CUPS 9asq3 POLICY RESEARCH WORKING PAPER 28 9 8 Relative Returns to Policy Reform Evidence from Controlled Cross-Country Regressions Alexandre Samy de Castro Ian Goldin Luiz A. Pereira da Silva The World Bank Development Economics Office of the Senior Vice President and Chief Economist October 2002 | POLICY RESEARCH WORKING PAPER 2898 Abstract De Castro, Goldin and Pereira da Silva aim at intervals of time). Once these controls are set, they contributing to understand the dispersion of returns explain the difference between average GDP growth from policy reforms using cross-country regressions. The rates during the country-specific post and the pre-reform authors compare the "before reform" with "after periods, relative to the average GDP growth of the reform" GDP growth outcome of countries that relevant control group. undertook import-liberalization and fiscal policy The explanatory variables in the regressions include reforms. They survey a large sample (about 54) of the standard growth-regression controls. The results are developing countries over the period 1980-99. The the following: benefits of openness to trade and fiscal prudence have * With a better measurement and timing of the policy been extensively identified in the growth literature, but reforms, the growth effect (the "returns on reform") is the evidence from simple cross-section analysis can generally smaller than in previous papers. sometimes be inconclusive and remains vulnerable to * There is evidence of contingent relationships criticism on estimation techniques, such as identification, between policy and growth, corresponding to the endogeneity, multicolinearity, and the quality of the country's size, its export profile, and its governance. data. * Within the group of policy reformers, some The authors use a different analytical framework that countries have exhibited a relatively weaker growth establishes additional controls. First, they construct a response. counterfactual control group. These are countries that- Overall, the findings suggest that more accurate under specific thresholds-did not introduce policy measurement and definition of the timing of reforms reforms under scrutiny. Second, the authors also try to does not strengthen the significance of the effects of use the most appropriate variable of policy reform, for reforms on GDP growth. In fact, the effects are weaker example, exogenous changes in import-tariffs instead of than indicated in most cross-section studies. This the endogenous sum of all trade flows. Third, the suggests that the policy implications to be derived from authors try to base the before-after reform comparison these relationships should be treated with even more on the most accurate date for the beginning of a policy caution than previously thought. reform period (instead of comparing averages over fixed This paper-a product of the Office of the Senior Vice President and Chief Economist, Development Economics-is part of a larger effort in the Bank to understand relative returns to policy reform. Copies of the paper are available free from the World Bank, 1818 H StreetNW, Washington, DC 20433. Please contact RoulaYazigi, room MC4-328, telephone 202- 473-7176, fax 202-522-1158, email address ryazigi@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted atadecastro@worldbank.org, igoldin@worldbank.org, or lpereiradasilva@worldbank.org. October 2002. (59 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 Relative Returns to Policy Reform: Evidence from Controlled Cross-Country Regressions Alexandre Samy de Castro, Ian Goldin and Luiz A. Pereira da Silva The World Bank, 1818, H street NW, Washington DC USA Adecastro@worldbank.org , Iqoldin @worldbank.orp , Lpereiradasilva@worldbank.org Key Words: Trade Reform, Import-Liberalization, Fiscal Policy, Growth, Cross-section regressions, evaluation of reforms JEL: F13, F43 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table of Contents Introduction and Overview pg. 4 Part l: The Effects of Import Liberalization on GDP Growth pg. 7 1.1. Introduction pg. 7 1.2. Methodology pg. 8 1.3. Data and Results pg. 10 1.4. Dating IL pg. 17 1.4. Conclusions of Part I pg. 20 Part Il: The Effects of Fiscal Policy Reform on Growth pg. 22 11.1. Introduction pg. 22 11.2. Methodology pg. 24 11.2. Results pg. 26 11.3. Conclusions of Part II pg..31 Part Ill: Final Remarks pg. 31 References pg. 59 Fiaures: 0. Average GDP growth, ratio of 1990-99 / 1980-89 pg. 33 I. Average Tariff Rate (Average 1990-99), in % pg. 34 II. Change in Average Tariff Rate, 1990-99 compared to 1980-89 (% points) pg. 35 Ill. Average Tariff Rates: 'Reformers' pg. 36 IV. Average Tariff Rates: "Non-Reformers' pg. 37 V. Changes in Gov. Consumption and Taxes, as % of GDP (1990-99 to 1980-89) pg. 38 page 2 Relative Retums to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT VI. Change in Government Consumption minus Change in Tax, as % of GDP (90's versus 80's) and Average Trade Taxes (% of Tax Revenues) pg. 39 VII. Change in Overall Budget Deficit, as % of GDP (90's versus 80's) pg.40 Tables: 0. Average GDP growth rates pg. 41 I. Some Studies on Openness and Economic Growth pg. 42 II. Correlation Matrices pg. 43 III. Description of the Data pg. 44 IV. Definition of Regional Trade Groups pg. 45 V. Control Group i: OLS Regression Results pg. 46 VI. Control Group Il: OLS Regression Results pg. 47 VII. Control Group Ill: OLS Regression Results pg. 48 VIII. Control Group IV: OLS Regression Results pg. 49 IX. Control Group V: OLS Regression Results pg. 50 X. Import Liberalization and Country Size. OLS Regression Results pg. 51 Xl. Table Xl: Average Tariff Reduction - "Reformers" versus "Non-Reformers" pg. 52 XII. Classification of Countries by Size (measured by GDP on 1995 US$) pg. 53 XiII. Control Groups: Criteria and Statistics pg. 54 XIV. Studies on Fiscal Policy and Economic Growth pg. 55 XV. Fiscal Policy Variables: Non-Control and Control Group pg. 56 XVI. Control Group I (Fiscal Policy): OLS Regression Results pg. 58 page 3 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Introduction and Overview Analyzing the Dispersion of Returns to Reform This paper investigates why is there such a dispersion in returns to policy reform among developing countries controlling for the standard growth-regression country-specific factors. Pulting it differently, why is it that some countries have achieved much higher post-reform GDP growth rates? When we look at the data in Table 0, the average growth rate of the 80 developing countries in our sample is slightly higher in the 1990s (3.6%) than in the 1980s (3.2%) but one of the most remarkable findings is that the STD of the difference 1990s-1980s is a high 2.8%. Except for a few early reformers, most countries implemented trade liberalization during the 1990's. Trade was not the only reform front: the agenda usually included reform of the financial domestic sector, capital account liberalization and fiscal reforms or consolidation, all affecting growth performance. This paper estimates the effects of these policy reforms on long- run growth performance. This paper uses a different methodology. Most studies make comparisons between reformers and non-reformers over time by conducting straighfforward cross-country analysis that regress GDP growth rates over sets of explanatory variables including the relevant policy one, e.g., trade openness. This paper adopts a different approach in many significant respects. First, this paper uses control groups (CGs). The control group is a set of countries whose growth performance will be used as a counterfactual. The aim is to avoid mixing in the comparison the growth performance of countries that did and did not engage in policy reforms. We aim to separate the countries that did not adopt structural policy reforms, and then examine the cross- country variation across those who did reform, in order to proceed with statistical inference on the policy parameters. Second the paper aims to provide improved proxies of "policy instruments". Third, the paper seeks to enhance the analysis' "timing" for the beginning of reforms and set the "before-after reform" comparisons accordingly. The policy reforms that we consider are reductions in import tariffs and fiscal consolidation. The paper is divided into two parts, each examining one of these reforms. We apply the same methodology in both cases, although the fiscal analysis involves additional complexity. We recognize that we have chosen only two sets of policy reforms (trade and fiscal) when in practice, structural reform should include a much wider set of reforms, as well as their pace and sequencing. However, severe data constraints at this stage limit the widening of cross- country analysis in broader directions. page 4 Relative Returns to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT Cross-country regressions of import-liberalization (henceforth IL) have been subject to severe criticism, as summarized in Rodriguez and Rodrik (2000) (RR). Besides criticizing the 'proliferation' of openness indicators, they suggest promising areas of investigation in the context of cross-country studies. We explore two of these suggestions regarding the measures of openness and the contingent relationships. First, in terms of measures of openness, we propose using Average Tariff Rates (ATR) data as a more accurate measure of policies leading to openness to imports. This type of data has the additional characteristic of allowing us to date import-liberalization events, i.e., defined as the time when a permanent reversion in the trend of Average Tariff Rates levels occurs. In a sample of 73 countries we date import-liberalization, and identify a subset - which we name 'Control Group' - that did not experience any such reduction in the sample period (1980 to 1999). That data treatment allows estimating growth regressions that explain differences in GDP growth acceleration between 'open' countries and 'closed' ones. After controlling for the differences in standard growth determinants between the two groups, we evaluate the impact of import-liberalization on growth. Regarding the second suggestion, we identify - under the framework discussed above - some of the contingent relationships, as suggested in RR, between import-liberalization policy and growth: do countries benefit differently from import-liberalization, depending on their comparative advantage, or on their size? Do regional trade and world trade play a role? We consider variables that capture some of those effects. Our results suggest that the positive effect of import liberalization on growth is greater for both countries with a comparative advantage in manufactures and higher-GDP levels, but further investigation is required before results can be conclusively validated. Overall, our results suggest that more accurate measurement (Average Tariffs Rates) and definition of the timing of reforms (date of reversal), does not strengthen the significance of the effects of import-liberalization on GDP growth. In fact, the effects are weaker than indicated in most cross-section studies. This suggests that the policy implications to be derived from these relationships should be treated with even more caution than previously has been identified. In our analysis, examining fiscal policy is a more complex task. Unlike trade policy - which we examine in terms of import-liberalization, fiscal performance consists of a long vector of components, including public investment, consumption, transfers to households, and the structure of the tax system. Theory suggests that we look at specific components of revenues and expenditures, in order to capture structural relationships between fiscal variables and productivity and thus, growth. But examining all these dimensions in a cross-country setting would be beyond the scope of our resources. Accordingly, we look at some aggregate measures of fiscal performance. That leads us to use a reduced form estimation, precluding the estimation of structural forms. Two other problems make fiscal policy analysis more problematic than that of trade: first, fiscal variables are strongly subject to endogeneity and multicolinearity; second, the data is less comparable in cross-country terms then in the case of import-tariffs. page 5 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Despite these difficulties, we adopt the same approach as In the case of the IL analysis, but data availability reduces our sample to 54 countries. The results suggest that growth is negatively related to govemment consumption; fiscal deterioration in larger countries exhibit a lower net marginal cost (i.e. a lower negative effect on GDP growth); and the worse the level of govemance, the higher is the marginal cost -in terms of GDP loss-- of a given budget deficit. The paper is divided into three parts: in the first part we will examine IL, and lay out the methodology used in the paper. In the second, we examine the retums to fiscal policy. In the last, we discuss what can be leamed from the results of the paper and identify some limitations of our approach. page 6 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT Part l: The Effects of Import Liberalization on Growth 1.1 - Introduction In this section, we estimate the effects of Import Liberalization (IL) on GDP growth in a cross-section of developing countries. We innovate on the traditional cross-section analysis in two directions: first, we adopt a proxy for IL which we believe is closer to revealing the actual degree of protection and the less inaccurate in terms of country-rankings. Second, we adopt an modeling strategy that relies on Control Groups (henceforth CG), which consist of a set of countries that, according to given criteria, are regarded as less open to imports. These control groups constitute a counterfactual for comparison of growth performance, conditional standard growth-determinants in the lIterature. Thus, variables enter the equation twice differenced: taking advantage of the data on Average Tariff Rates (henceforth ATR), we are able to date IL with reasonable accuracy for most countries. Once IL is dated, we estimate the model in first differences of average growth rates, before and after IL. These first-differenced variables (before versus after IL) are then differenced with respect to a CG, i.e., compared to the average GDP growth rates for a group of countries that did not undertake 'IL in the sample. The method is labeled differenced differences, henceforth DD '. That methodology and the procedure for dating IL will be discussed in the next section. We will be discussing the results of two different estimation exercises. In the first we will estimate the growth regressions, using DD, under a fixed IL date - set to 1990 - taken as a cut- off date for the beginning of policy reforms. In that case, the differences in growth rates correspond to 1990-99 averages subtracted from 1980-89 averages. The second exercise consists in first dating the IL on the basis of the evidence, and then computing twice-differenced variables according to each country's IL date. In this case, the CG component or counterfactual, will be different for each observation (country), depending on its respective IL episode. For example, suppose a country liberalizes in 1986. Differenced growth will be the average growth in 1986-1999 minus the average in 1980-1985. According to the methodology proposed, in both exercises we subtract those differences from the CG differences, for all variables in the equation. We consider alternative criteria to determine the CGs, as discussed below. The quality and suitability of our data is subject to obvious criticism: first, missing observations of ATR are filled-in by linear interpolation, which imply measurement errors. 