The cost of fuel, on average, was 6.3 percent of total costs; wages were 18.1 percent, and interest payments were 6.8 percent. The median values-which are significantly lower for cor- ruption but similar for fuel and wages-show that the variance on reported graft differs more than the variance in wage costs and fuel. Of the 167 firms for which data on both bribe payments and taxes are available, 70 percent reported higher bribe payments than corporate income taxes, with a median difference of US$800. This high number is partly driven by several small firms that do not pay corporate taxes. Still, the ratio of bribe payment to corporate taxes for the firms that paid corporate taxes averages 120 percent (and 31 percent at the median). Table 10.3 compares the size of reported graft and investment in machinery and equipment. A majority of Table 10.2. Comparison of Corruption and Other Costs (percent) Firms reporting positive graft Allfirms Category Mean Median Mean Median Corruption to total costs 7.9 1.0 6.4 0.5 Interest payments to total costs 6.8 0.0 8.3 0.0 Fuel to total costs 6.3 4.0 6.2 3.8 Wages to total costs 18.1 15.0 18.6 15.0 Note: Number of firms in sample of enterprises reporting positive graft = 132, and 164 firms in Hall firms" sample. Source: Author's calculations based on the 1998 enterprise survey. Table 10.3. Corruption and Investment (US$) Firms Firms reporting Firms reporting positive reporting positive investment Category Allfirms positive graft investment and graft Corruption (mean) 6,818 8,376 9,108 11,645 Investment (mean) 149,000 124,545 253,636 220,909 Corruption (median) 455 1,727 909 4,545 Investment (median) 1,136 418 27,273 37,273 Number of firms 172 140 101 79 Source: Author's calculations based on the 1998 enterprise survey. The Cost of Doing Business: Firms' Experience with Corruption 325 firms reported small or no investment in 1997. Consequently, almost 50 per- cent of the firms reported larger bribe payments than total investment. The distribution of bribes across firms is depicted in figures A10.2 and A10.3. Despite the careful data collection strategy, the sample likely has cases of misreporting, with the average graft numbers in particular being sensitive to such misreporting. However, this chapter is not concerned with the level of bribes per se, but rather on the correlates. The strategy used to collect infor- mation on graft should have minimized any systematic biases in the correla- tion between reported graft and the set of variables related to corruption. As mentioned earlier, evidence exists that firms that cannot avoid dealing ex- tensively with the public sector must pay bribes. Svensson (2000a) develops and tests a model in which the amount paid is a function of firm characteristics. The model's intuition is straightfor- ward. Thomas (1999) argues that the malfunctioning institutional system in many Sub-Saharan countries (lack of performance-based evaluations, discretionary dismissal powers) has given bureaucrats and office holders with hiring and firing power the opportunity to demand payments from those lower in the hierarchy (for a detailed analysis of the institutional sys- tem in several Sub-Saharan countries see Thomas 1999). Increased uncer- tainty of tenure has created strong incentives for those in government posts to quickly extract as much as possible to protect against impending unem- ployment or transfer to a less lucrative position. Consequently, many pub- lic institutions and bureaucrats act like a price discriminator with a focus of extracting rents. In such a system, a firm with higher current profits or expectations of higher profit in the future will be forced to pay higher bribes. Likewise, if the firm cannot credibly threaten to change its business activity or location, or import or export goods through other channels to avoid paying bribes, it will have to pay higher bribes. Firm characteristics such as profitability and the degree of reversibility of the installed capital stock thus determine the relative bargaining strength of the firm relative to the bureaucrats. Svensson (2000a) tests this hypothesis using data on current and expected profits, and a measure of the reversibility of the installed capital stock.5 Table 10.4 reports summary statistics on these variables. Firms reporting high cor- ruption (more than US$1,000) have significantly higher current profits, as well as higher expected future profits (proxied by employment size and capital stock), and use a production technology that would be costly to change (low 5. Expected future profits are proxied by the value of installed capital and em- ployment size. The opportunity cost of capital is the product of the resale value of capital times the degree of reversibility. The latter is measured as the difference be- tween resale and replacement value of capital after controlling for the age of the capi- tal stock. A negative value indicates that the firm's stock of capital is costly to move. 326 Jakob Svensson Table 10.4. Characteristics of Firms that Reported Positive Bribes Low bribe High bribe Variable Allfirms Allfirms' paymentsb paymentsa Profit (1997 US$) Mean 211,060 284,390 57,540 540,110 Median 27,270 27,270 11,230 95,690 Standard deviation 1,134,460 1,048,116 119,660 1,489,290 Bribes Mean 7,850 6,270 280 13,020 Median 910 910 180 9,090 Standard deviation 19,840 13,480 280 17,390 Capital stock (1997 US$) Mean 365,760 346,760 174,550 540,890 Median 90,910 90,910 38,640 227,270 Standard deviation 667,190 648,260 394,500 809,010 Employment (1997) Mean 119 109 36 192 Median 34 33 20 81 Standard deviation 262 251 53 346 Reversibility (log) Mean 0.001 0.001 0.002 0.000 Median 0.011 0.011 0.012 0.009 Standard deviation 0.034 0.034 0.033 0.035 Nunuber of observations 119 117 62 55 Note: Sample of firms for which data on corruption and other variables are available. Variable definition: profit = gross sales less operating costs and interest payments, capital stock = resale value of plant and equipment, reversibility = residual from the regressing of the ratio of resale to replace values of the capital stock to the average age of the capital stock and a constant (all variables in logs), employment = total employment. a. Excluding two extreme outliers. b. Low bribe payment is graft smaller than US$1,000. Source: Author's calculations based on the 1998 enterprise survey. reversibility). This is consistent with the rent-extraction hypothesis. How- ever, these are just partial correlates, and two valid objections are as follows: * Larger firms pay more bribes, but also make larger profits and have more capital installed; size is the determining factor. * Those firms that pay higher bribes receive valuable government fa- vors in return, and thus make larger profits; reverse causation is an influence. To check what mechanism best describes the data, Svensson (2000a) set up and tested the relationship between corruption, profit, capital stock, The Cost of Doing Business: Firms' Experience with Corruption 327 employment size, and the opportunity cost (defined as the degree of reversibility times capital stock) within a multiple regression framework.6 Table 10.5 depicts four of these corruption-level regressions, with different dependent variables (corruption in U.S. dollars, logarithm of corruption, cor- ruption per employee). Irrespective of specification, corruption is positively Table 10.5. Corruption Regressions Equation (I)a (2)b (3)c (4)cd Constant 8,701 8.83f 120.1f 112.89 (4,509) (0.892) (45.1) (54.5) Employment 11.399 0.0023f n.a. n.a. (4.76) (0.0004) n.a. n.a. Profit 0.0037f 5.5E-7f 0.0041f 0.0069f (0.0010) (1.OE-7) (7.4E-4) (0.0018) Opportunity cost -0.259f -3.5E-5f -0.2389 -0.260f (0.089) (1.3E-5) (0.091) (0.098) Capital stock 0.0059g 9.8E-7f 0.0042h 0.0037 (0.0023) (3.5E-7) (0.0022) (0.0024) Walde 29.63i 51.64i 36.20i 21.82i S.E. regression 12,168 1.74 123.0 128.2 Adjusted R2 0.18 0.35 0.21 n.a. Observations 117 117 117 117 n.a. Not applicable. Notes: All regressions are adjusted for selectivity (Heckman 1979), the inverse Mills ratio is not reported (see Svensson 2000a for details). Opportunity cost is the product of capital stock and reversibility. Variable definition: profit = gross sales less operating costs and interest payments in US$ (1997), capital stock = resale value of plant and equipment in US$ (1997), opportunity cost = product of the resale value of capital in US$ (1997) times the degree of reversibility, reversibility = residual from a regression of the ratio of resale to replace values of the capital stock on the average age of the capital stock and a constant (all variables in logs), employment = total employment (1997). a. Dependent variable is bribe payments in US$. b. Dependent variable is log of bribe payments in US$. c. All variables scaled by employment. d. 2SLS estimation. e. Test statistic for the hypothesis that the coefficient on employment, profit, opportunity cost, and capital stock is zero. f. Significant at the 1 percent level. g. Significant at the 5 percent level. h. Significant at the 10 percent level. i. Rejection of the null hypothesis of zero coefficients at the 1 percent level. Source: Author's calculation based on 1998 enterprise survey 6. Profit is defined as gross sales less operating costs and interest payments. The capital stock is measured as the resale value of plant and equipment, and labor force is total employment. All data are for 1997 and the monetary values expressed in U.S. dollars. 328 Jakob Svensson correlated with current profits, expected future profits, and the opportunity cost of capital. After controlling for size, firms with higher profits pay more in bribes and firms with better outside options pay less. The results also sug- gest that for most firms, more investment (through higher expected profits) implies that more bribes need to be paid. The last column in table 10.5 deals with the potential endogeneity prob- lem by instrumenting for profits using a set of firm-specific variables that arguably are uncorrelated with both the error term in the regression and reported bribes, but are correlated with firms' profit potential (and real- ized profits). The instrument set includes proxies of human capital, age of the firm, a measure of foreign ownership, distance to the main trading center (the capital Kampala), and the cost of security per employee. As shown by Reinikka and Svensson (chapter 7 in this volume), measures of human capital are correlated with productivity and profits. Distance to the main trading center presumably affects firms' operating costs, and risk arising from, for example, crime, has been found to be an important deter- minant of performance. The results in the final column support the claim that, on average, the level and rate of graft are influenced by firms' abilities to pay. The instru- ments perform well, picking up around 6 percent of the variation in profits across firms, and we cannot reject the null hypothesis of the validity of the instruments; that is, we find no evidence that the instruments for the profit rate belong in the corruption regression.7 These results do not prove that bribe-paying firms do not receive preferential government treatment. They may benefit, but the results suggest that the firm's ability to pay determine the price of this benefit. Table 10.6 shows the effects on corruption (bribe payment) of both a one standard deviation increase in the explanatory variables (column 1), and a 1 percent increase in the explanatory variables (column 2). The calculations show; for example, that a one standard deviation increase in profits per em- ployee is associated with roughly US$100 in additional bribe payments per employee (equal to 0.76 standard deviations), while a 1 percent increase in the capital stock results in a 0.22 percent increase in bribes paid. Effects So far the analysis has focused on who, why, and how much firms need to pay in bribes. A logical follow-up question is "What are the effects?" From the pre- vious two subsections it is obvious that evaluating the effects of corruption (such as on firm growth) is a tricky exercise. The problem is identification, because both growth and corruption are likely to be jointly determined. For example, 7. Svensson (2000a) also experiments with other sets of instruments. The results remain similar to those reported in table 10.5 (column 4). The Cost of Doing Business: Firms' Experience with Corruption 329 Table 10.6. Effects on Corruption of Changes in Firm Characteristics (l)a Change in bribe payment per employee due to a one standard (2)b Change in bribe deviation increase payment due to a 1 Equation in (US$) percent increase in (%) Capital stock per employee 25.5 n.a. (0.19) Profits per employee 104.2 n.a. (0.76) Reversibility index -42.1 -0.118 (-0.31) Capital stock n.a. 0.218 Profits n.a. 0.152 Employment n.a. 0.632 n.a. Not applicable. Note: Variable definition: profit = gross sales less operating costs and interest payments in US$ (1997), capital stock = resale value of plant and equipment in US$ (1997), reversibility = residual from a regression of the ratio of resale to replace values of the capital stock on the average age of the capital stock and a constant (all variables in logs), employment = total employment (1997). a. Calculations based on regression 1, Table 10.5, with standard deviations in parentheses. b. Calculations based on Svensson (2000a). Source: Author's calculation based on the 1998 enterprise survey. consider two firms of similar size and age in a given sector. One of the firms produces a good or brand perceived to have a favorable demand forecast, while the other firm produces a good with much less favorable demand growth. As- sume that both firms must clear certain business regulations and licensing re- quirements or require some public infrastructure services. Also assume that public servants have discretion in implementing and enforcing these regula- tions and services. A rational rent-extracting bureaucrat would try to extract as high a bribe as possible. In this arrangement, a bureaucrat would be expected to demand higher bribes from the firm producing the good with a favorable de- mand forecast, simply because this firm's expected profits are higher, and thus its ability to pay is greater. If the forecasts also influence the firms' willingness to invest and expand, a positive (observed) relationship between corruption and growth would be expected when comparing these two firms. Fisman and Svensson (2000) try to overcome this simultaneity problem by instrumenting for bribes using industry-location averages as instruments.8 They argue that if this problem is specific to firms, but not to industries or 8. Fisman and Svensson (2000) show that the IV-technique employed is likely to provide a lower bound (in absolute terms) of the effects of bribery on growth. 330 Jakob Svensson locations, netting out this firm-specific component yields a bribe measure that depends only on the underlying characteristics inherent to particular industries and locations. For example, in industries, the number of produced goods sold abroad, import reliance, and dependence on publicly provided infrastructure services all determine to what extent bureaucrats can extract bribes. Figure 10.1 illustrates the key findings in Fisman and Svensson (2000). The higher the average bribery-to-sales rate, the lower the growth rate. As evident, the effect is of considerable magnitude. A 1 percentage point in- crease in the rate of "required" bribe payments reduces a firm's annual growth rate by about 3.5 percentage points. Fisman and Svensson (2000) also compare the effects of corruption on growth with the effects of taxation on growth. They find that in the whole data set, the negative effect of bribery on firm growth is more than three times greater than that of taxation on growth. Moreover, after excluding out- liers, they find a much greater negative impact of bribery on growth and a considerably attenuated effect of taxation. This provides some validation for firm-level theories of corruption, which posit that corruption retards the de- velopment process much more than taxation. Figure 10.1. Corruption and Growth 25 20 - 1 15 10 0 0 0.02 0.04 0.06 0.08 Bribe rate Note: The bribe rate is the average bribery to sales rate. The bribe rate varies from 0 to 0.075 (7.5 percent) in the sample. The graph is based on the results reported in Fisman and Svensson (2000) and is evaluated at the mean of the controls initial sales (in logarithms), firm's age (in logarithms), and the average tax to sales rate. Source: Fisman and Svensson (2000). The Cost of Doing Business: Firms' Experience with Corruption 331 It is worth repeating that in reality, some firms may still benefit-and possibly a great deal-from corruption. This type of analysis identifies what is true on average, and the Ugandan data suggest that in general, there is a strong negative relationship between bribery payments and firm growth. Case Studies The experience of three typical firms-one trying to obtain public services, one involved in trade, and one paying a range of taxes-is described based on the survey data. These experiences are not based on one specific firm in each category, but on three average firms with these specific characteristics. Case Study 1: Getting Connected Although reported bribe payments are the key corruption variable frequently used, other methods exist for collecting objective data pertaining to corrup- tion. Specifically, cost data on providing homogeneous public services (goods) can reveal evidence of corruption. The survey collected information on two variables related to delivery of public services. The respondents were asked to report the total costs, including informal payments, of acquiring a connec- tion to the public grid and acquiring a telephone line. The fee for a telephone connection (around US$100) is supposed to be fixed. Thus, deviations from the given price typically reflect graft. Connec- tion costs to the public grid are more problematic, and are a complex func- tion of load requirements, necessary upgrades, and distance to existing volt- age connection. The complexity in determining the price of connection implies that the public electricity company has large discretion over the cost. Data on costs of acquiring a telephone line were obtained from 90 firms.9 Of those 90 firms, 83 percent (75 firms) reported costs above the fixed price. On average, a firm paid US$130 in addition to the fixed price, more than twice the stated cost to acquire a telephone line. The average firm had to wait approximately 13 weeks to get connected. No relationship exists be- tween connection cost and time waited. The simple correlation is 0.04. This stands in stark contrast to the "efficiency grease" hypothesis that predicts a negative correlation between bribes and bureaucratic delays, but is in ac- cordance with the basic hypothesis laid out earlier. If public sector employ- ers have discretion over implementation, delays are endogenously deter- mined to explicitly extract bribes. Figure A10.4 and table 10.7 present evidence that the excess cost paid by firms constituted informal payments (bribes). Of the 75 firms that reported excess connection costs, 13 did not report bribe data. For the remaining 62 9. Two extreme outliers (reporting errors) were dropped from the sample of firms reporting connections to the telephone system and the public grid. 332 Jakob Svensson Table 10.7. Partial Correlation between Connection Costs and Bribery Equation dependent (1)a,b Connection costs (2)c Excess cost of variable to public grid (log) telephone connection (log) Constant 9.162 10.75 (0.000) (0.000) Bribe payments (log) 0.508 0.068 (0.000) (0.001) Adjusted R2 0.44 0.15 Number of observations 25 62 Note: Standard errors adjusted for heteroskedasticity (White 1980). p values in parentheses. a. Regression 1 includes a proxy of informality (infrastructure service). b. Connection costs (public grid) has mean U Sh 6,330,400 and median U Sh 2,500,000. c. Excess cost of telephone connection has mean U Sh 155,600 and median U Sh 90,000. Source: Author's calculations based on the 1998 enterprise survey firms there is a high correlation between the excess cost and reported bribe payment (the simple correlation is 0.41), as illustrated in figure A10.5. Table 10.7, column (2), reports the simple regression of corruption on excess cost. Excess cost of connection is highly correlated with reported bribe payment. Of the 29 firms that obtained a connection to the public grid during 1995- 97,25 answered the question on bribes, and all 25 reported paying bribes. On average a firm paid US$5,540 for connection to the public grid, with the me- dian firm paying roughly US$2,700, and waited a little more than 12 weeks to get the connection. Part of the cost of connection may be caused by rea- sons other than corruption, in particular, the firm's distance from an existing voltage connection. The survey has no data on this, but used an infrastruc- ture service provision index indicating access to basic public services, such as water, electricity, telephones, waste disposal, and paved roads, as a rough proxy of the proper cost adjustment for location. The maintained hypothesis is that the infrastructure service provision index is likely to be highly corre- lated with distance to existing power connections. Table 10.7, column (1), displays the result of regressing reported bribe payment and the infrastructure service provision index on the cost of ob- taining a connection to the public grid. Both variables enter highly signifi- cant, thereby providing evidence that high cost of connection is linked both to location-specific characteristics and corruption. Figure A10.4 shows the partial correlation (controlling for location) between connection costs and bribes (0.67). Again, the time to get connected and the cost (controlling for location) is not correlated (partial correlation is 0.08). These findings are consistent with recent empirical results from other developing countries. Kaufmann and Wei (1999) examine the relationship between perception of corruption and management time wasted with bureaucrats. Contrary to the efficient grease argument, they find that firms that face more "bribe The Cost of Doing Business: Firms' Experience with Corruption 333 demand" are also likely to spend more management time with bureaucrats rather than less. These results have two clear implications. First, collecting data on pro- vision of homogeneous public services (goods) is a potentially fruitful way to collect evidence of corruption indirectly. The data reveal that the provi- sion of public services provides a powerful tool to extract bribes. Second, the data also suggest that clearer