POLICY RESEARCH WORKING PAPER 2906 Macroeconomic Effects of Private Sector Participation in Latin America's Infrastructure Lourdes Trujillo Noelia Martin Antonio Estache Javier Campos The World Bank World Bank Institute Governance, Regulation, and Finance Division and Latin America and the Caribbean Region Finance, Private Sector, and Infrastructure Unit October 2002 | POLICY RESEARCH WORKING PAPER 2906 Abstract Trujillo, Martin, Estache, and Campos provide empirical There is some degree of crowding-out of private evidence on the impact that private participation in investment resulting from greenfield projects in utilities, infrastructure has had on key macroeconomic variables and delayed crowding-in from concessions in transport. in a sample of 21 Latin American countries from 1985- * There is crowding-in of public investment by private 98. Specifically, they look at the effects on GDP per participation in utilities, while there is crowding-out by capita, current public expenditures, public investment, increased private investment in transport. and private investment, controlling for country effects * Private participation in utilities decreases recurrent and institutional factors. The authors also investigate the expenditures, while in transport it results in an increase. relevance of the specific contractual form of private The net effect on the public sector account is participation contracts on these variables and show uncertain, but this uncertainty is a major risk. The differentiated effects according to contract types. revelation of this risk may be the main contribution of The results suggest that: this paper since it is inconsistent with the fiscal gains * Private sector involvement in utilities and transport expected by many policymakers as they engage in have some, but not impressive, positive effects on GDP infrastructure privatization programs. per capita. This paper-a joint product of the Governance, Regulation, and Finance Division, World Bank Institute, and Finance, Private Sector, and Infrastructure Unit, Latin America and the Caribbean Region-is part of a larger effort in the Bank to increase understanding of infrastructure regulation. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Gabriela Chenet-Smith, room J3-304, telephone 202-473-6370, fax 202-676-9874, email address gchenet@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. Antonio Estache may be contacted at aestache@worldbank.org. October 2002. (26 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Research Advisory Staff Macroeconomic effects of private sector participation in Latin America's infrastructure Lourdes Trujillo" Noelia Martfin** Antonio Estache Javier Campos" We would like to thank Raquel Carrasco, Alex Galetovic, Luis Guasch, Marianne Fay, Danny Kaufman, Martin Rodriguez-Pardina and Luis Serven for useful discussions and suggestions. Department of Applied Economics, University of Las Palmas, Spain. The World Bank and ECARES. 1. Introduction The relevance of the design of institutions for the effectiveness of policies is now well recognized by policymakers from all boards (see The World Bank, 2002 for a recent survey of the debate and of the evidence). This design becomes particularly important where reforms have significantly changes the types and roles of players. The restructuring of the infrastructure sector with a view to increase competition and private sector participation in Latin America over the last decade provides a clear example of such an institutional change. Since the late 1980s, many Latin American countries have indeed progressively opened their infrastructure sectors to private operators, seeking a remedy to structural deficits and hoping to foster investment and growth. The literature on the impact of these reforms can be classified into three broad types. The first group focuses mostly on the macroeconomic effects of a macroeconomic view of the reforms - the "macro-macro" group -; the second, on the sector-specific effects of sector-specific reforms - the "micro-micro" group, and the third on the macroeconomic effects of sector specific reforms - the "micro-macro" group. The "macro-macro" category is by a wide margin the most populated. It has generally been supported by fairly detailed econometric analyses of the relatively good macroeconomic databases available (see McGillivray and Morrissey, 1999, for a relatively recent survey). The "micro-micro" group has generated fewer analytically strong studies partially because detailed relevant data is not easy to obtain. Most of the published papers have focused on Argentina and Chile, where enough time has gone by to generate reasonable time series (for a recent overview, see Guasch, 2001). A much more modest literature has focused the micro-macro group and looked at the macroeconomic effects of sector-specific reforms, even if these sectoral reforms have been key components of the overall macroeconomic restructuring agenda. Exceptions include the literature on the general equilibrium effects of reform or the literature on convergence (see De la Fuente, 2000, or Estache, Foster and Wodon, 2002). This paper contributes to the "micro-macro" literature by offering a first empirical assessment of the macroeconomic effects of the increased private sector participation in the 2 management and financing of the infrastructure sectors (PPI).1 Its main purpose is to provide empirical evidence for Latin America of the effects on several key macroeconomic variables of the increased role of "privatization", defined as the decision to rely on the private sector to 2 implement projects. In this analysis, we also isolate some institutional factors and control by country effects, recognizing that each country in the region may face different sources of risks. In view of these objectives, the paper suffers from at least two major drawbacks. First, the quality of the data available is a significantly restrictive factor and will force us to further limit the hope placed on being able to draw very strong policy conclusions. This drawback is however also a source of strength since it forces to highlight the main direction for additional work through the analysis. Second, it lacks an explicit theoretical model to justify the "micro-macro" effects of increased PPL.3 However, our goal here is not to test any specific theory but rather to provide - if possible - statistically significant evidence on the sign and size of the effects of PPI on the most common macroeconomic indicators. We specify very general empirical relationships between each of our macroeconomic variables and different subsets of instruments that summarize when and how private participation was introduced in each country and under which institutional environment. Our results, therefore, do not look for causal liaisons; they represent correlations that may hint what the macroeconomic impact of "privatization" (if any) has been so far in Latin America. In spite of these limitations, we manage to obtain useful results, relying on standard econometric technique. We, first, estimate pooled-data models ignoring country specific effects. These models provide both initial values for the "micro-macro" effects and a benchmark for comparisons. However, if unobserved individual heterogeneity (ie., country-specific effects) is relevant in our statistical relationships, its omission yields biased estimates. To overcome this problem, we also estimate panel data models that allow for an explicit testing of individual 'Siniscalco et aL (2001) provide a recent similar study for OECD countries. 2 The recent literature on regulation theory explicitly acknowledges that the term "private participation" is much more general that "privatization". The former encompasses many different forms that include divestures, concessions, management contracts, leases, etc. (see Laffont and Tirole, 1998). 