1 See Ravallion et al. (2001) .on how the DD methodology may be used in the literature on evaluation of public policies. page 7 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT However, if we assume the path of ATR is linear and monotonic, or, less strongly, that deviations from that path are i.i.d., then interpolation will not introduce systematic bias. Given that we will be averaging variables over long periods, that problem becomes even less severe. In our view, this assumption does not call for using a different proxy for IL, such as import-duties or worse yet, some aggregate ex-post indicator of openness to trade (such as the sum of imports and exports to GDP). Second, the irregular number of years used to averaging might introduce heteroskedasticity (greater variance the further the distance of IL date from 1990, because we will have less observations in one of the periods). In the estimates reported below, we find no evidence of heteroskedasticity, based on standard econometric tests. Also, taking into account country-specific dates allows more meaningful coefficient estimates, and eliminates the distortion that arises from averaging from a fixed and common date for all countries. Third, the scope of ATR is limited, since it does not include non-tariff barriers (NTB) and other trade restrictions, which may be predominant in several countries. The purpose of dating IL is to verify whether the first results, based on a fixed and identical date for all countries in the sample, will be corroborated when using a more accurate measurement of IL policy. If the answer is in the affirmative, then we will have found additional evidence the effects of IL on growth. Alternatively, if improving measurement reduces the explanatory power and the significance of the relationship between IL and GDP growth, then we will have an additional reason to be skeptical about cross-country studies of the relationship between openness and economic growth. 1.2 - Methodology We present a brief description of the differenced differences methodology. Begin with a standard production function for country i: Ya = F(A, Ki, Hi, ae't"KaH, (1 ) After time differencing the logarithm of the function and adding an i.i.d. shock, we obtain: Y.t K. H log alo[ Ka1 ] + 6.log[ Hi~ t I + ; + Ej (2) Yi.tl Ki.,-, Hijt_ or, in compact notation A log y,, = aA. log ki, + ,B.A log hi, + l9i + Ei , i = ,.,N (3) If we consider the same function for some subset of the countries in the sample (i.e., the CG), we obtain: page 8 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Alog yg, = a.Alogkg, + 8.Alogh., + Og + e , g = ,...,G N (4) Subtracting the two equations above and adding other terms, Alogy,, -Alogyg, =[9j -0g]+a.[Alogk,, -Alogkg,]+fi.[Aloghj, -Aloghg,]+ Molicy+Acontro1s+(e, -eg) ,Jig Equation (5) is the basic specification of our model in the two parts of this paper. Total Factor Productivity (TFP) differential is a constant, changes in capital stock are proxied by investment rates and human capital is proxied by primary education. The policy vector includes trade and non-trade policy variables: ATR, Government Consumption and 'Excess" money growth. The control vector includes: a Governance Indicator, Terms of Trade, regional dummies, regional trade shares and PTA-specific effects. Excess growth relative to the CG can be explained by differences in accumulation of factors of production. It also can be explained by the difference in TFP changes (the thetas). If we control for productivity differentials using openness to trade (more precisely, to imports), we are implicitly assuming that shocks that drive productivity's process in "open" countries are different from those in 'closed" ones. The constant in the regression will be capturing those shocks. It is however, impossible to identify those shocks, in the sense that other variables, unrelated to openness, may be affecting productivity. Our openness variable (ATR) will be capturing only a marginal effect. It is important to stress that we estimate a reduced form equation, since we do not model specifically the channels through which opening to imports affects productivity. Indeed, one of the problems with cross-sectional estimates is that theory predicts that almost anything can happen when it comes to openness and growth, depending on the structural assumptions related to each industry [see Tybout (2000)]. In this context, cross-country heterogeneity is not enough to identify productivity effects, since it fails to incorporate industry-level characteristics. Due to those limitations, the results presented below are merely an exploratory analysis of the data, and estimated partial correlations should not be given any structural interpretation. We are estimating reduced forms. Despite our imperfect measure of IL and sample size limitation, we believe that our cross- country study is valid: it examines some of the questions raised by RR, which relate to contingent relationships between trade policy and growth: 'Do trade restrictions operate differently in low versus high income countries? In small versus large countries? In countries with a comparative advantage in primaty products versus those with a comparative advantage in manufactured products? In periods of rapid expansion of world trade versus periods of stagnant trade?'. The model assumes that part of the productivity shock is attributed to the differences in trade regime [Equation (5)]. There is however, a source of potential bias arising from the page 9 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT correlation between the 'unexplained' part of productivity shock (the residuals) and capital accumulation. Indeed, if the model is not differenced, then certainly E[llogkjej] 0 but if the model is differenced, then there's no reason to believe E[Alogk, -Alogk8]'(e, -E;) * 0. In that sense, we believe OLS will be an appropriate estimation method - unbiased and efficient. The framework above resembles the DD methodology, which is a standard approach in non-experimental evaluation studies [see for example, Ravallion (2002)]. In such observational studies, the assignment of treatment to subjects (in our case, assignment of policies to countries) is non-random. Rosenbaum (2002) provides a thorough treatment of this topic.2 1.3 - Data and Results Before we discuss our results, we briefly review previous findings in the literature, so as to put our contribution in context. There are only a few studies using ATR as a measure of openness, mainly due to absence of data. Table I summarizes some previous results, on the effects of trade liberalization on economic growth. One noteworthy result is Wacziarg (1998). He estimates the indirect effects of trade openness (defined as a weighed average of import duties as a percentage of imports (ID), NTB coverage and Sachs Warners openness indicator), through the following channels: macro- economic policy, FDI, investment, manufactured exports, government consumption and 'distortions'. The way the index is constructed does not allow a meaningful interpretation of the total effect of IL on growth (e.g., a 15 %-point reduction in ID corresponds to 5.21 increase in the index) (Exports + Imports)/GDP = -34.73*1D - 0.217*(NTB) + 11 .262*(SW) Therefore, given that the effect of the index on growth is 7%, then a 15-percentage point reduction In ID would Increase growth by 5.21 x 0.07 = 0.36 percentage point per year. If we consider that ID underestimates real protection, then 15 percentage point reduction in ID corresponds to a greater reduction in ATR. Thus, the effect is probably larger then the above. 2 In our study, we'll be assuming that the assignment of treatment (policy) is random. One possible remedy for non-random assignment is the Propensity-Score Method. It estimates using a logit model, conditional probabilities of assignment to the treatment group, eliminating overt biases. Hidden bias may still be present however. Our DD estimation has some caveats: the treatment is not a binary variable, and thus it is not constant across treatment countries (some do more of the policy than others). Also, when we date IL, it is as if we were running a given experiment for subjects at different times. page 10 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT A second result is found in Pedroso and Ferreira (2000). They select an interesting choice of instruments for the growth determinants, namely, an indicator of openness and a social capability index. According to their cross-section results, a 15 percentage-point increase in openness would cause a 0.3 % increase in GDP per capita [see Table I]. Measuring the degree of openness of an economy is tricky, particularly using a low dimensionality vector. Barriers to imports include not only tariffs, but all sorts of NTB, such as quotas, AD duties, red-tape in customs, phyto-sanitary barriers, etc.. Because we preferably want a one-dimensional indicator of trade policy, we decided to use Average Tariff Rates3, non- weighed. A possible alternative is to use Import duties as a percentage of imports, but that variable underestimates real protection, for obvious reasons. More importantly, comparisons across countries reveal some severe distortions, in terms of country rankings. The series of Average Tariffs is incomplete. We use linear interpolation to fill-in missing values. This will cause measurement errors but we believe that this variables reflects the degree of import openness more accurately than Import Duties. In the first exercise, we assume a fixed, common date for IL reform: 1990. The differences are thus calculated as average in 1990-99 minus average in 1980-89. We consider three alternative CGs: GrouD I: countries that, on average over the period 1990-99, had Average Import Tariffs above 25% - Pakistan, India, Bangladesh, Sierra Leone, Sudan, China, Egypt, Mauritania, Nigeria, Thailand, Tunisia, Mauritius, Benin, Congo Dem. Rep., Ethiopia, Tanzania; GrouD 11: same set of countries in the CG for the IL-dated regressions (see section IV below); Group Ill: countries that did not reduce, or increased ATR in 1990-99 compared to 1980-89 - Mauritania, Zimbabwe, Saudi Arabia, Sierra Leone, Libya, Congo, Tunisia, Oman, Jordan, Syria, Bahrain, Paraguay, Israel, Singapore, Senegal, Malawi and Poland. We find it inappropriate to use indices that attempt to measure the quality of macroeconomic policy, such as that proposed by Wacziarg (1998) or indices that attempt to measure openness (such as Sachs Warners' or Dollar's - see Table I). RR provide reasons not to use them. These which are particularly compelling in the case of openness measures. In the case of the quality of macro policy the resulting country rankings are not in our view sufficiently robust to provide illuminating interpretation of the results. 3 The data is extracted from a compilation found at htto://wwwl.worldbank.org/wbieo/trade/TR Data.html. The primary sources include: WTO, IDB CD ROM database and Trade Policy Review -- Country Report, Various issues, 1990-2000; UNCTAD, Handbook of Trade Control Measures of Developing Countries -- Supplement, 1987 and Directory of Import Regimes, 1994; World Bank, Trade Policy Reform in Developing Countries since 1985, WB Discussion Paper #267, 1994, The Uruguay Round: Statistics on Tariffs Concessions Given and Received, 1996 and World Development Indicators, 1998-00; OECD, Indicators of Tariff and Non-Tariff Trade Barriers, 1996; IDB, Statistics and Quantitative Analysis data, 1998. The data was recently published in the Appendix of the WB's 'Development, Trade and the WTO - A Handbook', WB (2002). page 11 Relative Retums to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT 'Policy' variables such as Black Market Premium (BMP) or Ratio of External Debt to Exports listed above suffer from endogeneity with respect to output. As RR suggest, BMP reflects macroeconomic conditions broadly speaking, and have little to do with openness to trade or trade policies. They show that BMP is significantly correlated with inflation, debt to exports ratio and to quality of institutions. Indeed, the correlation matrices presented below are consistent with these conclusions, in that BMP is highly and negatively correlated with govemance measures and positively correlated with variance of terms of trade (TOT), in the 90's. In the 80's, BMP is highly and positively correlated with both excessive money growth and with debt to exports ratio. Table 11 presents simple correlation coefficients for some of the variables defined above. It suggests the following: 1- All governance variables are highly correlated amongst themselves, both in the 80's and the 90's; 2- Variance of TOT is positively correlated with BMP (90's); 3- In the 90's, GOV4 (regulatory framework), is negatively correlated with BMP. This is consistent with the idea the excessive or inappropriate regulation can go hand-to-hand with financial disintermediation. GOV4 is also negatively correlated with Tariffs, suggesting that the countries that have higher average tariffs are also prone to having 'bad" regulation. 4- In the 90's, the debt-to-exports ratio is positively correlated with excessive money growth. 5- In the 90's, GOV5 (Rule of Law) is positively correlated with M2/GDP. 6- In the 90's, Investment rate is highly and positively correlated with M2/GDP. 7- In the 90's, BMP is negatively correlated with excessive money growth and positively correlated with debt-to-exports ratio. 8- In the 80's, Investment rate is highly and positively correlated with M2/GDP ratio and negatively with Debt-to-exports ratio In the initial specification, we would like to include the variables related to growth (initial GDP, investment, lagged growth, human capital (education)]; policy variables (tariffs, excessive money creation, government consumption or debt or deficit); governance variable (rule of law); external shocks (terms of trade and/or its standard deviation); country characteristics (regional dummies, etc). page 12 Relative Retums to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT The difference of first-differenced GDP growth, relatively to the CG, depends, besides all the variables discussed above, on the difference between idiosyncratic shocks (other countries versus CG). If we regard such shocks as productivity shocks, then one interesting hypothesis to test is whether the countries that liberalized trade were subject to technological spillovers (or other effects of trade on growth) that do not occur in the CG. If it is true that such spillovers occur only (or mostly) in the manufacturing sector, then it is possible to account for that difference by interacting the tariff changes with the share of manufactures in total exports. The impact of a given effort towards liberalization on growth will be stronger the greater the share of manufactures on exports, precisely because of productivity spillovers4. The composition of exports is related to an issue that is very important in the context of import-liberalization among developing countries, namely, market access. Although the Uruguay Round achieved a significant reduction in the tariff gap between developing and developed countries, little progress has been achieved with respect to market access, as well as with the occurrence of peak tariffs and tariff escalation5. We acknowledge that such distortions affect every country differently, depending, among other factors, on export composition. Another control variable overlooked both in the present analysis, as well in the next section on fiscal policy, is financial liberalization. Reinhart and Tokatlidis (2002) provide empirical evidence that such reforms increase real interest rates and promotes financial deepening and lower aggregate investment (but not growth), but have an ambiguous effect on savings. The model allows us to evaluate the role of regional trade. We can compare, for each country, the evolution of trade flow (exports + imports) within a trade bloc, to that with the rest of the world. If a country is promoting IL and, at the same time, its regional trade is increasing faster than its trade with the rest of the world, then we may be observing some trade diversion effect. One possible way trade diversion might harm growth is by diminishing or ceasing technological spillovers or transfers. Initially, we follow an arbitrary classification for regional trade, which does not necessarily match legally established PTA's. Table IV describes the groups for which we calculate the trade shares, based on IMF's Direction of Trade Database. Growth of intra-region trade share isn't necessarily related to the establishment of a PTA itself. Therefore, we may want to follow an alternative approach to investigate the effects of regional trade. For that purpose, we will rely on the study by Foroutan (1998), who proposes a 4 Following the World Bank's Global Economic Prospects, 1996, Wacziarg (1998) uses the share of manufactured exports in total exports as a proxy for the extent of technological transmissions. r The IMF/World Bank (2001) manuscript provides an excellent assessment of those issues. it emphasizes that industrial countries exhibit: 1- high tariffs, tight tariff quotas, tariff peak and escalation in agriculture; 2- high protection (both tariffs and quotas) in labor-intensive sectors such as textiles and clothing. They also point that booming antidumping activity and GSP do not help increasing market access for poorer countries. page 13 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT criteria to discern non-effective regional trade blocs from effective ones. The criteria is as follows: the PTA is considered effective if the data reveals an increase in the share and or trade intensity of intra-group trade in the years following the formation of the PTA. Whether the observed trend - after the establishment of the PTA - is attributed to 'artificial statistical artifacts' such as for example, an increase/decline in world trade shares or to the PTA itself is arbitrary and subjective, as discussed in the paper. A dummy variable for every effective PTA, as well as one for all non- effective PTA, will enter the equations multiplied by twice-differenced average ATR. The study suggests the following PTAs are effective: - CACM: Costa Rica, El Salvador, Guatemala and Nicaragua; - Andean Pact: Bolivia, Chile, Colombia, Ecuador, Peru and Venezuela; - MERCOSUR: Argentina, Brasil, Paraguay and Uruguay; - CEAO/UEMOA (Communaute Economique de I'Afrique Occidentale/ Union Economique et Monetaire de l'Afrique Occidentale): Benin (joined in 1984), Burkina Faso, Cote d'lvoire, Mali, Mauritania, Niger and Senegal. - SACU (Southem African Customs Union): South Africa, Botswana, Lesotho, Namibia and Swaziland; - Zimbabwe, Kenya and Cameroon - NAFTA: Canada, Mexico and US; - Israel/US FTA The non-effective PTAs, among others, are: - ASEAN: Indonesia, Malaysia, Singapore, Thailand, Philippines and Brunei (from 84). Vietnam joined in 95. - GCC (Gulf Cooperation Council): Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and United Arab Emirates. - EU-Mediterranean Initiative: EU countries plus Algeria, Cyprus, Egypt, Israel, Jordan, Lebanon, Mafta, Morocco, Palestinian Autonomous Territories, Syria, Turkey and Tunisia. Another question we address is that of whether trade restrictions operate differently in low versus high income countries. To do so, we allow differenced differences of ATR to have distinct effects, depending on average GDP. We classify countries into small (three lowest deciles), medium and large (three highest deciles) average GDP in 1980-89, by splitting the sample into three (see Table XII). page 14 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT In the equations presented below, we use Current Government consumption expenditures as the fiscal policy variable, because both government deficit and debt as % of GDP do not contain enough observations. Since the sample of countries is not very large, we want to avoid restricting it even further. In Tables V, VI and VII we present the OLS regression results under the three different CGs. In order to account for the comparative advantage of the country, we include the manufactures' share in exports (or alternatively, the commodities' share in exports) as a 'weight' on the ATR reduction. We also consider dummies for PTAs, to capture possible region-specific effects. Regarding Group I, our main findings are [see Table VI: 1- Among the "growth" variables, GDP difference in the 80's has a positive effect in growth and is consistently significant across all specifications. Human capital, measured as primary education has a positive sign but is only marginally significant (around 15%) for most specifications. The coefficients on lagged investment are significant (marginally) only in the first four equations. 2- ATR, which is our variable of interest, is highly significant, but of small magnitude. The estimated coefficients, between -.05 and -.08, imply that a decrease of, say, 15 percentage points in the ATR will cause average GDP growth (per year) to rise between 0.75 and 1.2 percentage point. Relatively to a 3% p.a. baseline growth, after a decade, two countries with the same initial GDP would diverge at most by 12% due to a 15-point reduction in ATR6. 3- Whereas 'weighting' ATR reduction by agricultural products' share in export does not result in significant effects, the opposite is true if we use the share of manufactures7. Its estimates are around -.20. That means, a country with a 50%-share of manufactures in exports will increase average GDP growth by 1.5 percentage point, due to a 15-point reduction in ATR. The effect is slightly higher than that in 2 above, and will generate a 15%- gap on GDP after a decade, under similar assumptions. Countries with a very high share (close to one) will increase GDP growth by 3 %-points, leading to a 30% gap after a decade. 4- Allowing for PTA-specific effects (equations 4, 8 and 12), we find that, whereas ZKC and CEAO benefit from IL, that seems not to be the case for MERCOSUR countries. The effect for the complement-set of countries is very similar to that obtained in the equations without PTA-specific effects (equations 1, 5 and 9) - between -.05 and -.08. Due to limited degrees of freedom, we excluded all PTA-dummies that were non-significant. 6To simplify our interpretation, we are ignoring the CG component (or assuming it equal to zero) 7Caveat: the WDI data on the export shares in general do not add-up to one, with some even exceeding unity. page 15 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT 5- Macroeconomic policy and Governance variables enter significantly in most equations, and always with the expected signs: government consumption and excessive money growth both have a negative impact on growth whereas the impact of "good governance", measured by "rule of law", is positive. 6- Exogenous shocks - in the form of terms of trade shocks are marginally significant in the first four equations. Besides considering terms of trade alone, we tried to include a measure of exogenous shock as suggested by Rodrik (1998): it is calculated as the product of the variance of terms of trade and the degree of openness of the economy (exports plus imports as % of GDP). That variable was not significant and was excluded from all specifications. We define CG II as the same set of countries in CG V below, rather than selecting based on any given criteria. This is done for comparison purposes. Because of that, it contains both 'open' and 'closed' countries, i.e., countries with low and high average ATR in 1990-99. Whereas CG I is based on levels of openness, CG II is based on changes. The results - in Table VI - suggest the following: 1- as expected, the constant is non-significant in all equations. There is no evidence of a TFP growth differential, which should be expected in cases where more open economies' TFP grow faster. 2- ATR alone is significant only in the first equation. ATR weighed by share of manufactures in exports is marginally significant at near the 10% level, and the estimates are close to -5%. We were not able to capture significant PTA-specific effects in these equations 3- Looking at the equations together, it seems that under CG II, the effects of IL are significantly weaker then under CG I. This is true for all the coefficients: ATR alone, ATR weighed and allowing for PTA-specific effects. Finally, under CG Ill, we adopt a criteria as close as possible to that used in determining CG V - that is, we pick countries which did not exhibit a permanent reduction in ATR, from the 80's to the 90's. The findings are similar to the above [see Table VIl]: there is very little evidence that IL enhances growth. It is worth pointing that the present criteria might entail distortions. The reason being that since the ATR variable has been averaged out over fixed periods of time - 1990-99 and 1980-89 - the shifts in ATR (i.e., IL) are 'smoothed' out; therefore, selecting a CG based on percentage-point reduction isn't easy, since its dispersion is low (concentrated around zero): for instance, as can be seen in Figure II, whereas there's 16 countries with zero or positive change in ATR (the CG ll), there's another 16 which reduced ATR between 0 and 5 percentage- points. No clear threshold seems to exist. page 16 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT However, the coefficients on ATR weighed by agricultural exports are positive (and marginally significant at a 10% level), suggesting that primary products' exporters may in fact grow slower if they liberalize imports. Under fixed/common IL dates, overall results suggest that the criteria based on levels of openness (CG I), rather than changes (CG ll), is the most appropriate given the specification of the model: discrepant levels of openness are more likely to reflect productivity differences than discrepant changes. Perhaps the levels criteria (CG I) entails a conditional analysis, which translates into marginal effects of tariff reduction for countries that are already open (and thus, not in the CG). A couple variables were excluded from the model because they failed to be statistically significant: dummy variables for currency crisis and world imports. The former intends to capture possible effects of Balance of Payments crisis on growth performance. Our measure of BOP crisis is obtained from Reinhart (1999): it is derived from an index of market turbulence, defined as a weighed average of exchange rate changes and reserve changes8. If there is a possibility that a crisis is self-fulfilling, in the sense of Obstfeld (1994), then its occurrence isn't directly related to macroeconomic fundamentals, as reported for example, in Eichengreen, Rose and Wyplosz (1995,1996). In that case, controlling for such crisis could provide us additional explanatory power in our model. The latter is a measure of world trade. The purpose is to allow the IL effect to be influenced by the global trade, depending on the date countries liberalized imports. Here we haven't been able to answer RR's question of how trade restrictions operate depending on growth of world trade flows. 1.4 - Dating IL The next step is to consider the IL date in each country in the sample. All variables will be measured as the difference between average growth rates before and after trade liberalization, whenever it occurs. Just as in the previous section, we will consider alternative selection criteria for the CG, in order to allow comparisons with those results. Notice that the CG component will be different for every observation in the sample, since the differenced averages are time-specific, depending on the date of IL in each country. The specification of the equations will be the same 8The countries that suffered BOP crisis in the 80's are (number of crisis after the comma): Argentina, 4; Bolivia, 3; Brazil, 3; Philippines, 3; Venezuela, RB, 3; Chile, 2; Colombia, 2; Indonesia, 2; Israel, 2; Mexico, 2; Thailand, 2; Peru, 1; Turkey, 1; Uruguay, 1. Countries that suffered BOP in the 90's: Brazil, 2; Venezuela, RB, 2; Argentina, 1; Philippines, 1; Indonesia, 1; Mexico, 1; Thailand, 1; Turkey, 1; Malaysia, 1. The data is available at htto://www.o)uaf.umd.edu/facultv/-aDers/reinhartlDaoers.htm page 17 Relative Retums to Policy Refornm (Castro, Goldin and Pereira da Silva) DRAFT as before, allowing us to compare the results. Differencing is with respect to the two periods, determined by the date when the countries implemented IL. First, we use the same criteria as in CG I above, but selecting from a subset of the countries, which excludes the countries that exhibited no downward trend in ATR (see CG V below) - we name this CG IV. We impose that restriction simply because we are unable to divide the sample into two periods for those countries. CG IV is: Bangladesh, Egypt, India, Mauritius, Nigeria, Pakistan and Sudan. Next, we define CG V, as follows: instead of selecting countries where average tariff rate was greater than 25% on average over 90-99, or based on tariff reduction decade over decade, as before, we will simply include the countries that did not exhibit IL, that is, those which did not pernanently reverse the trend of ATR (see Figure II). In order words, countries that did not exhibit a clear downward trend in tariffs. We are not concerned with the level of ATR itself, but with changes. This group contains countries with a relatively low ATR throughout the sample (Israel and Cyprus, for example. See Table XI), as well as some with high ATR in the 90's. According to that criteria, the CG is: Bahrain, Congo, Dem. Rep., Cyprus, Ethiopia, Israel, Jordan, Libya, Malawi, Mauritania, Oman, Poland, Romania, Saudi Arabia, Senegal, Sierra Leone, Singapore, Syrian, Arab Republic, Tunisia and Zimbabwe. Group V is comparable to CG II, since both contain the very same countries. CG V is also comparable to CG III, for even though they aren't identical sets, they are based upon the same selection criteria. Table XIII in the appendix describes the selection criteria for the CGs and exhibits some statistics. Table Xl exhibits the dates when trade reform took place in every country in the sample, as detected through inspection of time-series of ATR. It also shows the average tariff before and after the reform. Figures II and III exhibit the path of ATR (normalized) for 'reforming' and 'non- reforming' countries. The number of missing observations varies according to the country, so that time averages are calculated with a different number of observations, even for countries that 'reformed' in the same year. Still, the patterns are clear: the reforming countries cut average tariffs by 15 % points on average, according to their reform schedule. On the other hand, the non- reformers (or CG) increased their average tariff by an average of 2 percentage-points in 1990-99 compared to 1980-89. And the dispersion of ATR among those countries is high in the final period (1990-99) - ranging from nearly zero (Singapore) to over 36% (Sierra Leone). Table VIII presents estimates of the model under CG IV. It is evident that the twice- differenced ATR coefficient is not as significant as under CG I and is smaller in magnitude. The same is true for the PTA-specific coefficients (Equations 40, 44 and 48). Regarding the export profile weights, results are somewhat contradictory: we find a strong (around -20%) and significant coefficient on weighed tariffs both for manufactures' share under CGI, and for page 18 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT agricultural share under CG IV. We should keep in mind that although the selection criteria is the same, the countries chosen are not. Table IX presents the results under CG V. That group is identical to CG II. Overall we find more significant effects of ATR under the former than under the latter. That applies not only to ATR alone but also for ATR weighed by export composition and allowing for PTA-specific effects. We may also compare the results under CG V with those under CG Ill, in which we adopt the same criteria but the outcome is a different CG. The same differences described above arise in that comparison and more sharply (because ATR is overall less significant in CG IlIl than in CG 11). When we use levels criteria [CG I and IV], there are no gains whatsoever in adopting IL- dated regressions. However, if the criteria is changes, then IL dating entails significant improvements in the significance of the relationship between IL and changes in GDP growth. The results suggest that under a more precise dating of IL, the effects of tariff reduction on growth are slightly stronger and much more significant than as revealed by the initial estimates, with fixed IL dates9(equations 13, 17 and 21 versus equations 49, 53 and 57). There are also significant improvements when we allow for PTA-specific effects (equations 16, 20 and 24 versus 52, 56 and 60). The improved significance of the share of manufactures' exports suggest that the export profile may be important in determining the potential benefits from trade liberalization. If we interpret this variable as an indicator of comparative advantage, then our result suggests that relatively capital abundant countries benefit from IL significantly more than labor-abundant ones. We cannot tell, however, whether this is due to technology spillovers in manufacturing, since we have not identified any structural parameters. Finally, when it comes to IL effects contingent on country size, the evidence is somewhat mixed. As shown in Table X, for CG I and IV [equations 61-63 and 70-72], we find that medium and low GDP countries benefit from ATR reduction, respectively. For large-GDP countries, ATR reduction is not statistically significant. However, for CG II, III and V, large-GDP countries seem to be the only group that benefits from ATR reduction. In other words, when we adopt a criteria based on levels of openness, either size effects are non-significant (Eq. 61 to 63) or only small countries benefit from IL. Conversely, when we adopt a changes criteria, it is becomes evident that large-GDP countries only are the ones who benefit from IL (Equations 64 to 69 and 73 to 75). 