rules can improve the situation from the firms' perspective. The relationship between bribe payments and excess cost of telephone connection is weaker than that between bribe payments and cost of getting connected to the public grid. However, clearer rules are not sufficient if no mechanisms exist for accountability of the public sector charged with providing public goods. Thus, even though a set price for a telephone connection theoretically exists, most firms must pay significantly more for a telephone line. More generally, the finding suggests that fight- ing corruption is not purely a technical problem. Although reforms of rules and regulations are important, the focus must be on creating a sustainable, credible, and ongoing system of accountability of public institutions and public servants. Case Study 2: Exporting and Importing Being engaged in trade, either exporting or importing, typically implies that a firm must pay bribes. In fact, 91 percent of the trading firms reported posi- tive bribe data, with an average level of graft equal to US$9,800 (the median was US$2,050). Consistent with these findings, the median exporter perceived corruption as a major problem (see figure A10.4). No evidence exists that bribes speed up the process of getting goods in or out of the country. For the average firm, imported goods require 66 days to arrive. It takes 30 days from the original shipping port (typically in Eu- rope) to Mombassa or Dar es Salaam, an additional 27 days from the port to the clearance point (Nakawa inland terminal), and 9 more days from the clearance point to the firm. Thus, firms involved in trade face additional costs because of both corruption and inefficient public services. Again, pro- vision of necessary services such as public transport and clearances gives corrupt civil servants a mechanism to extract bribes. Case Study 3: Paying Taxes Firms that pay fewer types of taxes also face a lower probability of paying bribes, particularly when controlling for tax exemptions. On average, the se- nior management in a firm that pays a majority of taxes spends almost 20 per- cent of its time dealing with government officials regarding taxes, permits, regulations, and so forth. The cost of accountants, lawyers, and auditors to deal with taxes and regulations cost the median enterprise nearly US$3,300 a year. As table 10.8 shows, the level and rate of bribes are significantly higher 334 Jakob Svensson Table 10.8. Differences in Tax Assessment and Corruption Tax assessment Tax assessment Category differs by 0-50% differs by 51-100% Graft rate (bribery US$/employment) 59.6 157.6 Graft level (US$) 4,530 14,450 Time spent dealing with taxes, etc. (%) 15.1 16.5 Number of firms 37 18 Source: Author's calculations based on the 1998 enterprise survey. for firms reporting large differences between their assessment of taxes to be paid and the tax authority's assessment. On average, a firm with a difference in tax assessment of more than 50 percent pays three times as much in bribes as a firm reporting a difference in tax assessment less than 50 percent. Ample anecdotal evidence suggests that the tax system provides bureau- crats with a potentially powerful tool to extract bribes. The firm survey evi- dence supports this assertion, although it is difficult to separate the benefits of lower taxes paid because of the bribe and the actual cost of the bribe. How- ever, it is safe to conclude that with respect to the tax system, the biggest loser is the public, because corruption deprives the government of income required to provide public goods and services. The relationship between number of taxes and corruption also has impli- cations for tax policy in general. Streamlining the number of taxes and sim- plifying the tax code can help mitigate the problem, and the tax reform en- acted in 1997 aimed at this (see chapter 9 in this volume). However, the simplification must be followed by auditing and accounting standards, and these standards must be applied both to the firms and to the tax authority. Conclusions Ugandan firms perceive corruption as one of the most serious impediments to conducting business. However, until recently it has been considered im- possible to measure corruption systematically. No data were available con- cerning the incidence and cost of corruption in the private sector or how much it affected firms' performance. With appropriate survey methods and interview techniques, however, quantitative data on corruption can be col- lected. The data show that firms typically must pay bribes when dealing with public officials whose actions directly affect the firms' business opera- tions. Such dealings cannot easily be avoided when exporting, importing, or requiring public infrastructure services. The data reveal that more than 80 percent of the firms must pay bribes during a typical business year. The amount paid could partly be explained by firm-specific characteristics, such The Cost of Doing Business: Firms' Experience with Corruption 335 as current and expected future profits and the reversibility of the capital stock. This suggests that the amount paid in bribes is not a flat fee for a given service provided by a public official, but a proportional tax on prof- its: the more the firm can pay, the more it will have to pay. In other words, the "price" for a given public service depends on ability to pay No evi- dence exists that firms that pay higher bribes, on average, receive more beneficial government favors in return. In fact, the rate of bribery is nega- tively correlated with firm growth. The negative effect of bribery on firm growth is more than three times greater than the effect of taxation on growth. The chapter has argued that clearer rules with respect to taxes and public service provision can help mitigate the problem. However, without institu- tionalized mechanisms for accountability of the public sector these changes will be insufficient. These mechanisms include both formal or government induced measures-it is important to select measures that are in line with Uganda's implementation capabilities-and measures to empower civil so- ciety and the private sector. Collective action or measures on the part of the business community could include the following: * Collecting and disseminating information about corrupt practices * Informing the private sector and the public about service standards, guidelines, and norms of major service providers * Increasing individual firm's ability to commit to no bribery * Recognizing those who are making efforts to resist corrupt practices. As Paul (1997) argues, corruption generally can be effectively tackled only when reform of the political process and restructuring of regulatory systems are complemented by a systematic effort to increase citizens' ability to moni- tor and challenge abuses of the system and to inform citizens about their rights and entitlements. Breaking the culture of secrecy that pervades the functioning of the government and empowering people to demand public accountability are two important components of such an effort. Recent reviews of the growth performance of Sub-Saharan Africa have identified recurring features of African politics that are likely to undermine the results of traditional institutional reforms such as tax reforms. These in- clude restricted civil society involvement, perceptions of the state as a ve- hicle of wealth accumulation, prevalence of patronage politics, and a small elite with close political connections. Although each may not be applicable to every country, a successful national anticorruption program must also tackle these fundamental determinants of corruption. 336 Jakob Svensson Annex 10.1. Ranking of Constraints and Payment of Bribes Figure A10.1. Ranking of Constraints to Investment by Firm Category, 1998 Allfirms Utility prices Taxes Poor utility . __ Corruption I Cost of finance _ ~ ~ ~ ~ ~ ~ ~~~~~I I II 1 2 3 4 5 No Small Moderate Major Severe problem problem problem problem problem 3 Mean * Median Largefirms Utility prices _ Corruption Taxes Poor utility Tax administration _ I ~ ~ ~ ~ ~~~~~I II 1 2 3 4 5 No Small Moderate Major Severe problem problem problem problem problem [1 Mean * Median (figure continues onfollowing page) The Cost of Doing Business: Firms' Experience with Corruption 337 Figure A10.1 continued Foreign firms Utility prices Corruption Taxes Poor utility _ Cost of finance _ I ~ ~ ~ ~ ~~~~I I I 1 2 3 4 5 No Small Moderate Major Severe problem problem problem problem problem O Mean * Median Exportingfirms Utility prices Taxes Poor utility Corruption Cost of finance _ ~ ~I - 1 TI~ 1 2 3 4 5 No Small Moderate Major Severe problem problem problem problem problem O Mean * Median Source: Author's calculations based on the 1998 enterprise survey. 338 Jakob Svensson Figure A10.2. Distribution of Firms According to Logarithm of Bribe Payments in U.S. Dollars 35 30 25 120 0 10 5 Bribe payment (log) Source: 1998 enterprise survey. Figure A10.3. Distribution of Firms According to Bribe Payments 70 60 50 0,40 - 40 j30- 20 - 10 10--_ I I I 1I_ I 1 I -I I I'|I 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 10,000 US$ Source: 1998 enterprise survey. The Cost of Doing Business: Firms' Experience with Corruption 339 Figure A10.4. Correlation between Graft and Excess Cost of Telephone Connection 21 - 19- 17 - U * 0~~~~~~ ; 15- * 13- * U 11 9 9 10 11 12 13 14 Excess cost of telephone connection Source: Author's calculations based on the 1998 enterprise survey Figure A10.5. Partial Correlation between Graft and Connection Costs to Public Grid 4 U U 2 .~~~ m 2 . ......................... ....... .................... ..... ...... * * -2 -4- l -4 -2 0 2 4 Connection costs to public grid (log) Source: Author's calculations based on the 1998 enterprise survey 340 Jakob Svensson References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Bardhan, Pranab. 1997. "Corruption and Development: A Review of Issues." Journal of Economic Literature 35(September):1320-46. Bliss, Christopher, and Rafael Di Tella. 1997. "Does Competition Kill Corrup- tion." Journal of Political Economy 105(October): 1001-23. Fisman, Raymond, and Jakob Svensson. 2000. "Are Corruption and Taxation really Harmful to Growth? Firm Level Evidence." Policy Research Working Paper no. 2485. Development Research Group, World Bank, Washington, D.C. Heckman, J. 1979. "Sample Selection Bias as a Specification Error." Econometrica 47: 53-161. Johnson, Simon, Daniel Kaufmann, and Andrei Shleifer. 1997. "The Unoffi- cial Economy in Transition." Brookings Papers on Economic Activity 2: 159-239. Kaufmann, Daniel. 1997. "Corruption: Some Myths and Facts." Foreign Policy (summer): 114-31. Kaufmann, Daniel, and Shang-Jin Wei. 1999. "Does Grease Money Speed up the Wheels of Commerce?" Policy Research Working Paper no. 2254. World Bank, Development Research Group, Washington, D.C. Paul, Samuel. 1997. "Corruption: Who Will Bell the Cat?" Economic and Politi- cal Weekly 32: 1350-55. Republic of Uganda. 1998. "National Integrity Survey. The Report to the In- spectorate of Government." Kampala. Ruzindana, Augustin, Petter Langseth, and Arthur Gakwandi, eds. 1998. Fight- ing Corruption in Uganda: The Process of Building a National Integrity System. Kampala: Fountain Publishers. Shleifer, Andrei, and R. W. Vishny. 1993. "Corruption." Quarterly Journal of Economics 108: 599-617. Svensson, Jakob. 2000a. "Who Must Pay Bribes and How Much? Evidence from a Cross-Section of Firms." Policy Research Working Paper no. 2486. Development Research Group, World Bank, Washington, D.C. . 2000b. "Foreign Aid and Rent-Seeking." Journal of International Eco- nomics 51(2): 437-61. Tanzi, Vito. 1998. "Corruption Around the World: Causes, Consequences, Scope, and Cures." IMF Staff Papers 45: 559-94. The Cost of Doing Business: Firms' Experience with Corruption 341 Thomas, Melissa. 1999. "The Incentive Structure of Systemic Corruption." World Bank, Washington, D.C. Processed. Wei, Shang-Jin. 1999. "Corruption in Economic Development: Beneficial Grease, Minor Annoyance, or Major Obstacle." Policy Research Work- ing Paper no. 2048. Development Research Group, World Bank, Wash- ington, D.C. White, H. 1980. "A Heteroscedasticity-Consistent Covariance Matrix Estima- tor and a Direct Test for Heteroscedasticity." Econometrica 50: 1-16. World Bank. 1998. "Uganda: Recommendations for Strengthening the Gov- ernment of Uganda's Anticorruption Program." Mission Report. Af- rica Region, Poverty Reduction and Social Development, Washing- ton, D.C. 11 Recovery in Service Delivery: Evidence from Schools and Health Centers Ritva Reinikka It is commonly held that Uganda had a well-functioning social service deliv- ery system in the 1960s. The subsequent economic and social decay all but decimated this system, however. Undoubtedly, institutional recovery is more complex than implementing policy reforms by "a stroke of the pen." While evidence on economic performance is fairly readily available, much less infor- mation exists on Uganda's institutional recovery during the past 15 years, ei- ther in terms of institutional assessments or systematic recording of perfor- mance indicators. This chapter sheds light on service delivery in education and health. The two subsequent chapters, which explore household responses to recent policy initiatives in these two sectors, complement this analysis. The principal motivation for the study reported in this chapter was the substantial increase in public spending on basic services, albeit from a small base, since Uganda's recovery started in the late 1980s, while several offi- cially reported outcome and output indicators remained stagnant. The most obvious disparity in output indicators was in primary school enrollments. Despite increases in budgetary allocations for education, officially reported enrollments increased only slightly during the first half of the 1990s. The hypothesis for the study was that actual service delivery, or output, was much worse than budgetary allocations implied because public funds, or inputs, were subject to capture by bureaucrats and did not reach the intended facili- ties (see, for example, Bardhan and Mookherjee 1998). To test this hypoth- esis, the study's author compared budgets and actual spending in the pri- mary education and health care sectors. This chapter draws onAblo and Reinikka (1998). Comments by jakob Svensson are greatly appreciated. 343 344 Ritva Reinikka While this chapter does not attempt a comprehensive analysis of public sector efficacy, the government's ability to translate budgetary allocations into actual spending at the facility level can be a useful proxy for it. As adequate public accounts are not available in many African countries, including Uganda, a diagnostic survey of schools and clinics was carried out to collect actual spend- ing data.' Survey work is typically limited to examining the effects of policies and interventions on households and firms, while inputs, such as flows of public funds, and outputs, such as primary enrollments, are left solely for official statistics or administrative records. As this study shows, a diagnostic survey of the supply side can provide a useful reality check when institutions are weak and official statistics are not a reliable guide for policymakers. While the Ugandan school survey results indicate some improvement in the input flow to service facilities during 1991-95, particularly in salary payments, they also confirm a serious lack of accountability. For example, only 2 percent of public nonwage education spending reached the schools in 1991, and only 20 percent in 1995. If efficiency of input flow is an indica- tor of the extent of institutional recovery, by 1995 this recovery was limited at best. The dismal situation revealed by this school survey sparked action by the central government, which began publishing information about monetary transfers to districts and demanding that transfer information be posted at schools and district headquarters. A recent replication of the school survey shows that schools now receive more than 90 percent of the nonwage spending intended for them, although often with delay (Republic of Uganda 2000). Hence, at least in some areas, institutional recovery in Uganda ap- pears to be accelerating. The 1996 school survey unearthed other important information critical to understanding the education delivery system and the efficacy of potential interventions. First, instead of being stagnant as official statistics indicate, primary enrollments increased by 60 percent in 1991-95. This indicates that, while input flow suffered from major problems, education system perfor- mance in the first half of the 1990s improved more than the information sys- tem that reports it. Furthermore, in 1997 the universal primary education initiative, discussed in chapter 12, resulted in a sudden increase in enroll- ment as households responded strongly to the president's election pledge of free education for four children per family. Second, the survey showed that public primary education was mostly funded by parents who, on average, contributed as much as 73 percent of total school spending in 1991 (42 percent at the median school). When the govern- ment retreated from funding and managing primary schools during the re- pressive Amin and Obote regimes, parents took over. The survey data show 1. In 1990, the government initiated efforts to develop and implement a finan- cial tracking system for primary education and health (Republic of Uganda 1990, 1992). These efforts bore little or no fruit. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 345 that by 1991 this situation had not changed much. However, the government's share increased during the survey period, and by 1995 parents financed 60 percent of total school spending on average (at the median school the parental share was roughly halved to 23 percent). Strikingly, parental contributions con- tinued to increase in real terms despite higher public spending. The health facility survey showed that these facilities did not keep sys- tematic financial or patient records in 1991-95. Therefore, assessing the flow of funds or services delivered was not possible. The public service facilities in the two sectors seem to vary their institutional behavior depending on the institutional context and incentives. However, limited, recent evidence from four districts shows that operations such as opening hours and staff avail- ability, as well as recordkeeping, have improved in health facilities since 1996 (World Bank 1999). The prevailing normative view of government assumes that once the right policy or intervention has been found-to correct market failure or externali- ties or to achieve a better distribution of income-the government imple- ments it as designed, and the desired effects will follow. Some view govern- ments as benevolent single agents, behaving in the same way everywhere in the world, and policymaking as a technical problem rather than a political process that varies between countries (Dixit 1996). New theoretical litera- ture, however, takes a more nuanced view by differentiating governments as providers of public goods. Svensson (1997), for example, finds that as society's polarization and degree of social conflict increase, the control of public policy is less effective. This results in more public spending, but fewer public goods. This emphasizes the importance of separating the effects of public capital on welfare from the effects of public spending on public capital. Pritchett (1996) argues that governments differ from the private sector in the degree to which they behave as profit-maximizing investors. If public in- vestment is guided by motives other than profitability, the cost of cumulated public capital is likely to be higher than its value in terms of future returns. Therefore, using investment cost to measure public capital across countries may be misleading. Similarly, as demonstrated in this chapter, using budget allocations to measure actual frontline service delivery may be misleading. Several recent empirical papers also highlight the divergence between the actual and potential impact of public spending on health outcomes in developing countries. Filmer and Pritchett (1999) find that 95 percent of cross- national variation in child mortality can be explained by factors not related to health policy, such as per capita income, income distribution, female edu- cation, and various cultural factors. Meanwhile, the impact of public spend- ing-typically measured by budget allocations-is very small and statisti- cally insignificant. The rest of this chapter is divided into four sections. The first section briefly describes the diagnostic survey carried out in Uganda in 1996. The next sec- tion examines official data on primary enrollments over time and compares them with the facility survey data for 1991-95. It presents the main results of 346 Ritva Reinikka the primary school survey with respect to actual public and private spend- ing at both the national and regional level. The chapter then explores service delivery and public spending on primary health care. Finally, the chapter concludes and summarizes the policy changes the government introduced following the survey findings. The concluding section also highlights recent evidence on improvements made since 1996. Diagnostic Survey Ideally, the public accounting system should provide timely information about actual spending on various budget items and programs. This is not often the case in many low-income countries. Because the revival of the accounting sys- tem has been slow in Uganda, a field survey was necessary to gauge the extent to which public resources actually filtered down to the intended facilities. A survey of 19 districts covering 250 government-aided primary schools and nearly 100 health clinics was carried out in 1996, covering the period 1991-95.' Apart from school and health unit income and expenditure, the objective of the survey was to collect data on primary enrollments and patient records at the facility level. From 10 to 20 schools were visited in each district.3 Of the districts sur- veyed, Bushenyi had the most primary schools, with 399 in 1994, while Bundibugyo had the least, with 59. In the districts with fewer than 100 government-aided schools, the enumerators visited 10 schools; in districts with 100-200 schools, they visited 15; and in those with more than 200 schools, they visited 20. The primary school-leaving examination results, 2. For the sample selection, the country was first divided into regions. To bring out regional differences more clearly, the traditional four regions (north, east, west and central) were reconfigured into seven regions, namely: northwest, north, north- east, east, central, southwest, and west. Kampala was treated as a separate region because it enjoys many advantages over the rest of the country. The 39 districts were then arrayed into 3 groups, based on the fiscal year in which a particular district first received a separate budget vote under the decentralization program which commenced in 1993. The objective was to pick one district per region in each successive phase of decentralization. In practice, only two districts were selected from the smaller re- gions. After some other minor adjustments, the following 19 districts were selected: Kampala; Arua, Moyo (northwest); Apac, Gulu (north); Soroti, Moroto, Kapchorwa (northeast); Jinja, Kamuli, Pallisa (east); Mukono, Mubende, Kiboga (central); Bushenyi, Kabale (southwest); and Kabarole, Hoima, Bundibugyo (west). Kiboga, which is a new district, had to be dropped subsequently because of limited data availability. 