3 The new growth theory literature is the most likely place to find such a model (see Aghion et al., 1999, for example). On addition, political economy models may help explain under what circumstances privatization policies can be a success or a failure (see Alesina and Perotti, 1996). 3 heterogeneity.4 The differentiation between these two types of model specifications yields evidence on the effects of the "privatization" policies for the region as a whole as well as for average country specific effects. The remaining sections of the paper are organized as follows. Section 2 discusses the methodology followed to draw a minimum set of robust policy implications on the effects of PPI. Section 3 describes the sample and defines the most relevant variables used. Section 4 summarizes the main results on the impact of private participation policies on each macroeconomic variable. Finally, Section 5 concludes with a quick summary of the main empirical implications. 2. Testing the macroeconomic effects of PPI There is no simple way of anticipating the overall macroeconomic effects of a policy opening infrastructure to the private sector in a particular country because many tradeoffs are at stake. The best we can do with the kind of data available is to focus on "reduced" forms which net out more structural positive and negative effects of the reforms on the key macroeconomic variables because we do not have enough data to separate them. From the viewpoint of private investment, for instance, many privatization policies are usually expected to bring about positive results on the medium or long run if the overall efficiency of the economy is improved as a result of the policy changes. However, the long run payoffs may be preceded by short-run costs, if increased competition reduces margins and profits, hampering the investment capabilities of private investors. Since we can only look at the sectors at a very aggregate level and sector specific reforms are difficult to pick up, we can only see the accumulated effects of the policy changes year after year, as new investments results from all prior reforms and cannot be assigned to any specific policy change. This is why we focus on a very limited concern which has not been studied so far. We focus on identifying the sign of the effects that can be genuinely attributed to the net effects of private sector participation in infrastructure projects. The size of the effects is also computed but are for now less interesting since it probably represents a large number of effects with opposite signs. 4 See Chamberlain (1984) for a survey and examples on the use of macroeconomic panel data. 4 With these limitations in mind, the paper proposes a formal test of the consequences of infrastructure "privatization" as defined specifically later on four selected macroeconomic variables: total gross domestic product (GDP) per capita, private investment, public investment and current public expenditures. The first dependent variable is measured in levels. By focusing on per capita figures, we also get a chance at a modest look at the impact on poverty though income levels. The other four are calculated as a percentage of the GDP. Within infrastructure, we will additionally distinguish between utilities (electricity, gas, water and sanitation and telecommunications) and transportation (which include airports, ports, railways and roads) to test for possible differences.5 Since the timing of the changes and the policy environment varies significantly across countries, we simultaneously control for a time trend and variables that represent the institutional framework. In addition, we specifically take into account and test the possible existence of (unobservable) country-effects.6 Formally, we handle the data in two separate ways. First, we use all available data, pooling together the whole information set into a single sample. In this pooled-data case, where each country and year is treated as separate observation (denoted by subscript i) and no individual heterogeneity is allowed, we specified and estimated the following linear relationship for each one of our five macroeconomic variables:7 yi = a + d,'p + x,,y + ej;. (2.1) The term y, represents the dependent variable, a is the intercept, d', is a vector of dummies that account for private participation and its starting year, and x', is a vector of control variables that include a time trend and others that characterize the country's institutional framework; finally, 4i is a normally distributed error term, uncorrelated with the regressors, and a, f3, y the (vectors of) parameters to estimate. 5 For a review on how different types of infrastructures affect macroeconomic fundamentals, see for example Munnell (1992) or Gillen (1996). 6 As usual, a strong misspecification risk is always present in this sort of ad hoc models. However, since our idea is to isolate partial correlations among the privatization variables and the macroeconomic one, the use of the time trend and the institutional variables is the easiest way of minimizing that risk in this kind of heterogeneous sample. A lagged dependent variable (tried at preliminary stages of the work) would have done something similar, but at the costs one degree of freedom and lower significance levels. 7Non-linear specifications were also discarded in preliminary estimations. 5 The variables included in x reflect the political and governance situation taking into account the degree of political stability of the country (approximated by the degree of internal conflict) and the strength of the governance structure of the country. According to the specification of x' it is possible to derive a separate models from expression (2.1). In the first one (Model 1), the macroeconomic variables are explained by two dummy variables that reflect the existence or not of some form of privatization in utilities and in transport (they will be labeled DU, DT respectively). The second model (Model 2) tests, in addition, for the effect of investment associated with a specific form of private sector participation on each one of the macroeconomic variables. We will distinguish three types of "privatizations" associated with private investment: divestures or sale of the assets (DMV), concessions (CONC) and greenfield projects (GP). Each of these variables is defined as the share of total investment from privatization associated with each contract type. The expected sign on these explanatory variables varies with the macroeconomic variable explained. If we are to believe the predictions of the advocates of privatization, for GDP per capita and for domestic investment as percentage of GDP, we should expect a net positive effect of infrastructure privatization, since these are some of the core macroeconomic promises of privatization. For the share of public investment in GDP and for the share of current expenditures, the a priori expectation would be a negative sign since we expect infrastructure privatization to reduce the overall size of the public sector. In addition, we also expect stronger institutions to generate better macroeconomic performances. The second model we test makes use of the panel characteristics of our sample, where a number of individuals (21 countries, denoted by subscriptj) are repeatedly observed through time (t=1985,...,1994). In the panel data case we can now specifically study whether there are country-specific effects not included in expression (2.1). For each of our five macroeconornic variables the linear relationship that we test becomes yj, =a+d'3+xx,,y+1j + ,j, . (2.2) Both dependent and independent variable have time-variability, but that a Model I and a Model 2 could be estimated again, using the same definitions of x provided above. The most significant difference between (2.2) and the pooled-data case is that a country-specific effect (labeled rY) is 6 explicitly accounted for, whereas the error term ,t, is again normally distributed and uncorrelated with the regressors. It is precisely the nonappearance of the country-specific term that may bias the estimates in the pooled-data case due to a standard "omitted-variable" problem (Amemiya, 1985). Panel data models allow for a method to correct this problem, either using a "fixed-effects" or a "random-effects" approach. In the first case, the (unobserved) individual heterogeneity is represented as a parametric shift in expression (2.2). It is "as if" a new intercept, aj = a + TV, time-invariant and particular to each country, were defined and the estimation by ordinary least squares (OLS) would explicitly consider it. In the random effects case, the individual heterogeneity term is assumed to be part of the error term, Ujt = flj + t#. The model becomes heteroskedastic and must be estimated by Generalized Least Squares (GLS). Unfortunately, both approaches do not always yield the same result, as observed by Hausman (1978). However, if the effects of omitted variables can be appropriately summarized by a random variable and the (unobserved) individual effects may also represent the ignorance of the investigator, it does not seem unreasonable to treat in one case the. source of ignorance as fixed (aj) and in the other case as random (Ujt). It appears that one way to encompass the fixed- effects and the random-effects models is to assume from the outset that the effects are random and use GLS to estimate them. The immediate check, summarized in the so-called Hausman test, would be then to contrast whether the heteroskedasticity of the model allows a fixed-effect approach or not.8 An Hausman test is used when there are two estimators of the parameter vector , (e.g. ,IGLS and PoLs)- Under the null hypothesis (Ho) individual effects are not correlated with the regressors, PIGLs is consistent and efficient, but I3oLs is inconsistent. Under the alternative HI, both estimators are consistent, but PGLS is inefficient. This allows a routinely performed comparison between fixed-effects and random-effects estimates. A final important question regarding model specification is related to potential dynamic effects in our estimated relationships. In general, unless the economies behave in a hyper-rational 8 This argument has been widely discussed in the panel data literature. For example, Arellano (1993) insists on the fact that in the fixed-effects model investigators make inferences conditional on the effects that are in the sample, whereas in the random-effects inferences are based on the population. But there is really no distinction in the nature of the effect: it is up to the investigator to decide whether to make one type of inferences or the other. 7 way and manage to internalize instantaneously the effects of reform policies, the optimal lag for the dummies included in vector d should be different from zero. It is natural to expect that privatization may not bring its full (positive or negative) consequences immediately. Instead, based on a simple look at the facts in the region, a reasonable lag of one or two years should be considered. We investigate these dynamic effects by estimating, for each of our dependent variables, for each data case (pooled vs. panel), and for each of our models (Model 1 and 2) slightly different variations on (2.1) and (2.2) where the dummies have been lagged one and two periods. The results of all these estimations, reported in Section 4 below, allow us to analyze the macroeconomic effects of privatizations, by type of process, considering the existence of country-specific effects, and taking into account short rum versus medium-run impacts. 3. The variables, the data and their limitations We have collected a sample that covers 21 Latin American countries, all from Mexico to Chile, excluding Caribbean states, Belize, Surinam and French Guyana. In principle, this geographical dispersion offers enough variety of infrastructure reform experiences and of income levels to yield useful policy conclusions. The time period covered stops in 1998, just before the effects of the Asian crisis started to have a major impact in the financing of Latin America's infrastructure. The specific sample size for each macroeconomic variable considered in this study varies across the models estimated. This is driven by the fact that we could not obtain comparable data for all variables for all countries. The largest samples cover all of the 21 countries. The smallest focuses only on 16 countries. Since the overall sample tracks the changes in the role of the private sector in infrastructure for 14 years (from 1985 to 1998), the econometrics can make use of panel of data approaches as described above. Since there are several variables for which no information was available for some years, our panel is unbalanced.9 The macroeconomic dependent variables - GDP per capita, total public investment, total private investment, total-gross domestic investment and current public expenditures, and public 9 However, this can easily be handled in the econometrics (Greene, 1995). 8 deficit - are from the World Development Indicators produced by The World Bank (1999), and are all expressed in 1995 US$ constant prices.10 Table 1: Main macroeconomic variables: levels and ranking GDP Public Investment GDI Current Flscal deficit Country Ranking Value (1) Ranking Value (2) Ranking Value (2) Raning Value (2) Ranking Value (2) Argentna 1 7065.76 20 1.49 14 19.04 17 8.29 10 -1.25 Bolivia 17 863.76 4 7.86 18 15.36 10 12.34 7 -1.53 Brazil 3 4269.94 19 2.32 11 21.38 5 16.00 1 -9.33 Chile 7 3444.01 7 5.02 6 23.87 14 10.44 16 1.19 Colombia 11 2182.56 5 7.82 10 21.61 9 12.56 8 -1.36 Costa Rica 9 2471.03 8 4.95 3 26.58 4 16.49 6 -2.25 Ecuador 14 1518.33 9 4.64 13 20.54 13 10.45 14 0.48 El Salvador 15 1488.07 13 3.65 17 15.56 12 10.79 - - Guatemala 16 1398.81 17 2.67 19 14.29 21 6.38 _ Guyana 19 671.65 1 15.91 1 29.64 2 19.00 - Haiti 21 440.21 10 4.59 21 10.45 20 7.84 4 -2.63 Honduras 18 697.03 6 7.61 4 25.43 11 11.81 - - Jamaica 13 1567.49 - - 2 29.23 7 14.85 - - Mexico 5 4102.54 16 3.23 7 22.70 15 9.51 3 -3.47 Nicaragua 20 476.69 2 12.40 5 23.97 1 24.01 2 -9.09 Panama 8 2847.40 15 3.54 9 21.94 3 17.66 13 0.44 Paraguay 12 1793.86 11 4.49 8 22.26 19 7.86 15 0.81 Peru 10 2391.83 14 3.62 12 21.31 18 8.23 5 -2.33 Trnidad and 4 4247.32 18 2.51 16 17.72 6 15.22 12 0.22 ,T obago__ _ _ __ _ _ _ __ _ _ __ _ _ __ _ _ _ __ _ _ Uruguay 2 4989.11 12 3.76 20 12.97 8 13.43 11 -1.08 Venezuela 6 3510.81 3 9.60 15 18.67 16 8.63 9 -1.29 Notes: (1) in US$ per capita. (2) In percentage of GDP. Source: World Development Indicators (1999) and own elaboraton. 10 Although initially tried, we ended up rejecting the models on the effects on the public deficit because we did not feel the variable to be reliable enough. For the interested reader, PPI in utilities, tends to be associated with an immediate increase the deficit while PPI in transport is associated with a delayed increase in the deficit. 