90f course, after controlling for extemal shocks, quality of govemance and other (non trade-related) macroeconomic policies. page 19 Relative Retrims to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT 1.5 - Conclusions for Part I The effects of IL on GDP growth were evaluated using a DD methodology. We compared results obtained from identical equation specifications, but using alternative measurements of the variables (fixed date versus dated IL). We propose alternative criteria to determine CGs under the two different measurements. The CGs contain countries that are considered not open to imports (levels criteria), or that have not exhibited a downward trend in ATR (changes criteria). The purpose of dating IL is to obtain a more sensible time-aggregation for variables in the model. GDP growth acceleration (or twice-differenced growth rate) depends on non-trade policies and exogenous shocks, which were included in the regressions. Our results seem to be highly sensitive to the CG choice. Under a levels criteria (CG I and IV), IL-dated results do not strengthen the case for advocates of openness as a growth booster, relatively to fixed-date results. Under a changes criteria however (CG 11, Ill and V), IL dating does strengthen the significance of the relationship between IL and GDP growth. The relevant issue then, seems to be that of determining which criteria is the most sensible to carry on our estimation exercises. We also find that the countries that are predominantly manufactures' exporters appear to benefit from IL more than those that are not. One conjecture is that such countries experience productivity spillovers. that increase total factor productivity and, as a byproduct enhance competitiveness, alleviating external borrowing requirements to growth. One potential problem is that the series of share of manufactures in exports may contain distortions due to the differences in classification of products across countries. Thus, it would be useful to examine the data more carefully. This result isn't unambiguous: in one of the CG (IV), we find that agricultural share- weighed ATR is highly significant, suggesting the opposite. The relationship between IL and growth, contingent on country size is sensitive to CG selection as well. Whereas under a levels criteria, small and perhaps medium-GDP countries only benefit from IL, under changes criteria it is evident that large-GDP countries only benefit from IL [Table X]. In theory, under standard assumptions, optimal tariff for large countries should be positive. But such assumptions (constant returns to scale, perfect competition, etc.) may not hold in reality. In any case, our model does not allow a formal test of the theory. Since our explained variable is difference in growth acceleration between 'reforming' countries and 'closed' ones, differences in TFP growth are supposedly captured by the constant term. In that case, the estimates of the constant represent differences in efficiency between the two groups, which may be related to the trade regime or not. The coefficient of tariffs is then just capturing a 'marginal' effect from liberalization. Under our framework, it is not possible to isolate page 20 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT the tariff effect, from the constant term, since that is, by construction, capturing differences in efficiency between open and closed economies. Although we are unable to identify such effects, we can look at what happens to the estimates of the constant when we exclude the ATR variable. If excluding it turns the constant significant, then we might suspect the constant accounts at least for a part of the differences in productivity that are related to the importing regime. On the other hand, if the exclusion of the tariffs does no affect the estimated constant, then we might suspect that the constant is capturing productivity differences that are unrelated to the importing regime. We find (not reported) that, whereas excluding ATR under levels criteria does not affect the constant, doing so under changes criteria tums the constant highly significant. Therefore, considering levels criteria, it appears that the identification problem is more severe, in the sense that we cannot isolate from other effects the IL effect on GDP growth. Levels criteria reduces the strength of tariff's marginal effect on the one hand, but increases the estimated productivity differential on the other hand. The discussion above reflects the limitations of the cross-sectional analysis: it is difficult to attribute the difference in the efficiency parameter, between 'open' and 'closed' countries, to IL alone. There could be other effects, which we may not have controlled for. In other words, we have not identified the mechanisms through which IL possibly enhances productivity. The reduced-form results allow for other interpretations too. The result leads us to believe that we cannot rely on the regression outcomes and we need to look at countries separately, either on a time series basis or using industry-level data, or both. page 21 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Part II: The Effects of Fiscal Policy Reform on Growth 11.1. Introduction We are cautious with respect to the evaluation of fiscal stance in cross-country comparisons. One reason is that the data on fiscal policy - i.e., Government Consumption Expenditures as % of GDP, Tax Revenues as % of GDP (GC and TAX, respectively, henceforth) and Budget Balance as % of GDP - is not subject to unique accounting principles across countries. In several cases, quasi-fiscal operations of the central bank and losses of public development finance institutions, as well as social security contingent liabilities, are not included in central government figures. Moreover, the fiscal data can be budgetary data, rather than consolidated, depending on the country. Overall, the fiscal data that is available should provide us with a somewhat blurred picture of the fiscal stance, not only in terms of its time path, but also in terms of cross-country comparisons. Nevertheless, we believe that our model may still provide some insights regarding the marginal benefits of public finance reforms, or rather, the marginal costs of not sustaining a sound fiscal policy over time. Before discussing the criteria to determine our control groups, we undertake an exploratory time-series analysis of the public finance variables, in order to detect any particular properties, possibly common to the whole sample of countries. In particular, there are two hypotheses we wish to test, for every country: first, whether the budget deficit is stationary, which translates into a test of the sustainability of the country's total liabilities. Second, we are interested in whether there is a causal relationship between expenditures and taxation. The results of our analysis, which will not be reported for the sake of brevity, suggest the following: 1- For the vast majority of countries, we have been unable to reject the null hypothesis of a unit root in the process of govemment budget balance (Overall Deficit); 2- mutual granger-causality tests performed on GC and TAX suggest that for most countries both variables tail to granger-cause one another. That is, shocks in expenditures do not lead, or help predict, changes in taxes and vice-versa'0. Point 1 above does not undermine the importance of unsustainability of debt. Financial liberalization, changes in the pattern of financing of public deficits or shifts in private capital inflows, for example, may have affected the overall budget deficits as put forth n the specific series we use. 10 The analysis has several shortcomings: the average sample size is not large (around 20 observations), which may induce small sample biases; it also precludes the inclusion of other desirable variables in the VAR, namely, rate of Inflation and GDP growth. The unit root test chosen (Augmented Dickey-Fuller) isn't robust to structural breaks and serial correlation. The lag length in the VAR was 3 years. page 22 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT Because GC is a more narrowly defined variable, we will discuss the returns to fiscal reform in terms of that variable, rather than in terms of deficit. In this way we hope to minimize discrepancies associated with measurement, and thus enabling less imprecise comparison across countries. Differently from the Trade Reform analysis, it is not trivial to identify a fiscal policy regime change, regardless of the "indicator" we choose: budget deficit, TAX and or GC and public investment. Unlike average tariff rates, fiscal deficits represent a much more complex time process. Indeed, we do not have deterministic and exogenous variables, such as import-tariffs. For example, a positive shock to government's expenditures, or a tax relief, can be compensated in the future by tax increases. In the long-run, if we think of a sustainable path for the debt, we should anticipate that revenues and expenditures are co-integrated, with a unit-elasticity. In other words, the two variables should not drift apart, and should, in the long run, grow at the same rate. The time-series analysis described above attempts to address some of those issues. Assuming debt sustainability, we cannot "date' fiscal reform, proxied by budget deficit, since that variable is covariance stationary. All innovations to deficit are temporary, and the probability that it will reverse to the mean in a finite period of time is equal to unity. Even though our sample (1980-99) might suggest that some countries have permanently reversed the deficit trend, it isn't compelling to assume they will stay on that path indefinitely. This makes it even harder to believe we could appropriately pinpoint structural, public-sector reforms. A temporary, endogenous 'bad" fiscal shock, either on TAX or GC (or a broader measure of expenditures), which has apparently reversed the trend, could lead us to mistakenly include a country in our control group. However, since our sample size is relatively small, there is not much we can do about it. It seems then, that the scope of the analysis of the returns to fiscal policy under a cross- country framework is relatively narrow. The analysis above also suggests that it is more reasonable to look at the data aggregated over long periods (1980-89 and 1990-99) rather than its short-run dynamics, in order to look at the countries that did not reform government finances. By doing so we will be "smoothing" out temporary shocks, and thus getting a clearer picture of the trend: countries that exhibit an persistent upwards trend in the deficit are more likely to be on an unsustainable path, and, because of that, might exhibit more severe macroeconomic imbalances, which may deter growth. In the light of the aforementioned considerations, the control group will include the countries that obey the following restrictions: page 23 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT 1- Change in GC minus change in TAX greater than five percentage- points 1 2- International Trade taxes (as % of total tax revenue) above 30% in 1990- 99. The group includes: Trinidad Tobago, Cameroon, Hungary, Congo, Oman and Brazil (subject to 1) plus Ivory Coast, Guinea, Mauritius, Sierra Leone and Ghana (subject to 2). Table XV presents the data on GC and TAX emphasizing the Control Group. If a given country has undertaken fiscal reforms, then we would expect that either GC did not rise and/or TAX did not fall. The exception would be countries that rely heavily on intemational trade taxes (in those cases, fiscal reform could widen the tax base, and thus induce a change in the level of GC. Moreover, it could push the lax system closer to an optimal taxation scheme, where the excess burden is equated across all taxes). So the difference between the two should provide us with a quite strong criteria for selection: if it is sufficiently large, the country has failed to meet any of the necessary conditions for a sustained fiscal adjustment. We seek to select the countries that exhibited significant deterioration of fiscal stance within the sample12. Figure V presents changes in GC and changes in TAX. Figure VI presents the difference between the two, together with trade taxes as a percentage of tax revenues. It provides us with a clear picture of the selection criteria of the control group. Some of the possible problems related to this methodology in the particular case of fiscal policy will be discussed below, when we comment our econometric results. 11.2 Methodology Evaluating the returns to fiscal reform is tricky. If we consider Barro (1990) as the benchmark model, the prediction is that per-capita GDP growth is non-linearly related to average tax rate (measured as government expenditure as % of GDP). The curve is U-shaped: growth is increasing in the tax for low levels of tax, but decreasing for high levels. This results from the interaction of the two channels through which government affects growth: a negative effect of taxation on the after-tax marginal product of capital and a positive effect of public services on this marginal product. We attempt to control for those channels using two variables: the share of international trade taxes and some govemance measure. If a country relies heavily on trade taxes, then its 11 Equivalent to a difference in average govemment deficit greater than 5 percentage-points (1990-99 compared to 1980-89) 12 We find it inappropriate to define any control group under a levels cnitena, based on, say, GC levels over the 90's. The reason is that, as Easterly and Rebelo (1993) point out, there's a strong correlation between fiscal variables (including GC) and the initial level of income, making it hard to "disentangle" the effects of the two variables. page 24 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT unlikely that the negative taxation-effect is predominant - thus it will be in the ascending portion of the curve. If a country has 'good" govemance indicators, then it is also more likely - ceteris paribus - to be in the ascending portion of the curve. Since o5k control group excludes countries with high a share of trade taxes, we are properly controlling for that effect. Graphically, we have per capita growth rate / XS~~~~~~~~f p/ pevornanre BYa Governance No shme of Trade Txes Lava share of Traoe Taxns It is important that we control for governance, in order to find a robust relationship between GC and per-capita growth. Our specification will allow both for the negative tax effect and the positive effect of public services on productivity. We will assume that the latter will be stronger the abetter" the govemance. If GC grows too much in a country that is badly govemed, then the productivity effect will be weak, and thus probably dominated by distortionary effect (taxation). A possible way of picking up such effects is to interact the fiscal variable with a govemance measure. Gerson (1998) emphasizes the quality of govemment expenditures as important determinant of its effects on productivity and thus, on growth rates. We review now rapidly the most recent findings of the extensive literature on fiscal policy and economic growth. There is great emphasis on composition effects of both tax and expenditure policy on long-run growth. In the case of taxes, results will depend on which factor is being taxed the most. In the case of expenditures, composition will matter too: the size of the externalities of public goods (defense, infrastructure, education, etc) will obviously depend on the allocation and quality of the expenditures. But that positive effect will be partially offset by a crowding-out effect. Since this study is not. primarily concerned with fiscal policy alone, we will not be considering composition effects. We will only test for aggregate effects of fiscal policy. An interesting, but not exhaustive approach of the subject can be found in Tanzi and Zee (1997). They discuss the relationship between fiscal variables, namely, tax, expenditure and budget policy according to three different channels: allocative efficiency, macro stability and page 25 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT income distribution. As far as this study is concerned, the first channel is the most important, since our "returns" variable is growth rate. Taxes are expected to affect negatively growth, because of the deadweight loss, which is a consequence of the distortion in incentives that agents face. Deadweight loss will occur even if taxes are optimally structured, in the sense that the deadweight loss is equated across all taxes13. They report that overall, the empirical cross- country studies have not been able to find conclusive evidence on the effects of both tax and expenditure policy on economic growth. One of the chronic problems is that to a great extent both taxes and expenditure are endogenous and there is multicolinearity with other policies and macroeconomic variables. Moreover, all the literature has been subject to the problem of reverse causation between tax and expenditures on the one hand and growth on the other. Last, but not least, as discussed above, the relationship between growth and fiscal variables is non-monotonic: public expenditure can strengthen growth at low levels, but weaken at higher levels, the same being true regarding taxes. Table XIV summarizes some of the most representative studies that measure the impact of fiscal variables on economic growth. 