3. At the time of the survey, there were about 8,500 government-aided primary schools, which were supposed to receive a large proportion of their funding from central and local governments. The rest of the schools, about 1,500, were either pri- vate or community schools. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 347 supplemented by information about school facilities, were used as criteria to select schools within a district. Both good and poor performers were included in the stratified random sample. Every district had many more schools than health facilities. Two districts- Kapchorwa and Kisoro-had no government health centers, while some had as many as 10. In some cases, missionary, private, or nongovernmental orga- nization (NGO) facilities compensated for the lack of government facilities. Enumerators visited five primarily government facilities in each district, such as two health centers, two dispensaries/maternal units, and one aide post, or some other combination. Most of the enumerators who collected the data from schools and clin- ics were former teachers and health workers who lived in those districts. They used standardized forms, and supplemented the quantitative data with qualitative observations. Enumerators were trained and closely su- pervised by a joint Ugandan-World Bank research team to ensure quality and uniformity of data collection and to assess the standard of recordkeeping in schools and clinics. Education and Public Spending Before the introduction of free universal primary education in 1997, official data indicates that primary school enrollment in government schools was almost stagnant for 10 years (table 11.1). Because the number of children of Table 11.1. Official Enrollment Data from Government-Aided Primary Schools, 1987-97 Number of Number of Number of Year schools teachers students (millions) 1987 7,627 72,970 2.31 1988 7,905 75,551 2.42 1989 7,684 81,418 2.53 1990 7,667 81,590 2.28 1991 8,046 78,259 2.54 1992 8,325 86,821 2.36 1993 8,430 91,905 2.67 1994 8,442 84,043 2.60 1995 8,531 - 2.64 1996 - 82,600 2.74 1997a 10,000 98,700 5.30 - Not available. a. These data are from a nationwide headcount of pupils and teachers in August. Source: Ministry of Education data. 348 Ritva Reinikka primary school age had increased along with high population growth, it fol- lows that net primary enrollment rates must have fallen.4 The official data cannot, however, be easily verified without going to the school level because the districts kept virtually no reliable educational statis- tics at the time. The well-developed recordkeeping of the 1960s broke down during the political and military turmoil of the 1970s and early 1980s, and had not recovered by mid-1990. The main source of official data for primary enrollments was the annual school census carried out by the Ministry of Edu- cation, which sent questionnaires to district education officers. The officers sent them on to schools, which returned the questionnaires through the same channel. Fieldwork by the school census staff was minimal. Chapter 12 discusses in more detail the free universal primary education for four children per family introduced in January 1997. This substantially increased primary enrollment, which rose to 5.3 million students, based on a nationwide headcount later in 1997, revealing a high private demand for education. Most of the increase was in the first grade (P1). Both underaged and overaged children entered P1 in 1997, producing an exceptionally large cohort of 2.1 million children. The school survey results, however, did not correspond with the trend in the official enrollment figures (table 11.2). Instead of being stagnant, pri- mary enrollment in the sample schools increased 60 percent between 1991 and 1995. The overall student-teacher ratio increased from 26:1 in 1991 to 37:1 in 1995. The survey results seem more plausible than the official fig- ures, given the continuous improvement in the political and socioeconomic environment and public finance since 1986. As the survey was based on a careful examination of individual school records, it suggests that the Table 11.2. Enrollment Data from Surveyed Schools, 1991-95 Year 1991 1992 1993 1994 1995 Number of students 81,318 90,330 109,063 119,919 129,087 Annual increase in students (%) 8 11 21 10 8 Number of teachers 3,077 3,312 3,663 3,897 3,498 Annual increase in teachers (%) - 8 11 6 -10 - Not available. Source: School survey. 4. The 1992/93 integrated household survey recorded an average gross primary enrollment of slightly more than 90 percent, while net enrollment (the proportion of children between 6 and 12 years of age enrolled in school) was 67 percent nationally. The net enrollment rate among the lowest expenditure quintile was only 46 percent, and 59 percent for the second lowest quintile, compared with 81 percent for the highest quintile. High dropout and repetition rates were also common (World Bank 1996a,b). Recovery in Service Delivery: Evidencefrom Schools and Health Centers 349 officially reported enrollment statistics grossly understate the progress made in the 1990s. Determining where in the delivery system the incentive to underreport was the highest or how it might have changed over time is difficult. At the school level, it would have meant fewer tuition fees remit- ted to the district, while at the district level underreporting would have required smaller transfers of capitation grants to schools. Availability of Data on Public Spending The total budgetary allocation for recurrent expenditure on education almost tripled in real terms during 1991-95 (table 11.3). Neither functional nor spatial disaggregation of education spending is easy, however. First, at the central government level, data were not available on salaries paid to primary school teachers either by district or by school in 1991-95. The only data available were the aggregate salary payments, which lumped together payments to teachers in the primary, secondary, and tertiary levels, as well as those made to non- teaching staff. This made systematic comparison of budget allocations for teacher salaries and actual spending at the school level impossible. Also, some teachers were not on the central government payroll, which further compli- cated efforts to track salary spending. Additional teachers were hired directly by schools and funded by parent-teacher associations (PTAs). The only sys- tematic spending data available at the central government level were capita- tion grants for nonwage spending. Second, initially the intention was to track public spending through the entire delivery system, which included the central government, districts, and schools. The field survey revealed that the district-level records for both nonwage and wage spending were even worse than at the central gov- ernment level. The quality of available information both on transfers from the center and disbursements to schools was so poor-both before and Table 11.3. Recurrent Budget Allocation for Education, 1991-97 (1991 prices) Year U Sh (millions) Index 1991 19,202 100 1992 30,002 156 1993 24,569 128 1994 32,258 168 1995 51,891 270 1996 49,027 255 1997 68,081 355 Note: The exchange rate ranged from U Sh 960 to U Sh 1,200 per U.S. dollar during 1991-95. Data are from fiscal years. Source: Ministry of Finance and Economic Planning data. 350 Ritva Reinikka after decentralization-that districts were excluded from the expenditure tracking exercise. School records were relatively comprehensive, however. Presumably parents who contributed substantially to school income before 1997 demanded financial information and accountability from the school. Therefore a detailed comparison of budgetary allocations and actual spend- ing could only be made between the central government outlays for nonwage spending and the equivalent school income.5 Actual Spending at Primary Schools Table 11.4 presents a summary of the sources of income for the 250 sample schools (both cash and in-kind). During 1991-95, the central government's financial contribution to primary education consisted of three components: primary teacher salaries, capital expenditure, and capitation grants. Teacher salaries was the largest item, consistent with the finding that public spending choices tend to favor teacher salaries over their actual con- tribution in producing educational outputs (Pritchett and Filmer 1997). Capital expenditure was limited almost entirely to rehabilitation rather than Table 11.4. Summary of School Income Data, 1991-95 (1991 prices in millions of U Sh) Income 1991 1992 1993 1994 1995 Teachers' salary payments by government 213.9 214.7 381.3 748.6 914.6 Capitation grants received by schools 4.2 15.8 58.0 60.9 58.3 Other government funding 73.8 62.5 73.6 118.7 147.1 Total government contribution 291.9 293.0 512.9 928.2 1,120.0 Tuition collected 55.4 96.8 116.6 136.2 141.3 Amount of tuition retained by schools 2.2 7.4 10.6 23.7 50.3 PTA levies 591.1 609.6 775.2 934.9 1,032.7 PTA salary payments 125.8 134.1 196.0 300.7 475.9 Total parent contribution 772.3 840.5 1,087.8 1,371.8 1,649.9 Source: School survey. 5. Donor assistance for primary education has come in two main ways. First, financing has been made available for textbooks and other scholastic materials. Sec- ond, donors have provided substantial financing for school construction. With the exception of one major donor-funded project, tracking of donor and NGO expendi- tures was difficult in the absence of any disaggregated data at the center. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 351 new construction.6 The capitation grant for nonwage expenditure is a pay- ment per student enrolled and is a 50 percent matching government contri- bution against the mandated tuition fees paid by parents. The capitation grant is intended to defray part of the costs of textbooks and other learning materials, as well as general school running costs. The survey confirmed that the main sources of income for government- aided primary schools were, in order of importance, (a) PTA levies collected from parents by the school, (b) central government transfers and PTA contri- butions for teacher salaries, (c) government funding for capital expenditures and capitation grants, and (d) retained tuition fees. PTA funds are under the full control of the schools, and the PTA executive committee oversees their use. Because their level depends on the ability of parents to pay, these levies vary widely between schools and across regions. The government's total contribution to the funding of primary schools almost quadrupled during 1991-95 in real terms, albeit from a negligible base. This is proportionately more than the overall increase in education spend- ing. Despite an increase in government spending, spending by parents doubled during the same period. The average parental contribution per stu- dent increased by 35 percent in real terms between 1991 and 1995, while the average government contribution more than doubled (table 11.5). Table 11.6 shows total expenditure by parents and government at the median school during 1991-95. A comparison of the means and medians shows that the distribution of parent expenditure at the school level is highly asymmetric, with the median only a fraction of the mean. Hence the median is a better measure of the general tendency in parent expenditure. The distri- bution of government spending is much less asymmetric, although the me- dians are lower than the means. Parent expenditure per student doubled during 1991-95 at the median, while the increase in government spending was almost fivefold during the same period. Table 11.7 shows the proportion of school income from parents and gov- ernment during 1991-95. Although declining in importance during the sur- vey period, parental contributions were clearly the mainstay of finance in 6. Since the 1970s the central government had virtually abandoned its responsi- bility for classroom construction. In principle, the provision of classrooms became the responsibility of local governments. As the local government tax base needed to support school construction is underdeveloped, local governments in turn passed the responsibility for classroom construction on to parents. To shoulder this and other school-related financial obligations, PTAs increasingly resorted to PTA levies. In ad- dition, the central government is responsible for counterpart funding, which is the government's share of the cost of donor-financed development projects. The central government also incurs expenditure on teacher training, examinations, and school inspections, which have a separate allocation. Table 11.5. Mean Parental and Government Contribution to School Income Per Student, 1991-95 (1991 prices in U Sh) Parents Government Year Tuitionfees collected PTA levies PTA salaries Total Capitation grant Salaries Other Total 1991 682 7,269 1,547 9,498 68 2,630 908 3,606 1992 1,072 6,749 1,484 9,305 118 2,377 692 3,187 1993 1,069 7,108 1,797 9,974 280 3,496 675 4,451 1994 1,136 7,796 2,507 11,439 352 6,243 990 7,585 1995 1,094 8,000 3,687 12,781 330 7,085 1,139 8,554 Source: School survey. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 353 Table 11.6. Median Parental and Government Contribution to School Income Per Student, 1991-95 (1991 prices in U Sh) Year Parents Government 1991 1,173 1,639 1992 1,631 2,215 1993 1,792 4,179 1994 2,209 4,467 1995 2,291 7,729 Source: School survey. Table 11.7. Parental and Government Contribution to Total School Income, 1991-95 (percent) Parents Government Year Mean Median Mean Median 1991 73 42 27 58 1992 74 42 26 58 1993 68 30 32 70 1994 60 33 40 67 1995 60 23 40 77 Source: School survey government-aided primary schools. In 1991-92 parental contributions ac- counted for more than 70 percent of school income on average; by 1995 the share had declined to 60 percent. However, for the median school, parental financing was less important, declining from 42 percent in 1991 to 23 percent in 1995. This indicates a highly skewed distribution of spending. Without an adequate breakdown of the salary data at the central gov- ernment level, one of the key questions this study sought to answer was how much of the nonwage expenditure (capitation grants) made available by the central government actually reach the schools. The government's stated policy was to disburse the grant in full to the schools either in cash or in-kind through the district education officers. The capitation grant was set in 1991 at the nominal rate of U Sh 2,500 per child enrolled in grades P1- P4 and U Sh 4,000 per child enrolled in grades P5-P7. These rates remained the same until 1997, although they grossly underestimated the cost of pro- viding scholastic materials and maintaining the physical facilities. Infla- tion, although moderate since 1993, eroded the real value of the grant. Thus, the real increase in total recurrent expenditure over time (table 11.3) was 354 Ritva Reinikka not reflected in nonwage spending on primary education. To compensate for the inadequacy of the central government provision for nonwage (and wage) expenses, school administrators resorted to PTA levies. Table 11.8 indicates the amount of capitation grant disbursed by the cen- tral government and the average amount received by the schools (in 1991 prices).7 While the central government's contribution in real terms was at its highest in 1991, the schools received on average only 2 percent of this grant. However, even if 1991 and 1992 are viewed as extreme cases, the figures for 1994-95, although higher, are still extremely low. In the best year the schools received, on average, one-fifth of the capitation grant (zero at the median). Recent evidence from a similar school survey shows that the situation has improved greatly since 1995. With increased transparency, 90 percent of the capitation grant is now released to the schools (Republic of Uganda 2000). Interviews during the 1996 school survey confirmed that local govern- ment authorities retained the bulk of the grant. Some districts apparently disbursed the grant on the basis of how many students had paid tuition. The funds intended for children who had enrolled but not paid tuition fees were typically retained by the urban or district councils. This practice cer- tainly hurt poorer communities the most, because in these communities parents are more likely to default on the payment of tuition fees. Some lo- cal governments reported that the discrepancy was used to cover the ex- penses of the district education officer. In some districts the funds retained by the local authorities were spent for purposes unrelated to education. In addition, part of the intended grant apparently remained at the center, as the government budgeted and disbursed the grant on the basis of the 1991 enrollment figures. As enrollment increased over time, the grant per stu- dent actually disbursed to the districts certainly decreased. During the survey period, parent contributions toward financing primary education consisted of (a) tuition fees at the nominal rate of U Sh 2,500 per child in grades P1-P4 and U Sh 4,000 per child in P5-P7 to match the capita- tion grant paid by the government, (b) PTA levies that varied from district to district and from school to school, and (c) contributions to teacher salaries. Tuition fees collected by the schools were not remitted to the central govern- ment. Rather, each district determined how the funds raised should be redis- tributed among the schools. In some districts, the schools were allowed to retain a certain percentage or a fixed amount of the tuition fee collected per student, with the balance transferred to the district education officer. In other districts the tuition fees collected were all remitted to the district headquar- ters. Subsequent disbursements to schools, either in cash or in-kind, may or may not have taken place. Collection efficiency of tuition fees was very low in 1991, but has improved since 1992 (table 11.9). 7. The average capitation grant was based on the assumption that 70 percent of students were in grades P1-P4 and 30 percent were in grades P5-P7. Table 11.8. Average Capitation Grant Per Student, 1991-95 (1991 prices) Schools actually received Mean Maximum Intended grant Percentage of Percentage of >, Year amount (U Sh) U Sh intended amount Minimum Median U Sh intended amount 1991 3,100 68 2 0 0 2,509 26 1992 1,966 118 9 0 0 1,916 47 1993 1,869 280 28 0 0 1,867 67 1994 1,850 352 27 0 0 1,826 69 1995 1,737 330 26 0 0 1,734 56 Note: 997 observations; 71 observations omitted from the sample as outliers. Source: School survey. Table 11.9. Average Tuition Per Student, 1991-95 (1991 prices) Tuitionfees retained by schools Mean Maximum Tuitionfees Percentage of Percentage of Year collected (mean) U Sh fees collected Minimum Median U Sh fees collected 1991 682 27 4 0 0 256 38 1992 1,072 82 8 0 0 395 37 1993 1,069 97 9 0 0 398 37 1994 1,136 197 17 0 0 605 53 1995 1,094 390 36 0 0 546 50 Source: School survey. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 357 In 1991 schools received, on average, 4 percent of the tuition collected. By 1995 this had improved considerably, but schools still only retained 36 per- cent of the average tuition fees. Hence, as shown in table 11.9, local govern- ment authorities not only retained the bulk of the capitation grant, but also kept a large portion of the tuition fees paid by parents. Variation between districts was also substantial. Despite anecdotal evidence that teacher salary payments suffered from de- lays and other problems in the flow of funds, interviews during the survey indi- cated that government salary payments mostly reached the schools.8 Because of the lack of annual disaggregated data at the center, salaries could not be tracked through the system, but the school survey provides other useful information. Teachers derived salaries from three sources: the government, PTAs, and others such as NGOs (table 11.10). In 1991 and 1992 nearly half of teacher salaries came from sources other than government. From 1993 on, the gov- ernment contribution rose significantly, thanks to a presidential directive that called for annual salary increases for teachers. Increased budgetary alloca- tions were reflected in higher salary payments at the school level, but this alone is not adequate to determine the extent to which budgetary allocations translated to actual spending. Parental contributions fluctuated from a quar- ter to a third of the total wage bill during the survey period. Note that the share of total PTA contributions used for teacher salaries increased from 16 percent in 1991 to 29 percent in 1995, despite the quadrupling of government spending on salaries. Total spending on instructional materials and other nonwage items by schools increased only by 20 percent in real terms between 1991 and 1995, while the equivalent spending on salaries (government and parents combined) tripled during the same period and more than tripled per teacher. Not only did public spending choices favor teacher salaries over nonwage spending, but teachers may have exerted a disproportionate influence over PTAs as well.9 However, the starting point was extremely low (U Sh 11,360, or around US$12 per month, in 1991). This was less than a quarter of what the civil service reform program considered a minimum living wage at the time. Sur- vey interviews confirmed that absenteeism was a serious problem, as teach- ers were compelled to make a living outside their profession. Although the targeted living wage had not yet been attained by 1995, the situation had improved considerably from the teachers' point of view. While teacher salaries were given priority over instructional materials and other nonwage items, a major pay increase was perhaps warranted to 8. The only systematic way of misappropriating funds was by having "ghosts" on the payroll. A total of 15,000 ghost teachers (around 20 percent of all teachers) was removed from the payroll in 1993. 