9 Table 1 summarizes the ranking of the countries covered by the sample, for the sample time average and for each one of the macroeconomic variables. At first glance, the table shows the lack of consistency of countries in ranking, suggesting that there are enough differences in behavior across variables to justify a separate analysis of each macroeconomic variable individually. The table also shows the mains sources of imbalances in our data panel. The fiscal deficit is the least complete variable since values for El Salvador, Guatemala, Guyana, Honduras and Jamaica are missing. The data quality issues already referred to start here. Indeed, there may be a measurement problem in the definition of several of these macroeconomic variables in relation to the concerns addressed here. According to the World Development Indicators database, public deficit, public expenditure and public investment all refer to the central government alone. However, much of the infrastructure-related activities are usually developed by public enterprises that may finance themselves outside the central government's budget. Even if most Latin American public enterprises get financed through government transfers and decline in infrastructures financial needs but be matched by a corresponding decline in the government budget, our data could produce some imprecise results. For example, they fail to capture much of the impact on deficits and public investment of utilities privatization if the current and/or capital expenditures on utilities prior to privatization used to be made by public firms rather than the government. Furthermore, this effect may be different for different types of infrastructures (e.g., telecoms and power usually belonged to public enterprises, while roads and ports were typically under the central government). Fortunately, there is not a similar problem in the case of activities financed by regional government in Latin America. With the major exception of Brazil (in the case of roads, for example), the most relevant privatization transactions in the region always involved the central govermnent. The second matter of concern with the variables is the specific definition of "privatization". We rely on a set of "infrastructure privatization dummies" (d in the econometric model), constructed from The World Bank PPI database on private participation in infrastructure projects as follows: * DU: takes a value of 1 starting on the first year there is a private utility project in a specific country (e.g. a private power generator or a private cellular operator). 10 * DT: takes a value of 1 starting on the first year there is a (significant) private operator of transport infrastructure in a specific country. Table 2 shows the first year in which each dummy takes the value of 1. Table 2: Starting year for private participation in utilities and transport Country Utilites Transport Country Utilites Transport Argentna 1990 1991 Honduras 1994 Bolivia 1987 1996 Jamaica 1990 - Brazil 1985 1985 Me,dco 1991 1991 Chile 1987 1995 Nicaragua 1993 - Colombia 1991 1994 Panama 1996 1994 Costa Rica 1989 - Paraguay 1992 - Ecuador 1985 1985 Peru 1985 1985 El Salvador 1995 - Trlnidad and Tobago 1991 - Guatemala 1994 1997 Uruguay 1992 1993 Guyana 1991 - Venezuela 1985 1985 Haiti 1995 _ Notes: (1) Average. Dummies DUand DTtake value 1 from the startIng year onwards. Source: The World Bank, Intemational Country Risk Guide and own elaboration The main problem with this variable is that it reflects the start of the reliance on some form of project finance scheme rather than a major effort to restructure the sector and to rely systematically on private finance and operation for most of the sector. The correlation between the variable constructed this way and a variable that would focus on major policy changes is strong but far from perfect. We decided to stick to this approach because project finance data is more closely related to actual investment levels expected to influence the levels of macroeconomic indicators, in particular for the public sector. In addition, we also have tried to distinguish between contract types associated with each project. To do so, we constructed the following variables associated with the three types of infrastructure "privatization": * DIV: is the number of divestitures or asset sales contracts in each year due to infrastructure privatizations for each of the two broad subsectors for each country. * GP: is the number of greenfield projects contracts in each year due to infrastructure privatizations for each of the two broad subsectors for each country. * CONC: is the number of concessions contracts in the database in each of the two subsectors. Each contract type variable is multiplied by the relevant dummy to ensure that the contract type only kicks into the regression after the first privatization in utilities and transport has started. This is recognized by a DT and DU suffix attached below to each contract type in the tables summarizing the results., Table 3: Average value of the institutional variables between 1985 and 1998 Country Political Stabillty Corruption Country Political Stability Corruption Argentina 9.9 3.4 Honduras 5.8 2.1 Bolivia 5.9 2.1 Jamaica 9.1 2.6 Brazil 8.9 3.6 Mexico 9.4 3.1 Chile 7.4 3.2 Nicaragua 5.3 4.7 Colombia 5.4 2.7 Panama 8.0 2.1 Costa Rica 9.3 4.9 Paraguay 9.4 1.2 Ecuador 9.8 3.1 Peru 5.1 3.0 El Salvador 4.7 2.5 Trinidad and Tobago 8.8 2.8 Guatemala 6.3 2.5 Uruguay 8.3 3.0 Guyana 7.7 1.7 Venezuela 10.4 3.0 Haiti 4.7 1.4 NtegIs (1) Political stability is measured from 1 (low) to 12 (high). Corruption goes from 1 (bad) to 6 (clean). Source: Intemational Country Risk Guide. Table 3 shows the institutional explanatory variables used as regressors in the model specifications (2.1) and (2.2) above are the following. Two institutional variables (labeled by x in the model) have been obtained from the International Country Risk Guide (ICRG). The index of political stability (D) is approximated by the inverse of the degree of violence and its impact on the ability of the government to govern. The countries are ranked on a scale of 1 to 12 with the lowest rating is allocated to the most unstable countries (e.g. countries during a civil war) and the highest rating does to the stable countries. The quality of the political system of the country (F) is 12 also approximated by a ranking on a scale of 1 to 6. A ranking of 1 is allocated to the most corrupt countries. A value of 6 is allocated when a country is perceived to be corruption-free." l 4. The results We relied on the LIMDEP v.7.0 econometric software to obtain OLS and GLS estimates of the linear specifications (2.1) and (2.2) described above. For each dependent variable (GDP per capita, GDI, public investment, public expenditure and public deficit) we first provide a table for Model 1 (where privatization dummies are separated into transportation and utilities) and then a second table for Model 2 (where contract types for transportation and utilities are separately identified). Each table is divided into two main columns that allow an explicit comparison between the pooled data case (i.e., not taking into account the presence of country-specific effects) and the panel data case. Finally, each column presents the results distinguishing whether the privatization dummies are simultaneous (zero lag) or are lagged one or two periods in order to identify delays or adjustments in the "macro-mnicro" effects. All estimated coefficients are accompanied by the standard goodness of fit statistics (t-coefficients at a 95% of confidence, adjusted R2 values and the corresponding log-likelihood ratios).'2 In general, panel data results (which specifically account for the presence of individual heterogeneity) correspond to the random-effects specification, except when the result of the Hausman test suggests that fixed- effects could be more appropriate. Finally, the ultimate comparison between pooled-data estimates and panel data estimates (i.e., whether country-specific effects are relevant or not) can be carried out through a general specification test on the heteroskedasticity properties of the panel residuals. There are different tests for this purpose in the literature. We have chosen the standard LM-test proposed by Breusch and Pagan (1979) because its calculation is simpler. The LM statistics, whose null hypothesis in this case implies that individual effects are not relevant, are showed Tests are carried out in the final rows of each table. "This modeling strategy has been used before. See Fosu (2001), for example. 12 Goodness of fit measures in GLS models are usually controversial since R2 measures losses effectiveness under heteroskedasticity. All the existing alternatives must be carefully interpreted, since they cannot be reliably used to compare models. 