Aid, Policies, and Growth Reconsidered* ** Hirohisa Kohama Graduate School of International Relations, University of Shizuoka, Yasuyuki Sawada*** Graduate School of Economics, University of Tokyo, and Hisaki Kono Graduate School of Economics, University of Tokyo Revised: October 20, 2003 Abstract Burnside and Dollar (2000) found that the impact of aid on growth of recipients is positive with good policies. However, Easterly, Levine and Roodman (2003) showed that this conditional linkage between aid and growth breaks down once data is extended from 1970–93 to 1970–97. In this paper, we futher investigate the reason of these seemingly inconsistent findings by decomposing their aid variable into different components, i.e., loan and grant. By employing this augmented framework, we found that the disappearence of the aid and growth nexus is generated by a combination of different effects of loans and grants on economic growth. Interestingly, we found that loans seem to have growth effects in the extended period whereas grants may be effective in the shorter data. This may indicate a structural change in aid provisions in the late 1990’s. * A paper to be presented at an experts’ research workshop on "Quantifying the Impact of Rich Countries' Policies on Poor Countries" at the Center for Global Development (CGD), October 23-24, 2003. **A part of this paper is financially supported by the Foundations for Advanced Studies on International Development (FASID). We would like to thank Craig Burnside and David Roodman for helping our replication of their results. We also thank Yujiro Hayami and seminar participants at FASID for their constructive comments. Of course, we are fully responsible for any remaining errors in this paper. *** Corresponding author: Graduate School of Economics, University of Tokyo, 7-3-1, Hongo, Bunkyo-ku, Tokyo, 113-0033, Japan. Email: sawada@stanfordalumni.org 1 1. Introduction How would aid affect growth? This question became the focal point of research and policy on international aid. One of the most influential papers in this area is Burnside and Dollar (2000), hereafter BD, which found that the impact of aid on growth of recipients is positive with good fiscal, monetary and trade policies but has little effect for recipients with poor policies. Their findings suggest that aid would be more effective if it were more systematically conditional on good policy responses of recipients. The BD findings are practically important, too. In fact, their findings have been used as a rationale for the selective provisions of foreign aid. However, this conditional linkage between aid and poverty reduction through economic growth breaks down once data is extended. Easterly, Levine and Roodman (2003), hereafter ELR, augmented their data from 1970–93 to 1970–97, as well as filling in missing data for the original period 1970–93. Then they find that there is no significant relationship between the amount of aid and economic growth of the recipient countries even after controlling for policy variables. In sum, the BD finding is not robust once the data set is refined. In this paper, we futher investigate the reason of this seemingly inconsistent findings of BD and ELR by decomposing the aid variable into different components, i.e., loan and grant. By employing the augmented BD-ELR framework, we found that the disappearence of the aid and growth nexus is generated by combined different effects of loans and grants on economic growth. Interestingly, we found that loans seem to have growth effects in the extended period whereas grants may be effective in the shorter data. The reminder of this paper is organized as follows. In Section 2, we replicate the BD and ELR findings. Section 3 presents our results by estimating the augmented BD-ELR framework. In the final section, we summarizes concluding remarks 2 2. The Replication of the BD-ELR findings Data With respect to data, OECD defines Official Development Assistance (ODA) as a net sum of grants, including tied to technical assistance, and