9. To some extent, donor funds compensated for slow growth in nonwage spend- ing, but only in some schools. Table 11.10. Contributions to Teachers' Salaries, 1991-95 (1991 prices) Government PTA Other Total Year U Sh (million) Percent U Sh (million) Percent U Sh (million) Percent U Sh (million) Percent 1991 213.9 51 125.8 30 79.7 19 419.4 100 1992 214.7 52 134.1 33 61.5 15 410.3 100 1993 381.3 59 196.0 30 72.4 11 649.7 100 1994 748.6 66 300.7 26 86.7 8 1,136.0 100 1995 914.6 61 475.9 32 104.7 7 1,495.3 100 Source: School survey. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 359 reduce absenteeism and restore the quality of teaching. Some evidence sug- gests that this strategy worked, given the finding that enrollment increased by 60 percent. At the same time, a more balanced spending pattern between salaries and instructional and other materials might have produced an even better result. Regional Differences As national averages conceal regional variations, it is useful to explore actual spending in the subregions in the survey Table 11.11 shows government ex- penditures per student that reached the schools by subregion (in 1991 prices). The western region appears to have the lowest per student public spending at the school level, possibly indicating worse inefficiency in the transfer system between the center and the schools than in other subregions. As schools are not larger in the west than elsewhere, a lower unit cost is not likely to result from a higher student-teacher ratio and a resultant lower wage bill.'0 The opposite is probably true in the north and northeast, where classes are smaller and the per student expenditure is therefore higher. To explore regional differences in efficiency further, the capitation grant is a good proxy, as this was intended to be the same amount per student across the country. When the share of the capitation grant spent on the intended purpose is re- gressed on a regional dummy variable (using ordinary least squares), only the north (Apac and Gulu districts) entered negatively and highly signifi- cantly (at 1 percent). The north is one of the poorest regions in Uganda, as measured by household expenditure, and continues to suffer from conflict. Parent expenditure per student has a much larger spatial spread than public spending (table 11.12). The level of private spending is the highest in the better-off central region and Kampala, while the three poor northem sub- regions and the west have extremely low spending levels per student." Impact of Decentralization Before fiscal decentralization, which began gradually in mid-1993, the bulk of public funds came from the central government. The Ministry of Education 10. This appears to be the case in Kampala, where the share of public funding is the smallest and classes are large. 11. The district-level (Spearman rank) correlation coefficient between public spending on primary schools and poverty measured by household expenditure is -0.228. Poorer districts seem to benefit from a somewhat higher level of public spend- ing per student available to the schools. However, this may also reflect a lower student- teacher ratio, as households in those districts can afford to send fewer children to school. There is a positive correlation (0.56) between household expenditure and pri- vate spending on primary education. Table 11.11. Average Government Contribution Per Student Reaching Schools by Subregion, 1991-95 (1991 prices in U Sh) Year Northwest North Northeast East Central Kampala Southwest West c 1991 1,623 4,866 2,599 3,546 5,878 1,067 5,718 1,958 ° 1992 1,772 3,972 2,781 3,315 4,220 2,348 4,392 2,488 1993 3,964 4,664 5,138 4,516 6,122 3,535 6,285 3,307 1994 7,384 7,526 8,405 8,048 10,120 6,438 7,962 6,235 1995 12,811 8,151 7,748 8,179 10,318 8,636 7,300 5,977 Source: Sdcool survey. Table 11.12. Average Parental Contribution Per Student by Subregion, 1991-95 (1991 prices in U Sh) Year Northwest North Northeast East Central Kampala Southwest West c 1991 1,345 1,048 839 6,932 27,545 49,084 3,064 1,480 1992 976 991 1,195 4,709 20,134 65,829 3,436 1,559 1993 1,107 1,763 1,175 5,500 22,176 46,170 4,440 1,988 1994 1,880 2,074 1,070 7,196 27,576 41,792 6,053 2,189 1995 2,034 2,277 999 8,522 31,568 37,286 6,520 1,795 Source: School survey. 362 Ritva Reinikka played a major role in primary education, controlling nearly all the recurrent budget allocations for the sector. The district administrations, however, chan- neled these funds to schools even before decentralization. Following decen- tralization, district authorities and the district and urban councils gradually gained control of the funds provided by the central government for primary education. In 1996, estimates indicate that the ministry controlled only about a quarter of the total recurrent spending on primary education. The standard capitation grant is a good proxy for exploring the impact of decentralization on the flow of public funds to schools, as it was supposed to be the same (nominal) amount per student throughout the study period in all districts. Using ordinary least squares, the share of the capitation grant reaching the schools is regressed on time dummies and a decentralization dummy variable. The latter takes the value one when the district where the school is located was decentralized; otherwise it is zero.12 As table 11.13 shows, the input flow at the school level improved at a statistically significant level over time, albeit modestly. The decentraliza- tion variable (DECEN) enters significantly negative, indicating that decen- tralization adversely affected the flow of funds to schools. The schools af- fected by decentralization received, on average, 9 percentage points less of the intended capitation grant per student than their counterparts in Table 11.13. Impact of Decentralization on the Flow of Capitation Grants to Schools, 1991-95 Year Coefficient t-statistic 1991 0.022 1.204 1992 0.060 3.332 1993 0.149 8.767 1994 0.221 11.617 1995 0.224 12.079 DECEN -0.093 -3.862 R2 0.085 n.a. Number of observations 997 n.a. n.a. Not applicable. Note: Ordinary least squares estimation. Dependent variable is the share of the capitation grant that reached the school, 1991-95 are time dummies, and DECEN is a binary variable taking the value one if the school is located in a fiscally decentralized district, zero otherwise. Source: School survey. 12. For example, schools located in the districts that were decentralized first in mid-1993 take the value one from the beginning of the following school (calendar) year. The second phase of fiscal decentralization occurred in mid-1994 and the last phase was in mid-1995. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 363 nondecentralized districts. Instead of receiving 22 percent, they received 13 percent in 1995. The deterioration in decentralized districts may be tempo- rary, but it serves as a reminder that decentralization could come with an adjustment cost in terms of service delivery. Health Care and Public Spending As in primary education, limited official data exist from central government health services at the time of the survey. Contrary to the education sector, how- ever, the health unit survey found little systematic facility-level information on financial flows or outputs, such as the number of inpatients or outpatients. One explanation for such a marked difference in facility-level behavior be- tween the two sectors could be that the PTAs that financed most of the school- level expenditure in 1991-95 demanded basic recordkeeping and accountabil- ity, while users in health clinics exerted no such pressure. A long-term relationship between providers and beneficiaries that characterizes primary education-in contrast to health care, where the relationship is typically short and more ad hoc-clearly favors better organization on the demand side. At the design and pilot stage of the survey, the researchers did not fully anticipate the lack of almost any financial information at the facility level and the heavy reliance on in-kind measures throughout the system. As the data gathering proceeded, any hope of systematic tracking of expenditure on the basis of data from primary health facilities faded. Many of the re- sources received by health units were in-kind with no value indicated, and hence not easy to compute. Although user fees are collected and retained at the health facility level, records on their use were either not available or patchy. Unlike in primary education where school income and expendi- tures could be related to pupils enrolled, records on patients were extremely poor and unreliable.'3 Availability of Data on Health Spending For the survey period, reliable health spending figures are available only for 1992/93 because of the difficulty of obtaining information about annual do- nor flows. Public spending was only US$4.38 per capita (including donor assistance) in 1992/93, while private spending was US$5.36. Although the level of health spending is low in Uganda, this study attempts to examine the flow of those minimal public funds from the center to service facilities. 13. Based on a cross-section of 61 developing countries, Uganda's health outcomes are worse than expected given its level of overall gross national product (GNP). An infant mortality rate of only 71 (compared with 97) would be predicted for Uganda given its per capita GNP in 1994 (Demery and Dorabawila 1997). 364 Ritva Reinikka Poor efficacy magnifies the negative impact of a low level of spending on health outcomes. Recurrent budgetary allocations for health increased 2.5 times between 1991 and 1995 (table 11.14). However, donors finance the bulk of public expenditure on primary health care. According to the data for fiscal year 1992/93, donors financed 77 percent of health spending, while the government's share was only 23 percent (World Bank 1996b). For hospitals, the distribution was reversed, with donors funding 36 percent and the gov- ernment funding 64 percent. With decentralization, nonwage recurrent ex- penditure on primary health care became part of the block grant, but drugs and other supplies funded largely by donors continue to be delivered from the center. The central government's main responsibilities were the salaries of health workers. Most primary health workers were central government staff seconded to local authorities; direct hiring by the local authorities was limited. Despite decentralization, this was still the case during the survey period. As in education, data on staff salaries were not disaggregated ei- ther by district or health facility for 1991-95. At the district level, locally recruited health workers are paid out of the district's own resources, but this information was patchy. Similarly, donor funding could not be disag- gregated either by district or by facility. As public resources dwindled from the mid-1970s, government health facilities at all levels increasingly resorted to various informal charges for drugs, meals, consultation, treatment, and operations. Attempts were made in 1989 to formulate a national policy on user fees for public health services, but were soon abandoned for another decade. The new national policy on user fees Was adopted in 1999. Before then districts theoretically could set user fees for their health services, although in practice, the imposition of charges was left to each facility. The Ministry of Health issued fee-for-service guidelines that allowed up to 50 percent of fees collected to be spent on staff incentives; up to 25 percent on drugs and supplies; and the rest on mainte- nance, supervision, and outreach. Table 11.14. Recurrent Budgetary Allocations for Health, 1991-97 (1991 prices) Year U Sh (millions) Index 1991 6,381 100 1992 9,109 143 1993 8,863 139 1994 14,429 226 1995 16,819 264 1996 16,470 258 1997 19,925 312 Note: These are fiscal years. Source: Ministry of Finance and Economnic Planning data. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 365 Qualitative Survey Results Although the enumerators found little or no reliable quantitative output and financial information at the health facilities they visited, they provided the fol- lowing qualitative observations of the situation in health care facilities in 1995. * Drugs were the main nonwage recurrent input into the primary health care delivery system. They are supplied quarterly, directly to the health units from the center. The central delivery system ensures that the drugs reach health units with little or no leakage. * Clinic compliance to user fee guidelines provided by the Ministry of Health was minimal. * Salaries for seconded staff generally reached the intended facilities, although remuneration of health workers was low, resulting in un- ethical conduct that adversely affected delivery of and access to pri- mary health care. Local recruited staff were paid less, and less regu- larly, which caused additional problems. * Health workers devoted very little time to the activities of health units. * Health workers had a high rate of attrition. * Rural health units did not attract qualified health workers. While the survey found that in-kind inputs into the health care system pro- vided by donors and the government mostly reach the intended facilities, an- other study carried out around the same time sheds more light on other prob- lems regarding efficacy of services at the facility level (McPake and others 1999). Researchers studied 12 health units in the Bushenyi and Iganga districts in depth, using focus groups, exit polls, and direct observation, to determine the socioeconomic survival strategies of health workers and their implications for formal health financing policy. One of the findings was that health workers in all but two facilities routinely charged users above the formally agreed levels, and the drugs supplied by donors or the government were routinely used as a source of additional income. The leakage estimate ranges from 40 to 94 percent of the public supply of drugs to the facilities. McPake and others (1999, pp. 61- 62) summarize the findings of their study as follows: The situation described by the preceding results suggest that almost all elements of the system which were once public have been incorporated into the private business activity of the health workers. More than half the drugs supplied to public health units had become the private property of health workers. The estimated drug leakage rate of the median facility was 78 percent! The re- sult is that very few free services were delivered in the public health facilities, and almost none at all were delivered to the poor. Little information exists about improvements in the health service in the latter half of the 1990s. Although limited, a recent rapid assessment of data availability for a new public expenditure tracking survey in four districts (Bushenyi, Iganga, Ntungamo, and Tororo) indicated that both operational 366 Ritva Reinikka efficiency of the health facilities and information on inputs and outputs had improved compared with 1994-96 (World Bank 1999). Conclusions and Policy Changes This study was inspired by the observation that officially reported primary enrollments did not improve in the first half of the 1990s, despite substantial increases in budgetary allocations for the education sector. The hypothesis was that without institutional recovery and improvement in accountability, public funds were subject to capture before they reached the schools. A diag- nostic survey of 250 schools and 100 health units was carried out in 1996 to measure the actual outputs and public spending at the facility level to proxy public sector efficacy by its ability to translate budgetary allocations into ac- tual spending at the level of service facilities. From the perspective of institutional recovery and accountability, the sur- vey provided three major findings. First, the behavior of public service provid- ers varies considerably between sectors, depending on the institutional con- text and incentives faced. Primary schools kept relatively good records on enrollments and financial flows, while health clinics had an almost complete void of information both on outputs, such as inpatients and outpatients, or financial information, such as user fees and cash and in-kind transfers of pub- lic resources. This survey and other evidence from Uganda indicates that edu- cational institutions improved faster in funding accountability than health in- stitutions. Such a marked difference could occur because in primary education parents financed most of the public school system. PTAs contributed as much as 73 percent of the total school expenditure in 1991 and 60 percent in 1995, and are likely to insist on accountability and exert pressure on the schools to provide services in return for their contributions. (Parental contributions were 42 percent at the median school in 1991, which indicates a highly skewed spend- ing distribution.) However, parents seemed to have little control over public spending, which was dominated by central and local government bureaucrats. Users of public health clinics were likewise unable to exert much pressure on these services. As shown in chapter 13, Ugandans, including the poor, most often opted for private services. Second, instead of being almost stagnant as the official data indicated, primary enrollment increased by 60 percent between 1991 and 1995. While the survey results cast serious doubt over the reliability of the officially reported data, they also point to a considerable improvement in the per- formance of the system at the school level. Increasing enrollment rates seem plausible, given the improvement in Uganda's political and socio- economic conditions. Third, while the survey results indicate some improvement in input flow- such as in teacher salaries, the main public spending item in education- they also confirm that serious accountability problems remained in 1995. Only 2 percent of public nonwage education spending had reached the schools in 1991, and four years later this share had increased only to about 20 percent. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 367 Although this is a significant improvement, the share remained abysmally low. District authorities captured most of the nonwage public funds intended for schools. Regression analysis also shows that decentralization negatively affected input flow in the delivery system, at least temporarily. In health care, drugs and medical supplies were transferred in-kind with- out records of their value, making it impossible to generate systematic quan- titative information about public funding reaching the facilities. Qualita- tive observation during the field survey generally confirmed that drugs and other supplies reached the health units directly from the center. How- ever, a study by McPake and others (1999) suggests that, unlike in educa- tion, the leakage occurred at the health unit level, where the staff siphoned off 78 percent of the drugs and supplies to compensate for their low pay. Although not fully comparable, recent evidence suggests some improve- ment in health facilities since then. The central government initiated the following immediate measures in 1996 in response to the survey findings to improve information flow and transparency: * Monthly transfers of public funds for wage and nonwage expendi- ture to districts are now regularly published in the main newspapers and broadcast by radio. e All district headquarters and government primary schools are required to maintain public notice boards and post monthly transfers of funds. * Measures to enhance accountability and dissemination of accounting information were incorporated in the 1997 Local Government Act. * Districts are required to pay all conditional grants for primary educa- tion directly on individual school accounts. School-based procurement also replaced the highly inefficient central supply of construction and other materials. * A renewed effort is under way to put in place basic budgeting, ac- counting, and auditing systems for the public sector, including local governments. The school survey was replicated in 1999, showing that the flow of funds to schools has improved dramatically since 1995. This resulted from the cen- tral government's initiative to disseminate information monthly on transfers through newspapers and radio and to insist that all schools post information on the funds released to them (Republic of Uganda 2000). Schools now re- ceive more than 90 percent of the intended capitation, on average, although apparently with considerable delays because of inefficiencies in the districts and the banking system. The median receipts of the capitation grant are also around 90 percent.'4 14. A preliminary analysis of the data shows that considerable variation remains in what individual schools receive per student. In particular, variation can be ex- plained by schools in Kampala having an advantage over rural areas. 368 Ritva Reinikka Nevertheless, this represents a welcome improvement, particularly as the universal primary education initiative of 1997, covered in chapter 12, has substantially increased resource flows to the districts from their previous lev- els, including the capitation grant. Overall, this study demonstrates that improvements in institutions and accountability are much more difficult to achieve than macroeconomic re- form. Although improved since the beginning of the 1990s, service delivery continued to suffer from major inefficiencies in the mid-1990s. Compared with the evidence presented in chapter 3 on the adverse effects of cash man- agement of the budget on expenditure programs, it appears that the volatil- ity the "cash flow" system creates for some spending items may be relatively insignificant compared with gross inefficiencies caused by lack of account- ability. At the same time, Uganda's experience shows the power of informa- tion and transparency-publishing and posting of resource flows-in im- proving accountability and service delivery. References The word "processed" describes informally reproduced works that may not be commonly available through library systems. Ablo, Emmanuel, and Ritva Reinikka. 1998. "Do Budgets Really Matter? Evi- dence from Public Spending on Education and Health in Uganda." Policy Research Working Paper no. 1926. World Bank, Development Research Group, Washington, D.C. Bardhan, Pranab, and Dilip Mookherjee. 1998. "Expenditure Decentraliza- tion and the Delivery of Public Services in Developing Countries." Working Paper (November). University of California, Department of Economics, Berkeley. Demery, Lionel, and Vajeera S. Dorabawila. 