13 4.1. Effects of private participation in infrastructure (PPI) on GDP per capita Table 4 summarizes the estimates of Model 1 using GDP per capita as the dependent variable, both for the pooled data case and the panel data case. Since the comparison tests show that panel data estimates (using the fixed effects approach) are preferred to pooled data estimates, the results to consider are those in the final columns. Moreover, and even though the goodness of fit measures are to be taken cautiously in panel estimations, the values of the adjusted R2 are relatively high. Overall, the preferred regression suggests that the trend matters strongly and that the institutional variables are highly significant with the expected sign, even when lagged dummies are considered into the regression. Table 4. Effects of PPI on GDP per capita (Model 1) POOLED DATA CASE PANEL DATA CASE (Fixed-effects) Unlagged Lag =1 Lag =2 Unlagged Lag =1 Lag =2 Coet. tratlo Coeo. t-ratlo Coot. t-ratlo Coof. t-ratlo Coet. t-ratlo Coof. t-ratio One 158.661 0.52042 174.878 0.51919 205.508 0.54850 TIME -43.922 -1.641 -52.253 -1.74169 -58.1843 -1.71985 18.7264 3.36999 17.9955 3.01499 21.3374 3.18398 D 242.708 6.94567 252.367 6.73199 257.547 6.42126 28.3979 3.18444 25.2048 2.78471 23.7823 2.50583 F 264.676 3.20049 249.947 2.83498 236.825 2.47938 -106.00 -3.8653 -82.446 -3.00623 -67.1605 -2.31178 DU -641.87 -2.532S -29.439 -0.5787 DT 1433.55 6.90419 489.852 8.40341 OU-1 -591.80 -2.25789 3.21975 0.6350 DT-1 1453.3 6.51564 504.02 8.31114 DU-2 -530.48 -1.95242 30.2842 0.57773 DT-2 1465.98 6.04391 480.864 7.00114 Adj. R' 0.3273 0.31729 0.3025 0.98041 0.9821 0.9823 Log Lr -2546.68 -2368.4 -2190.09 -2016.3 -1860.7 -1715.89 Comparison tests Unlagged Lag =1 Lag =2 Panel LM test d.f. Prob. LM test d.t. Prob. LM test dJ. Prob. PoolVd 492.75 1 .0000 432.92 1 .0000 386.69 1 .0000 Fixod Hausman d.f. Prob. Humn d.f. Prob. Hausman d.f. Prob. vs. tesort dtJ.eProb. test d J._Prob. Random 10.99 5 .05167 18 5 .0029 21.34 5 .0006 Note: Dependent variable Is GDP per capita. t-ratios are calculated at 95% level of confidence. As for the main focus of this paper, the coefficients on the PPI dummies, DU and DT, suggest that only PPI in transport infrastructure seems to have a positive (and significant) effect on GDP per capita, both when considered unlagged and when a lag of 1 or 2 periods is included. 14 These results are somewhat surprising, but imply that the effect of PPI on growth varies across infrastructure types in Latin America. The lagged dummies do not alter the signs or size of these effects very much, suggesting that the impact of PPI in transport may be distributed over time. Table 5 summarizes the results for Model 2, in which the dummies are separated by type of PPI (divestures, DIV, greenfield projects, GP, and concessions, CONC). The estimation methodology is very consistent with Table 4, since Hausman test suggests that fixed effects are preferable and the LM test does not allow to discard the existence of country-specific effects. Table 5. Effects of PPI on GDP per capita (Model 2) POOLED DATA CASE PANEL DATA CASE (Fixed-effects) Unlagged Lag =1 Lag = 2 Unlagged Lag =1 Lag =2 Coet. t-ratlo Coet. t-ratio Coet. t-ratlo Coot. It-ratio Coet. t-ratio Coet. t-ratio One 523.698 1.8647 678.623 2.23009 841.8 2.52653 TIME -109.31 -4.8668 -118.527 -4.67099 -128.067 -4.41238 16.954 3.34287 18.8212 3.41408 23.0818 3.76959 D 225.598 7.08405 229.748 6.7498 228.998 6.27975 23.5891 2.57865 20.6748 2.19548 17.6267 1.82271 F 260.585 3.44418 235.477 2.94337 220.828 2.56155 -73.6671 -2.47859 -61.1194 .2.05414 -58.2651 -1.94686 DIVDU 128.391 4.99402 12.9868 2.22143 GPDU 99.394 2.51293 38.75 4.03708 CONCDU 209.51 1.43128 21.8197 0.77787 DIVDT -836.1 -0.5994 7Z2.373 2.6597 GPDT 476.72 0.43957 -9.805 CONCDT 165.07 1.07172 25.503 DIVDU-1 106.818 3.41141 13.1279 1.8994 GPDU-1 98.48 2.49363 36.4376 3.86373 CONCOU-1 183.028 1.21337 17.097 0.61229 DIVDT-1 -867.062 40.61826 672.495 2.54702 GPDT-1 -2125.38 -0.90207 -178.181 4.3913 CONCDT-1 404.275 1.50967 29.2965 0.54961 DIVDU-2 118.364 3.95182 18.0522 2.77911 GPDU-2 102.304 2.54982 31.2365 3.34042 CONCDU-2 128.024 0.80218 29.884 1.04156 DIVDT-2 -841.789 4.5937 663.142 2.55595 GPDT-2 -15Z.47 4.61293 83.771 0.18168 CONCDT-2 323.96 1.13581 -6.28027 -0.1131 Adj. R2 0.4159 0.4169 0.4111 0.9795 0.9811 0.9821 Log Lr -2523.87 -2344.90 -2166.67 -2020.16 -1866.01 -1715.32 15 Comparison tests Unlagged Lag ,1 Lag =2 Panel vs. LM test df. Prob. LU tt d.f. Prob. LM test d.f. Prob. Pooled 566.59 1 .0000 493.99 1 .0000 419.37 1 .0000 Hausman d. Pb Hausman d . Prob. Hausman d.f. Prob. Fixed vs. test test test Random 16.95 9 .0495 17.36 9 .043 14.89 9 .0939 Note: Dependent variable Is GDP per capita. t-ratios are calculated at 95% level of confidence. The institutional variables in the panel data estimations are, respectively, positive and negative for D and F, with the same interpretation as above. However, the disaggregated effects of PPI types shows several new results. First, divestitures and greenfield projects have significant and positive effects for utilities (even when lagged one and two periods). Concessions, on the other hand, do not yield significant coefficients. For transport, only divestitures seem to have a relevant impact on GDP per capita. Divestitures are sometimes viewed as the strongest form of commitment to let the private sector take care of the delivery of the services. What this suggests, at least in a first analysis, is that only the strongest commitments to a private sector role have an impact on GDP per capita. 4.2. Effects of infrastructure PPIs on private investment Table 6 shows the results from Model 1 using private investment (as directly reported by the World Development Indicators database) as the dependent variable. We added an interest rate variable, LR (the lending rate listed in the IMF statistics) to ensure a better specification of the model. Both for the pooled data case and the panel data case. As in Table 4 above, country-specific effects are very relevant, according to Breusch and Pagan's LR test, but now the Hausman tests suggests that random effects, instead of fixed ones are the preferable way to specify n. In general, it seems that this is not as good a model to explain what happens to private investment. 16 Table 6. Effects of PPI on Private Investment (Model 1) POOLED DATA CASE PANEL DATA CASE (Random effects) Unlagged Lag =1 Lag = 2 Unlagged Lag =1 Lag =2 Cool. t.ratlo Coof. t-ratlo Coof. t-ratlo Coof. t-mtlo Coof. t-ratlo Coot. t-ratlo One 11.861 8.02367 12.0409 6.98812 0.36487 2.20199 14.9078 8.81839 14.7674 8.14323 0.47517 3.80217 TiME 0.29529 2.41093 0.32733 2.28893 0.4425 2.02643 0.36130 3.68471 0.43668 3.961 0.29703 1.58399 D 0.67260 3.95373 0.54518 2.80695 0.86932 1.55013 0.45569 2.9258 0.43663 2.50088 -0.40788 -0.64676 F 0.60011 1.37261 0.82739 1.67031 -0.00188 -2.16141 -0.61882 -1.23564 -0.44978 -0.80148 0.00069 1.07466 LR -0.0007 -2.62139 -0.00126 -2.48127 40.58515 -0.50375 0.00014 0.67633 0.00022 0.58817 0.1647 0.17563 DU 0.74026 0.66619 0.69279 0.81816 DT -2.7784 -3.15398 0.48218 0.51042 OU-1 -0.70641 40.62162 4.13708 4.14876 DT-1 0.01114 0.19090 4.01072 4.2771 DU-2 0.04304 0.71746 0.01851 0.47977 DT-2 _ 12.3933 6.27925 _ 15.5332 7.94757 Ad). Fe 0.2131 0.15799 0.1311 0.6789 0.66016 0.6741 Log Lr -703.43 -647.51 -585.539 -590.903 -5 15 -484.37 Comparison tests Unlagged Lag =1 Lag =2 Panel vs. LM test d.f. Ptob. LM test d.t. Prob. LM test d.O. Prob. Pooled 337.56 1 .0000 353.10 1 .0000 344.24 1 .0000 Hausman dO. Prob. Hausman dfl. Prob. Hausman d.f. Prob. Flxed vs. test test test Random 9.86 6 0.13 9.34 6 0.1553 6.58 0.361 Note: Dependent variable is private investment. t-ratios are calculated at 95% level of confidence. The trend continues to be a significant factor as is the degree of political stability. The measure of corruption used does not perform well as it does not appear to have a statically significant effect. Most interesting from our viewpoint is the fact that the PPI dummies (except for DT when lagged two periods) are never significant. Table 7 tells us a very similar story. Again, panel data (with random effects) are preferable to pooled data, but the overall significance of the model is lower than for GDP per capita. As for our variable of concern, the emerging story is interesting. It suggests that greenfield projects can make a difference but do so with a negative sign, implying some crowding out of other private investment projects. The results also shows, somewhat expectedly, that concession contracts in transport have a positive lagged effect on private investment. As well known by the specialists of investment promotion programs, good transport services are crucial to attract investment. These results confirm their experience. 