11.3. Results The empirical results will rely on the same equation specification as the analysis in Part I (see section 1.2). The basic equation Is equation (5). The difference is that the Control Group, as discussed in the previous section, is now based on fiscal performance, not import- liberalization. The basic equation includes: factors of production, proxied by investment rates and human capital levels; policy variables, including monetary and tariff rates; and other control variables, such as govemance. The same sort of objection also applies in the present analysis: the equations are not meant to identify structural parameters that describe the mechanisms through which fiscal policy affects growth rates. As discussed above, structural modeling would require a much more complete dataset, as well as more reliable and homogeneous measurement across countries. We must acknowledge that the present analysis is even more subject to problems related to endogeneity, multicolinearity and selection bias in the choice of the control group. The reason is that fiscal policy variables are to a large extent more endogenous then tariff rates: not only may they interact amongst themselves but also with other policy and macroeconomic variables, such as money growth, inflation and GDP. 13 Our selection criteria for the control groups includes countries that have a high share of trade taxes on tax revenues precisely because they are more likely to be further away from an optimally structured tax policy, i.e., most likely, the excess burden of trade taxes will be much larger than that of say, income or consumption taxes. page 26 Relative Retums to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFr Our main findings are14: I- GC has a negative and highly significant impact on per-capita growth; TAX on the other hand, is non-significant. II- after controlling for country size (measured by GDP), it seems that GC reduces growth for small countries only, conversely, TAX increases growth for large countries only (marginally significant estimate). III-differently from GC and TAX, budget deficit is significant only when interacted with GC - the effect is strong and significant. That is to say, the higher the levels of GC, the more deficits hurt per-capita growth. The equations presented in Table XVI include two country-specific dummy variables (for Bolivia and Nicaragua) and one region-specific (Middle East and North Africa). We tested the inclusion of other regional dummies - including ones for Sub-Saharan Africa, Latin America and South East Asia - but they were non-significant and thus, excluded from the equations. We also tried the inclusion of a governance variable and, it was not significant. That is a curious finding specially in face of the results on trade reform, which suggested that governance was an important factor underlying growth performance. One plausible explanation for that fact is that revenues and/or expenditures may be collinear with govemance levels. In the lines below, we discuss the results in detail, equation by equation. Equation I: TAX and GC enter the equation separately. Because GC does not include investment, we have fewer reasons to expect a positive effect on growth. The reason is that the externality of public services on output is largely due to investment, not GC. The result suggests that an increase in GC actually hurts growth performance. TAX, on the other hand, is only marginally significant, with a positive sign. One possible interpretation here is that it is directly related to some measure of good governance, as for example rule-of-law (indeed, taxes and govemance have a correlation of .45). We must also note the possibility that the correlation may be reflecting reverse causation, from growth to taxes. Equation II: we weigh taxes and expenditures by the log of ratio of GDP to Control Group's GDP (average in the 80's). The result is that only GC is significant. Note the if GDP80 equals C_GDP80 then the impact is zero. If the GDP is low enough (i.e., below control group's average) then GC will reduce output. Conversely, if GDP is sufficiently high, then its impact will be positive. The result is consistent with the idea of scale effects, as identified by Easterly and Rebelo (1993). They suggest that there may be non-convexities in the benefits of publicly provided goods and services. If for example, there are high fixed costs and low marginal costs to provide public goods, then large countries would be able to provide a given amount of such services at a lower average cost. 14 In this section, we will be referring to the variables in the equations without explicitly mentioning the transformation used. For example, when we mention simply 'per capita growth" we are actually referring to the differenced first difference of per capita growth. page 27 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Equation Ill: we continue investigating size-effects by introducing a dummy variable for country size (measured by GDP level), instead of interacting on relative GDP. The only significant coefficient is that of the expenditures for non-large countries. The result is consistent with that of the previous equation for it suggests that the inability of exploiting economies of scale in the provision of public goods renders small countries no benefit from GC. Equation IV: we consider the difference between GC and TAX, which is a 'partial' budget deficit15 (which does not include public investment). We obtain a negative and significant coefficient (-0.10), meaning that government deficits have a negative impact on per-capita GDP growth. A six percentage-point reduction In the average deficit (compared to the control group) Increases average per-capita growth by 0.6 percentage-point. We have: agrowth _ 10 adeficit Equation V: still using our 'partial' deficit measure, we allow for a different impact of deficit depending on the country size (proxied by GDP - see Table Xli for explanation). The estimated coefficients are only marginally significant, suggesting a lack of evidence in favor of size-effects arising from budget deficits. Whereas scale-effect-results, obtained in Equations II and IlIl above, possibly relate to economies of scale in the provision of public goods or services, in this case it appears that the absence of such effect is perfectly consistent with the 'instability' channel effect suggested by Tanzi and Zee (1997): persistent deficits may raise suspicion that the government recurs to monetary financing, bringing about inflation expectations and thus, macro instability. As it is widely recognized on both theoretical and empirical grounds, inflation levels and variance have deleterious effects on long-run growth rates. Assuming the instability channel is the one that matters, then there is no prior reason to believe its impact would depend on country size. Equation VI: we interact our partial, calculated deficit with an increasing monotonic transformation of the governance index, but the variable is statistically non-significant. The purpose was to check whether govemance could affect perhaps the instability channel of budget deficits. Equation VIl through IX: we include the Overall budget deficit series extracted from WDI, instead of expenditures and revenues separately'6. That variable is not significant either alone or allowing for country-size effects. This equation uncovers a result which is consistent with that of Easterly and Rebelo (1993) [see Table XIV]: they reject Ricardian Equivalence: under that 15 That is defined as the difference between General govemment final consumption expenditure (% of GDP)and Tax revenue (% of GDP) [Notation: GCxx - TAXxx ]. 16 That variable is WDI's Overall budget deficit, including grants (% of GDP) [Notation: DEFxx]. page 28 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT assumption, deficits should leave private savings unchanged and thus should have no impact on growth whatsoever. The Overall deficit is significant only when we interact it with GC (Equation Vill). That means deficits are good for countries that have low GC levels, but bad otherwise. Interestingly enough, we find, without reporting, that the deficit, when interacted with (a monotonic transformation of) governance (relative to control group), becomes significant: countries with high 'sufficiently good' governance benefit from deficit increase, the opposite being true for those with 'sufficiently bad' governance17. We have considered an alternative criteria to determine the control group, based on the change in budget deficit . That group contains the two highest deciles of the distribution of changes In overall budget deficit" (1990-99 compared to 1980-89). The results, not reported, are not much different from those obtained from the first control group: the deficit does not affect per-capita growth, except when we interact it with governance. Since the overall budget deficit is a broader concept, involving more blocs (consumption, investment, several government sectors), it is likely that cross-country comparisons will suffer from lack of homogenous standards. Moreover, the potential estimation problems aforementioned might be magnified. Because of that, we find that, at the margin, additional results will contribute little to our analysis. 11.4. Conclusions of Part II We have estimated the effect of fiscal policy variables - TAX, GC and overall budget deficit - on per-capita GDP growth across a sample of developing countries. We adopted the same approach as in the analysis of import-liberalization reform discussed in the first part of this paper. The purpose was to evaluate relative growth-performance, which we regard as one of the measure of returns to reform, vis-a-vis a group of countries that has not undertaken and/or sustained a sound fiscal policy - either in terms of fiscal revenues, expenditures or consolidated deficit. Differently from the analysis of trade reform, in the case of fiscal policy, there are greater potential problems related to the selection of control groups. Those are basically due to the endogeneity of fiscal variables. As in the case of the import-liberalization analysis, we do not propose estimating a structural model'9. 17 Again, this result is consistent with the stability channel-type of argument: if countries under good govemance should be able to run fiscal deficits without the threat of spreading uncertainty about future policies and destabilizing the macroeconomic environment. Gerson (1998) stresses the importance of the quality of public investment, which is, arguably, positively correlated with govemance. I That corresponds to a group of 11 countries: ROM, TTO, ZAR, BHR, ETH, TUR, BGD, HUN, BOL, CMR and ZAF. See Figure VIl. 19 Say, the effect of public expenditures on the marginal product of capita, or the effect of taxation on factor retums and supply. As Tanzi (1987) has long recognized, wealthier countries rely more on income and payroll taxes. Therefore, the effect of taxes are hard to be separated from initial income levels. In our analysis however, our explanatory variable isn't tax levels, but its changes, so we do not expect to face such page 29 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Despite methodological shortcomings, we have reached conclusions that are in line with the literature. Our conclusions, however, do not provide unambiguous evidence on the relationship between fiscal performance, as modeled in our analysis, and long-run growth. Our basic findings are: 1- an acceleration of Government Consumption Expenditures is harmful for growth; the same being true for the fiscal deficit (including only consumption on the revenue side); 2- country-size effects appear to be important, revealing significant economies of scale in the provision of public goods; that conforms to the literature; 3- the overall budget deficit seems not to affect growth performance; measurement problems related to that variable preclude it from constituting a robust criteria to select a control group; 4 -Governance levels seem to be an important element in determining how large a response from fiscal policy a given country will be able to obtain in terms of growth performance; "poor" fiscal results, i.e. budget deficits, coupled with good levels of governance may in fact streng. , growth performance; this could be explained both by the high quality of expenditures and a better tax system. a problem. As Gerson A1998) suggest, the literature on aggregate taxes and growth are mostly ad hoc, not relying on structural models. In that sense we follow the 'tradition'. page 30 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Part Ill: Final Remarks and Overall Conclusions The cross-country estimation results discussed in the paper suggest that there are significant correlations between GDP growth and both IL and fiscal policy reforms. The paper's DD methodology, however - as well as the benter measurement techniques in the case of IL - do not strengthen the results of the exisiting literature. In the case of IL, GDP growth effects are weaker than those of previous studies. In the case of fiscal reform, the robustness can be questioned, both because of data problems and adequacy of the proposed estimation method. Our results suggest that country-size, regional or neighborhood effects, governance and export-orientation are significant in determining the returns to reform. We also would have liked to examine the effects of institutional and governance reforms using our DD methodology, but the lack of adequate data allowed us to treat those reforms only as country-specific effects. The analysis of the two types of policy reforms were performed independently. The question that naturally arises is to what extent the pace and sequencing of the reform agenda might affect the outcomes. That has been a recurrent issue in the literature on liberalization of both trade and capital flows, but there has been "widespread disagreement regarding the optimal pace and sequencing of the structural reforms", according to a survey by Bhattacharya (1997).20 In fact one of the reasons for such dissent, as he argues, is that 'the initial political and economic conditions facing the reforming economy, and in particular the preexisting macroeconomic situation are of crucial importance". Subsuming a specific policy reform into a single variable is not an easy task. In the context of the standard cross-section approach, making two or more of such variables interact in order to elicit the cross-policy effects is not feasible. Therefore, the analysis of the dynamics and the mutual causation between different policy dimensions would have to be performed in country- specific analysis and studies. 