1997. "Health Outcomes, Pov- erty and Health Spending: Uganda in International Perspective." World Bank, Poverty Reduction and Economic Management Network, Wash- ington, D.C. Processed. Dixit, Avinash K. 1996. The Making of Economic Policy: A Transaction-Cost Poli- tics Perspective. Munich Lectures in Economics. Cambridge, Massachu- setts: MIT Press. Filmer, Deon, and Lant Pritchett. 1999. "The Impact of Public Spending on Health: Does Money Matter?" Social Science and Medicine 49(10): 1309- 23. McPake, Barbara, Delius Asiimwe, Francis Mwesigye, Matthius Ofumbi, Pe- ter Streefland, and Asaph Turinde. 1999. "The Economic Behaviour of Health Workers in Uganda: Implications for Quality And Accessibility of Public Health Services." Social Science and Medicine 49(7): 849-65. Recovery in Service Delivery: Evidencefrom Schools and Health Centers 369 Pritchett, Lant. 1996. "Mind Your P's and Q's. The Cost of Public Investment is Not the Value of Public Capital." Policy Research Working Paper no. 1660. World Bank, Development Research Group, Washington, D.C. Pritchett, Lant, and Deon Filmer. 1997. "What Educational Production Func- tions Really Show. A Positive Theory of Education Spending." Policy Research Working Paper 1795. World Bank, Development Research Group, Washington, D.C. Republic of Uganda. 1990. The Financial Tracking System for Primary Health Care (PHC) and Primary Education (PE). Ministry of Planning and Eco- nomic Development, Kampala. . 1992. The Implementation Programme for the Financial Tracking System (FTS) and the Reform of Local Authorities Budget Process. Ministry of Fi- nance and Economic Planning, Kampala. . 2000. "Tracking the Flow of and Accountability for UPE Funds." Re- port by International Development Consultants, Ltd. Ministry of Edu- cation and Sports, Kampala. Svensson, Jakob. 1997. "The Control of Public Policy: Electoral Competition, Polarization, and Endogenous Platforms." World Bank, Development Research Group, Washington, D.C. Processed. World Bank. 1996a. Uganda: The Challenge of Growth and Poverty Reduction. A World Bank Country Study, Washington, D.C. . 1996b. "Access to Education and Health Care in Uganda." Eastern Africa Department and Poverty and Social Policy Department, Coun- try Operations Division, Washington, D.C. Processed. . 1999. "Rapid Assessment of Data Availability in Health Core Units." With the Makerere Institute of Social Research. Washington, D.C. Pro- cessed. 12 What Can We Expect from Universal Primary Education? Simon Appleton President Museveni's 1997 election pledge to provide free primary educa- tion catapulted education issues up the policy agenda. Until that point edu- cation had arguably been a low and declining priority for the government. In 1987, when the Museveni government had first come to power, it set up the Education Policy Review Commission to report on the state of education. The commission's most notable recommendation was to attain universal pri- mary education by 2000. However, the government was slow to implement the measures and commit the resources needed to meet this goal. The response to the universal primary education (UPE) initiative of 1997 was strong-leading to a near doubling of officially recorded primary school enrollments-and the government's modest funding increase was insuffi- cient to meet demand. To deal with this dramatic expansion, the Education Sector Investment Program of 1998-2003 envisioned a 50 percent increase in expenditures on primary schools and a doubling of resources to second- ary schools. The apparent about face in 1997 has many possible explanations, includ- ing short-term political considerations. The change in policy stance could, however, also reflect a more fundamental shift in thinking about education both as an intrinsically desirable goal (part of human development) and as an investment in economic success (part of human capital). Uganda's The author is grateful to the Bureau of Statistics for the use of the data and to Marcel Fafchamps, Dominique Guillaume, and Francis Teal for comments. The text has benefited from useful comments by the editors and anonymous reviewers. A sec- tion of this chapter draws heavily on the unpublished work of Kim Otteby (1999). 371 372 Simon Appleton economic recovery has given the government the confidence that resources can be found to meet the large long-term commitment implied by the UPE initiative. Furthermore, rapid economic growth has transferred attention away from purely economic measures of development and toward other indica- tors (see, for example, the critique of Uganda's success by the United Na- tions Development Programme [UNDP 1997]). School enrollment is the so- cial indicator perhaps most directly affected by government policy. Education is also instrumental in the attainment of other social development targets, such as health. Apart from the human development aspect, government opin- ion may have shifted toward education as a productive investment. Before 1997 many officials questioned whether education-primary education, in particular-was productive, especially when compared with investing in physical infrastructure, such as roads. This chapter focuses on three aspects of the UPE initiative: (a) the impli- cations for the equity of educational access, (b) the likely affect on household income generation, and (c) the implications for school quality and academic performance. Consider first the equity of educational access. A central part of the argu- ment for UPE is that user charges curtail the enrollment of girls and the off- spring of less-educated parents. Perhaps the most remarkable feature of the UPE initiative to date is the particularly large increase in overall enrollments following the removal of fees. Similar enrollment surges occurred as a result of UPE initiatives in Kenya and Tanzania in the 1970s, and more recently in Malawi in 1994, when enrollments rose 50 percent following the abolition of primary school fees (Reddy and Vandermoortele 1996). These increases stand in stark contrast to the assumption by advocates of increased user charges for social services that charges would not greatly reduce, and could increase, enrollment. Support for this assumption comes from conventional cross- sectional estimates-which can often be misleading-that show small price elasticities for education demand (Jimenez 1987). Using data from Uganda before the removal of fees, small-and indeed perverse estimates of the price elasticities for enrollments are obtained when, in fact, the actual re- sponse of enrollments to the abolition of fees was strong. Increases in educational enrollments, particularly among disadvantaged children, are desirable for many reasons. For a poor country like Uganda, the hope is that educational expansion will increase the productivity of workers and, hence, foster economic growth. The extent to which this hope is likely to be realized is taken up in the section on returns to education. The value of education-primary education, in particular-is largely based on estimates of returns to wage employment. This chapter provides a more comprehensive estimate of the effect of education on household earnings by looking at the effects of education in all income-generating activities: wage employment, farm- ing, and nonfarm self-employment. It also traces the effect of education in reallocating family labor among wage employment, farming, and nonfarm self- employment. This integrated approach provides a more reliable and rather What Can We Expectfrom Universal Primary Education? 373 different picture of the returns to education than the existing studies of either urban wage employment or agricultural production functions. When assessing the possible income benefits of UPE, we take current as- sociations between education and earnings as our guide. This may be peril- ous for many reasons; one, in particular, is that UPE may reduce the quality of education and, hence, weaken the economic benefits of schooling. The third section of this chapter explores this issue using information gathered from preliminary research on the impact of UPE in two districts of Uganda. The research corroborates the hypothesized benefits of UPE in terms of en- hancing the equity of educational access, but it also documents the deterio- ration of conventional indicators of school quality, such as student-teacher ratios. We obtain tentative estimates for the effects of school quality on stu- dent academic performance by simulating academic testing before and after UPE. Unlike much of the literature on educational production functions, these results show a negative association between student-teacher ratios and aca- demic performance. Consequently, by overstretching educational resources, UPE risks a decline in student academic performance. Access to Education Prior to the UPE Initiative This section presents a multivariate model of school enrollment before the UPE initiative that identifies the characteristics of children not attending school. Because the UPE initiative has led to the enrollment of almost all primary school-age children, the analysis will reveal the characteristics of children who will most likely access primary education as a result of the policy change. The research uses data from the 1992 Integrated Household Survey because it was the largest household data set available at the time of writing and is the richest in terms of potential explanatory variables. Despite evidence that school enrollment increased between the 1992 sur- vey and the UPE initiative, the survey represents the situation before UPE quite well (see chapter 11 in this volume). The sample of school-age children (ages 5 to 14) includes 4,122 who have never been to school, 8,857 who are currently attending school, and 967 who have left school, for a total of 13,946. After weighting the sample by population multipliers in order to yield na- tionally representative results, 61 percent are in school, 32 percent never en- rolled, and 17 percent left school. A multinomial logit models the probability that a child has a particular educational status (currently attending, never enrolled, or left school).' The 1. The logit model is the standard statistical technique for analyzing an unor- dered discrete variable, although it suffers from the limitation of imposing the "irrel- evance of independent alternatives." In other words, the relative probabilities of two outcomes are unaffected by the presence of a third. This assumption is relaxed in the multinomial probit, which is computationally more difficult to estimate. 374 Simon Appleton model can be interpreted as the outcome of a process where the utility from each choice is a linear function of the explanatory variables and a stochastic term. The explanatory variables chosen for the model are personal character- istics (age and sex), parental education; demographic characteristics of the household (size and number of boys and girls), characteristics of the house- hold head (age and sex), dummy variables for piped water and for use of firewood, household income per capita, school fees, distance to schools and district centers, and regional dummy variables. Table A12.1 presents the results, but the coefficients are hard to interpret. Consequently, table 12.1 presents predicted probabilities from the model, evaluating at the means of the explanatory variables. The baseline figures show that, at the mean of all explanatory variables, a child aged 5-14 has a 69 percent probability of being in school, a 25 percent probability of never hav- ing been in school, and a 6 percent probability of having left school.2 Natu- rally, these probabilities are highly age dependent. At five, a child has only a 15 percent chance of being in school. By 10 this chance increases to 82 percent and then decreases with age.3 By 14 the probability of never having attended school falls to 8 percent. Substantial gender inequalities exist. At the mean of the other variables, girls are less likely to be in school, with a predicted prob- ability of 65 percent compared with 73 percent for boys.4 Parental education has powerful effects on school enrollment. If both parents have no formal education, the probability of their children never attending school evaluated at the mean of the other explanatory variables is 49 percent (figures not reported in tables). If both parents have postprimary education, the corresponding probability is 5 percent. Each increment in parental education reduces the probability of a child never having been to school. Paternal literacy has a large impact on the probabil- ity of enrollment. At the means of other variables, children with illiterate fathers have a 39 percent chance of never having been in school, compared with 26 percent if their fathers are literate. The literacy of mothers appears not to have a marked effect, although some primary schooling does. With the exception of literacy, maternal education generally has stronger effects on school enrollment than paternal education. Household income per capita raises the probability of attending school. To avoid endogeneity problems, we use only the income of those over 20 years of age. If household incomes are twice the average, the probability of being in school is 79 percent, ten percentage points higher than the baseline. The effect of shortfalls below the mean appears less pronounced: The probability of 2. These probabilities at the mean of the explanatory variables are different from the mean proportions because of the nonlinearity of the logistic model. 3. Figures not reported in table 12.1 but are available from the author on request. 4. For comparative international statistics on gender inequalities in 41 countries, see Filmer (2000). What Can We Expectfrom Universal Primary Education? 375 Table 12.1. Predicted Probabilities from the Logit Model for School Enrollment (percent) Never having Category attended school Left school In school Baseline (mean of all variables) 25 6 69 Boy 22 5 73 Girl 29 6 65 Father uneducated 39 6 56 Father literate 26 9 65 Father some primary 25 7 68 Father full primary 20 4 76 Father postprimary 15 4 81 Mother uneducated 34 6 60 Mother literate 32 8 61 Mother some primary 21 5 74 Mother full primary 14 4 82 Mother postprimary 9 6 85 Male head 26 6 68 Female head 22 6 72 Head aged 33 26 6 68 Head aged 53 25 5 70 Half mean income 29 6 65 Double mean income 16 5 79 No firewood 22 7 70 UJse firewood 25 5 69 No piped water 27 4 69 Piped water 25 6 69 Primary school in each village center 22 6 73 Central urban 17 8 75 Central rural 19 6 75 Western rural 23 5 71 Western urban 20 7 73 Eastern rural 26 5 69 Eastem urban 26 6 68 Northern rural 37 5 57 Northern urban 29 6 66 Source: Author's calculations from the 1992 Integrated Household Survey. being in school if household incomes are half the mean is 65 percent, four percentage points lower than the baseline. The magnitude of these pre- dicted income effects is perhaps smaller than might be expected; however, they are pure income effects. Actual differences in enrollments by income are larger because income differentials are associated with variances in other ex- planatory variables such as parental education and regional location. 376 Simon Appleton The model makes perverse predictions of the effects of school fees and parent teacher association (PTA) contributions levied by local primary schools.5 Such charges predict an increase in the probability of attending school: the coefficients on both the fee and the PTA contribution variables are positive (table A12.1). In table 12.1 we do not report how much abolish- ing fees is estimated to reduce enrollments, as such estimates are clearly implausible in the light of subsequent events. Class sizes also had positive effects in preliminary regressions. These results presumably reflect endogeneity problems. If demand for education in a locality is high, schools will be able to levy high charges and have large classes. Large classes may also reflect high demand for a particularly high-quality local school. High charges may be a cause of high demand, for example, if they permit an in- crease in school quality. However, the reverse causality appears more plau- sible: high demand enables schools to levy high charges. These perverse re- sults are cautionary, given the large response of enrollments when the government abolished charges in 1997. These results show the danger of re- lying on cross-sectional estimates for policy purposes. An analysis based on the 1992 data would not identify charges as a major constraint to attendance.6 Endogeneity problems may also arise with the variables for distance to schools. One might expect schools to be located closer to where demand is high. Unlike endogeneity biases on charges and class sizes, the likely endogeneity bias on distance to school will reinforce the true structural ef- fects. Distance to schools could be associated with low demand either be- cause distance reduces demand or because high demand encourages the es- tablishment of nearby schools.7 Consequently, one must be cautious in interpreting the negative effects of distances to primary and secondary schools on the probability of attending school. Only distances to primary schools 5. We could not reject the restriction that fees and PTA charges have equal effects by using a likelihood ratio test and, hence, we imposed these restrictions on the model. 6. A referee queried whether the analysis was set up as a "straw man" and com- mented that a more nuanced conclusion would be that cross-sectional analysis that does not control for problems of endogenous fees (or school placement) can easily lead to incorrect inference. However, many such cross-sectional analyses exist in the literature, for example, those cited by Jimenez (1987). Such studies were influential in the policy debate about user charges for social services in the 1980s and, indeed, are praised for their rigor even by critics of user charges for basic social services (Reddy and Vandermoortele 1996). The reason why such studies are the norm is that it is very hard to find convincing instruments for user charges, school placement, and school quality in cross-sectional data sets. Only a few studies with longitudinal data have attempted to address the endogeneity of public programs. 7. A reviewer suggested that schools might be placed where they are needed most, biasing the estimated effect of distance to school towards zero. However, it seems more likely that schools in Uganda have been established where there is local demand. What Can We Expectfrom Universal Primary Education? 377 have statistically significant effects. The Ugandan data on distance is limited in that it refers to distance of the school from the center of the village (LC1), not the distance of the child's home from the school. Distances from village centers to primary schools are not great for most communities-one kilome- ter or less for two-thirds of the children. Having a school in the center of each village is predicted to raise school attendance from 69 to 73 percent. A number of demographic characteristics of the household significantly affect outcomes. Children from households with female heads are more likely to be in school than are those from households with male heads, after con- trolling for other characteristics (72 percent compared with 68 percent). The age of the household head has a nonmonotonic effect on school attendance. Up to age 34, the probability of being in school decreases with the age of the household head; thereafter, it increases. The effects are modest: a household head aged 53 (10 years older than the mean) raises the probability of being in school by only 1.5 percentage points. Additional household members increase, not always significantly, the probability of school enrollment, controlling for per capita income, consistent with the presence of economies of scale in house- hold consumption. Extra children cause more of an increase than adults, con- sistent with children having less costly needs than adults, for example, need- ing fewer calories.8 Girls in the household increase the probability of school enrollment more than boys, because girls' lower enrollment rates, on aver- age, provide less competition for educational funding within the household than boys. The use of firewood and the absence of piped water are hypoth- esized to increase the demand for child labor. While the absence of piped water and the use of firewood do not materially alter the probability of chil- dren attending school, they do raise the probability of not attending school and reduce the probability of dropping out. This finding suggests that they delay school enrollment, but do not prevent it.9 The significance of the dummy variables for location affect educational outcomes in ways that cannot be explained in terms of parental education, income, or the included infrastructure variables. However, the model is fairly successful in explaining differences in enrollment rates in terms of observ- able variables. For example, the actual proportion of children in school is 83 8. In theory, it would be possible to adjust the household income variable to make some allowance for economies of scale. In practice, we have no agreed method of estimating such economies of scale, and different approaches can lead to dramati- cally varied estimates (see Lanjouw and Ravallion 1995). 