17 Table 7. Effects of PPI on private investment (Model 2) POOLED DATA CASE PANEL DATA CASE (Random) Unlagged Lag =1 Unlagged Lag =1 Coof. t-ratlo Cooe. t-ratio Coef. t-rntio Coof. t-ratlo One 11.8566 7.89852 12.1531 7.17573 15.0516 8.14836 15.1976 7.9948 TIME 0.28293 2.42325 0.32373 2.36962 0.48999 5.68235 0.53941 5.47164 D 0.57704 3.49916 0.52336 2.80214 0.54424 3.45748 0.55977 3.23 F 0.78758 1.76644 0.77469 1.54584 -0.960S0 -1.79418 -1.12983 -1.89907 LR -0.0007 -2.57747 -0.00123 -2.41655 0.00016 0.77739 0.00037 1.00481 DIVDU -0.0347 -0.28951 0.05864 0.48231 GPDU 0.05293 0.27260 -0.30279 -1.91108 CONCDU -0.1222 -0.19009 0.10B65 0.25257 DIVDT 2.06112 0.34429 1.75513 0.45052 GPDT 6.08447 1.2658 1.52561 0.49100 CONCDT -0.9697 -1.27965 0.12912 0.23612 DIVDU-1 -0.04318 -0.28148 -0.04177 -0.26587 GPDU-1 -0.07583 -0.36051 -0.50265 -2.8282 CONCDU-1 -025176 -0.34985 -0.35248 -0.71361 DIVOT-1 3.20066 0.51858 1.59119 0.39855 GPDT-1 0.14229 0.01056 -18.8738 -1 .99786 CONCDT-1 -0.13469 -0.0807 2.77924 2.22362 AdJ. Fe 0.1757 0.1453 0.6785 0.6787 Lag Lr -706.84 -646.96 -588.73 -535.99 Comparison tests Unlagged Lag =1 Panel vs. LM test d.f. Prob. LM test d.t. Prob. Pooled 376.60 1 .0000 364.74 1 .0000 Fixed vs. Hausman test d.f. Prob. Hausnian test d.f. Prob. Random 7.73 10 .655 11.67 10 .3079 Note: Dependent varable Is private Investment. t-rabos are calculated at 95% level of confidence. Since we had some concern on the quality of the dependent variable used, we also ran the models by redefining private investment as the difference between total investment and public investment. This analysis is carried out in Table 8, where Model 1 (and Model 2 but not reported here) has been re-estimated using this new definition of the clependent variable. The estimates - once more, panel data ones - score slightly better, particularly the institutional variables (same signs that in Tables 4 and 5) and the utilities PPI dummy, but again the overall significance of the model is not as good as we had hoped. They do suggest however that there is a lagged crowding 18 out taking place during the 1980-1990s as a result of the increased presence of private sector participation in utilities. Table 8. Effects of PPI on private investment (dermed as Gross Domestic Investment-public) (Model 1) POOLED DATA CASE PANEL DATA CASE (Random) Unlagged Lag =1 Lag = 2 Unlagged Lag =1 Lag =2 Coeo. t-rato Coof. t-ratlo Cool. t-ratlo Coot. t-ratlo Cool. tatlo Cool. t-ratlo Ono 8.52524 5.90699 12.0482 7.60141 8.56549 4.64182 11.9965 7.70941 8.18617 5.11725 12.717 7.2551 TIME 0.38858 3.33634 0.38012 4.13288 0.42191 2.74017 0.33188 3.94077 0.42688 3.25328 0.36512 3.4953 D 0.48252 2.88952 0.41659 2.91702 0.38643 1.88699 0.37075 2.77623 0.46902 2.59491 0.31643 2.0644 F 0.35229 0.83851 -1.06257 -2.25479 0.37264 0.71691 -1.07941 -2.46355 0.43905 0.96630 -1.1380 -2.104 LR -0.0006 -2.26562 6.109e-05 0.19924 0.0015 -2.01302 0.00010 0.55941 .0.0010 -2.34242 0.0004 0.8560 DU -2.3802 -2.24367 -0.56833 0.78091 DT -0.05089 -0.05994 1.0974 1.35159 DU-1 -1.13005 -1.4864 -2.7763 -2.67919 DT-1 -0.00707 -0.22991 0.03528 0.67958 DU-2 2.1426 -2.02563 -0.130 -0.167 DT-2 0.0391 0.72729 -0.002 -0.084 Adl. R2 0.1353 0.1322 0.1308 0.720 0.720 0.7365 Log Lr -624.24 -574.47 -522.14 -47.00 456.39 -407.24 Comparison tests Untagged Lag =1 Lag 2 Panel vs. LA test d.f. Prob. LM test df. Prob. Ll tost d.f. Prob. Pooled 411.83 1 .0000 372.38 1 .0000 341.30 1 .0000 Hausman df. Prob. Hausman df. Prob. Hausman d.f. Prob. Fixed vs. tooes_ __ _ _ _ _ __ _ _ _tt _ _ _ _ _ _ _ _ _ _ _ _ tost _ _ _ _ _ _ Random 8.14 6 .22 10.14 6 .1188 6.79 6.3402 Note: Dependent vanable is private Investment defined as GDI - Public Investment 4.3. Effects of infrastructure PPIs on public investment The estimates in Table 9 summarize the effects of PPI policies on public investment. The overall statistical results are similar to those of previous tables (particularly, again panel data is preferred and political stability is the strongest institutional explanatory variable). 19 Table 9. Effects of PPI on public investment (Model 1) POOLED DATA CASE PANEL DATA CASE (Random effects) Unlagged Lag .1 Lag a 2 Unlagged Lag -1 Lag =2 Coot. t-ratlo Coot. t-ratlo Cooe. t-rtio Coof. t-atlo Coof. -ratio Coot. t-rtlo One 3.66962 3.94788 3.9936 3.75475 4.06308 3.38307 4.44897 4.20102 4.83884 4.1948 5.20166 4.32134 TIME -0.07364 -0.98179 4.0624 -0.7266 40.0049 -0.05000 -0.05740 -1.38683 4.0541 -1.24162 -0.0266 -0.60620 D 0.02823 0.26252 -0.01665 -0.13596 -0.11209 -0.8315 0.14989 2.27065 0.10937 1.61114 0.06459 1.01417 F 0.44421 1.64164 0.44173 1.44382 0.61026 1.77114 0.01922 0.08736 0.02644 0.11457 0.11744 0.49584 DU 2.28888 3.35006 0.97727 2.75014 DT -1.50248 -2.74791 4.76619 -1.85432 DU-1 2.23534 3.14047 0.85183 2.41344 DT-1 -1.28836 -2.14511 4.65454 -1.57607 DU-2 1.73967 2.40797 0.31974 0.98976 DT-2 4.95966 -1.45039 4.55866 -1.33359 Adj. R2 0079 0.0603 0.0393 0.8314 0.8560 0.8888 Log Lr -533.17 486.19 -435.03 -347.08 -300.381 -244.885 Comparison tests Unlagged Lag =1 Lag =2 LU test d.f. Prob. LU test d.L Prob. UkeliRtod d.t. Prob. Panel vs _ _ _ _ _ _ _ _ __ _ _ _ _ _ _ _ _ Ratio T oot _ _ _ _ _ _ _ _ _ _ Pooled 479.3 1 .0000 395.18 1 .0000 341.15 1 .0000 Hausman d.f. Pro Hausman d.f. Prob. Hauaman d.t. Prob. Fixed vs. test __ _ _ _ _ _ _ _ _ _ _ test _ _ _ _ __ _ _ _ _ __ teat _ _ _ _ _ _ Random 3.33 6 .7664 3.20 6 .7838 4.15 6 .6563 Note: Dependent variable Is public Investment t-ratios are calculated at 95% level of confidence. The coefficients of the policy variables reveal several notable differences. First, the unlagged PPI dummies are significant and have the strongest statistical significance but the impact of PPI is still strong with a one year lag as well. Second, and much more interestingly, the PPI in utilities and transport infrastructures has a different sign (positive and negative, respectively). PPI in utilities complement or crowd-in public investments, while PPI in transport substitutes for or crowds -out public investment. What this may reflect is the fact that reforms in the utilities sector are used by governments to raise matching resources from private operators for the sector where as for transports, private investments allow governments to reduce its commitments to the sector in terms of expansion at least. These results hold however at the very aggregate level because we are not able to draw similar, albeit more subtle, conclusions from a dis-aggregation of contract types. 