Wps dcaql POLICY RESEARCH WORKING PAPER 2041 Aid Allocation and Poverty In the "efficient" allocation of aid, aid is targeted Reduction disproportionately to countries with severe poverty and adequate policies. For a Paul Collier David Dollar given level of poverty, aid tapers in with policy reform. In the actual allocation of aid, aid tapers out with reform. Aid now lifts about 30 million people a year out of absolute poverty. With a poverty-efficient allocation, the same amount of aid would lift about 80 million people out of poverty. The World Bank Development Research Group Office of the Director and Macroeconomics and Growth January 1999 l POLICY RESEARCH WORKING PAPER 2041 Summary findings Collier and Dollar derive a poverty-efficient allocation of adequate policies - the type of country where 74 aid and compare it with actual aid allocations. percent of the world's poor live. In the actual allocation, They build the poverty-efficient allocation in two such countries receive a much smaller share of aid (56 stages. First they use new World Bank ratings of 20 percent) than their share of the world's poor. different aspects of national policy to establish the With the present allocation, aid is effective in current relationship between aid, policies, and growth. sustainably lifing about 30 million people a year out of Onto that, they add a mapping from growth to poverty absolute poverty. With a poverty-efficient allocation, this reduction, which reflects the level and distribution of would increase to about 80 million people. Even with income. They compare the effects of using headcount political constraints introduced to keep allocations for and poverty-gap measures of poverty. India and China constant, poverty reduction would They find the actual allocation of aid to be radically increase to about 60 million. different from the poverty-efficient allocation. Reallocating aid is politically difficult, but it may be In the efficient allocation, for a given level of poverty, considerably less difficult than quadrupling aid budgets, aid tapers in with policy reform. In the actual allocation, which is what the authors estimate would be necessary to aid tapers out with reform. achieve the same impact on poverty reduction with In the efficient allocation, aid is targeted existing aid allocations. disproportionately to countries with severe poverty and This paper - a joint product of the Office of the Director, and Macroeconomics and Growth, Development Research Group - is part of a larger effort in the group to examine aid effectiveness. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Emily Khine, room MC3-347, telephone 202- 473-7471, fax 202-522-3518, Internet address kkhine@worldbank.org. The authors may be contacted at pcollier @worldbank.org or ddollar@worldbank.org. January 1999. (42 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 Policy Research Dissemination Center Aid Allocation and Poverty Reduction Paul Collier and David Dollar Development Research Group, World Bank Notfor citation without permission of the authors. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. We thankAartKraay and Martin Ravallion for useful advice and Charles Chang and Dennis Tao for excellent research assistance. 2 1. Introduction The allocation of aid among countries can legitimately reflect multiple objectives. Aid may be used to rebuild post-conflict societies, or to meet humanitarian emergencies. However, the core objective is most commonly poverty reduction. In this paper we estimate the allocation of aid that would maximize the reduction in poverty and compare it to actual allocations. We find that the present allocation of aid lifts around 30 million people out of poverty each year, whereas a fully poverty-efficient allocation would have more than double the impact. This apparent inefficiency with respect to poverty reduction does not necessarily imply that aid allocation is sub-optimal. First, our estimates do not incorporate all of the country-specific information which may be available to donors. Secondly, aid allocations may legitimately trade-off poverty reduction against other objectives. However, our poverty-efficient allocation can serve as a benchmark against which deviations can be justified, whether by superior information or by other objectives. The paper is organised in three steps. In Section 2 we estimate the effect of aid upon income. We first show that whether aid raises income is dependent upon the policy environment. We utilise a new World Bank measure which rates twenty aspects of the policy environment for 144 countries as of 1997/98 and relate it to recent growth performance. We then esfimate diminishing returns to aid. Combining these, we arrive at the marginal efficiency of aid in terms of increases in income. In Section 3 we derive the conditions for the optimal allocation of aid for poverty reduction. This depends on the policy environment as well as on the mapping from changes in income to changes in poverty. Conditional upon policy, a one-year increase in 3 the aid flow will raise the growth rate for one year, thereby sustainably increasing the level of income. A sustained increase in the level of income will, conditional upon the initial level of income, its distribution, and the long-term growth rate, sustainably reduce poverty. For each country this yields a function in which aid reduces poverty but is subject to diminishing returns. A poverty-efficient allocation of aid is one in which the marginal cost of poverty reduction is equalised across recipient countries. The efficient allocation is, of course, conditional upon the overall aid budget constraint: for each global aid availability there is a poverty efficient allocation. A corollary is that for each efficiently allocated global aid availability there is a unique marginal cost of poverty reduction. To find the optimal allocation of aid we need a mapping from income to poverty, and we explore several different approaches. We begin with the average elasticity of the poverty headcount with respect to mean income from a large number of surveys. We then refine this measure in three ways. First, we add information on the distribution of income, which is available for 60 countries. The estimated elasticity of poverty is then a function both of the level of mean income and of its distribution. Secondly, we switch the dependent variable from being the headcount of poverty to being the