highly concessional loans with grant element of at least 25%. Yet, Chang et al. (1998) pointed that the net flow nature of net ODA underestimated the aid content of disbursed flows by netting out amortization payments. Also, 25% grant elements for concessionality of loan overrepresent of loans with high concessionality and under-represent loans with low concessionality. Hence, we employ the values of Loans and Grants for the 1975-95 period which are extracted from the World Bank’s database, the Effective Development Aid (EDA) compiled by Chang et al. (1998). Note that BD and ELR also employ the EDA data. In order to construct the aid data set for 1970-74 and 1996-97, we took the following two step procedure. First, by using the 1975-95 data set, we regress the loan variable from the EDA data set on variables such as net loan, grant, and technical assistance. Then, with the estimated regression coefficients, we have constructed loan data for 1970-74 and 1996-97. With respect to grant data, we employed the OECD (2001) data directly, since grant data in EDA is the same as that in OECD (2001). Note that EDA does not contain data on technical assistance (TA) since donors benefit from payments received in return for the TA supplied, which may greatly reduce the donor’s net financial cost. On the other hand, we also estimate the TA’s effect on economic growth explicitly by using the OECD (2001) data as well. All variables other than aid variables are the same as those of the ELR data set. We exclude the outliers identified by ELR. The Replications of the ELR specifications By using the data set between 1970 and 1997, we can replicate the regression results by ELR. The first three columns in Table 2, i.e., the specifications (1) and (2) 3 show our replications. Strictly speaking, the specification (1) is closely related to the ELR’s results by using the full sample between 1970 and 1997. As we can see, the coefficients on aid and aid-policy interaction are positive and negative, respectively, but are both statistically insignificant. This replicates the results of ELR which found that the BD results do not hold when we extend the analyses through to 1997. We also conduct the 2SLS to cope with the possible bias generated by endogeneity of the aid variables. The results, which are not reported in this paper but available upon request, are also consistent with the ELR’s results. The Replications of the BD-ELR specifications Once we restrict the sample between 1970 and 1993, we obtain the replicated regression results of BD and ELR. By the specifications (1) and (2) in Table 4, we have shown the replication results. In the specification (2) of Table 4, the coefficient of the aid-policy interaction variable is positive and statistically significant at the 95% level of significance. This suggests that aid promotes economic growth if the recipient has sound economic policies. We also conduct the 2SLS to cope with the possible bias generated by endogeneity of the aid variables. We do not report the results, which are also consistent with the BD and ELR’s results. 3. Estimation Results of the Augmented BD-ELR Model ELR criticized the BD findings since the results are not robust once they extended the data. We hypothesize the two reasons for the seemingly inconsistency between findings of BD and ELR. First, both BD and ELR utilized an aggregate variable of foreign aid. Yet, the simple summation of grant and loan will be misleading, since the way grant and loan work to generate economic returns are quite different. Second, there may be a structural break in the 90’s with respect to the effectiveness of foreign aid which leads to the ELR findings of ineffective aid on growth once they extended the data set. 4 Aid Modality In the specifications (3)-(8) of Table 2, we show the results which include the effects of loans/GDP and grants/GDP on economic growth by using the extended data. We found that the coefficients of loans are positive and statistically significant when we include the loan-policy interactions. The