20 Table 10, where Model 2 estimates are presented, suggests that disaggregating the PPI dummies by contract type (DIV, CONC, GP) does not only reduces the overall significance of the panel data model, but also eliminates the validity of individual coefficients in all cases. Table 10. Effects of PPI on public investment (Model 2) POOLED DATA CASE PANEL DATA CASE (Random) Unlagged Lag =1 Unlagged Lag =1 Coot. t-ratlo Coot. t-ratio Coot. t-ratio Coot. t-ratio One 3.00518 3.17881 3.05074 2.83953 4.38753 3.86902 4.81907 3.85598 TIME 0.070658 0.972953 0.086508 1.02017 -0.00185 -0.04831 0.004202 0.102901 D 0.084315 0.833599 0.049902 0.435871 0.177649 2.61672 0.159847 2.28561 F 0.579541 2.12111 0.579349 1.88307 -0.00192 -0.00797 -0.08577 -0.33593 DIVDU -0.10843 -1.40861 -0.00826 -0.15074 GPDU -0.03527 -0.29531 -0.05281 -0.76188 CONCDU -0.30248 -0.78691 0.008682 0.046816 DIVOT -0.41943 -0.11882 -0.34932 -0.21648 GPDT -1.08498 -0.38211 -0.1971 -0.15255 CONCDT -0.40700 -0.90236 -0.14526 -0.62068 DIVDU-1 -0.08516 -0.91265 -0.03976 -0.54395 GPDU-1 0.011500 0.091964 -0.07574 -1.06111 CONCDU-1 -0.15569 -0.30543 -0.01017 -0.04515 DIVDT-1 -0.50780 -0.14025 -0.24170 -0.15809 GPDT-1 6.09788 0.769687 -1.61612 -.44201 CONCDT-1 -1.18086 -1.19824 0.091974 0.186214 AdjR2 0.0525 0.03386 0.5199 0.5475 Leg Lr -534.039 -486.68 -352.592 -303.393 Comparison tests Unlagged lag=1 Panel vs. LM test d. . Prob. LM test d. . Prob. Pooled 490.14 1 .0000 413.20 1 .0000 Fixed vs. test d.f Prob. test d.. Prob. Random 5.12 10 .8829 4.34 1 0 .9306 Note: Dependent variable is public investment. t-ratios are calculated at 95% level of confidence. 4.4. Effects of infrastructure PPIs on recurrent public expenditures The effects of PPI on recurrent public expenditures summarized in Tables 11 and 12 follow a particularly interesting pattern, in particular when contrasted with the pattern seen for the effect 21 of PPI on public investment. From an overall statistical viewpoint, the unlagged panel case with fixed effects provides the best results according to the values of the comparison tests and as usual by now, political stability matters. The time trend has been eliminated as there was a multicollinearity problem with the institutional variable. Table 11. Effects of PPI on public expenditures (Model 1) POOLED DATA CASE PANEL DATA CASE (Fixed-effects) Unlagged Lag =1 Lag = 2 Unlagged Lag ml Lag =2 Coot. t-ratlo Coot. t-ratio Coet. t-ratio Coet. t-ratlo Coot. t-rato Coot. t-ratlo One 11.3039 10.7462 10.9635 9.89115 10.3882 9.28352 D -0.1788 -1.34699 -0.21252 -1.53208 -0.21873 -1.58207 -0.53505 -4.72158 -0.60036 -5.10504 -0.55101 -4.79349 F 1.76807 5.82027 1.69406 5.34853 1.67279 5.18957 0.37483 1.02608 0.46611 1.23573 0.55759 1.48251 DU -2.3251 -2.88953 -1.90985 -3.45204 DT -0.4033 -0.49698 1.60745 1.91227 DU-1 -1 .5276 -1.84674 -1.03453 -1.78406 OT-1 -0.33347 -0.39110 1.7084 1.86185 DU-2 -1.04329 -1.27273 -0.60165 -1.03656 DT-2 -0.11565 -0.1333 2.22864 2.32731 AdJ.R2 0.128 0.1040 0.0933 0.676 0.6648 0.6687 Log Lr -86.74 -821.45 -746.58 -731.53 -677.59 -610.06 Comparison tests Unlagged Lag =1 Lag =2 LM Test d.f. Prob. LM Test d.f. Prob. LM Test d.t. Prob. Panel va. Pooled 492.75 1 .0000 432.92 1 .0000 386.69 1 .0000 Hausman d.t. Prob. Hausman d.. Prob. usman d.t. Prob. Fixed vs. test __ _ _ _ _test _ _ _ _ _ _ _ _ _ __ test Random 22.12 5 .0004 1 9 5 .0019 16.48 5 .0055 Note: Dependent variable is public expenditure. t-rabios are calculated at 95% level of confidence. The coefficients of the policy variables reveal several new elements. First, the PPI dummies for utilities has a declining impact over time (as seen in the declining t ratios for the lagged variables) while it is increasing for the transport dummies. are significant and have the strongest statistical significance but the impact of PPI is still strong with a one year lag as well. Second, the PPI in utilities and transport infrastructures has a different sign (negative and positive, respectively). PPI in utilities reduces recurrent public expenditures while PPI in transport seem to increase these recurrent expenditures. For transport, this reflects the common wisdom among practitioners that investments in the sector are only viable when the operation of the services allowed by the investment are subsidized. For utilities, it may reflect the fact that PPI 22 often lead to significant cost reductions and that subsidy levels tend to decline once private operators take over operations. Table 12 however suggests that this result does not hold for divestitures in the utilities sector since it seems that when PPI takes place with type of contract, recurrent expenditures increase. Table 12. Effects of PPI on public expenditures (Model 2) POOLED DATA CASE PANEL DATA CASE Unlagged Lag =1 Lag = 2 Unlagged Lag =1 Lag =2 Coof. t-atlo Coot. t-ratlo Coof. t-ratlo Coof. t-rlo Coof. tHelo Coot. t-rato One 11.8751 11.0155 11.5392 10.3069 10.8341 9.75749 13.4454 7.97494 D -0.3618 -2.879 -0.33320 -2.56723 -0.29091 -2.27076 -0.69229 -6.24888 -0.70928 -6.18977 -0.55700 -5.19081 F 1.59759 5.24564 1.58883 5.06543 1.64259 5.21548 0.31248 0.84395 0.56458 1.48095 0.62041 1.75473 DIVDU 0.10125 0.95671 0.16150 2.10343 GPDU 0.02751 0.18406 0.04272 0.36408 CONCDU -0.0648 -0.11170 0.44645 1.24778 DIVDT -2.8262 -0.51358 -1.82768 -0.53024 GPDT 4.98035 1.16415 -1.97114 -0.7388 CONCDT 0.16325 0.2677 0.32307 0.84723 DIVDU-1 0.03432 0.27820 0.18662 2.02477 GPDU-1 0.03225 0.22135 0.10929 0.92870 CONCDU-1 -0.24189 -0.4143 0.44321 1.22012 DIVDT-1 -3.16602 -0.58592 -2.00054 -0.5856 GPDT-1 -0.31654 -0.03476 2.97191 0.50511 CONCDT-1 1.00619 0.96937 0.52354 0.945324 DIVDU-2 -0.04841 -0.43473 0.10595 1.25544 GPDU-2 0.03067 0.22158 0.15917 1.48609 CONCDU-2 -0.37117 -0.64509 0.32673 0.90769 DIVDT-2 -3.90711 -0.76645 -1.82049 -0.56524 GPDT-2 -10.1595 -1.13554 -3.7521 -0.6523 CONCDT-2 2.32176 2.25521 0.78765 1.14032 Adj. R. 0.091 0.0895 0.1064 0.672 0.6683 0.6724 Log Lr 990.77 921.57 -742.71 -731.29 -673.94 -606.39 Comparison tests Unlagged Lag =1 Lag =2 LM Test d.f Prob. LM Test | d. Prob. Lu Test | d. Prob. Panel vs. Pooled 566.59 1 .0000 493.99 1 .0000 419.37 1 .0000 FixedHausman d. Prob. Hausman d. Prob Hausma d. Prob. Fixed vs test __ _ _ _ _ _ _ _ _ test --test P___ _b_ Random 21.30 9 .0113 18.96 9 .0255 12.7 9. .1767 Note: Dependent variable is public expenditure. t-ratlos are calculated at 95% level of confidence. 23 5. Conclusions This paper provides empirical evidence on the impact that private participation in infrastructure has had on key macroeconomic variables in a sample of 21 Latin American countries during the 1985-1998 period. Specifically, we look at the effects on GDP per capita, current public expenditures, public investment and private investment, controlling for country effects and institutional factors. The most interesting initial conclusions focus on the sign of these effects of the average macro effects of these micro reforms as estimated from model 1. Table 13 summarizes the main results with respect to the statistically significant signs we have been able to identify. Table 13: Summary of signs of average macro effects of PPI PPI In utilities PPI In transport GDP/capita Not significant + Private Investment Not significant Public Investment + Recurrent Public Expenditures + The first obvious fact to emerge from Table 13 is that transport and utilities "privatization" should not be expected to have the same macroeconomic effects. Transport has a significant positive effect on per capita income, utilities has none observable. Second., PPI, at best, leaves private investment constant but in the case of utilities tends to crowd it out, which is the opposite of the effect it has on public investment. Indeed, the third result to emerge is from a public sector perspective. Utilities lead to increases in public investments but reduce recurrent expenditures. The opposite holds for transport. In other words, there is crowding-in of public investment for PPI in utilities and crowding-out for transport. Also, while private transport investments require a matching commitment to operational subsidies, the arrival of private utilities operators reduces the burden of these operational subsidies. The results generated by model 2 are in general less interesting. Indeed, the disaggregation of PPI per contract type yielded few statistically significant results. The most interesting ones are that Divestitures, the strongest form of commitment to private sector has clear positive effects on GDP per