20 There is however, a relatively high amount of consensus in two topics: the need to undertake fiscal reforms prior to liberalization of domestic financial market, so as to avoid inflationary pressures - Chile being the good example and Brazil and Argentina, the bad; and that liberalization of the capital account of the balance of payments should be the last step in the reform process, for not only does it tend to appreciate the exchange rate, but demand efficient regulation and supervision of the financial sector as well. page 31 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Table 0: Average GDP Growth Rate (%) 1980-1989 1990-1999 Difference I Ratio ____________________ (A) (a) (BXA) (A Peru 0.3 3.2 2.9 9.4 Saudi Arabia 0.4 2.7 2.2 6.0 Sudan 1.4 6.9 5.5 5.0 Trinidad and Tobago 0.6 2.7 2.1 4.7 Uruguay 0.7 3.3 2.5 4.6 Guatemala 1.0 4.1 3.1 4.2 Bahrain 1.4 5.5 4.1 3.9 Nigeria 0.9 3.1 2.1 3.3 Costa Rica 2.2 5.4 3.2 2.4 Malawi 1.7 4.2 2.5 2.4 Uganda 3.0 6.9 3.9 2.3 Syrian Arab Republic 2.8 6.3 3.5 2.2 Ethiopia 1.7 3.7 2.0 2.2 Ghana 2.0 4.3 2.3 2.1 Mauritania 2.2 3.4 1.2 1.5 Chile 4.4 6.5 2.1 1.5 Mexico 2.3 3.4 1.1 1.5 Israel 3.7 5.3 1.6 1.4 Benin 3.1 4.5 1.4 1.4 Tunisia 3.6 5.1 1.5 1.4 Philippines 2.0 2.8 0.8 1.4 Senegal 2.5 3.3 0.8 1.3 Sr Lanka 4.1 5.3 1.1 1.3 Malaysia 5.9 7.2 1.4 1.2 Jordan 4.0 4.9 0.9 1.2 Maurtius 4.3 5.3 0.9 1.2 Nepal 4.1 4.8 0.8 1.2 Bangladesh 4.3 4.9 0.6 1.1 Jamaica 1.4 1.5 0.1 1.1 Singapore 7.4 7.7 0.2 1.0 China 9.4 9.8 0.3 1.0 India 6.0 5.7 -0.3 0.9 Guinea 4.5 4.1 -0.5 0.9 Turkey * 4.3 3.8 -0.5 0.9 Belize 5.5 4.7 -0.9 0.8 Colombia 3.4 2.9 -0.5 0.8 Tanzania 3.8 3.1 -0.6 0.8 Korea, Rep. 7.6 6.2 -1.3 0.8 Ecuador 2.4 1.9 -0.5 0.8 Taiwan, China 8.1 6.4 -1.8 0.8 Indonesia 6.4 4.8 -1.5 0.8 Egypt, Arab Rep. 5.9 4.5 -1.4 0.8 Cyprus 6.1 4.4 -1.7 0.7 Thailand 7.3 5.2 -2.1 0.7 Morocco 3.9 2.7 -1.2 0.7 Iceland 3.2 2.3 -1.0 0.7 Oman 9.0 6.2 -2.8 0.7 Brazil . 3.0 1.8 -1.1 0.6 Pakistan 6.9 4.0 -2.9 0.6 Paraguay 4.0 2.3 -1.7 0.6 South-Africa 2.3 1.3 -1.0 0.6 Algeria 2.8 1.6 -1.2 0.6 IZimbabwe 5.2 2.9 -2.3 0.6 Kenya 4.2 2.1 -2.1 0.5 , Antgua and Barbuda 6.7 3.3 -3.5 0.5 Barbadbs 2.2 * 0.7 -1.5 0.3 Zambia 1.4- 0.3 -1.1 0.2 Cameroon - 4.0 0.4 -3.6 0.1 Hungary 1,5 0.0 -1.4 0.0 Bunundi,. 4.3 -1.2 -5.5 -0.3 X Guyanax -2.8 4.8 7.6 -1.7 Sierra Leone 1.9 -4.0 -5.9 -2.2 ElSalvador -1.9 4.9 6.8 -2.5 Romania 0.9 -2.4 -3.3 -2.6 Congg, Dem. Rep. 1.8 -5.1 -6.9 -2.8 I Nicaragua -0.8 2.9 3.6 -3.7 i 'Argenlina 4-0.7 4.5 5.2 -6.2 Bolivia -0.4 4.0 4.4 -9.7 Cote d'lvoire -0.2 2.8 3.0 -11.7 Venezuela, RB -0.2 2.4 2.6 -14.8 -Average, -3 ;- -. .21= - 3.1 , -, = .4 . _______________ . ..;-2.5 2.5 2.. 8 . page 32 PER - - SAU SDN rro URY GTM BHR " NGA CRI UGA SYR ETH GHA MRT CHL MEX ISR ' BEN t TUN PHL SEN JOR 0 MUS NPL I BGD S JAM 0) SGP0 CHN - IND 0) GIN TUR o BLZ COL 0 ~~~~~~~~TZA l KOR l ECU a TW0 IDN EGY 1 CYP THA c MAR ISL OMN - BRA co ZAF 1 DZA ZWE KEN ATG BRB ZMB CMR HUN BDI GUY SLE SLV ROM ZAR NIC ARG __ BOL CIV= VEN - Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Figure 1: Average Tariff Rate, % (Average 90-99) 60 . 40- 30 20 |n- ------------ ------------ ------------ ------------ 10 0 page 34 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Figure II: Change In Average ATR, 1990-99 compared to 1980-89 (% points) 20 w -20 (,o~~0 cDozz -0- -lIl -40- -50 35 Relative Returs to Policy Refornn (Castro, Goldin and Pereira da Silva) DRAFT Figure III Average Tariff Rates, "Reformers" .EiJ~~~~~~h - ~ ~ ~ ~ ~ im .. . . .... . .. . . .... ............ . . .. . ~~~~~~~~~~~IU . ..p IF.. -d -ed n ..E.IE. 36 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Figure IV: Average Tariff Rates, "Non-reformers" Jilin Jim I I 11111 III I .5 1. 45 A I - ATR_UI Ir lff - TR _ | I - T--R r l ff AT_R -MI- TRo I 37 Relative Returns to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT Figure V: Changes In Govemment Consumption and Taxes, as % of GDP (1990-99 to 1980-89) 15 I*TAX-CH UGCCH; 10- E 0 -5 0~~- 0 L a: CD -JU a -10 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~~~x- I-~~~~~-- 38 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Figure VI: Change In Govemment Consumption minus Change In Taxes and Trade Taxes as % Tax Revenue 70 - Z U GO-TAX 60 - , TAXTR90 50 co~~~~~~~a 40- -20 39 ATG ARG IHR BGD - BRB BLZ BEN BOL BRA BDI CMR CH CHN COL ZAR CRI CIV ECUP CE EC EG) ETH SL GA HUN ISL~~~~~S JOR- LBY ~ ~ ~ L US MEK NPL MAR - - ~~~NIC OMN lb PAK S PRY PRA POL PH ROM SAU C SEN c SGF ZAFC L fA SR *SDN AZ Tro~~~~H TUN UGA URY ZMB 0~~~~~~~~~~~~W Relative Returns to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT Table l: Some Studies on Openness and Economic Growth Study Effect of IL on growth Openness Measurement Methodology and sample size Trade volumes (-0.039) 7 adjusted trade Income of the bottom quintile as a function of GDP per- Dollar and Kraay volumes (-0.038), SW Index (-0.07), ID capita and openness measure. Includes regional (2001) as % of imports (-0.16), WTO dummy dummies. Sample: 137 to 223 countides. ___________________ ~~~~~~~~(0.02). Cross-section and Panel estimates. Per-capita GDP as a "a 1 %-point increase in the function of Social Capability (Adelman-Morris Index), (expqrt+imports)/GDP ratio has ~~~~and openness. Estimate using instruments - distance Pedroso e an impact in the GDP per worker (export+imports)/GDP from Equator, % of population that speaks European Ferreira (2000) arond impact6 in thsGDPser workrn);+mors/D idioms; and 'geographic' trade volume (given by a 0 005% (panel data) gravity model). Two-stage LS, 59 countries. Panel-data: 0,005% (panel data) fixed effects, 1 and 5-year averages; 1400 and 280 observations respectively. On Edwards: Problems with identification assumption WDR=-0.0126 and choice of weight for WLS. Use log(GDP) instead of Rodrik and HF=-0.02 levels. Include HF's Index of Property Rights Protection. Rodriguez (1999) Trade Duty=-1 .837 Only 3 indicators remain significant, but still, they are [review someof ~~~~~~~~~~~~~~~~~~~subject to rankidng problem, failing some consistency [review some of checks. the most cited On Doliar: robustness check indicates that only variability studies: distortion=-0.008 (ns); variability= is robust to indusion of regional dummies. Include also Dollar(1992), Ben- -0.099 initial income and schooling. Use updated Summers- David (1993), I Heston database, which include more countries (112) Sachs Wamer Item 4 of Index is equivalent toadumfoSb-hrn Afia.Ie5 Sachs Warner ~~~~~~~~dummy for Sub-Saharan Africa. Item 5 (199) and Edwards OPEN=2.12 (considering 4 and 5 is strongly related to macroeconomic On SW: the index core amounts to just the last two (1998)1 only) and political variables [inflation, components (monop.exp. and BMP) debt/exports, quality of institutions (ICRG)] SW Index, WB's classification of Trade SW= 0.0094 Strategies (WDR 1987), Leamer's WDR=0.0075 (1988) index, BMP, Average Import Tariffs=-0.045 Duties and NTB coverage (both by TFP, 1980-90, as a function of initial income, schooling Edwards (1998) BMP=-0.0217 Barro and Lee (1994)), Heritage and nine altemative indicators of openness. Estimation Foundation's Index of Intemational Method: WLS (weight=GDP per capita in 1985) HF=-0.0074 Trade, ratio of export and import duties Trade duty=-0.48 to total trade and Holger Wolf's index of import distortions for 1985. Wacziarg (1998) 0.071 Fitted (X + M)/GDP as a linear Panel data, simultaneous equations; estimation based on 41 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT combination of ID, NTB coverage and Zellner and Theil (1962) - 3-stage LS, 5-year averages SW indicator over 1970-89; 57 countries; Two indices of Trade Uberalization: TR Vary depending on time I: Papageorgiou et al. (1991) - uses Cross-section and Panel data with fixed effects; Very Harrison (1996) aggregation (annual data or 5- comnmer inaliol in 1 8 onl small country coverage for openness measures - 15 and Harriso (1996) year averages). See p.436 cmeiapocyn1908,nl 5 28 countries respectively. Evaluates levels and changes 0.02 (TR II, levels) countries; TR II: Thomas et al (1 991) - in trade policy. based on tariff and non-tariff barriers in 1979-88, only 28 countries. Dummy - A country is not OPEN if any Growth of GDP per capita, 1970-89, as a function of an d OPEN= 2.44 [open countries of the following is true: 1- ATR >40%; 2- openness indicator. Regression includes: initial GDP, Sachs and grow two and a ha % points NTB coverage>40%; 3- socialist investment rate, govemment consumption as % of GDP, Warner (1995) more than non-OPEN onest] regime; 4- state monopoly of major assassinations per capita, schooling, revolutions and mrtnn.] exports; 5- BMP>20% in the 70's or coups, deviation from world investment prices. Sample: 80's. 79 countries. DIS=-0.02; VAR= -0.07 Reducing Cross-section: GDP growth in 1976-85, as function of DIS to Asian level adds 0.7 %- RER distoroion (DIS) and variability outward orientation and Investment rates. Includes Dollar (1992) point to Latin American growth (VRER dsotn(DSadvribit' dummies for African countries. 95 countries. OLS. Does and 1.8 to African. Reducing (VAR). not include basic controls such as initial income, VAR would do 0.8 and 0.3 education and other regional dummies. 42 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT Table II: Correlation Matrices TAR80 CGD80 GC80 BMP80 SDTOT80 M2GDP80 M280 EDX80 GOV1 GOV2 GOV3 GOV4 GOV5 GOV6 INV80 CGD80 -0.24 GC80 -0.30 0.54 BMP80 -0.12 0.13 0.47 SDTOT80 0.00 -0.03 0.18 0.28 M2GDP80 -0.23 0.28 0.36 0.78 0.28 M280 -0.07 0.39 0.60 0.00 0.00 -0.12 EDX80 -0.03 0.21 0.41 0.91 0.26 0.86 -0.09 GOV1 0.03 0.22 0.17 0.02 -0.31 0.11 0.32 0.02 GOV2 -0.01 -0.03 0.20 -0.02 -0.19 -0.02 0.38 -0.06 0.72 GOV3 -0.12 0.04 0.13 -0.11 -0.40 -0.11 0.43 -0.10 0.54 0.71 GOV4 -0.11 0.00 0.07 -0.08 -0.12 -0.06 0.28 -0.13 0.65 0.69 0.59 GOV5 -0.12 -0.04 0.16 -0.15 -0.36 -0.26 0.47 -0.24 0.60 0.81 0.83 0.63 GOV6 -0.10 0.08 0.14 -0.21 -0.41 -0.19 0.37 -0.20 0.59 0.67 0.87 0.57 0.83 INV8O -0.14 0.13 0.18 -0.05 -0.16 -0.24 0.60 -0.19 0.34 0.35 0.33 0.43 0.44 0.22 TOT80 0.05 -0.09 0.13 0.14 0.86 0.16 0.03 0.09 -0.23 -0.08 -0.36 -0.07 -0.21 -0.35 -0.04 TAR90 CGD90 GC90 BMP90 SDTOT90 M2GDP1 M290 EDX9O GOVI GOV2 GOV3 GOV4 GOV5 GOV6 INV90 CGD90 0.22 GC90 -0.15 0.30 BMP90 0.19 0.01 0.13 SDTOT90 0.16 0.27 0.06 0.22 M2GDP90 -0.14 0.30 0.04 0.24 0.14 M290 0.15 -0.04 0.37 -0.13 -0.18 -0.43 EDX90 0.23 0.37 -0.08 0.30 0.43 0.18 -0.33 GOV1 -0.32 -0.24 0.05 -0.35 -0.07 -0.13 0.16 -0.41 GOV2 -0.12 -0.24 0.01 -0.35 -0.22 -0.28 0.37 -0.24 0.65 GOV3 -0.23 -0.06 0.08 -0.46 -0.14 -0.17 0.35 -0.12 0.50 0.64 GOV4 -0.55 -0.27 -0.12 -0.60 -0.25 -0.24 0.08 -0.41 0.62 0.54 0.47 GOV5 -0.16 -0.19 0.17 -0.26 -0.16 -0.37 0.45 -0.18 0.57 0.76 0.74 0.40 GOV6 -0.23 0.00 0.15 -0.28 -0.12 -0.13 0.25 -0.22 0.65 0.63 0.83 0.43 0.75 INV90 0.01 -0.42 -0.02 -0.05 -0.27 -0.42 0.63 -0.55 0.10 0.22 0.15 0.10 0.37 0.08 TOT90 -0.02 -0.22 0.27 0.09 0.14 -0.14 0.32 -0.47 0.24 0.16 -0.10 0.14 0.16 -0.04 0.38 43 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table Ill: Description of the Data Notation: suffix xx = 80. if averaee 1980-89: suffix xx = 0, if average 1980-year before IL; =90, if average 1990-99 =1, iaverage year IL-1999 Variable Description BMPxx Black Market Premium. Source DEFxx Overall budget deficit, including grants (% of GDP) EDxx Extemal Debt (% GNI) GCxx General govemment final consumption expenditure (% of GDP) GDP80-LARGE Impulse dummy = 1, if average GDP (1980-89) is greater than 35.75 (1995 US$ bn) - see Table XII = 0, otherwise GROWTHxx GDP growth (annual %) INFxx Inflation, consumer prices (annual 9%o) INVxx Gross capital formation (% of GDP) M2GDPxx M2 growth[Money and quasi money (M2) (current LCU)] minus real GDP growth (annual %) M2xx Money and quasi money (M2) as % of GDP MDUTxx Import duties (% of imports) OPENxx Imports plus Exports of goods and services (% of GDP) PCGROWTHxx GDP per capita growth (annual %Yo) PExx School enrollment, primary (% gross) SDINFxx Standard Deviation of Inflation, consumer prices (annual %) SExx School enrollment, secondary (% gross) TARxx Average Tariff, % TAXxx Tax revenue (% of GDP) TOTxx Terms of trade (goods and services, 1995 = 100) XAGR_CHxx Growth Rate of Agricultural raw materials exports (% of merchandise exports) XAGRxx Agricultural raw materials exports (% of merchandise exports) XDUTxx Export duties (% of exports) XMAN_CHxx Growth Rate of Manufactures exports (% of merchandise exports) XMANxx Manufactures exports (% of merchandise exports) Govemance Indicators GOV1 Voice and Accountability GOV2 Political Stability/Lack of Violence GOV3 Govemment Effectiveness GOV4 Regulatory Framework GOV5 Rule of Law GOV6 Control of Corruption Regional Trade (see spreadsheet 'trade.ds') RTSHxx Average Trade share, (exports + Imports within region)/(exports + imports to the world) RTSH_CHxx Average Growth rate of RTSHXX BOP and Banking Crisis (data only until April 98) BANKxx Number of Banking crises in period XX BOPxx Number of Balance of Payments crises in period XX PTA Dummy Variables CACM Costa Rica, El Salvador, Guatemala and Nicaragua CEAO Benin, Burkina Faso, Cote d'lvoire, Mali, Mauritania, Niger and Senegal LAC Latin America and the Caribbean MENA Middle East and North Africa MERCOSUR Argentina, Brasil, Paraguay and Uruguay SA Southeast Asia SOUTHA South Africa ZKC Zimbabwe, Kenya and Cameroon 44 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table IV: Definition of Regional Trade Groups MPrmncDsuI Smjibth Ampraca Fnat AMin, innimrfingp .lapa :uih aharan Af ica ARG ARG CHN BDI BRA BOL IDN BEN PRY BRA KOR CIV URY CHL MYS CMR COL PHL ETH !:nr th ASia ECU SGP GHA BGD PER THA GIN IND PRY TWN KEN NPL URY MRT PAK VEN MUS LKA MWI haribbean Mideldi Fant A N AfricaEa-ste.rn .urnpe NGA ATG BHR HUN SDN BLZ DZA POL SEN BRB EGY ROM SLE CRI ISR TZA GTM JOR UGA GUY LBY ZAF JAM MAR ZAR MEX OMN ZMB NIC SAU ZWE SLV SYR TTO TUN VEN Notes 1. Argentna, BrasD, Paraguay and Uruguay's shares were those to mercosur, even though they're included in other South American countries 2. Venzuela's share is to South America 3. Cyprus, Iceland and Turkey' shares are relative to European Union 45 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table V: Control Grouo r - OLS Regression Results. Dependent Variable: [GROWTH90-GROWTH8O]-[A-C-GROWTH90-A_C_..GROWTH8O]- Eq.1 Eq.2 Eq.3 Eq.4 Eq.5 Eq.6 Eq.7 Eq.8 Eq.9 Eq.10 Eq.11 Eq.12 C 0.412 0.377 0.380 0.567 0.355 0.302 0.389 0.528 0.246 0.160 0.280 0.453 0.112 0.192 0.133 0.036 0.172 0.298 0.119 0.041 0.326 0.520 0.234 0.081 GROWhtH8-&A_C_GROWTH8O -0.418 -0.412 -0.425 -0.398 -0.451 -0.441 -0.445 -0.428 -0.435 -0.444 -0.433 -0.392 0.001 0.008 0.001 0.002 0.003 0.012 0.002 0.004 0.002 0.006 0.001 0.003 NVSO-A_CCJNV80 0.062 0.019 0.058 0.064 -0.029 -0.034 -0.030 -0.022 -0.007 -0.016 -0.006 -0.002 0.181 0.726 0.161 0.139 0.549 0.514 0.548 0.659 0.857 0.701 0.877 0.957 -PoPE80A_C_PE90+A_C_PE8O 0.050 0.038 0.035 0.057 0.025 0.021 0.014 0.033 0.011 0.006 0.005 0.021 0.010 0.037 0.040 0.003 0.156 0.207 0.375 0.044 0.534 0.674 0.733 0.239 TARS0 TARWA_CTAR9O+A.CffAR9O -0.079 -0.066 -0.050 0.014 ao2s 0Q046 XAGRtV(rAR90TAROA-CC-TAR90+A.-C-TARyOOM7t *0.132 -0.079 -0.095 0.168 a445 0.373 XMAN(TAR90 TARBSO-&C.-TAR90+&.C.TARIO)lW01 -0.232 -0.205 -0.176 0.015 aoo4 0.009 (ZKC+CEAOr(TAR90-TARtO-C_-TAR9t+.C_TARtO) -0.137 -0.142 -0.107 0.166 0.216 0.135 MERCOSUR-TAR90-TAeA&.C3TAR9+A_CJTARtSO) 0.126 0.223 0.183 0.016 0.001 0.002 (1-CACM-ZKC-CEAO-MERCOSURr -0.083 -0.079 -0.055 (rAR90-TARSOA-C-TAR90+A-C-TARSO) 0.011 ao10 ao32 GC9GGCSO-A_C_GC90+A_C_GC80 0.239 -0.168 -0.226 -0.247 -0.164 -0.121 -0.171 -0.177 -0.162 -0.118 -0.173 -0.172 0.003 0.031 0.004 .02 0.042 a0116 0.034 a033 0a29 0.075 a0o20 0.026 M2GDP90-M2GODPO-A-C.