9. We could argue that firewood use is endogenous. However, the likely bias would be to exaggerate the negative effect of firewood use on enrollment. Unobserv- able factors associated with households using an inferior fuel source such as fire- wood may also be associated with educational disadvantage. Reverse causality- households using firewood because they have children available out of school-would induce the same direction of bias. 378 Simon Appleton percent in central urban areas and 45 percent in northem rural areas. Evalu- ating at the means of the explanatory variables, the proportions change to 75 and 57 percent, respectively. Thus, just over half of the gap in enrollments between the two areas has been explained in terms of the determinants in- cluded in the model. However, northem rural data have particularly atypi- cal educational outcomes after controlling for observable determinants such as education and income. The model shows relatively small differences be- tween other areas after such controls. For example, in the raw data 83 per- cent of the sample in the central urban region attend school compared with 70 percent of the sample in the central rural area. Nonetheless, in the model, rural and urban areas of the central region are predicted to have the same proportion in school, at the mean of other explanatory variables. The model therefore explains all the mean differences in enrollments between rural and urban areas of the central region in terms of observable determinants, such as parental education and income. The model also predicts small urban-rural differentials-other things being equal-in western and east- ern regions. Only the sizable urban-rural differential in the northem re- gion is unexplained by observed household characteristics. In summary, the multivariate analysis of school enrollment in 1992 iden- tifies inequalities that have disappeared or considerably diminished with UPE. Taken individually and holding other things constant, differences in paren- tal education, gender, household income, and region can by themselves lead to pronounced differences in the probability of attending school. Often such effects will be combined: for example, northem rural households are likely to have below average parental education and household income. Inequali- ties in education are likely to be of concem for many reasons. The next sec- tion focuses on the instrumental economic importance of education. Retums to Education: Productivity and Labor Allocation Effects If, as is often claimed, returns to primary education in Africa are high, then that is prima facie evidence that market failures-such as credit market im- perfections-are preventing the realization of those retums. The size of the economic retums to education thus has some bearing on the efficiency of policies like UPE to expand access and provides evidence on the likely im- pact of such initiatives on economic growth. A survey of the literature-mostly conventional studies that measure the effects of education on wage eamings-reports social retums to primary edu- cation in Sub-Saharan Africa of 24 percent (Psacharopoulos 1994).10 Bennell (1996) has questioned these estimates because they are strongly influenced 10. Social returns are 18 percent to secondary education and 11 percent to tertiary education. Private returns are 41 percent to primary education, 27 percent to second- ary education, and 28 percent to tertiary education. What Can We Expectfrom Universal Primary Education? 379 by a few studies that used poor data." Recent estimates of Mincerian returns to education (that is, wage premiums to a year of education) produced dis- tinctly different results.'2 A study of 2,174 urban wage employees in Uganda in 1992 found Mincerian returns to education of 4 percent at the primary level, 8 percent at the secondary level, and 28 percent at the tertiary level (Appleton, Hoddinott, and Knight 1996).'3 These Mincerian returns are not comparable to the widely cited rates of return summarized by Psacharopoulos (1994). In particular, they may underestimate returns to primary school because they implicitly assume students forego wages to attend school, and they may over- estimate returns to tertiary education as they ignore the direct costs of the education. However, the Mincerian returns do suggest modest gross benefits to primary education. They appear to provide some support for the view ex- pressed by some in the Ugandan government prior to 1997 that students who left primary education had acquired few useful skills, and so no great eco- nomic return could be expected from additional expansion of such schooling. This section, in common with standard microeconomic estimates of rates of return to education, provides estimates of the benefits of education based on cross-sectional partial associations between education and earnings. These associations are unlikely to be a fully reliable guide to real returns. Perhaps the most common concern is that they may be subject to omitted variables bias; that is, educated adults may be more productive for unob- served factors (higher preschool ability, better family background, and the like) rather than because of their schooling. A few studies outside Uganda have attempted to control for such problems by trying to measure preschool ability (Knight and Sabot 1990) or by using difference estimates from samples of twins to remove the effects of family background (Ashenfelter 11. One such study of Uganda in 1966 estimated a 66 percent return to primary education, but relied purely on government wage scales for the educated. As the study had virtually no hard data on incomes of the uneducated, the 66 percent was effectively an assumption rather than an empirical observation (see Knight 1968, for an early critique of the study). 12. Mincerian returns to education are the wage premiums to a year of education (Mincer 1974). They correspond to the private return to education on a number of strong assumptions, notably that (a) there are no pecuniary costs to education, (b) the opportunity cost of education is the wage, and (c) the individual lives forever. 13. These returns show a similar pattern, but at somewhat lower levels than those obtained elsewhere in Sub-Saharan Africa. Since 1980, returns have averaged 5 per- cent for primary school, 14 percent for secondary school, and 37 percent for univer- sity (Appleton 1999). In 1990, a study of 298 employees from Kampala produced dif- ferent returns: 9 percent for primary education, 3 percent for secondary education, and 11 percent for tertiary education (Bigsten and Kayizzi-Mugerwa 1999). The au- thor was unable to reconcile the two sets of results for Uganda because the data un- derlying the latter estimates have apparently been lost. Given the difference in sample sizes and representativeness, the 1992 estimates appear more reliable. 380 Simon Appleton and Rouse 1998). These studies suggest that the ability biases are not large. The twin studies-albeit restricted to the United States-find no signifi- cant difference in returns to education from conventional estimates. Simi- larly, Knight and Sabot (1990) did not find a large independent effect of their preschool ability measure in urban wage determination in Kenya and Tanzania in 1980 (see also Glewwe 1996; Moll 1998). A more serious limitation of cross-sectional estimates is that they provide only a snapshot picture of current returns, when in reality returns to educa- tion accrue over decades. Moreover, nonmarginal changes in the provision of education-such as UPE-will reduce the scarcity value of education and lower its returns. For example, the expansion of secondary schooling in Kenya appears to have dramatically lowered conventional estimates of returns to education during the last two decades (Appleton, Bigsten, and Kulundu Manda 1999).14 If the Kenyan experience is typical, and the verdict is still out on this, it may help reconcile the large social returns based largely on pre- 1980s data (Psacharopoulos 1994) with the more modest Mincerian returns found in Sub-Saharan Africa in more recent studies. This caveat must be borne in mind when interpreting the results presented later in this section. This study goes beyond the conventional approach of estimating the ben- efits of education in terms of wage earnings. In Uganda and many other de- veloping countries, most workers are self-employed; only a minority of the labor force is in wage employment. In such a context positive correlations between wages and education do not necessarily reflect productivity effects from education. On average, wage employees who are more educated re- ceive higher wages in Uganda as in most other labor markets, but it is less clear whether more education benefits farmers or self-employed workers. Existing estimates of the returns to education in Uganda imply lower rates of return in agriculture than in wage employment."5 For this reason a broader approach to returns in terms of income-generating activities is taken-wage employment, farming, and nonfarm self-employment-using the household rather than the individual as the unit of analysis. In the case of self- employment, this approach overcomes the problem of assigning individual earnings when more than one member of a household works in a household enterprise.'6 Therefore, individually assigned wage earnings are aggregated 14. Primary retums do not seem to have been affected, although conventionally estimated returns to primary education are of limited relevance in Kenya, because all recent cohorts of urban wage employees have primary education. Returns to tertiary education in Kenya have also not fallen, and may even have risen. 15. Four years of primary education are associated with a 7 percent rise in agri- cultural productivity (Appleton and Balihuta 1996). This rate of return is typical of developing countries (Phillips 1994). 16. Although the survey reports individual income, it makes a rather unconvinc- ing distinction between unpaid helpers on family enterprises and the self-employed. It also assigns income equally among household members working on an enterprise. What Can We Expectfrom Universal Primary Education? 381 to the household level for ease of comparison with earnings from self- employment and farming. Reduced Form Estimates of the Return to Education One simple approach to estimating the returns to education is a reduced form approach. Household income is modeled as a function of the educa- tion of adult household members and other exogenous determinants of earnings. The focus is on the household's earned income, the sum of earn- ings from wage employment, and farming and nonfarm self-employment. The household education measure is the average education of adult non- students (students are excluded, as they are unlikely to be contributing significantly to household income), with a distinction between average years of primary education, average years of secondary education, and attendance at university. We also include as an explanatory variable the proportion of nonstudent household adults who went to university (be- cause data on years of university education are not available). The aver- age age of the adults and the proportion of women are included as con- trols. Other hypothesized determinants of earnings include quantities of the household factors of production: labor (number of adult nonstudents), cultivable land, and productive capital. Although the household's hold- ing of the factors of production could be considered endogenous, these factors are treated as exogenous due to a lack of good instruments. We include controls for whether a woman heads the household, how many years the household has lived in the area, and whether the father of the household head was a farmer. The coefficient on average years of primary education is 0.043 (see table 12.2). This implies that an extra year of primary education for each nonstu- dent adult in a household is associated with earned income that is 4.3 per- cent higher, other things being equal. A year of secondary education brings a greater increment to earnings than a year of primary education. We cannot precisely estimate the increment from a year at university. However, assum- ing university attendance takes three to four years, the increment per year is larger than that of schooling. These results would imply that the rate of return is higher for postprimary education if we make the Mincerian assumptions that there are no pecuniary costs to education and that the opportunity costs are foregone adult earnings. These assumptions are not useful, however, when comparing the returns to different levels of education. Pecuniary costs are higher for postprimary edu- cation, particularly university education. Opportunity costs are also likely to be higher for postprimary education. Indeed, attending primary school may not lead to a significant loss of earnings-the students may be too young to be generating significant income outside of school. The opportunity costs of edu- cation depend partly on the productivity of child labor. Child wages for agri- cultural work in Uganda are less than half of the adult wage. However, few children work for wages and a more relevant estimate of productivity may be 382 Simon Appleton Table 12.2. Reduced Form Household Earnings Functions with Community-Level Fixed Effects Variable Coefficient Characteristics of household workers Average years of primary education 0.043a Average years of secondary education 0.091a Average been to university 0.440a Average age 0.0457a Average age squared -0.000591a Proportion women 0.155a Factors of production (quantities) Log number of household workers 0.489a Log capital 0.042a Log cultivable land 0.273a Other determinants Log years resident in location 0.011 Female-headed household -0.262a Head's father had nonagricultural work 0.054a Note: Controls for missing values of land and capital not reported. a. Significant at the 1 percent level. Source: Author's calculations from the 1992 Integrated Household Survey. gained from analysis of the marginal products to family labor."7 Research ex- tending the analysis of household earnings from agriculture (discussed later) to disaggregate labor into adult and child labor suggests that child labor can be just as productive as adult labor on famnily farms (Angemi 1999).18 How- ever, children of primary school age who are not enrolled in school typically do not work full time. Therefore, the household income foregone by those at- tending school is likely to be substantially less than a full-time wage (or mar- ginal product).19 For example, in the 1992/93 survey, only 38 percent of chil- dren ages 7 to 14 not attending school reported helping with family enterprises (almost exclusively farms) and virtually none worked for wages. Of the third who did work on the farm, the average number of hours worked per week 17. Out of 10,459 children aged 7 to 14 covered by the integrated household sur- vey of 1992/93, only 52 worked for wages and had usable data on wage rates. 18. Ordinary least squares estimates imply that adult labor is approximately 10 percent more productive than child labor. When child labor is instrumented for, it appears twice as productive as adult labor. 19. From a welfare standpoint, labor supply considerations would not matter if one valued the child's leisure at their marginal productivity, but such a valuation is controversial. What Can We Expectfrom Universal Primary Education? 383 was 33. Conversely, 22 percent of children in school helped with household enterprises, working, on average, 16 hours per week. A simple comparison of these statistics implies that school attendance reduces the amount of child la- bor by only about eight hours per child per week. This is probably an underestimate, because the study fails to control for age differences between children in and out of school, and does not consider work on nonincome-generating activities, such as domestic work. However, the comparison does suggest that the Mincerian assumption-that the op- portunity cost of primary school is a full-time wage (or marginal product)- is inappropriate. Given a typical working week of 40 hours, the Mincerian assumption overstates the opportunity cost of schooling by a factor of five (40 * 8 hours of child labor lost). Conversely, the true monetary returns to primary education may be five times greater than the 4 percent productivity benefit estimated. What this suggests is that although primary education should not be expected to give a large boost to output, nonetheless, it may be an attractive investment, even from a narrowly monetary perspective.20 As expected, all three factors of production-labor, capital, and land- have a significant positive effect on household earned income. A house- hold with more adults is associated with higher earnings, but the increase is not proportional. The elasticity is around 0.5, implying that doubling the number of household adults reduces their average earnings by around a quarter. The stock of productive capital held by the household has a rela- tively small coefficient. However, the median level of capital is low, rela- tive to median earnings, so the coefficient implies an extremely high rate of return to physical capital.2" Holdings of cultivable land have a significant positive association with earnings. Aside from education, a number of other characteristics of house- hold workers affect earnings. An inverse U-shaped relation exists between household earned income and the average age of the adults. The tuming point of the relationship occurs at age 39. Up to that point, household earn- ings rise with the average age of their adults; thereafter, they fall. We obtain mixed results relating to gender and earnings. Other things being equal, a higher proportion of adult women is associated with higher earnings, but a female household head is associated with 23 percent lower earnings. The latter results echo the findings of an earlier study of female- headed households in Uganda using the same data set (Appleton 1996). Female-headed households as defined in the survey include some de facto 20. Further adjustments would be required for the pecuniary costs of primary education and the fact that recipients do not have infinite lives. However, the rate of retum is likely to remain reasonable even after these adjustments. 21. Median earnings are U Sh 343,200 per year, and median holdings of produc- tive capital are U Sh 6,200. Together with the coefficient on the log of capital in the reduced form, this implies a 232 percent return to capital (0.042 x 343,200/6,200). 384 Simon Appleton women-headed households that receive substantial remittances, sometimes from migrant husbands. As a group, female-headed households in the sur- vey have lower earned incomes than male-headed households do, although high remittance receipts prevent them from having lower total household income and expenditure. Among the other variables included, there is no effect of length of residence on earnings. Lastly, if the father of the household head is engaged in nonagricultural work, the prediction is that total earnings will be 5 percent higher. The reduced form results provide a simple overall assessment of how education is associated with higher earnings, controlling for household en- dowments of factors of production and various characteristics. It does not reveal the channels or mechanisms through which education raises income. One channel is through education raising productivity within particular income-generating activities. Such direct productivity effects can be estimated through earnings (or production) functions for particular activities, for ex- ample, wage employment or farming. A second channel is through educa- tion, increasing the likelihood of the household engaging in higher return activities or "entry effects," because education is hypothesized to affect the probability of a household entering particular sectors. In what follows we try to estimate the various effects of education. One motivation is to compare conventional estimates of returns to education based on wage earnings with broader estimates. Do we obtain misleading results by focusing solely on wage earnings? Are returns to education lower in agricul- ture and nonfarm self-employment than in wage employment? These factors will be important in assessing the benefits of education to children in rural areas where opportunities for wage employment are limited. A second moti- vation is to assess whether many of the benefits of education come in the form of access to wage employment. For example, take the extreme case where edu- cation does not raise returns in either farming or wage employment, but merely allows workers to enter the higher return wage sector. In such a case, the pri- vate benefits of education may not lead to corresponding social benefits.21 Entry Effects This section decomposes the effect of education in entry effects and direct productivity effects. The decomposition of entry effects identifies three house- hold income-generating activities: wage employment, farm self-employment, and nonfarm self-employment. It is assumed that the income-generating ac- tivities for a household depend on the number of adult members, their edu- cation, their other characteristics (age and sex), the parental background of the household head, and the region in which the household lives. Adults are 22. Much would depend on whether education increases total employment in the higher-return wage sector or merely rations a given number of jobs. What Can We Expectfrom Universal Primary Education? 385 defined as those over the age of 15 who are not full-time students.23 We did not include household holdings of land and productive assets, as these are endogenous with respect to the household's engaging in a particular activ- ity. Independent probits are used to model whether a household earns in- come from a particular activity.24 Table A12.2 reports the full results. Table 12.3 reports some predictions of the model evaluating at the population- weighted mean of the all the explanatory variables. Table 12.3. Probabilities of Engaging in Income-Generating Activities Nonagricultural Wage Category Farming self-employment employment Baseline (mean of all variables) 92 27 33 Characteristics of all adult household members (nonstudents) No education 94 23 29 Complete primary 91 33 31 Four years secondary 78 22 60 Al men 84 22 53 All women 96 32 19 One adult 85 23 27 Three adults 95 31 38 Occupation offather of household head Farmer 94 26 29 Nonagricultural self-employed 91 36 30 Government employee 93 25 39 Private employee 89 27 38 Other variables Male headed 92 27 31 Female headed 89 28 38 10 years resident in area 89 28 36 50 years resident in area 95 25 27 Note: Predictions from probit models evaluating at the mean of the explanatory variables. Also controlled for, but not reported, are dummies for location (region by urban-rural), for parental education, and for missing values. Sample size: 9,078. Source: Author's calculations from the 1992 Integrated Household Survey. 23. One could argue that adult nonstudent members are endogenous in this model. However, we do not have good instruments for educational attendance. 