coefficients of this interaction terms are all negative and mostly significant. However, the total effect of loan on economic growth will be positive in the relevant range, at least if we evaluate the total effect at the average policy level, i.e., 1.423. Coefficients of the loan-policy interactions are mostly negative in Table 2. This means the impact of loans on the income growth is lower in the country with good policies. This result is partly due to the characteristics of a policy variable. BD-ELR policy variable is based on the regression results of income on fiscal surplus, inflation rate, and openness. Considering the "flypaper" effect of aid money, new loans lead to net increases of fiscal surplus. In fact, the existing studies on fungibility issues in foreign aid suggest that there is a greater flypaper effect with concessional loans than with grants [World Bank (1998), pp. 64-66]. Therefore, loan-policy interactions could be regarded as the squared loan amounts. If this is a case, negative coefficients of loanpolicy interactions imply the decreasing marginal impact of loans on growth. This interpretation is consistent with the findings of Burnside and Dollar (2002) and Collier and Dollar (2002) which showed that there are diminishing returns of the impact of aid on growth. In the context of Japanese foreign aid, it has been commonly argued that the requirements of loan payment prevent the recipients from investing in ineffective projects and therefore impose discipline on project management. Accordingly, loans are more cost-effective than grants. It is possible to understand this mechanism as an analogy to the work requirements in workfare programs which can be used as selfsorting and incentive enhancement device [Besley and Coate (1992)]. Of course, there is an opposite disincentive effect of loan particularly when the amount of outstanding debt becomes excessive. This is known as the debt overhang problem in the sovereign debt literature [Krugman (1988)]. Our results suggest that the former incentive effect 5 dominates the latter disincentive effect in the context of foreign aid.1 Interestingly, the coefficients on the technical assistance (TA) indicate that aid in the form of TA enhance economic growth provided that policies of the recipient countries are sufficiently sound [specifications (7) and (8) of Table 2]. For example, if the level of the policy variable is above two, then the technical assistance will increase per capita real GDP growth rate definitely. 1 To further investigate the effectiveness of loans, we also included the lagged variable of loans. Loans are typically used for infrastructure projects so the impacts of loans realize with time lags. The fact that data of loans are disbursement-based might also introduce the lagged effects. The results which are not reported here are also consistent with the results of Table 2. 6 Structural Change in the 1990’s Surprisingly, once we limit the sample between 1970 and 1993, the positive effects of loans disappear. Alternatively, the coefficient on the grant-policy interaction becomes statistically positively significant [specification (4) of Table 4]. This implies that unless the policy variable is negative, aid in the form of grant will increase economic growth rate. The specification (8) of Table 4 indicates that the interaction term of squared technical assistance (TA) and policy has a positive coefficient, showing that TA can increase economic growth when policies of the recipient countries are sufficiently sound. The change from the results of Table 2 to those of Table 4 implies that there may be a structural change of the effectiveness of foreign aid in the late 1990s. In this period, loans became more effective than grant in promoting economic growth. This change is parallel to the shift of the focus in foreign aid from macroeconomic stabilization to poverty reduction. 