capita. The second interesting result is that concession contracts 24 and Greenfield projects in transport have significant payoffs in terms of future investments. Finally, divestitures in utilities and transport concessions tend to increase recurrent expenditures. These results, however limited, provide a first set of econometric evidence on the macro effects of micro reforms for the region in which PPI policies have been the most active. Much better data is needed to draw more specific and more robust policy conclusions. Much more ambitious econometric analysis is also needed. In particular, causality has not been tested and an optimal lag structure has not been identified because of data limitations. As the PPI experience progresses and more and better data becomes available, we should be able to refine these results. But for now, these results provide already enough reasons to be concerned about a good assessment of the macro and in particular the fiscal effects of private participation in infrastructure. The fact that the effects on GDP/capita at neutral at worse and most probably positive is good news but it comes at a risk with respect to its effects on the public sector accounts. The revelation of this risk may be the main contribution of this paper since it is inconsistent with the fiscal gains expected by many policymakers as they engage in infrastructure privatization programs. References Aghion, P., E. Caroli and C. Garcia-Pefialosa (1999): "Inequality and economic growth: the perspective of the new growth theories", Journal of Economic Literature, 37, 1615-60. Alesina, A. and R. Perotti (1996): "Income distribution, political instability, and investment", European Economic Review, 40, 6, 1203-28. Amemiya, T. (1985): Advanced Econometrics, Blackwell, Oxford. Arellano, M. (1993): "On the testing of correlated effects with panel data", Journal of Econometrics, 59, 87-97. Breusch, T.S. and A.R. Pagan (1979): "A simple test for heteroskedasticity and random coefficient variation", Econometrica, 47, 1287-94. Chamberlain, G. (1984): "Panel Data", in Griliches, Z. and M. Intriligator (eds.), Handbook of Econometrics, vol. II, North-Holland, Amsterdam. De la Fuente, A. (1996): "Growth and Infrastructure: A Survey", WBI, mimeo. Estache, A., V. Foster and Q. Wodon (2002), "Accounting for Poverty in Infrastructure Reform - Lessons from Latin America's Experience", WBI Development Studies 25 Fosu, A.K. (2001): "Political instability and economic growth in developing economies: some specification empirics", Economic Letters, 70, pp. 289-294. Gillen, D. W. (1996): "Transportation infrastructure and economic development: a review of recent literature", Logistics of Transportation Review 32, pp. 39-62. Gramlich, E. M. (1994): "Infrastructure investment: a review essay", Journal of Economic Literature, 32, 3, 1176-96. Greene, W. (1995): LIMDEP v. 7.0 User's manual. Econometric Software, Inc. Guasch, J.L. (2001): "The impact on performance and renegotiation of concession design: lessons from an empirical analysis of ten years of experience", World Bank, mimeo. Hausman, J.A. (1978): "Specification tests in econometrics", Econometrica, 46, 1251-71. Kaufmann, D., Kraay A., Zoido-Lobat6n P. (2000): "Governance Matters", mimeo., The World Bank, Washington DC. Laffont, J.J. and J. Tirole (1998): A Theory of Incentives in Procurement and Regulation, The MIT Press, Cambridge, Ma. McGillivray, M and 0. Morrissey (1999): Evaluating Economic Liberalization, St Martin's Press, New York. Munnell, A. (1992): "Infrastructure investment and economic growth", Journal of Economic Perspectives, 6, 4, Fall, 189-98. Siniscalco, D., Bortolotti, B. and M. Fantini (2001): "Privatisation around the World: new evidence from panel data", FEEM Working Paper No. 77.2001. The World Bank (1999): World Development Indicators, Washington DC. The World Bank (2002): World Development Report: Building Institutions for Markets, Washington, DC 26 Policy Research Working Paper Series Contact Title Author Date for paper WPS2881 Returns to Investment in Education: George Psacharopoulos September 2002 N. Vergara A Further Update Harry Anthony Patrinos 30432 WPS2882 Politically Optimal Tariffs: Dorsati Madani September 2002 P. Flewitt An Application to Egypt Marcelo Olarreaga 32724 WPS2883 Assessing the Distributional Impact B. Essama-Nssah September 2002 0. Kootzemew of Public Policy 35075 WPS2884 Privatization and Labor Force Alberto Chong September 2002 H Sladovich Restructuring around the World Florencio Lopez-de-Silanes 37698 WPS2885 Poverty, AIDS, and Children's Martha Ainsworth September 2002 H. Sladovich Schooling: A Targeting Dilemma Deon Filmer 37698 WPS2886 Examining the Feasibility of Jerry R. Skees September 2002 E. Laguidao Livestock Insurance in Mongolia Ayurzana Enkh-Amgalan 82450 WPS2887 The Demand for Commodity Alexander Sarris September 2002 M Fernandez Insurance by Developing Country 33766 Agricultural Producers: Theory and an Application to Cocoa in Ghana WPS2888 A Poverty Analysis Macroeconomic Luiz A. Pereira da Silva September 2002 R. Yazigi Simulator (PAMS) Linking Household B. Essama-Nssah 37176 Surveys with Macro-Models Issouf Samak6 WPS2889 Environmental Performance Rating Hua Wang September 2002 Y. D'Souza and Disclosure: China's Green- Jun Bi 31449 Watch Program David Wheeler Jinnan Wang Dong Cao Genfa Lu Yuan Wang WPS2890 Sector Organization, Governance, Antonio Estache September 2002 G. Chenet-Smith and the Inefficiency of African Water Eugene Kouassi 36370 Utilities WPS2890 Sector Organization, Governance, Antonio Estache September 2002 G. Chenet-Smith and the Inefficiency of African Water Eugene Kouassi 36370 Utilities WPS2891 Trends in the Education Sector from Nga Nguyet Nguyen September 2002 E. Khine 1993-98 37471 WPS2892 Productivity or Endowments? Hiau Looi Kee September 2002 P. Flewitt Sectoral Evidence for Hong Kong's 32724 Aggregate Growth WPS2893 Banking on Foreigners: The Behavior Maria Soledad Martinez September 2002 A. Yaptenco of International Bank Lending to Peria 31823 Latin America, 1985-2000 Andrew Powell Ivanna Vladkova Hollar WPS2894 Telecommunications Sector Jean-Paul Azam September 2002 P. Sintim-Aboagye Reforms in Senegal Magueye Dia 38526 Tchetche N'Guessan Policy Research Working Paper Series Contact Title Author Date for paper WPS2895 Telecommunications Reform in Jean-Jacques Laffont September 2002 P. Sintim-Aboagye C6te d'lvoire Tchetche N'Guessan 38526 WPS2896 The Wage Labor Market and John Luke Gallup September 2002 E. Khine Inequality in Vietnam in the 1990s 37471 WPS2897 Gender Dimensions of Child Labor Emily Gustafsson-Wright October 2002 M. Correia and Street Children in Brazil Hnin Hnin Pyne 39394 WPS2898 Relative Returns to Policy Reform: Alexandre Samy de Castro October 2002 R. Yazigi Evidence from Controlled Cross- Ian Goldin 37176 Country Regressions Luiz A. Pereira da Silva WPS2899 The Political Economy of Fiscal Benn Eifert October 2002 J. Schwartz Policy and Economic Management Alan Gelb 32250 in Oil-Exporting Countries Nils Borje Tallroth WPS2900 Economic Structure, Productivity, Uwe Deichmann October 2002 Y. D'Souza and Infrastructure Quality in Marianne Fay 31449 Southern Mexico Jun Koo Somik V. Lall WPS2901 Decentralized Creditor-Led Marinela E. Dado October 2002 R. Vo Corporate Restructuring: Cross- Daniela Klingebiel 33722 Country Experience WPS2902 Aid, Policy, and Growth in Paul Collier October 2002 A. Kitson-Walters Post-Conflict Societies Anke Hoeffler 33712 WPS2903 Financial Globalization: Unequal Augusto de la Torre October 2002 P. Soto Blessings Eduardo Levy Yeyati 37892 Sergio L. Schmukler WPS2904 Law and Finance: Why Does Legal Thorsten Beck October 2002 K. Labrie Origin Matter? Ashl DemirgO,c-Kunt 31001 Ross Levine WPS2905 Financing Patterns Around the World: Thorsten Beck October 2002 K. Labrie The Role of Institutions Asl1 DemirgOc9-Kunt 31001 Vojislav Maksimovic