24. The use of independent probits is a simplification. An altemative approach would be to model the choices to enter each activity jointly, for example, using a multi- nomial logit. Households could be modeled as falling into one of six categories: farm only, nonfarm self-employment only, wage employment only, farm and nonfarm 386 Simon Appleton In Uganda, primary education alone appears to do little to increase ac- cess to wage employment. Average primary education of the adults has no significant effect on the probability of receiving income from wage employ- ment. However, primary education significantly reduces the probability of the household receiving any income from farming and increases the prob- ability of receiving income from nonfarm self-employment. The latter effect is large. At the mean of the other explanatory variables, if the household has no adults with primary education, it has only a 22 percent probability of obtaining income from nonfarm self-employment. If all the adults in the household have primary education, the probability of engaging in nonfarm self-employment rises to 33 percent. By contrast, secondary education has powerful effects on the probability of obtaining income from wage employ- ment but reduces the probability of receiving income from nonfarm self- employment and farming. Table 12.3 shows that households where all adults have a secondary education would be twice as likely as uneducated house- holds to receive some wage income, other things being equal. The other determinants of participation in activities are worth mention- ing, although they are not the primary focus of this chapter. For all three ac- tivities, the probability of receiving nonfarm earned income rises with the number of workers in the household. This may reflect some benefit in re- duced risk from the diversification of income made easier with several work- ers. Given an underdeveloped land market, there may also be limits to the amount of family labor that can be gainfully employed on a family farm, so that additional workers must look for employment elsewhere. The sex ratio among adults in the household affects participation in different activities. An increase in the proportion of women workers in the household is associated with a rise in the probability of receiving self-employment income, including farming, and a fall in the probability of receiving income from wage employ- ment. Comparing an otherwise average household composed entirely of women with one composed entirely of men, table 12.3 shows that the women- only household is nearly 50 percent more likely to engage in nonfarm self- employment, but only 40 percent as likely to obtain wage income. This may reflect the situation that self-employment is more compatible with child rear- ing or, alternatively, it may reflect discrimination in employment. Neverthe- less, having a female head reduces the probability of receiving income from farming and increases the probability of receiving income from wage employ- ment. This may reflect an endogeneity bias: women may be more able to head a household if employed. Alternatively, problems with women owning land may limit the feasibility of farming in female-headed households. The aver- age age of workers in the household enters as an inverse U-shaped quadratic self-employment, self-employment and wage employment, and engagement in all three types of activity. We use the independent probit approach for simplicity, be- cause distinguishing between all six categories is not the focus of the chapter. What Can We Expectfrom Universal Primary Education? 387 in all three probits. The turning point of the quadratic for the probit for farm- ing is so high (97 years), however, that age effectively has a monotonic posi- tive effect on the probability of the household farming. The turning points for the quadratics for nonfarm self-employment and wage employment are 35 and 38 years, respectively. Extended residence in a particular area is positively associated with farm- ing and negatively associated with both nonfarm activities, especially wage- employment.25 The main occupation of the household head's father exerts an independent effect on the activities of the household. If the head's father was self-employed in nonagricultural work, it raises the probability of the house- hold receiving an income from nonagricultural work, other things being equal, and lowers the probability of it engaging in farming. lf the household head's father was employed by government or the private sector, the probability of the household's receiving income from wage employment increases. Productivity Effects This section models earnings from household activities. Factors of production used to estimate a Cobb-Douglas production function for agriculture include labor, land, and capital.26 The function also controls for the characteristics of the household members working in agriculture and cluster fixed effects.27 Earn- ings from nonagricultural self-employment are similarly modeled using a pro- duction function, in which any land among the business assets of the enter- prise is included in capital. We cannot estimate earnings from wage employment as a production function per se, because it must be estimated at the firm level. Instead, we model household earnings from wage employment as a function of labor input and characteristics of the workers.28 The estimates are made af- ter allowing for community-level fixed effects and for the endogeneity of both labor input and the characteristics of the workers. As instruments we use the 25. One could make a strong case for a reverse causation interpretation here. Households are likely to migrate in order to obtain wage employment. 26. We do not include variable inputs such as seeds, fertilizer, and pesticides, because these variables are endogenous, and we lack good household-level instru- ments for them. Part of the effect of education and other factors may work via use of variable inputs (see Appleton and Balihuta 1996). 27. Given that the sample was drawn from many different areas, there may be differences in location conditions that affect earnings. For example, some clusters may enjoy better agroclimatic conditions for farming; others may enjoy higher de- mand for nonagricultural labor. One way to control for such differences is to allow for unobserved differences in mean earnings between clusters. These differences are sometimes termed cluster fixed effects. 28. Including enterprise capital may lower the return to education in the earn- ings function (see Bigsten and others 2000). The private returns to worker education, however, are those estimated without controlling for enterprise capital. 388 Simon Appleton number of adult nonstudents in the household and their characteristics. That is to say, we take the household's endowment of labor and its characteristics as given, but model the allocation across activities as endogenous. In many cases, allowing for endogeneity does not alter the results markedly. Often worker characteristics in the household map almost one-to-one onto worker charac- teristics for income-generating activities. In general, the qualitative results are fairly robust to controls for community-level fixed effects and for endogeneity.29 As with the reduced form earnings function, treating capital and land as en- dogenous is not possible in the absence of appropriate instruments. It could be argued that capital and land holdings change slowly and, hence, are less sub- ject to short-run endogeneity problems than, for example, labor allocation. We also make no control for the selectivity of activity choice because we cannot identify a priori any factors that may affect choice of activity that may not also influence returns within that activity.30 Table 12.4 estimates the determinants of earnings from each activity for the subsamples receiving any income from such activities. The similarity in the estimated effects of education on returns in all three activities is striking. Each average year of primary schooling of the household workers raises earnings from both farming and wage employment by 4 percent; for secondary school- ing, the corresponding increases are approximately 6 percent. If all household workers had been to college, returns to both farming and wage employment would be more than 40 percent higher. The returns to nonfarm self- employment are broadly similar: somewhat larger for secondary schooling and smaller for university. The rough similarity between the effect of education earnings functions for the wage employed and the nonagricultural self-em- ployed has also been found for samples from urban Kenya in 1978 and 1986 (Appleton, Bigsten, and Kulundu Manda 1999). Many other studies, however, have found stronger effects of education on returns within nonfarm activities than from on-farm activities (see, for example, Fafchamps and Quisumbing 1999). Particularly stark is the contrast between the findings of Appleton and 29. Controlling for community-level fixed effects reduces the estimated effect of education in agriculture, increases it in nonfarm self-employment, and has no effect in wage employment. Controlling for the endogeneity of worker education raises the effect of primary schooling in all cases. For secondary education, the effects of con- trolling for endogeneity are more varied: lowering returns in wage employment, rais- ing them in farming, and having no effect in nonfarm self-employment. Perhaps the most noticeable effect of endogenizing labor and its characteristics is to increase the coefficient on the labor-input variable in all three activities. 30. The occupation and education of the household head are perhaps the most promising instruments for activity choice. We have seen in the previous section that these variables do affect the probability of a household engaging in different activi- ties. However, a priori, it is hard to rule out productivity effects. For example, paren- tal education may enhance children's learning in school, while children may become proficient in a trade by learning from their parents. What Can We Expectfrom Universal Primary Education? 389 Table 12.4. Household Earnings Functions with Community-Level Fixed Effects and Endogenous Labor Variables Nonagricultural Farming self-employment Wage employment Category earnings (log) earnings (log) earnings (log) Characteristics of household workers Average years primary A 0.038b 0.056b 0.041b Average years secondary A 0.058b 0.073b 0. 057b Proportion been to university A 0.426b 0.278 0.424b Average age A 0.0259b 0.0568b 0.0338b Average age squared A -0.000341b -0.000713b -0.000476b Proportion women A 0.152b 0.314a 0.420a Factors of production (quantities) Log hours of work A 0.552 0.673b 0.579b Log capital 0.091b 0.027a n.a. Log cultivable land 0.261b n.a. n.a. Other determinants Log age of enterprise 0.031b -) n.a. Log years resident in location 0.054b (-) 0.001 Female-headed household () 0.394b -0.488b Head's father had nonagricultural work (-) (-) 0.030 n.a. Not applicable. A Treated as endogenous. (-) Not included due to insignificance. Note: Controls for missing values of land and capital not reported. a. Significant at the 5 percent level. b. Significant at the 1 percent level. Source: Author's calculations from 1992 Integrated Household Survey. Balihuta (1996) and the crop production functions estimated here, because both were based on the same data set. Appleton and Balihuta discovered a zero effect of secondary education and modest returns to primary education (2.8 percent). This apparent discrepancy is discussed later. The average age of workers has an inverse U-shape in all earnings func- tions, with returns peaking shortly after age 35 (38 in farming, 40 in nonfarm self-employment, and 36 in wage employment). Surprisingly, the proportion of women workers positively affects returns, although this finding is one of 390 Simon Appleton the few that is not particularly robust to the estimation method. Without con- trolling for endogeneity, women appear less productive than men in all three activities. The apparent endogeneity bias suggests that households relying on the labor of women face unobservable factors that are less favorable to generating income. Female-headed households appear to receive much less income, other things being equal, from the nonagricultural activities they engage in. The labor input variable is also fairly sensitive to estimation method, having a markedly lower coefficient when treated as exogenous. Again, this suggests that households work more when unobservables deter- mining income are unfavorable. This may seem counterintuitive; we might expect higher labor input when returns to labor are higher. However, where income and consumption are not separable, the finding may reflect income effects on the supply of family labor. Of the nonlabor determinants, capital has a rather low earnings elasticity in both agricultural and nonagricultural earnings functions (0.09 and 0.03, respectively). Land has an agricultural earnings elasticity of 0.26. Unlike Benjamin's (1992) findings for Thailand, controlling for community-fixed ef- fects does not noticeably alter the estimated productivity of land. Following Deininger and Okidi (chapter 5 in this volume), we have included variables for the age of the enterprise and years of residence in the location as proxies of informal human capital accumulation in farming. Similar to Deininger and Okidi, both are positive and significant in raising agricultural earnings. These two variables did not, however, significantly affect returns within nonagricultural activities.31 We also tried including whether the household head's father had been a farmer. Although this variable is positive and sig- nificant in regressions where labor is measured by number of workers, the effect is effectively zero when labor is measured in hours worked. This sug- gests that the variable works by raising labor input to the farm, rather than by raising total factor productivity. Putting It All Together Now that we have estimates on how education affects the type of income- generating activity a household engages in and how it affects returns within these activities, we can use this information to analyze the overall effect of education. We use a method similar to Fafchamps and Quisumbing (1999), who studied rural households in four districts of Pakistan. They looked at di- rect productivity effects of education in both wages and agriculture, together 31. We first estimated a general model including all explanatory variables. We excluded some variables that were wholly insignificant if the variables were not of great interest a priori. The exclusions did not materially affect the coefficients on the remaining variables. What Can We Expectfrom Universal Primary Education? 391 with the effect of education in reallocating labor between these activities.32 Table 12.5 reports the decompositions of the effects of primary and secondary edu- cation. In both cases, we look at the outcome of giving all adults in the house- hold an extra year of schooling. We assume that the models previously esti- mated would continue to be valid after expanding education, so implicitly we are concerned with a marginal expansion of education. This is a serious limita- tion to the analysis, because a larger-scale expansion-such as that initiated by UPE-is likely to alter both the probabilities (probits) for participation and returns within activities. However, it is hard to factor in this effect given that our data provides only a snapshot of Uganda in 1992/93. Consider first the effect of primary education on the probability of house- holds engaging in different income-generating activity (table 12.5, row 2). Pri- mary education reduces the probability of a household receiving income from farming and increases the probability of receiving nonfarm earnings, especially from self-employment. The key point is that the changes in the probabilities do not sum to zero. Primary education increases the probability of a household receiving nonfarm earnings by twice as much as it reduces the probability of receiving earnings from farming. For this reason, the overall entry effects of pri- mary education are positive. These effects are given in row 3, by weighting the marginal effects (row 2) by the average earnings of those engaged in each sector (row 1). Entry effects account for just over one-fifth of the overall return to pri- mary education (row 11). This implies that even if primary schooling had no direct productivity effects, it would raise household earnings by encouraging households to engage in nonfarm self-employment. Surprisingly, the overall entry effects of secondary schooling are less pronounced. It is true that secondary edu- cation greatly increases the likelihood of a household receiving wage earnings; however, this is almost fully offset by reductions in the probability of receiving earnings from farming and nonfarm self-employment. There is no support for the hypothesis that much of the return to secondary education in Uganda comes from switching people into wage employment out of other activities. 32. Fafchamps and Quisumbing (1999) found no productivity effects of educa- tion in agriculture, but they did find substantial labor allocation effects. In their analy- sis, the effect of education in reallocating labor from farming into wage employment accounts for about one-quarter of the total effect of education on household earnings. They did not consider the effects of education in determining whether a household receives any income at all from a particular activity. That is to say, they do not allow for the entry effects of education laid out in this chapter. Such entry effects may or may not be significant in Pakistan, but they clearly are significant in Uganda. Education has some consequence on earnings by increasing the probability of households ob- taining off-farm income. These entry effects are distinct from the effects of education in reallocating labor between activities. Other related studies include Coulombe and McKay (1996) and Vijverberg (1993). Table 12.5. Decomposing the Effect of Education on Expected Household Earnings Nonagricultural Wage Total earned Category Farming self-employment employment income 1 Mean conditional earnings (U Sh) 305,650 419,176 336,277 482,015 2 Effect on probability of receiving income (percent) p: -0.9 p: 1.6 p: 0.4 n.a. s:-3.6 s:-2.6 s:6.7 3 Entry effect (U Sh) = (1) x (2) p : -2,742 p: 6,695 p: 1,453 p: 5,406 s: -11,113 s: -10,765 s: 22,588 s: 710 4 Mean probability receiving income (percent) 83.26 27.64 33.20 n.a. 5 Productivity effect directly via education (percent) p: 3.5 p: 5.1 p 4.0 n.a. s:4.7 s:7.5 s:6.1 6 Productivity effect via labor supply (percent) p: -0.7 p: 1.4 p: 1.7 n.a. s:-4.6 s:2.2 s:3.3 7 Productivity effect via other worker characteristics (percent) p: -0.0 p: 0.1 p: 0.2 n.a. s:-0.3 s:0.4 s:1.5 8 Percent effect on mean conditional earnings: = (5) + (6) + (7) p: 2.8 p: 6.7 p: 5.9 n.a. s:-0.3 s:10.1 s:10.9 9 Weighted productivity effect (U Sh) = (4) x (8) x (1) p 7,042 p : 7,724 p: 6,617 p: 21,383 s -658 s: 11,726 s: 12,132 s:23,200 10 Total effect (U Sh) = (3) + (9) p: 4,300 p: 14,419 p: 8,069 p 26,789 s: -11,771 s 961 s: 34,719 s: 23,910 11 Percent total effect = (10)/(1) p :5.6 s : 5.0 n.a. Not applicable. p: Effect of increasing the average amount of primary education held by adult nonstudents in the household by one year s: Corresponding effect for secondary education. Source: Author's calculations from 1992 Integrated Household Survey. What Can We Expectfrom Universal Primary Education? 393 Secondary education appears to bring benefits by raising returns within activities rather than by allowing households to enter higher-return activi- ties. The direct productivity effects of education are given in row 5. These effects depend heavily on the educational coefficients estimated in the three earnings functions in table 12.4.33 However, raising education also has indi- rect effects on productivity by altering the hours spent in different activi- ties (row 6) and the noneducational characteristics of the workers engaged in different activities (row 7). The latter effects are typically small. One ex- ception is the effect of secondary schooling on the characteristics of wage employees, which is driven by secondary schooling substantially increas- ing female wage employment. However, the former-labor allocation-ef- fects are substantial. These labor allocation effects broadly mirror the entry effects discussed earlier. Primary education increases the amount of hours a household devotes to off-farm work at the expense of farming. Secondary education increases time spent in wage employment at the expense of both farming and nonagricultural self-employment.34 The distinction between these labor allocation effects and the entry effects is that the labor alloca- tion effects are conditional on the household's activity choice. For example, the negative labor allocation effects of secondary education on farming show that, even if households continue to farm, secondary education will reduce farm eamings by reducing labor input to the farm. Interestingly, this nega- tive labor allocation effect almost cancels out the positive direct effect of secondary education on farm earnings. Consequently, secondary schooling appears to have almost no effect on conditional earnings from agriculture and, in fact, lowers them marginally (see row 8). Conversely, the indirect effect of secondary education in increasing household labor allocated to wage employment amplifies the direct productivity benefits by over 50 percent. Even for primary schooling, the effect on agricultural earnings is only half of the effect on nonfarm earnings. Secondary schooling appears to have substantially larger effects than primary schooling on retums within nonagricultural self-employment and wage employment. These discrep- ancies are largely attributable to the labor allocation effects of education. 33. Typically, the education of all workers raises the education of workers in a particular activity on an almost one-to-one basis. The direct effect of secondary edu- cation on agriculture is the main exception. Increases in the secondary schooling of all household workers is predicted to lead to proportionately smaller increases in the secondary schooling of those household workers allocated to work on the farm. 34. The only difference in the sign of entry and labor allocation effects concerns secondary education and nonfarm self-employment. Secondary education reduces the likelihood of a household engaging in nonfarm self-employment (a negative en- try effect). However, if a household does engage in nonfarm self-employment, sec- ondary education increases the amount of labor allocated to it, creating a positive labor allocation effect. 