-M2GDP9+A-C-aM2GOPO -0.357 -0.343 -0.348 -0.359 -0.312 -0.318 -0.307 -0.315 -0.349 -0.349 -0.343 -0.363 0.000 aoo4 0.001 0.000 a016 0.019 0.021 0.016 0.002 aoo5 a02 0.001 GOVS-A-C-GOVs 0.804 0.598 0.687 0.705 0.932 0.678 0.898 0.863 0.839 0.683 0.829 0.685 0.013 0.111 0.049 0.24 a012 0.088 0.026 0.013 0.017 0.075 0.026 0a.28 TOT9oTOTS-A-CJTOT9OA.CTOTSO 0.019 0.014 0.020 0.019 -0.002 0.000 0.001 -0.002 -0.001 -0.001 0.002 - 0.001 0.101 0.251 0.071 0.160 0.833 0.964 a918 a843 0.943 ago6 a02t 0.877 MERCOSUR -2.498 -2.260 -1.567 -3.652 -3.010 -2.757 -2.898 -2.507 -2.231 0.001 0.016 0.016 0.000 .QO1 0.001 0.001 a02 0.002 SOUTHA 1.175 0.832 1.136 1.019 0.831 0.975 -1.443 1.280 1.398 0.014 a0161 0.029 0.9 0234 a 127 a01 0.044 0.019 SA 1.207 1.114 1.230 1.510 1.333 1.499 1.295 1.223 1.427 0.001 0.005 0.001 a0002 0.004 0a02 aoo7 0.014 0oo0 (RTrSH-CHsoRTSH-rCH8tA-C-RTSH-CH9O+&C.AT5H.CHOM)10 -0.043 -0.029 -0.040 -0.042 0.009 0.069 0.010 0.008 (FTSH90-RTSHt0A-C-RTSH9so0+C-HRTSO)-100 -0.055 -0.023 -0.055 -0.062 a306 0.670 0.318 0.279 R-squared 0.91 0.88 0.90 0.90 0.88 0.86 0.89 0.88 0.88 0.86 0.88 0.87 Adjusted R-squared 0.86 0.82 0.86 0.86 0.83 0.80 0.83 0.83 0.83 0.81 0.84 0.83 S.E. of regression 1.06 1.19 1.07 1.05 1.16 1.24 1.14 1.14 1.12 1.18 1.09 1.14 Sum squared resid 26.75 33.71 27.44 27.62 33.58 38.52 32.27 33.76 37.70 41.64 35.76 39.96 Log likelihood -46.50 -50.78 -46.97 -47.09 -51.57 -54.18 -50.82 -51.67 -57.33 -59.41 -56.22 -58.55 Durbin-Watson stat 2.13 2.08 2.25 2.31 1.78 1.95 1.74 1.88 1.94 2.08 1.98 1.73 Sample Size 37 37 37 37 38 38 38 38 42 42 42 42 Countries with average ATR greater than 25% in 1990-99: Pakistan, India, Bangladesh, Sierra Leone, Sudan, China, Egypt, Mauritania, Nigeria, Thailand, Tunisia, Mauritius, Benin, Congo, Ethiopia and Tanzania p-values below estimates 46 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table VI: Control GrouP II' - OLS Regression Results. Dependent Variable: [GROWTH90-GROWTHBO]-[A-C-GROWTH9-A_C_GROWTH801 Eq.13 Eq.14 Eq.15 Eq.16 Eq.17 Eq.18 Eq.19 Eq.20 Eq.21 Eq.22 Ea23 Eq.24 c 0.241 0.703 0.546 0.399 -0.230 -0.003 -0.164 -0.135 -0.196 0.000 -0.172 -0.146 0.575 0.140 0.222 0.371 0.598 0.995 0.695 0.763 0.645 0.999 0.671 0.721 GROWTH8O-A_C_GROWTH8O -0.499 -0.460 -0.506 -0.459 -0.433 -0.405 -0.452 -0.357 -0.440 -0.403 -0.456 -0.355 0.000 0.008 0.001 0.002 0.012 0.033 0.009 0.038 0.009 0.031 0.006 0.018 INVWO-A_CJNVS0 0.133 0.088 0.111 0.108 0.024 0.007 0.026 -0.002 0.030 0.007 0.033 0.001 0.018 0.187 0.059 0.077 0.672 0.921 0.656 0.970 0.562 0.921 0.522 0.977 PE9OPES A_C_PE9O+A.C PESO 0.035 0.033 0.029 0.027 -0.016 -0.012 -0.016 -0.023 -0.011 -0.011 -0.011 -0.022 0.158 0.165 0.232 0.308 0.583 0.669 0.571 0.365 0.692 0.684 0.691 0.205 TAR90-TARSODA.C_TAR9O+A_C_TARSO -0.050 -0.015 -0.010 0.015 0.340 0.558 XAGRSrNrAR90-TARMA.-cTAR9o+A.-cTAR8OY)OO 0.061 0.127 0.150 0.779 0.559 0.496 XMAN6TrAR9O-TAr5.A..C-TARM90+A.C_TASORflOo -0.058 -0.043 -0.034 I a i~~~~~~~~~.24 0.123 0.2 12 (ZKC4CEAO)(TAR9O-TAREOA_C-TAR9O+A C_TARSO) -0.002 0.056 0.067 0.971 0.369 0.277 MERCOSUR (TARso-TAR80A-CJARso+A,C-TARSO) -0.053 -0044 -0.031 0.114 0.179 0.291 (1CACM-ZKC-CEAOMERCOSUR)-(TAR90-TAR80 A_C-TAR90.+AC_TAR8b) -0.033 -0.012 -0.008 0.077 0.497 0.625 GC90-GCOA_C._GCs0+A. C_GC8O -0.125 -0.121 -0.103 -0.148 -0.075 -0.096 -0.065 -0.107 -0.072 -0.100 -0.064 -0.108 a049 0.180 0.103 0.029 0.286 0.325 0.364 0.154 0.317 0.273 0.371 o.130 M2GDP90-M2GDPBO-A.C.M2GDP9O+A.C_M2GDP0SO -0.390 -0.369 -0.375 -0.390 -0.390 -0.386 -0.385 -0.391 -0.395 -0.389 -0.393 -0.398 0.ooo 0.002 0.o0 0.000 0.007 0.007 0.007 0.006 0.005 0.005 0.004 0.002 GOV5-A_C_GOVs 0.453 0.375 0.349 0.314 0.554 0.470 0.510 0.305 0.508 0.451 0.474 0.246 0.190 0.312 0.344 0.351 0.157 0.245 0.186 0.392 0.164 0223 0.199 0.436 TOT90-T0TDA_C_TOT9A0_c_TOT8O 0.038 0.033 0.031 0.038 0.008 0.010 0.006 0.012 0.009 0.010 0.008 0.014 0.001 0.005 0.009 0.001 0.455 0.372 0.499 0.227 0399 0.335 0.427 0.179 MERCOSUR -0.012 0.358 0.001 0.284 0.501 0.183 0.066 0.463 -0.050 0.982 0.728 0.999 0.703 0.626 0.798 0.917 0.633 0.932 SOUTHA -0.530 -0.528 -0.548 0.651 0.640 0.630 SA -0.120 0.626 -0.167 0.649 -0.952 0.324 0.694 0.976 0.413 a852 0.252 0.828 0.363 0 139 0.679 0.330 0.144 0.592 (RTSH-CHso.RTSH_CHa80C-sH.-.CHso+A C_RTSH-CHS0) 1O -0.049 -0.039 -0.043 -0.046 0.001 0.004 0.003 0.002 (RTSH90-RTSH8O-A-C-RTSH9O+A-C-RTSHSO)^1O -0.031 -0.011 -0.034 -0.036 0.559 .849 0519 0.501 R-squared 0.90 0.88 0.89 0.90 0.84 0.84 0.85 0.85 0.84 0.85 0.85 0.85 Adjusted R-squared 0.86 0.84 0.85 0.85 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.80 S.E. of regression 1.04 1.11 1.07 1.05 1.22 1.22 1.21 1.20 1.21 1.21 1.20 1.18 Sum squared resid 28.93 33.22 31.18 29.50 44.85 45.01 43.97 43.50 45.47 45.08 44.76 44.51 Log likelihood -49.51 -52.21 -50.98 -49.90 -60.97 -61.05 -60.56 -60.33 -62.22 -62.03 -61.88 -61.76 Durbin-Watson stat 2.12 2.32 2.25 2.19 2.30 2.30 2.27 2.50 2.27 2.24 2.25 2.53 Sample Size 39 39 39 39 42 42 42 42 43 43 43 43 Same countries as in Control Group V 47 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table VIl: Control Group 1III - OLS Regression Results. Dependent Variable: [GROWTH90-GROWTH80HA C_GROWTH9N-A C_GROWTH8O] Eq.25 Eq.26 Eq.27 Eg.28 Eq.29 Eq.30 Eq.31 Eq.32 Eq.33 Eq.34 Eq.35 Eq.36 c 0.090 0.862 0.379 0.231 -0.464 -0.045 -0.434 -0.379 -0.421 -0.034 -0.429 -0.354 0.851 0.143 0.447 0.627 0.256 0.921 0285 0.411 0.281 0.939 0.273 0389 GROWTH8D-ACGROWH8O -0.361 -0.274 -0.340 -0.338 -0.312 -0.267 -0.315 -0.272 -0.325 -0.270 -0.329 -0.276 0.014 0.058 0ao26 0.023 0.061 ao92 0.058 0.080 0.041 0.079 0.037 0.047 INV8O-A_C_NV8O 0.079 0.028 0.060 0.064 -0.011 -0.035 -0.013 -0.021 -0.003 -0.034 -0.004 -0.017 0.119 0.553 0232 0.199 0.831 0.492 0.799 0.646 0.938 0.477 0.931 0695 PE90-PE8-A_C_PE90+A_C_PE8O 0.030 0.031 0.027 0.021 -0.017 -0.014 -0.017 -0.021 -0.011 -0.012 -0.011 -0.019 0.225 0.154 0266 0.409 0.540 0.615 0.536 0.363 0.676 0.629 0.673 0.228 TAR90S-TAR8O-A-C-TAR90+A.C_TARSO -0.037 -0.007 -0.002 0.167 0.730 0.921 XAGRS(O(TAR90-TAR80.A-C_TAR90A_C-TARSO)fOO 0.392 0.357 0.374 0.079 0.080 Q084 XMAN8rAR90-TARSA-CTAR9O+A_C-TARSYO100 -0.020 -0.017 -0.010 0.570 0527 0.696 (ZKC cEAOr(rAR9o-TAR8O-A_C_TARSo+A.C TARso) 0.011 0.056 0.065 0824 0.340 0.230 MERCOSUR'(TAR90-TAR8D-A..C-TAR9DOA.C-TARtS) -0.064 -0.052 -0.040 0.088 0.139 a0175 (1-CACM-ZKC4EAMERCOSURr(rAR90oTARso-A-C-TARg0+AC_TAR8o) -0.024 -0.007 -0.004 0.257 0.718 0.836 GC90-GC8O-A_C_GC0+A_C_GC8O -0.190 -0.248 -0.183 -0.198 -0.136 -0.191 -0.135 -0.145 -0.130 -0.192 -0.129 -0.143 0.016 0.010 0.024 .01 1 0.075 0.028 .085 0.058 0.094 0.025 0. 103 0.055 M2GDP90-M2GDPSO-A_C_M2GDP90+A_C_M2GDPSO -0.421 -0.390 -0.411 -0.423 -0.415 -0.402 -0.414 -0.420 -0.422 -0.404 -0.423 -0.424 0.000 0.000 0000 0.000 0.003 Q002 0.003 0.002 0.002 0.002 .002 0.001 GOVS-A_C_GOVS 0.402 0.116 0.293 0.249 0.466 0.274 0.452 0.234 0.430 0.258 0.427 0.187 0.165 0.703 0.310 0.407 0.170 0444 0a179 0481 0.19o 0.447 0.193 0.517 TOT90-TOTSO-A-C-TOT9DOA_C_TOTSS 0.031 0.032 0.028 0.032 0.005 0.010 0.005 0.010 0.007 0.010 0.007 0.011 0.007 0.004 0.0 16 0.007 0.533 0.275 0.563 0.249 0.411 0.225 0.429 0.169 MERCOSUR 0.787 2.299 1.142 1.244 2.007 1.219 0.870 1.906 0.815 0.423 Q025 0.231 0280 0.078 0255 0.362 .048 0353 SOUTHA -0.215 -0.234 -0.213 0.847 0.816 0.848 SA -0.126 0.925 0.103 0.577 1.100 0.446 0.618 1.121 0.517 0.871 0.078 0.896 0.420 Q069 0568 0.385 0.075 0.499 (RTS_.CH904RTSH-CH80 A_C.RTSH-CH90+AC_RTSH.CHOr100 -0.045 -0.041 -0.041 -0.043 0.002 .003 0.005 0.003 (RTSH90-RTSH8O-A_C.RTSH90+A.C.RTSHSS)'IO -0.038 -0.012 -0.039 -0.030 0.532 0.839 0.536 0.621 R-squared 0.88 0.89 0.87 0.88 0.84 0.85 0.84 0.85 0.84 0.85 0.84 0.84 Adjusted R-squared 0.84 0.84 0.83 0.84 0.79 0.80 0.78 0.79 0.79 0.80 0.78 0.80 S.E. of regression 1.11 1.10 1.16 1.11 1.23 1.20 1.25 1.21 1.22 1.19 1.24 1.19 Sum squared resid 35.84 33.65 37.55 35.45 48.41 44.95 48.26 47.12 49.30 45.04 49.22 47.75 Log likelihood -55.42 -53.30 -55.49 -55.19 -64.53 -61.97 -63.50 -63.94 -65.91 -62.95 -64.90 -65.19 Durbin-Watson stat 3.11 2.75 2.97 3.04 2.48 2.31 2.46 2.35 2.46 2.30 2.45 2.30 Sample Size 41 40 40 41 44 43 43 44 45 44 44 45 Countries that did not reduce or Increased ATR (average ATR in 1990-99 compared to average ATR in 1980-89) 48 Relative Returns to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table Vill: Control Group IV'. Dependent Variable: [GROWTHI-GROWTHO-[C_GROWTHI-C-GROWTHO] Ea.37 Eq.38 E9.39 EqAi_0 Ea.41 Eq.42 Eq.43 Eq.44 Ea.45 Ea.46 Eq.47 EqA48 c -0.909 -0.961 -1.035 -0.922 -0.754 -0.747 -0.858 -0.766 -1.023 -0.935 -1.042 -1.066 0.188 0.170 0.163 0235 0.216 0.238 0.209 0.234 0.068 0.098 0.083 0.069 GROwrHO-C_GROWTHO -0.664 -0.609 -0.614 -0.668 -0.637 -0.574 -0.582 -0.638 -0.573 -0.555 -0.568 -0.581 0.001 0.002 0.003 0.001 0.001 0.002 0.004 0.001 0.003 0.004 0.006 0.006 INVO.CJNVO 0.052 0.037 0.022 0.060 0.040 0.010 -0.003 0.050 -0.004 -0.004 -0.012 -0.004 0.373 0.549 0.717 0.350 0.431 0.857 0.961 0.376 0.940 0.933 0.814 0.950 PE1-PEO-C_PE1+C_PEO 0.050 0.044 0.034 0.054 0.053 0.040 0.030 0.061 0.003 0.005 0.004 -0.002 0.153 0.245 0.374 0220 0.117 0266 0.414 0.153 0.928 0.844 0.885 0.940 TAR1-TARO-C_TAR1+C_TAR0 -0.046 -0.046 -0.011 0.066 0.049 0.593 XAGRo(rTAR1-TARO-C-TARI+C-TARO)00 -0.217 -0.214 -0.144 0.065 0.049 0.103 XMAN(r(TAR1-TARO.C_TAR1+C-TAROY100 -0.044 -0.030 0.005 0.599 0.692 0.945 (CACM)(TARI-TARO.C_JAR1+C_TARO) -0.035 -0.042 0.004 0.568 0.504 0.943 (ZKC+CEAO)(TAR1-TARO-C-TARI+CJTARO) -0.012 -0.009 -0.035 0.859 0.885 0.654 MERCOSUR-(TAR1-TARO-C-TAR1+C_TARO) -0.045 -0.044 -0.035 0.169 0.221 0.118 (1-CACM-ZK-CEAO-MERCOSURr(TAR1-TARO.CJAR1q+C_TARO) -0.059 -0.054 -0.008 0.136 0.114 0.721 GC14CO-C.GC1+C_GC0 -0.104 -0.091 -0.079 -0.095 -0.106 -0.099 -0.080 -0.101 -0.071 -0.074 -0.066 -0.062 0.099 0.132 0220 0.164 0.106 0.117 0227 0.140 0260 0218 028 0.342 M2GDP1U2GDPO-C M2GDP1+C_M2GDPO -0.255 -0.253 -0.242 -0.255 -0.258 -0.261 -0.249 -0.259 -0.268 -0.272 -0.262 -0.261 0.029 0.024 0.041 0.038 0.029 0.023 0.039 0.031 0o26 0.016 0.027 0041 GOVS6C_GOVS 0.780 0.687 0.853 0.853 0.796 0.732 0.882 0.843 0.793 0.723 0.822 0.760 0081 0.138 0.097 0.099 0.059 ao86 0.0Q59 0.076 0.062 0.087 0061 .098 TOTI-TOTO.C_TOT1UC.TOTO 0.016 0.013 0.010 0.017 0.015 0.010 0.007 0.016 0.000 0.000 0.000 0.006 0.121 0.198 0291 0.142 0.097 0296 0.400 0.112 0.492 0472 0.583 0.515 (RTSH-CHI-RTSH-CHOC_RTSH-CHI+C_CRTSH-CHO)-100 -0.005 -0.013 -0.011 -0.001 0.652 0.243 0.339 0.939 (RTSH14-TSHO-CARTSH1+C..RTSHO)100 0.002 0.020 0.005 0.000 .955 0.639 0.909 0.999 R-squared 0.829 0.827 0.816 0.831 0.828 0.824 0.813 0.830 0.808 0.812 0.806 0.810 Adjusted R-squared 0.770 0.768 0.753 0.743 0.773 0.767 0.753 0.748 0.756 0.762 0.754 0.732 S.E. of regression 1.426 1.434 1.479 1.508 1.388 1.406 1.449 1.462 1.425 1.409 1.431 1.494 Sum squared resid 52.864 53.501 56.883 52.297 53.945 55.343 58.823 53.455 60.955 59.524 61.445 60.275 Log likelihood -58.00 -58.21 -59.32 -57.80 -60.58 -61.06 -62.22 -60.40 -64.05 -63.58 -64.20 -63.83 Durbin-Watson stat 1.78 2.04 1.81 1.74 1.64 1.89 1.71 1.59 1.85 2.01 1.89 1.92 Sample Size 36 36 36 36 38 38 38 38 39 39 39 39 Countries vwith ATR higher than 25% in the second period (i.e., after IL): Bangladesh, Egypt, India, Mauntlus, Nigeria, Pakistan and Sudan 49 Relative Returms to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table IX: Control Group V^. Dependent Variable: [GROWTHl-GROWTHOJ-IC.-GROWTH1-C-GROWTHOI Eq.49 Eq.50 Eq.51 Eq.52 Eq.53 Eq.54 Eq.55 Eq.56 Eq.57 Eq.58 Eq.59 Eq.60 C 0.397 0.713 0.603 0.078 0.257 0.584 0.447 -0.063 0.274 0.509 0.419 0.046 0.259 0.035 0.047 0.829 0.491 0.105 0.163 0.874 0.376 0.085 0.127 0.889 GROWTHIOC.GROWTHI -0.645 -0.595 -0.650 -0.667 -0.608 -0.563 -0.624 -0.631 -0.568 -0.540 -0.596 -0.578 0.000 000 0.000 0.000 0.001 0.003 0.001 0.001 0.00 0.01 o.000 aoo1 INVOCJNVO 0,049 0.031 0.031 0.063 0.017 0.004 0.007 0.031 0.006 -0.002 0.001 0.013 0368 0.596 0.573 0272 0.736 0.946 0.891 0.583 0.897 0.966 0.978 a0774 PE1-PEO-C_PE1+C_PEO 0.024 0.027 0.017 0.016 0.011 0.018 0.006 0.005 0.000 0.002 -0.003 -0.006 0.470 0451 0.642 0.679 0.751 0.643 0.884 0.905 0.990 0.917 0.894 0.794 TARI-TARO0C_TAR1+C_TARO -0.034 -0.030 -0.026 ao58 0.082 0.061 XAGRO0(TAR1-TARO.C-TAR1+C_TARO)1OO -0.108 -0.072 -0.090 .512 0.717 0524 XMAW(TAR1-TARO0C-TAR1+CJTAROyIOO -0.047 *-0.046 -0.044 0.078 0.087 0.029 (CACM)-(rAR1-TARO-C_TARI+C_TARO) -0.226 -0.240 -0.203 0.037 0.041 .031 (ZKC4CEAO)(TAR1-TARO-CTAR1+C_TARO) -0.067 -0.050 -0.035 0.094 0.234 0.380 MERCOSURW(TAR1-TAR0-C-TAR1+C_TARO) -0.030 -0.017 -0.024 0.266 0632 0257 (1-CACM-ZKC-CEAO4MERCOSUR)-(TAR1-TARO-C_TAR1+C_TAR0) -0.047 -0.043 -0.035 0.013 Q015 0017 GC1-GCO-C_GC1+C_GCO -0.075 -0.064 -0.075 -0.034 -0.076 -0.073 -0.076 -0.040 -0.077 -0.068 -0.075 -0.046 a.153 0.262 0.164 0557 0.178 0270 0.177 0.478 aQ113 0.202 0.125 0.404 M2GDP1-A2GDPO-C0_M2GDP1+C_M2GDPO -0.261 -0.262 -0.254 -0.258 -0.271 -0.270 -0.263 -0.266 -0.278 -0.279 -0.270 -0.270 0.009 0.013 0.013 0.009 .009 0.012 013 0009 0.005 0007 0.008 .007 GOv5-C-GovS 0.980 0.959 0.945 1.122 1.025 0.995 0.988 1.153 0.963 0.948 0.947 1.040 0.012 0.022 0018 0.013 0.007 0.014 0010 0013 0.008 0.013 0.009 .011 TOT1-TOTO-C_TOT1+C_TOTO 0.009 0.007 0.006 0.007 0.006 0.004 0.003 0.003 0.000 0.000 0.000 0.002 0219 0337 0.388 0337 0430 0.517 0640 0629 0.460 0.573 0.635 0.685 (RTSH_CH1RTSH_CH04C_RTSH-CH1+C_RTSItCHOr100 -0.016 -0.014 -0.014 -0.016 0.160 0219 0.222 0213 (RTSH1-4TSHOC_RTSH1+C_RTSHO)-100 -0.007 0.003 -0.005 0.020 0.869 0957 .909 0718 R-squared 0.843 0.832 0.841 0.853 0.835 0.826 0.836 0.846 0.831 0.824 0.833 0.840 Adjusted R-squared 0.798 0.783 0.795 0.790 0.788 0.776 0.788 0.781 0.793 0785 0.796 0.786 S.E. of regression 1.291 1.336 1.299 1.315 1.322 1.358 1.322 1.344 1.255 1.281 1.247 1.277 Sum squared resid 51.647 55.313 52.335 48.417 54.204 5Z7137 54.157 50.582 56.743 59.074 55.954 53.812 Log likelihood -62.91 -64.31 -63.18 -61.59 -63.90 -64.98 -63.88 -62.48 -69.07 -69.97 -68.75 -67.88 Durbin-Watson stat 2.00 1.82 2.01 2.08 1.96 1.80 2.00 2.01 1.90 1.80 1.95 1.94 Sample Size 41 41 41 41 41 41 41 41 45 45 45 45 * Countries without a downward trend in ATR: Bahrain, Congo, Cyprus, Ethiopia, Israel, Jordan, Lybia, Malawi, Mauritania, Oman, Poland, Romania, Saudi Arabia, Senegal, Sierra Leone, Singapore, Syria Tunisia and Zimbabwe. 