4. Discussions In this paper, we augment the framework of BD and ELR, which investigate the aid and growth nexus, by decomposing the aid variable into loan and grant. By employing the augmented framework, we found that loans seem to have growth effects in the extended period whereas grants may be effective in the shorter data. Technical assistance might be growth-enhancing provided that policies of the recipient country are appropriate. Our results suggest that the seemingly disappearence of the effectiveness of aid on growth is generated since the ELR model combines different effects of loans and grants. Then the relevant question is: how can we implement the policy implications of our empirical results? In reality, effective aid should satisfy two conditions. The first condition is that the recipients utilize the received aid to enhance productivity of the economy. Yet, a very influential paper by Boone (1996) found that aid has no effects on investment and human development indicators, while aid does increase the size of government. A number of other studies such as Alesina and Weder (2002) also 7 concluded that the aid quantity does not alter the quality of policies of recipient countries. Alesina and Weder (2002) found that an increase in aid is likely to increase corruption, probably because an unexpected transfer will induce rent-seeking activities. Second, donors should allocate foreign aid to the countries which can utilize the aid effectively and efficiently. In fact, there is a substantial controversy over the motivation behind aid provisions. Aid donor countries may be concerned with such issues as mutual benefits, potential economic and political benefits for themselves, poverty reduction, equity, and international security. Based on statistical tests of a rigorous theoretical model of ODA, Trumbull and Wall (1994) found that foreign aid allocations are determined by the needs of the recipients represented by infant morality and political-civil rights. Sawada and Yamada (2003) applied the poverty targeting framework of Besley and Kanbur (1988) based on the Foster=Greer=Thorbecke (1984) poverty measure to investigate poverty reduction effects of aid provisions. They found that in the late 1990s, grant allocations of Japan, the Netherlands, U.K., Canada, Norway, and Sweden as well as aid allocations of mutilateral institutions, were consistent with the necessary condition of optimal poverty targeting. Yet, there are the recent studies which found that donor countries largely seem to be motivated by strategic considerations, rather than the altruism or real needs of the receving countries [Alesina and Dollar (2000)], confirming findings by Maizels and Nissanke (1984). Collier and Dollar (2002) also support this view, finding that the actual aid allocation is far from efficient in terms of poverty reduction. Moreover, Aelsina and Weder (2002) documents that there is no evidence that donors allocate more aid to less corrupt governments. Interestingly, Lahiri and Raimondos-Moller (2000) argues that lobbying by ethinc groups in the donor country enhances aid provisions to its country of origin. 2 These realistic aspects should be taken into account when we operationalize the implications of the aid-growth regressions. 2 On the other hand, motivation for multilateral aid can be said to be more transparent. Multilateral agencies are largely apolitical and more exclusively concerned with development and/or poverty reduction [Cassen et al. (1994); Maizels and Nisanke (1984); Frey and Schneider (1986); Sawada (1996); Sawada and Yamada (2003)]. 8 References Alesina, A. and Dollar, D. (2000) “Who Gives Foreign Aid to Whom and Why?