394 Simon Appleton One of the surprising results of the earnings functions in table 12.4 was the finding that education had comparable direct productivity effects in agricul- ture to those in nonf arm self-employment and wage employment. This is in contrast to most of the literature on the returns to education, which tends to find more modest effects of education on agriculture than on wage employ- ment. For example, on the same data set, Appleton and Bahihuta (1996) and table 12.5 report that an extra year of primary education for each farmer raised agricultural production by 2.8 percent and that secondary education appeared to have no effect. The decomposition exercise in table 12.5 is able to explain this apparent contradiction. In particular, we can see that the overall produc- tivity effects of education on agricultural earnings (table 12.5, row 8) are close to those estimated on the same data set by Appleton and Balihuta (1996). Many differences exist between the agricultural earnings function in table 12.4 and the crop production estimated by Appleton and Balihuta (1996). The depen- dent variable is different (agricultural earnings compared to gross crop out- put, the functional form is different (Cobb-Douglas compared to translog), the exogeneity assumptions are different (education and worker characteristics are treated as endogenous here and not in Appleton and Balihuta 1996), and some of the determinants are also different. Some of these differences-par- ticularly regarding the dependent variable and endogeneity-do matter for the estimated effects of education. However, perhaps the most important point for the present purposes is that Appleton and Balihuta (1996) measured labor input by number of workers rather than hours of work. Secondary education lowers the amount of hours worked on a farm more than the number of work- ers. Consequently, when the number of workers is controlled for in a pro- duction or earnings function, secondary education appears less directly ben- eficial than if the number of hours worked is controlled for. Lin particular, if the agricultural earnLings function in table 12.4 is estimated with labor input measured in workers rather than hours, the coefficient on average years of secondary education falls from a highly insignificant 5.6 percent to an insig- nificant 1.9 percent.35 As most studies of the impact of education on agricul- tural productivity tend to measure labor input in workers rather than hours, these raise the possibility that the direct productivity effects of education on agriculture have been substantially underestimated. Examining the combined entry and labor input effects on earnidngs is in- teresting because the distinction is somewhat arbitrary (both revolve around the effects of education on labor allocation between activities). At the pri- mary level, the combined effect is substantial and accounts for approximately two-fifths of the total effect of primary schooling on expected household earn- ings. This is due, in roughly equal parts, to entry and labor allocation effects. 35. Surprisingly the labor-input coefficient is entirely robust to its measurement (remaining at 0.37 under both specifications) as are many other variables. The one exception, whether the household head's father was a farmer, only has an effect when labor input is measured in number of workers, not hours worked. What Can We Expectfrom Universal Primary Education? 395 This implies that studies of the impact of primary education that do not ac- count for these effects-for example, conventional estimates of rates of re- turn based on wage earnings-seriously underestimate the benefits. In con- trast to primary schooling, the combined effects are negative and reduce the total effect of secondary education on expected household earnings by one- fifth of what it would be otherwise.36 Effects of UPE on School Quality Uganda did not experience the postindependence educational expansion characteristic of most Sub-Saharan African countries. Official statistics on school enrollment ratios in Uganda appear particularly problematic, but, as far as we can rely on them, they show no sustained rise in the two de- cades after independence.37 The gross primary school enrollment ratio in 1980 was 50 percent, effectively the same as in 1960. By contrast, the other Sub-Saharan African countries for which data are available almost doubled their primary school enrollment ratios from 40 to 77 percent.3 The country's poor record on education does not appear to be wholly attributable to the Amin years in the 1970s-there was no sustained rise in enrollments dur- ing the 1960s either. After Amin's overthrow, however, Uganda caught up with the rest of the subcontinent, which was entering a period of stagnant enrollment rates, attaining a gross primary enrollment rate of 73 percent in 1985. Thereafter, official enrollment rates did not increase, remaining at 73 percent in 1995. Secondary enrollment rates rose slightly from 1960 to 1980 (from 3 to 5 percent). Rates then doubled to 10 percent in 1985, but in 1995 only stood at 12 percent. Consequently, in terms of official statistics, the first decade of the Museveni government appears to have achieved little in terms of school enrollment, 36. We notice that the total estimated effect of education-particularly secondary education-is somewhat lower using this structural approach than when using the reduced form model reported in table 12.3. This seems to be due to a violation of the assumption of log-normality of the error terms in earnings functions. In particular, if the earnings functions are estimated using a linear, rather than log-linear, functional form, the correspondence between the structural and reduced form estimates of the education is mathematically exact. 37. The 1992 integrated household survey implied much higher enrollment fig- ures than those officially reported, with the gross primary enrollment ratio being es- timated at 93 percent compared with an official figure of 73 percent (World Bank 1996). A survey of 250 government schools in 19 districts in 1996 implied a 60 percent increase in primary enrollments between 1991 and 1995, while official enrollment rates were stagnant (chapter 11 in this volume). Even the 73 percent official gross primary enrollment ratio is hard to reconcile with the reported doubling of primary school enrollments during UPE. 38. Sub-Saharan African figures are the cross-country average from 39 states, in- cluding South Africa (World Bank 1998). 396 Simon Appleton especially at the primary level.39 However, these quantitative indicators may provide an incomplete picture. It is widely argued that the educational sys- tem in Uganda had a high reputation for quality at independence, but this was destroyed during the chaos of the 1970s, partly through decaying school infrastructure and partly through the exile of many well-educated Ugandans. The first decade of the Museveni government had seen a period of educa- tional reconstruction that partially restored the quality of services. The situation changed dramatically during the campaign for 1996 elec- tions, when the incumbent, President Museveni, promised to provide free primary education to four children in every family. Although the govern- ment of Uganda had long declared the attainment of universal primary education a policy goal (for example, in the 1992 "White Paper on Educa- tion"), this promise was the first significant step toward attaining that goal. Following his re-election, President Museveni addressed the nation in De- cember 1996 and announced that he would implement his election promise starting in January 1997. The initial public response to this initiative was impressive, with primary school enrollment rising from 2.9 million in 1996 to 5.3 million in 1997. The key element of the initiative has been the aboli- tion of tuition and PTA fees (some discretion still exists in urban schools that continue to levy fees). These fees had assumed increasing importance in the 1970s and 1980s as state funding of education declined. By 1991 pa- rental contributions to primary schools constituted 70 percent of total fund- ing (43 percent at the median, see chapter 11 in this volume). Equity aspects of the initiative include a 1:1 gender balance requirement when identifying the four children per household to benefit from free education and funding the education of all orphans. The main concern about UPE is that it is likely to lead to deterioration in the quality of education provided. At least in the short term, government educational expenditures will not be able to rise sufficiently both to offset the abolition of PTA fees and the consequent expansion of student numbers. Although it is rather early to evaluate the effects of UPE in Uganda, some interesting exploratory evidence is provided by a study of 22 primary schools in Mukono and Kampala districts in 1998 (Otteby 1999). Otteby's study fo- cused on three dimensions of the impact of UPE: (a) the composition of the student intake, (b) school quality, and (c) academic performance.40 39. Figures on tertiary enrollment rates are incomplete but show a much greater proportionate expansion: from 0.1 percent in 1965 to 0.8 percent in 1985 and 1.5 per- cent in 1994. 40. The Uganda National Examination Board (1999) carried out a post-UPE testing of learning outcomes in mathematics and English using the same test administered in 1996. These data, when analyzed, will provide a more definite picture of the impact on learning and school quality. What Can We Expectfrom Universal Primary Education? 397 How Has UPE Affected the Composition of the Student Intake? In both the Mukono and Kampala schools, a large increase in enrollments followed UPE: 110 percent in rural schools and 30 percent in urban schools. Surprisingly, enrollment of boys increased more than that of girls- partly due to higher re-entry of boys at the third year (P3) and above. At the lower grades (P1 and P2), the proportion of girls increased following UPE. Otteby (1999) used an indirect approach to assess how UPE had changed the composition of the school intake in other dimensions. She sampled 20 second-year (P2) students and 20 fifth-year (P5) students from each school. The P2 cohort was selected to reflect a post-UPE intake, be- cause most additional enrollments following UPE were concentrated in the lower year groups. By contrast, students in the P5 year group had entered school prior to UPE. (Some students may have re-entered P5 following UPE.) A comparison of the characteristics of P2 and P5 students in 1998 should, therefore, shed some light on how UPE has changed the composition of the student intake. The material wealth of students' parents was calculated us- ing a 12-point scale measure of housing quality and possession of consumer durables. The average wealth of those in P2 and P5 grades in urban schools- both averaging about eight on the scale-was not markedly different. How- ever, in rural areas, the P5 cohort averaged more than five on the wealth scale, whereas P2 cohort averaged four. Moreover, in rural areas, those in P2 included students with low parental wealth (two points on the scale), whereas the P5 group had no similarly disadvantaged students. Likewise, the proportion assigned only three or four points on the wealth scale in P2 was almost double that in P5. How Has UPE Affected School Quality? Otteby (1999) inquired about school quality before and after UPE. The av- erage student-teacher ratio in rural schools after UPE rose from 30 to 51, and in urban schools the student-teacher ratio rose from 50 to 66. How- ever, actual class sizes were typically much larger than student-teacher ra- tios calculated at the school level. For example, in urban schools, the student- teacher ratio for those in P2 rose from 41 to 77, while the average class size increased from 96 to 136. The increase in student numbers exacerbated an existing shortage of chairs and desks. In the rural schools visited, half the sample had no chairs for students before or after UPE. Before UPE 10 per- cent of rural schools had enough chairs; after UPE none of them did. In urban schools visited, the proportion reporting sufficient chairs fell from 70 percent before UPE to 31 percent after. The number of textbooks per student did not decline; indeed, the number increased slightly in the sample schools because of a USAID project. However, national statistics imply that this finding is atypical. 398 Simon Appleton How Has UPE Affected Student Performance? Comparable test scores for pre- and post-UPE student cohorts were not avail- able. Nevertheless, Otteby (1999) tentatively estimated the likely impact of UPE on average student performance. As a first step, current (post-UPE) P2 students were tested in mathemat- ics and English. They also took a test on nonverbal reasoning designed to be independent of schooling (Raven, Raven, and Court 1991). Performance in the mathematics and English tests were then modeled as a function of non- verbal reasoning and indicators of school quality. These models (or educa- tional production functions), used in the final analysis were parsimonious. The statistically significant variables were retained only after a stepwise elimi- nation of other variables from a more general specification. In the final edu- cational production functions, performance in English and mathematics de- pended on three explanatory variables. Academic performance depended positively on nonverbal reasoning, negatively on the student-teacher ratio, and positively on indicators of school quality (school facilities in the case of English, textbook-student ratios in the case of mathematics). Performance in the tests of nonverbal reasoning, in turn, was modeled as a positive function of parental wealth and education. The most important finding was the negative relationship between student-teacher ratios and performance in both English and mathematics. The finding is noteworthy given the possible positive endogeneity bias of good schools being oversubscribed and the observation that better-funded urban schools have larger classes than rural schools. This finding contradicts most of the literature on educational production functions, which more often than not finds class size to be insignificant (Fuller 1987; Hanushek 1986). However, nearly all the existing literature relates to class sizes below fifty and provides little guidance on what happens when class sizes rise to the high levels now observed in Uganda. Otteby (1999) used the estimated edu- cational production functions to simulate the impact of academic performance on the rise in student-teacher ratios after UPE. Consider, for example, the estimated impact of a rise in student-teacher ratios from 41 to 77 as was ob- served in P2 in urban schools. If this rise had not occurred, the predicted average test scores would be 11 percent higher in mathematics and 6 percent higher in English. Clearly, these are substantial effects and cause for concern. Otteby's (1999) educational production functions were used to predict how P5 students would have scored if, pre-UPE, they had sat for the same mathematics and English tests in P2. The purpose of this simulation is to gauge what is likely to happen to average academic performance as a re- sult of UPE. Clearly, prior to UPE students are likely to have scored better in the tests than the current P2s. This is partly because of the increasing student-teacher ratios noted earlier. However, a decline in average academic performance is also likely for "compositional" reasons. For example, UPE What Can We Expectfrom Universal Primary Education? 399 has increased enrollment in rural areas more than in urban areas, and rural students tend to do not as well academically in Uganda. Moreover, even within rural and urban areas, we have seen that UPE has led to greater educational access for poorer students, who again tend to perform less well on average. Taking all these factors into account, Otteby (1999) predicts that, had she been able to test the P5 students when they were in P2, their scores would have averaged 10.1 in English and 7.5 in mathematics. The actual test scores of the post-UPE cohort currently in P2 averaged 8 in En- glish and 6.6 in mathematics. Consequently, this simulation predicts that UPE will lead to a fall in mean academic performance of 21 percent in En- glish and 11 percent in mathematics. Changes in the mean parental background of the pupils account for a relatively small part of these simulated declines: 11 percent in English and 5 percent in mathematics. Much more important is the fact that, after UPE, a higher proportion of students will come from schools with low indicators of school quality. Inferior school facilities account for 70 percent of the predicted decline in performance in English. In mathematics, 77 percent of the fall is due to a rise in the student-teacher ratio. In English, the rise in the student- teacher ratio accounts for 20 percent of the predicted fall in performance. These results imply that UPE will lead to a fall in mean academic perfor- mance, primarily through a worsening of school quality (student-teacher ra- tios and school facilities). The analysis of the effects of UPE on quality is indicative; such a small sample is not nationally representative. The simulation of the likely effects of UPE on average performance requires a number of strong assumptions.41 None- theless, the analysis does suggest what casual reasoning would imply: that UPE is likely to lead to an observed fall in average academic performance. Most of this fall will be compositional. The typical student will come from a less advantaged background. More importantly, the expansion in student num- bers will be disproportionately concentrated in already disadvantaged rural schools. Arguably, these compositional effects are not a cause for concern; if average performance falls after UPE only because of these effects, no student will be worse off. However, the estimated educational production functions also imply that by increasing student-teacher ratios in individual schools, UPE is likely to lead to substantial falls in academic performance. 41. These assumptions include the following: that primary school performance in general can be captured by testing those in P2; that the cross-sectional estimates of the parameters of the educational production function are unbiased; that UPE has not led to a "structural break" in the educational production function; that the char- acteristics of P5 students and their schools provide a good estimate of what the char- acteristics of P2 students would have been without TUPE; and that UPE has not al- tered the unobservable characteristics of students and schools. 400 Simon Appleton Summary and Conclusions Although the data used here to model determinants of educational attain- ment and returns to education come from the 1992 Integrated Household Survey and were collected prior to the UPE initiative of 1997, they still pro- vide useful insights into its likely effects. The data show that before UPE, particular types of children were less likely to attend school: girls, children from households with poorly educated parents; children from extremely poor families; and children from certain regions, such as rural northern areas. To the extent that UPE is successful, children with these types of characteristics will be able to benefit from education. Assessing UPE's impact on economic efficiency is much harder than in- ferring its equity effects. Analysis of the determinants of household earnings suggests that for each year of education, eamings increase by about 4 per- cent. This is a modest benefit, although it may still imply a high rate of return if the opportunity costs of attending primary school are low. The most sur- prising finding of the analysis is that education has similar proportional pro- ductive benefits in all three income-generating activities: farming, nonfarm self-employment, and wage employment. This is what one might expect if human capital were allocated efficiently across activities. However, it does run contrary to the common belief-and much supporting evidence-that education is rewarded more in wage employment and brings only small re- tums in farming. We suggest that the relatively small returns to education conventionally found in agricultural production functions may partly arise from a failure to control properly for the input of labor to farming. The re- sults of the study imply that education may bring tangible benefits to the poor in Uganda, who typically are not wage earners. Estimates of the extent to which education brings returns by reallocating labor rather than by direct productivity benefits were rather surprising. One would expect a principal indirect benefit of secondary education to be in- creased access to wage employment. However, in our decompositions this benefit was wholly offset by the loss of income associated with withdrawal from farming and nonfarm self-employment. Indeed, the estimates for the combined entry effect and labor allocation effects of secondary education were mildly negative. By contrast, an important channel through which pri- mary education appears to benefit households is in encouraging entry to nonfarm income-generating activities and reallocating labor out of farming. The efficiency effects of the UPE initiative will depend partly on how it alters the quality of education. Exploratory research of two districts in Uganda implies that average academic performance may fall sharply under UPE. Much of this will be purely compositional, however, as more students from disad- vantaged backgrounds are now enrolling, and they are enrolling dispropor- tionately in lower performing rural schools. Nonetheless, other grounds for concem exist. Although most studies of educational production functions im- ply no adverse effects of class sizes on academic performance, the exploratory research reported here suggests that this generalization may not remain valid What Can We Expectfrom Universal Primary Education? 401 in the context of the extremely large class sizes apparent in post-UPE Uganda. Higher class sizes and generally stretched resources are likely to reduce the amount children learn at school. The challenge of the UPE initiative is to com- bat these potentially adverse consequences. Annex 12.1. Models Table A12.1. Multinomial Logit Model for School Attendance for Children Ages 5-14,1992 Dropped out of school In school Category Coefficient t-ratio Coefficient t-ratio Constant 2.9767 1.32 0.9710 0.45 Female -0.0110 -0.12 -0.3704 -6.57 Father literate 0.8339 2.53 0.5708 2.34 Father some primary -0.3098 -0.95 0.0461 0.19 Father full primary -0.1627 -1.43 0.3362 5.18 Father postprimary 0.1328 0.76 0.3151 3.20 Mother literate 0.2880 0.81 0.0695 0.28 Mother some primary 0.0353 0.10 0.6133 2.49 Mother full primary 0.1645 1.02 0.5189 5.78 Mother postprimary 0.7170 2.63 0.4498 2.58 Female head 0.1384 1.49 0.1672 2.89 Age of head -0.0419 -2.45 -0.0206 -1.79 Age head squared 0.0005 2.98 0.0003 2.48 Log income -0.8789 -2.20 -0.5861 -1.54 Log income squared 0.0489 2.69 0.0475 2.81 Number of boys 0.0387 1.49 0.0524 3.35 Number of girls -0.0616 -1.43 -0.0199 -0.77 Firewood -0.1144 -2.49 0.0480 1.76 Piped water -0.3890 -2.79 -0.1099 -1.23 Fees 0.3423 2.44 0.0628 0.73 PTA charges 0.0146 2.16 0.0105 2.18 Distance to primary school -0.0819 -3.59 -0.1331