50 Relative Retums to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Table X: IL and Country Size. Dependent Variable: [GROWrHXX-GROWTHXXH-C_GROWTHXX-C_GROWTHXX] FRxed-date Regresslons IL-Dated Regressions Control aroup I Contl Group 11 Control Group IU Control Group IV Control Group V Eg.61 Eq.62 Eq.63 Eq.64 EQ.65 Eq.66 EQ.67 EQ.68 Eq.69 Eq.70 Eq.71 E1.72 Eq.73 Eq.74 Eq.75 c 0.447 0.293 0.259 0.463 0.059 -0.089 0.200 -0.468 -0.505 -1.097 -0.981 -0.855 0.732 0.602 0.466 0.117 0.339 0.331 0.238 0.886 0.820 0.618 a246 0.152 0.064 0.138 0.178 0.042 0.079 0.171 aRowrKUxA.cGRoWTHz -0.405 -0.440 -0.419 -0.453 -0.375 -0.348 -0.381 -0.334 -0.319 -0.656 -0.746 -0.663 -0.607 -0.570 -0.536 0.001 0.005 0.002 0.001 0.018 0.017 0.004 0.019 0.014 0.001 aoo1 0.001 0.0o0 0.002 0.0o1 IVX(-A&C_.NV.X 0.079 -0.001 0.006 0.064 -0.048 -0.039 0.033 -0.044 -0.037 0.035 0.078 0.064 0.000 -0.032 -0.028 o.o96 a083 0.890 0.344 0.491 0.567 0.493 0.374 0.405 0.534 0.222 o.25o o.999 0.606 0.592 PMa.4C.PEXX.c_PUxx 0.045 0.020 0.010 0.036 0.003 -0.005 0.032 -0.002 -0.007 0.026 0.031 0.055 0.039 0.027 0.004 0.035 0.401 0.585 0.037 0.884 0.757 0o0s 0.938 0.675 0.443 0.361 0.120 0.219 0.404 0.860 awDxxYSMALLTARXTAPU-C_-TAPR.A-C_TAM -0.029 -0.030 -0.007 0.030 0.080 0.065 0.018 0.031 0.030 -0.059 -0.132 -0.078 0.040 0.045 0.025 0.503 .576 a871 0.377 0.059 0.127 0.306 o. 196 0.185 0.053 0.008 0.019 0.565 0.505 0.674 ssPUWxA_T APMA-C-TARxxAC TAxx -0.080 -0.035 -0.040 -0.014 0.012 0.012 -0.029 -0.028 -0.024 0.011 0.008 -0.004 -0.011 -0.007 -0.010 a040 a447 0.362 0.508 0.575 0.545 0.122 0.287 0.278 0.720 a0806 0.894 0.541 a695 0.551 awOXXLAROE(TARXX T-PXX-AC-TARX.A,C-TARXX -0.075 -0.041 -0.039 -0.050 -0.026 -0.024 -0.059 -0.042 -0.037 0.029 -0.042 -0.041 -0.039 -0.035 *0.032 0203 as46 0.291 0.017 0.213 0.228 ao11 0.074 ao59 0.474 0.302 0.362 a045 a059 ao67 CxX4cu&c_a.cxx.AcacxX -0.251 -0.161 -0.158 -0.160 -0.107 -0.119 -0.207 -0.147 -0.150 -0.077 -0.097 -0.100 -0.111 -0.111 -0.102 0.002 a063 0.052 ao11 0.178 a0107 0.003 0.070 0.046 a222 a0150 0.144 0.076 0.098 0.088 M2acPD14 DPXX&_C &.c.PXX mAC_PXX -0.396 -0.369 -0.398 -0.381 -0.386 -0.409 -0.412 -0.420 -0.433 -0.277 -0.307 -0.276 -0.241 -0.250 -0.267 aoo1 0.011 o0.02 aooo ao0 aoo1 aooo o.001 0.ooo ao14 aoo3 0.015 a012 0.011 ao06 aovc-a~CovO 0.623 0.665 0.583 0.460 0.574 0.490 0.436 0.478 0.419 0.852 1.100 0.939 1.050 1.096 1.007 0.079 0.129 0.099 0.141 0.099 0.104 a101 0.156 a14o 0.054 a.21 0.043 0.008 0.004 0.006 ToTxx.TOTxxM_.. CTOTXX.A.. 0.029 0.009 0.009 0.034 0.007 0.009 0.030 0.008 0.008 0.000 0.009 0.014 0.006 0.003 0.000 a015 0.390 a0363 0.001 0.398 a30s aoo2 0.328 0.251 0.327 0.386 0.134 0.418 0.734 0.666 mRTSHcHXXRTHxx-A ,C_TSNRCHxx,&c..RTBHc-sHxx) -0.049 -0.046 -0.044 0.020 -0.016 ao4 0.001 aooo 0.126 a0174 (wfTaffMX&xx_&c.TMSa-C_RwHxxr1Oo -0.027 -0.039 -0.027 -0.004 -0.009 a673 0.432 0.625 0.934 0.837 R-squared 0.89 0.84 0.85 0.91 0.87 0.86 0.90 0.86 0.85 0.83 0.86 0.85 0.86 0.85 0.84 Adjusted R-squared 0.84 0.78 0.79 0.88 0.82 0.82 0.86 0.81 0.81 0.77 0.80 0.78 0.80 0.79 0.79 S.E. of regression 1.15 1.33 1.25 0.95 1.12 1.13 1.01 1.16 1.15 1.38 1.32 1.37 1.27 1.31 1.26 Sum squared resid 31.61 44.04 46.60 24.62 37.53 40.75 29.62 42.99 44.69 53.21 41.96 48.72 47.06 49.41 53.77 Log likelihood -48.74 -55.72 -60.80 -46.37 -57.23 -59.86 -51.51 -61.92 -63.70 -61.40 -53.84 -58.64 -61.00 -62.00 -67.86 Durbin-Watson stat 2.25 1.85 1.88 2.04 2.44 2.43 2.92 2.42 2.26 1.50 1.58 1.50 1.82 1.82 1.92 Sample Size 36 36 36 39 42 43 41 44 45 39 36 38 41 45 45 51 Relative Returns to Policy Reformn (Castro, Goldin and Pereira da Silva) DRAFT Table XI: Average Tariff Reduction - 'Reformers' versus 'Non-Reformers' country Yeaw of Average Tariff Rots (1k 'Reformw Before After Diference Aigerla 1983 37.9 24.2 -13.7 Guineoa 1984 48.3 12.2 -38.1 Chile 1985 35.0 13.5 -21.5 MoroCoo 198 41.0 23.8 -17.4 Iloaknd 1987 9.0 4.0 -4.9 MWACO 1987 24.4 12.5 -11.9 Ghaa 1988 29.4 15.6 -13.8 South Africa 1988 25.4 9.6 -15.8 Taiwan. China 1988 28.9 10.2 -16.8 BuundiM 1989 37.3 15.6 -21.8 Hungary 1989 18.5 12.2 -6.'3 Korea. Rep. 19819 22.3 1 1.8 -10.7 MaurlitLus 199388 27.6 -9.2 Cameroon 1990 29.7 20.2 -9.6 Indonesia 1990 30.4 15.7 -14.7 Peru 1990 35.5 16.4 -19.1 Sudan 1990 52.3 32.8 -19.5 Unuguay 1990 37.1 14.0 -23.1 Venezuela, RB 1990 30.5 14.1 -16.3 Argentina 1981 27.6 11.5 -16.1 olIvia 1991 17.4 9.7 -7.6 Brazil 1991 45.4 15.0 -30.4 Colormbia 1991 38.1 12.9 -23.2 Cote dlvoira 1991 27.0 22.2 -4.6 Ecuador 1991 32.3 12.3 -20.0 Sri Lanka 199 3.2.2 -7.9 Guatermala 1992 17.3 10.7 -6.6 India 1992 87.2 40.7 -46.5 Nepal 1992 21.1 14.6 -6.2 Philippines 1992 29.9 17.1 -12.7 Bangladesh .1993 90.2 32.1 -58.1 Costgk Rica 1993 17.0 10.0 -7.1 Egypt. Arab Rep. 1993 41.6 29.0 -12.6 El Salvador 1993 17.4 9.4 -8.1I Kenya 1983 39.3 17.0 -22.3 Antiguia and Barbuda 19413.6 10.9 -2.7 Berdn 1994 41.4 16.6 -24.6 Nigeria 1994 33.6 26.0 -7.5 Pakistan 1994 68.6 48.8 -22.0 Paraguay 1994 12.0 8.3 -2.7 ThaIland 1994 39.4 21.1 -18.2 Barbados 1995 19.5 13.2 -6.3 China 1995 41.4 22.0 -19.4 Jant'alca 1995 18.5 11.7 46.9 Majayala 1995 14.0 6.9 -5.1 Nicargua 1995 14.0 6.6 -5.2 Tanzania 1995 29.0 21.7 -7.3 Tuorkey 1995 26.3 13.9 -12.4 Uganda 1995. 19.a 13.7 -6.1 Beliza 1996 16.3 8.5 -6.6 Guyarna 1996 18.4 12.4 -8.0 TrinIdad and Tobago 1996 16.2 10.7 -7.8 Zan1bla 1996 27.3 11.3 -16.0 I=iotru 1980-89 1990-99 Diference Bahrain -3.8 4.3 0.6 Contgo. Dem. Rep. -22.8 27.1 4.3 Cyprus -13.6 11.1 -2.4 EthlOpa -29.3 26.6 -2.4 Israel -7.8 7.9 0.1 Jordan -15.3 18.6 1.3 LIbya -18.1 24.2 6.1 MatawIA 19.4 19.2 -0.2 Mauritania -21.2 30.2 9.0 Oman -2.5 5.0 2.5 Poland 13.7 12.9 -0.8 Romania -16.6 12.3 -4.2 SaudI Arabia -4.4 12.5 8.1 Senegal -13.2 13.1 0.0 Sierra Leone -29.6 3.8. 7.3 Singapore -0.3 0.3 0.0 Syrian Arab Republic -14.6 15.6 1.0 Tunisa -25.9 28.5 2.6 Zirnbab,ve -9.4 18.3 6.9 52 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFr Table XII: Classification of Countries by Size (Measured by GDP In 1995 US$)* Small Medium Large Antigua and Barbuda Bangladesh Algeria Bahrain Cameroon Argentina Barbados Chile Brazil Belize Congo, Dem. Rep. China Benin Costa Rica Colombia Bolivia Cote d'lvoire Egypt, Arab Rep. Burundi Ecuador Hungary Cyprus El Salvador India Ethiopia Guatemala Indonesia Ghana Iceland Israel Guinea Kenya Korea, Rep. Guyana Morocco Malaysia Jamaica Nigeria Mexico Jordan Oman Peru Libya Pakistan Philippines Malawi Paraguay Romania Mauritania Sri Lanka Saudi Arabia Mauritius Syrian Arab Republic Singapore Nepal Trinidad and Tobago South Africa Nicaragua Tunisia Taiwan, China Poland Uruguay Thailand Senegal Zimbabwe Turkey Sierra Leone Venezuela, RB Sudan Tanzania Uganda Zambia 'Small: less than 5 US$ bil., average 1980-89 Medium: between 5 and 35.75 US$ bil., average 1980-89 Large: more than 35.75 US$ bil., average 1980-89 53 Relative Returns to Policy Reforn (Castro, Goldin and Pereira da Silva) DRAFT Table Xil: Control Groups: Criteria and Statistics j Criteria Control Group Countries IGROWTH INV I PE ATR I GC | M2GDP i GOV5 TOT RTSHCHi 1 1 2 11 I 1 2 :1 1 1 2 1 1 2 il 1 1 2 1 1 11 21l 1T 21 Fixed-date Rearesslons ___________ Pakistan, India, Bangladesh, ATR >25% Senra Leone, Sudan, China, Group I in 2nd T T , Mauriis 4.02 3.74 20.81 20.92 77.04 80.35 43.59 33.49 13.75 11.9 -3.89 -4.11 -0.32 -0.32 110.89 98.49 0.11 0.09 period* Benin, Congo Dem. Rep.. ____Ethopia. Tanzania Control Group 1 same as V below 2.85 3.2 24.27 20.43 83.97 87.28 14.8 16.99 20.1 17.99 -2.51 -1.37 0.13 0. 13 108.53 100.09 0.19i 0.06 Mauritania, Zimbabwe. Saudri Aabia, Sierra Leone, Libya, Contrl chage incong.Tuniia,omn,Joran"I.9 3.9 202 052 803 117 37169 20.75 17.89 -2.53 -1.38 0.04 0.04 110.32 99.77 0.21 0.06 Group IlI ATR>=-0 Syria. Bahrain. Paraguay. Israel29 .9 2.22.2 8.39.7 1. 66 Singapore, Senegal, MalavA and Poland IL-dated Rearessions same as I Control above, and Bangladesh, Egypt, Indiai Group IV excluding Maurilius,Nleria, Pakistanand 4.22 4.99 20.1 21.16 78.32 86.79 55.55 37.72 12.28 10.53 -3.99 -4.77: -0.37 -0.37 131.41 102.77 0.04 0.08 Gru Vcountries In V Sudan below No donadBahraln. Congo. Dem. Rep.. downward Cyprus. Ethiopia. Israel, Jordan. trn ii bya, Malawl, Mautritna, CGontrl V TFi(See Ama, Plnd. Lom ne, 2.99 3.15 23.90 20.78 84.07 88.36 15.53 15.10 19.67 17.17 -2.44 -2.38 0.13 0.13 108.71 98.95 0.18 0.07 equivalent Singapore. Syrian. Arab toca Republic, Tunisia and tcange Zimbabwe __ _ _ in A TR>O _ _ _ _ _ _ _ _ 54 Relative Retums to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT Table XIV: Studies on Fiscal Policy and Economic Growth Effect of Tax on Effect of Effect of Study Growth* Expenditures on Deficit on Methodology and Sample ____________ ~~~~~~Growth Growth ___________________ Engen and Growth of tax rate: Growth of GC: 86 developina countries. Explained: GDP growth, Ekingen and92 -0.06 -0.105 - average 1970-85. Controls: investment rate, Skinner (1992) ;137 1.67 population growth, levels of GC and tax rate, initial (1.37) (1.67) GDP. Uses 2-stage instrumental variables. GC net of education 105 countries. Explained: Per-capita growth and and defense: Investment Rate, average 1970-88. Controls: initial Non-tax revenue: -0.075 GDP, initial PE and SE, assassinations, revolutions, 0.106 (-1.56) war casualties, M2/GDP and trade share in 1970. Easterly and (1.14) -0.129 Tests 13 tax variables, one at a time: only 'marginal' Rebelo (1993) 'Marginal' income Expenditure on (2.22) income tax is signfficant. Stresses the strong tax: general public correlation of fiscal variables with income levels. -0.05 services: Taxes are not significant in general. Other policy (-1.19) -0.24 variables make the fiscal variables even less (-1.78) significant. Ratio of GC (net of spending on defense 87 countries. Explained: per-capita growth, average Barro and and education) to 65-75 and 75-85. Use Barro and Lee (1994) dataset. Sala-l-Martin GDP: -0.11 Controls: initial GDP, PE and SE (male and female), (1995) (average over life expectancy, investment rate, BMP and political several equations, instability. Uses SUR and instrumental variables. highly significant) *t-statistics in parenthesis page 55 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table XV: Fiscal Policy Variables: Non-Control and Control Group 1 Government Government Consumption Expendtures % Tax Revenue as % of GOP Consumptlan minus of GOP JTaxes (ditferenced C ountry______________________________ I_______________________________ ditherence) 1980-89 1990-99 difference 1980-89 1990-99 difference (A) (B) (C)=(B)-(A) (D) (E) (F)=(E)-(D) (G)=(F)-(C) Non-Control Group Syria 20.5 13.0 -7.5 13.1 17.3 4.2 -11.6 Romania 5.9 13.5 7.6 10.3 28.0 17.8 -10.2 Jordan 27.3 23.8 -3.4 14.4 19.7 5.4 -8.8 Nicaragua 29.6 20.3 -9.3 26.2 24.9 -1.3 -8.0 Uruguay 13.7 12.2 -1.5 21.3 26.0 4.7 -6.2 Bangladesh 8.9 4.5 -4.4 5.6 7.0 1.4 -5.8 Ecuador 12.5 9.9 -2.6 12.8 15.5 2.7 -5.3 Kenya 18.3 16.4 -1.9 18.7 21.6 2.9 -4.8 Thailand 12.1 10.1 -2.0 13.9 16.2 2.3 -4.4 Uganda 9.9 9.6 -0.2 6.2 10.1 3.9 -4.1 Zimbabwe 20.1 17.5 -2.6 20.9 22.3 1.5 -4.0 Peru 10.3 9.2 -1.1 11.8 14.1 2.3 -3.4 Malaysia 15.5 12.1 -3.5 20.5 19.8 -0.7 -2.8 Morocco 16.6 17.2 0.6 20.8 24.0 3.3 -2.6 Malawi 17.5 15.5 -2.0 17.5 17.2 -0.3 -1.7 Philippines 8.5 11.3 2.9 11.3 15.6 4.3 -1.4 Bolivia 11.6 13.3 1.8 8.8 12.0 3.2 -1.4 Indonesia 10.2 7.8 -2.5 17.0 15.9 -1.1 -1.4 Singapore 11.3 9.5 -1.8 16.9 16.2 -0.6 -1.2 Korea, Rep. 10.7 10.4 -0.3 15.3 16.1 0.8 -1.1 Cyprus 15.0 17.8 2.8 20.1 24.0 3.9 -1.1 Egypt 16.2 10.5 -5.8 24.5 19.6 -4.9 -0.9 Nepal 8.8 8.8 0.0 7.0 7.8 0.8 -0.8 Tunisia 16.5 16.1 -0.4 24.5 24.8 0.3 -0.7 Costa Rica 15.8 13.6 -2.2 19.2 17.7 -1.5 -0.6 Paraguay 6.7 7.3 0.6 8.7 9.3 0.6 0.0 Iceland 18.4 21.4 3.1 22.6 25.6 3.0 0.1 Pakistan 12.5 12.5 0.0 13.1 13.0 -0.2 0.2 Bahrain 20.7 21.7 1.0 6.9 7.6 0.8 0.2 Turkey 8.9 12.3 3.4 12.4 15.5 3.1 0.3 Chile 12.7 10.3 -2.4 21.5 18.7 -2.8 0.4 South Africa 17.4 19.7 2.3 22.3 24.0 1.7 0.6 Venezuela, RB 11.1 7.7 -3.4 19.0 14.8 -4.2 0.8 India 11.1 11.4 0.3 10.0 9.3 -0.6 0.9 Ethiopia 15.7 13.4 -2.3 13.4 10.2 -3.2 0.9 Burundi 9.3 12.8 3.4 13.6 15.9 2.3 1.1 Sri Lanka 9.1 9.9 0.8 17.5 17.0 -0.5 1.3 Mexico 9.3 10.1 0.8 14.2 13.0 -1.2 2.0 Israel 35.5 29.1 -6.4 43.3 33.6 -9.7 3.2 Argentina 4.5 10.0 5.5 10.3 12.0 1.7 3.8 Jamaica 17.4 13.3 -4.1 28.4 20.4 -8.0 3.9 Colombia 10.1 14.9 4.9 10.4 10.9 0.5 4.3 Control GrOUD LQ2> . 30% ___ U li t ; ' - - *.8.9 7.0 -2.0 :11.4 110 -04 .. -; 1 A,Qotedlvolre X- 16.5 , ,: ,-13.6 I -2.9 21.1 -20.5 06 , : 23 :'Sirra Leone 86 10.5' 1.9 7.0 8.5 1.5 04 MiaZiUus 127 11.9 -0.7 19.1 18.9 -0.2 -Ghaa-. 9.0 12.5 3.5 -8.5 12.1 , 3.8 -01 page 56 Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Table XVI: Control GrouI I- Dependent Variable: PCGROW90-PCGROW8O-C_PCGROW90+C_PCGROWSO Eq.1 Eq2 Eq.3 Eq.4 E9.5 ES.6 Eq.7 Eq.8 Eq.9 c -1.526 -1.153 -1.388 -1.646 -1.576 -1.063 -1.029 -0.969 -1.098 0.0W2 102 0.002 Q.001 0 0.01 0.035 0.066 0.12 PCGROW80C-PCGROW90 -0.371 -0.341 -0.328 -0.335 -0.317 -0.344 -0.320 -0.306 -0.314 0oo2 00w4 0.07 0aw 0.004 0.007 0.012 0.006 n01 INV8O-C-INVO 0.057 0.072 0.045 0.044 0.023 0.039 0.042 0.029 0.045 0.1SO 0.064 0228 a193 0.43 0218 0.187 0.318 Q136 SE90-SEeoC_SE90OCSE80 0.046 0.043 0.046 0.047 0.040 0.030 0.028 0.018 0.031 0.17 0013 aos2 0o12 0.092 0.069 n070 0.204 a119 GC904GCIO-c_GC90SC_GC8O -0.155 (GO C9 C.OCGC90eC.SC O(GOPSOARGE) -0.073 0.500 (GC904C8-C.GC90SC.GCSOr(1-GOMP-LARGE) -0.192 TAX9O-TAXWCOTAX90+CoJAX8Q 0.079 0.072 (TAX90-TAXSCTAxoCTAXeC)LOG(GoPWCw_GoPs0) -0.004 0851 (rAX90-TAX8 C-TAX90+CjTAX80)(GDPaO-ARGE) 0.077 0.237 (TAX90-TAXSO-C-TAX90OC_TAXO)(t GDP10-LARGE) 0.001 a097s (GC90CS0-C.GC90SC.GCSaOTAX90-TAX00 C_cAXSOCTAX5O) -0.100 ((GC90.GC8C.GC9O-C.AC8DPTAXSOTAX8D CATAX9O.C..XAXBO)P80ULARGE -0.112 0079 ((GC9D4C8DC.GC90_G GCaHTAX9O.TAX8DC.TAX#0C.TAXaO)r(1-GDPU.jLARGE) -0.073 a 120 ((GC90 GC08 GOC GCiO0 TAXOO-' AX8OC CTAX9O.C TAX8O)TEXP(G0V5-C GOV5) -0.005 Qs27 DEF90-DEF80 C DEF90+C.DEFO -0.038 0.345 (DEP90-DEF&C-DEFBGO+CDEFSOr(GCaO.C_GC501)MO -0.804 0.026 (DEF90-DEF80DCDEF90SC.DEF80) GDP80_-ARGE -0.051 01333 (DEF9O.EFOOC-DEFso9C-DEFO)Y140GDPF7ULARGE) -0.020 0753 TAR9o0TARSO-CTARso+c_TARSO -0.008 -0.019 -0.013 -0.006 -0.009 -0.012 -0.009 -0.002 -0.010 0487 a 186 0310 Q615 aS1 1 a486 O589 0.871 aS01 t2GoP9O-M2GDP80.C.M2GDP9o+C.M2DPU) -0.644 -0.590 -0.614 -0.674 -0.665 -0.626 -0.645 -0.647 0.653 0060 a 164 0.000 0.09 0 0.000 0o.oo 0 n0ooo Q0000 R-squared 0.913 0.905 0.910 0.905 0.897 0.883 0.886 0.898 0.887 Adjusted R-Squared 0.883 0.872 0.875 0.877 0.866 0.853 0.857 0.871 0.853 S.E of egresion 0.778 0.814 0.804 0.800 0.834 0.873 0.862 0.817 0.873 Sum squared resid 17.546 19.224 18.117 19.192 20.847 23.629 23.020 20.688 22.888 Log likelihood -40.276 -42.103 -40.917 -42.069 -43.724 -46.230 -45.707 -43.571 -45.593 Durbin-Watson tat 1.597 2.097 1.967 1.746 1.885 2.006 1.943 2.226 1.967 *Countrieswit:changein lG-TAX greoter atn 5% ponts (average 1990-99 compared to 198009); orwth trade taxes >. 30% in 1990-99: TrO. CAM. HUN. OMN. BRZ CIV, GIN, MUS, SLE and GHA Relative Returns to Policy Reforrn (Castro, Goldin and Pereira da Silva) DRAFT Vil - References: Ben-David, D. (1993) 'Equalizing Exchange: Trade Liberalization and Income Convergence", Quarterly Journal of Economics, 108(3). Barro, R. (1990) 'Government Spending in a Simple Model of Endogenous Growth", Journal of Political Economy, Vol.98, pp.103-125. Barro, Robert J. and Xavier Sala-l-Martin (1995) "Economic Growth" (New York: McGRaw-Hill) Battacharya, R. (1997) "Pace, Sequencing and Credibility of Structural Reforms", World Development, Vol.25, No.7, pp.1045-1061. Dollar, D. (1992) "Outward-Oriented Developing Economies Really Do grow More Rapidly: Evidence from 95 LDCs, 1976-85" Economic development and Cultural Change, pp. 523- 544. Easterly, William and Sergio Rebelo (1993) "Fiscal Policy and Economic Growth: An Empirical Investigation" Journal of Monetary Economics, Vol.32 (December), pp. 417-58. Engen, Eric M., and Jonathan Skinner (1992) "Fiscal Policy and Economic Growth" NBER Working Paper No. 4223 )Cambridge, Massachusetts: National Bureau of Economic Research). Foroutan, F. (1998) "Does Membership in a Regional Preferential Trade Arrangement Make a Country More or Less Protectionist". Mimeo, World Bank, Washington DC. Gerson, Philip (1998) "The Impact of Fiscal Variables on Output Growth" IMF WP/98/1 (Washington: International Monetary Fund). Harrison, A. (1996) 'Openness and Growth: A Time-Series, cross-country analysis for developing countries". Journal of Development Economics Vol.48, pp 419-447. IMF/World Bank (2001) "Market Access for Developing Countries' Exports". Manuscript, April 27, 2001. Pedroso, A.C., P. C. Ferreira (2000) "Abertura Comercial e Disparidade de Renda Entre Paises: Uma Analise Empfrica". Texto Para Discussao N.728 Instituto de Pesquisa Economica Aplicada. Rio de Janeiro. Ravallion, M., Emanuela Galasso, Teodoro Lazo and Emesto Philipp (2001) "Do Workfare Participants recover Quickly from Retrenchment?", Policy Research Working Paper 2672, The World Bank, Washington, DC Relative Retums to Policy Reform (Castro, Goldin and Pereira da Silva) DRAFT Reinhart, C and loannis Tokatlidis (2002) "Before and After Financial Liberalization", paper presented at the WB/CFS/GWU Joint Conference: 'Financial Globalization: A Blessing or a Curse?", May 2002. Rodrik, D. (1998) "Where did all the growth go? External shocks, social conflicts and growth collapses". NBER Working Paper 6350. Rodrik, D and Fransisco Rodriguez (1999) "Trade Policy ands Economic Growth: A Skeptics Guide to the Cross-National Evidence". NBER Working Paper 7081.. Rosenbaum, P. (2002) "Observational Studies" Second Edition (Springer Series in Statistics), New York, NY. Sachs, Jeffrey and Andrew Wamer (1995), "Economic Reform and the Process of Global Integration", Brooking Paper on economic Activity, pp.1 -118. Tanzi (1987) "Quantitative Characteristics of the Tax System of Developing Countries" in The Theory of Taxation for Developing Countries, ed. by D. Newberry and Nick Stem (Oxford: Oxford University Press). Tanzi, V. and Howell H. Zee (1997) 'Fiscal Policy and Long-Run Growth". IMF Staff Papers, Vol.44, No. 2 (June 1997) Tybout, James (2000) "Manufacturing Industries in Developing Countries: How well they do and Why?" Joumal of Economic Literature, 38 (1): 11-45. Wacziarg, R. (1998) 'Measuring the Dynamic Gains from Trade" Mimeo, World Bank. Washington DC, May 1998. World Bank (2002) "Development, Trade and the WTO - A Handbook" Bemard Hoekmam Aadityia Mattoo and Philip English , editors. The World Bank: Washington, DC. 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 Samake 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 Tchetch6 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 Cote 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