“ Journal of Economic Growth Vol. 5, 33-64 March. Alesina, A. and Weder, B. (2002) “Do Corrupt Governments Receive less Foreign Aid?” American Economic Review 92(4), 1126-1137. Besley, T. and S. Coate (1992), “Workfare vs.Welfare: Incentive Arguments for Work Requirements in Poverty Alleviation Programs,” American Economic Review 82(1), 29-261. Besley, T. and Kanbur, R. (1988) “Food subsidies and poverty alleviation,” Economic Journal Vol. 98, 701-719. Boone, P. (1996) “Politics and the effectiveness of foreign aid,” European Economic Review 40: 289-329. Burnside, C. and Dollar, D. (2000) “Aid, Policies, and Growth,” American Economic Review Vol.90, 847-868. Burnside, C. and Dollar, D. (1998) “Aid, the Incentive Regime, and Poverty Reduction,” Policy Research Working Paper 1937. World Bank, Development Research Group, Washington, D.C.. Cassen, Robert and Associates (1994), Does Aid Work? - Report to an Intergovernmental Task Force, Second Edition, Oxford University Press. Chang, Charles, C. and Eduardo Fernandez-Arias, and Luis Serven (1998), Measuring Aid Flows: A New Approach,” Policy Research Working Paper # 2050, DECRG, the World Bank. Collier, P. and Dollar, D. (2002) “Aid Allocation and Poverty Reduction,” European Economic Review 46, 1475-1500. Dowling, J.M. and Heimenz, U. (1985) “Biases in the allocation of foreign aid: some new evidence,” World Development Vol.13: 535-541. Dudley, L. and Montmarquette, C. (1976) “A model of the supply of bilateral foreign aid,” American Economic Review Vol.66:132-142. Easterly, William, Ross Levine, and David Roodman (2003), “New Data, New Doubts: Revisiting “Aid, Policies, and Growth,” Center for Global Development Working Paper 26. Foster, J., Greer, J. and Thorbecke, E. (1984) “A class of decomposable poverty 9 measures,” Econometrica Vol52, 761-766,1984. Kohama, H. (1995) “Japan's Development Cooperation and Economic Development in Asia.” in Takatoshi Ito and A. O. Krueger, eds., Growth Theories in Light of the East Asian Experience, University of Chicago Press, 201-226. Krugman, P. R. (1988), “Financing vs. Forgiving a Debt Overhang,” Journal of Development Economics 29, 253-268. Lahiri, Sajal and Pascalis Raimondos-Møller (2000), “Lobbying by Ethnic Groups and Aid Allocation,” Economic Journal 110, 62-79. Maizels, A. and Nissanke, M. (1984) “Motivations for Aid to Developing Countries,” World Development 12:879-900. OECD (2001) International Development Statistics (CD-ROM). Sawada,Y. (1996) “Aid and Poverty Alleviation: An International Comparison,” IDS Bulletin 27, January 100-108. Sawada, Y. and H. Yamada (2003), “Is Aid Allocation Consistent with Poverty Indicators? A Cross-Donor Comparison,” mimeo, Faculty of Economics, University of Tokyo. Trumbull,W. N. and Wall, H. J. (1994) “Estimating Aid-Allocation Criteria with Panel Data,” Economic Journal 104:876-882. Wall, H. J. (1995) “The Allocation of Official Development Assistance,” Journal of Policy Modeling 17(3): 307-314. World Bank (1996), Annual Report, the World Bank. World Bank (1998), Assessing Aid: What Works, What Doesn’t, and Why, the World Bank. World Bank (2001) World Development Indicators 2001. (CD-ROM). World Bank (2002), Poverty Monitoring, the World Bank. (http://www.worldbank.org/research/povmonitor/) 10 Table 1 Summary Statistics for the ELR Model Sample period: 1970-1997 Variable name GDP/capita growth ln(initial GDP) Policy Ethnic Frac. Assassinations Ethnic*Assas. Institution M2/GDP Aid Loan Grant Technical Assistance Obs Mean 344 344 344 344 344 344 344 344 344 344 344 344 1.353 7.499 1.423 0.468 0.492 0.189 4.324 26.175 1.189 0.366 0.823 0.463 Std. Dev. 3.435 0.759 1.048 0.297 1.274 0.606 1.543 14.478 1.478 0.490 1.184 0.537 Min Max -12.693 5.429 -4.740 0.000 0.000 0.000 1.580 4.580 -0.049 -1.224 -0.002 0.009 10.076 9.339 3.720 0.900 11.500 7.360 8.140 120.308 7.312 2.382 6.696 2.831 11 Table 2. Reconsideration of the ELR Model Sample period: 1970-1997 ln(initial GDP) Policy Ethnic Frac. Assassinations Ethnic*Assas. Institution M2/GDP Sub-Saharan Africa East Asia Aid Replication of ELR (1) (2) -0.42 -0.37 (-1.12) (-0.96) 1.04*** 1.26*** (5.25) (5.54) -0.01 0.03 (-0.02) (0.04) -0.34 -0.37 (-1.34) (-1.45) 0.12 0.20 (0.19) (0.31) 0.32** 0.32*** (2.63) (2.56) 0.00 0.00 (0.23) (0.21) -1.61*** -1.65*** (-2.92) (-2.96) 1.35*** (2.62) -0.04 (-0.25) Aid*Policy 1.12** (2.18) 0.24 (0.83) -0.18 (-1.24) Loan (3) -0.38 (-1.01) 1.05*** (5.34) -0.02 (-0.03) -0.32 (-1.27) 0.10 (0.16) 0.33*** (2.69) 0.00 (0.27) -1.64*** (-2.95) (4) -0.38 (-0.98) 1.19*** (5.49) 0.09 (0.13) -0.34 (-1.31) 0.15 (0.23) 0.33** (2.60) 0.00 (0.06) 1.63*** (-2.93) 1.28** (2.51) (5) -0.36 (-0.90) 1.17*** (5.17) 0.08 (0.12) -0.33 (-1.27) 0.13 (0.21) 0.33** (2.57) 0.00 (0.00) 1.62*** (-2.80) 1.32** (2.48) (6) -0.39 (-1.02) 1.05*** (5.32) -0.03 (-0.04) -0.32 (-1.26) 0.10 (0.16) 0.33*** (2.68) 0.00 (0.26) -1.60*** (-2.68) (7) -0.42 (-1.09) 1.07*** (5.00) 0.10 (0.15) -0.34 (-1.31) 0.14 (0.22) 0.33*** (2.66) 0.00 (0.34) -1.40** (-2.32) (8) -0.52 (-1.35) 1.44*** (5.74) -0.12 (-0.18) -0.35 (-1.29) 0.14 (0.21) 0.28** (2.27) 0.01 (0.48) -1.09* (-1.84) 1.34** (2.60) 1.34** (2.56) 0.96* (1.82) 0.34 (0.66) 1.29* (1.67) -0.72* (-1.72) 0.35 (0.65) 1.57* (1.94) -0.88** (-2.03) -0.14 (-0.73) -0.18 (-0.44) 0.02 (0.11) 1.40* (1.69) -0.54 (-1.43) -0.20 (-0.79) -0.20 (-0.47) 0.06 (0.21) 0.00 (-0.05) -0.13 (-0.50) 0.54 (0.96) -0.40 (-1.38) -0.08 (-0.12) -2.25** (-1.97) 1.25** (2.17) 344 0.33 344 0.35 1.74* (1.95) -0.73* (-1.70) -0.25 (-1.01) 0.28 (0.52) 0.08 (0.18) -0.06 (-0.88) -2.08* (-1.96) -1.07 (-1.32) 1.02*** (3.18) 344 0.37 1.34*** (2.62) Loan*Policy Loan2*Policy Grant Grant*Policy Grant2*Policy Technical Assistance(TA) TA*Policy TA2*Policy Observations R-squared 344 0.33 344 0.33 344 0.33 344 0.34 344 0.34 Notes: The dependent variable is real per capita GDP growth. We present t-statistics in the parentheses. * Significant at 10-percent level. ** Significant at 5-percent level. 12 Table 3 Summary Statistics for the BD-ELR Model Sample period: 1970-1993 Variable GDP/capita growth ln(initial GDP) Policy Ethnic Frac. Assassinations Ethnic*Assas. Institution M2/GDP Aid Loan Grant Technical Assistance Obs Mean 268 268 268 268 268 268 268 268 268 268 268 268 1.328 7.535 1.346 0.459 0.465 0.181 4.396 24.210 1.027 0.366 0.662 0.438 Std. Dev. 3.477 0.711 1.280 0.300 1.265 0.614 1.519 11.393 1.285 0.486 0.952 0.521 Min Max -12.693 5.743 -3.910 0.000 0.000 0.000 1.580 4.580 -0.049 -1.224 -0.002 0.009 10.076 9.339 3.720 0.900 11.500 7.360 8.140 81.641 6.682 2.382 4.607 2.831 13 Table 4. Reconsideration of the BD-ELR Model Sample period: 1970-1993 ln(initial GDP) Policy Ethnic Frac. Assassinations Ethnic*Assas. Institution M2/GDP Sub-Saharan Africa East Asia Aid Aid*Policy Replication of BD-ELR (1) (2) -0.04 -0.19 (-0.08) (-0.41) 1.00*** 0.70*** (5.73) (3.86) -0.53 -0.50 (-0.73) (-0.66) -0.51* -0.54* (-1.82) (-1.69) 0.82 0.92* (1.83) (1.60) 0.34** 0.37*** (2.56) (2.72) 0.02 0.02 (1.09) (1.40) -1.53** -1.45** (-2.28) (-2.39) 0.79 1.17** (1.34) (1.98) 0.23 -0.20 (1.03) (-0.64) 0.35** (2.39) Loan (3) -0.07 (-0.17) 0.99*** (5.66) -0.53 (-0.71) -0.56* (-1.84) 0.93* (1.84) 0.34** (2.53) 0.02 (1.10) -1.49** (-2.37) 0.79 (1.33) (4) -0.19 (-0.40) 0.72*** (3.94) -0.39 (-0.52) -0.54* (-1.75) 0.85 (1.65) 0.37*** (2.66) 0.02 (1.07) -1.57** (-2.47) 1.28** (2.14) (5) -0.22 (-0.47) 0.78*** (4.25) -0.51 (-0.67) -0.53* (-1.71) 0.84 (1.62) 0.33** (2.37) 0.02 (1.17) -1.42** (-2.19) 1.13* (1.89) (6) -0.08 (-0.17) 1.02*** (5.74) -0.55 (-0.73) -0.58* (-1.89) 0.97* (1.91) 0.34*** (2.63) 0.02 (1.02) -1.18* (-1.75) 0.72 (1.20) (7) -0.23 (-0.51) 0.71*** (3.79) -0.42 (-0.55) -0.55* (-1.79) 0.87* (1.68) 0.37*** (2.70) 0.02 (0.96) -1.04 (-1.50) 1.25** (2.08) (8) -0.34 (-0.71) 0.97*** (4.75) -0.44 (-0.59) -0.56* (-1.81) 0.86* (1.69) 0.32** (2.34) 0.02 (0.88) -0.92 (-1.37) 0.92 (1.54) -0.07 (-0.13) 0.48 (0.59) -0.19 (-0.57) 0.01 (0.02) 0.92 (1.03) -0.39 (-1.06) 0.35 (1.18) -0.40 (-0.85) 0.48*** (2.68) 0.56 (0.65) -0.05 (-0.13) -0.08 (-0.83) -0.64 (-1.25) 0.10 (0.29) 0.16 (1.52) 0.60 (1.52) 0.19 (0.32) 0.23 (0.72) -0.78 (-1.00) -2.01 (-1.61) 0.68 (1.21) 268 0.39 268 0.41 1.10 (1.15) -0.42 (-1.05) -0.06 (-0.44) -0.04 (-0.06) 0.71 (1.40) -0.10 (-0.61) -1.94 (-1.57) -1.40 (-1.58) 0.98** (2.38) 268 0.42 Loan*Policy Loan2*Policy Grant Grant*Policy Grant2*Policy Technical Assistance(TA) TA*Policy TA2*Policy Observations R-squared 268 0.38 268 0.39 268 0.38 268 0.40 268 0.40 Notes: The dependent variable is real per capita GDP growth. We present t-statistics in the parentheses. * Significant at 10-percent level. ** Significant at 5-percent level. 14