poverty gap. Thirdly, we experiment with a $1 per day poverty line and a $2 per day poverty line. We show that the "poverty-efficient" allocation of aid is very similar, regardless of which poverty line is used or which poverty measure. If the present global flow of aid were efficiently allocated, the marginal cost of lifting a person permanently above the $2 per day threshold would be $665. Because of diminishing returns to aid, this marginal cost exceeds the average cost. The total annual 4 reduction in poverty achieved by an efficiently allocated aid program of $36bn would be 82 million people, so that the average cost of taking a person permanently out of poverty would be only $445. Explicitly or implicitly, developed country governments work with an acceptable unit cost of reducing domestic poverty which far exceeds this figure. In a globalising world it can be expected that there will be increasing awareness of poverty as an international phenomenon, and so a tendency for the present massive differences in the unit costs of reducing domestic and international poverty to diminish. The fourth section of the paper examines the extent to which the actual allocation of aid deviates from the poverty-efficient allocation. First, we show that it is dramatically different from an efficient allocation. In particular, aid has the "wrong" relationship with policy, after controlling for poverty. In precisely the range of policy in which aid becomes increasingly effective in poverty reduction, aid is currently lower the better is policy. In short, aid is being tapered out with reform, when it should be tapered in with reform. One way of expressing the inefficiency is as the consequence for the average cost of reducing poverty. On the current allocation of aid the average cost of poverty reduction is $1,205, or around three times as high as it would be were aid efficiently allocated. Another way to measure the inefficiency is to estimate the gain that would result from shifting from the current allocation to the optimal: an additional 51 million people would be lifted out of poverty through this improvement in the allocation. With the current allocation, a four-fold increase in the total volume of aid would be required to achieve this same reduction in poverty. While radical revision of aid allocation may seem to be politically difficult, it is probably much less politically difficult than quadrupling 5 aid budgets. Indeed, it is reasonable to suppose that the move to an efficient allocation of aid would itself strengthen the case for expansion in aid budgets. For example, consider a budgetary choice which would face the world's most generous aid donor, Norway. Currently, the Norwegian government accepts a per capita cost of around $15,000 to link inhabitants of its islands to the mainland.' If, for each inhabitant linked to the mainland around twenty-five people could be permanently lifted out of severe poverty, Norwegians might choose to increase their already generous aid budget. The final section concludes. While global reallocation of aid budgets would be remarkably effective in reducing poverty, this is not the true choice facing a national aid agency such as NORAD, or an international agency such as the World Bank. Global aid allocation is not the result of a collective decision but of many agency-specific decisions. The relevant consideration for an individual agency is what would be the impact of more efficient allocation of its own funds taking as given the present global allocation. Because many countries in which aid is highly effective in reducing poverty currently get far less aid than would be implied by an efficient allocation, there are remarkable unexploited opportunities for poverty reduction. Given the present allocation of aid, the marginal cost of reducing poverty with efficiently allocated incremental aid is only $333. Hence, the true decision facing Norwegian society is whether linking one inhabitant to the mainland is more or less desirable than removing 45 people permanently out of severe poverty. We started by acknowledging that poverty reduction is not the only legitimate rationale for aid, and so is not the only criterion on which aid allocation should be based. Could the typical aid agency therefore reasonably take the view that the current allocation of aid reflects an appropriate balance of considerations between poverty reduction and 6 other objectives? Our work sheds light on this question by providing a quantitative estimate of the opportunity cost - in terms of foregone poverty reduction - of pursuing other objectives with foreign aid. These other objectives reduce the annual poverty reduction impact from 82 million people to 31 million people. Primafacie is not consistent with the declared primacy of poverty reduction in aid programs. For example, the World Bank's Annual Report 1998 opens with the statement that its "purpose is to help borrowers reduce poverty and improve living standards through sustainable growth and investment in people." And the British Government White Paper on international development states that "We shall refocus our international development efforts on the elimination of poverty and encouragement of economic growth which benefits the poor." (UK 1997, p. 6). . The Mappingfrom Aid to Growth Our objective in this section is to arrive at estimates of the current impact of aid on growth for a large number of countries, as a first step toward estimating the impact of aid on poverty reduction. Burnside and Dollar (1997) have shown that the impact of aid on growth depends on the quality of the incentive regime.2 However, their study has three limitations for the practical application of the results to aid allocation. First, Burnside and Dollar confined their measurement of policies to three readily quantifiable macroeconomic indicators. It is implausible that these are the only policies which matter for growth and, as acknowledged by the authors, they are likely to be proxying a much broader range of policies for which comparable quantitative measures were lacking. We address this problem by utilizing a new data set, the World Bank's Country Policy and Institutional Assessment for 1997. This measure is composed of 20 7 different components covering macroeconomic and sectoral policies, as well as issues such as the rule of law and corruption. Each of the twenty components is rated ordinally by country specialists, on a scale of 1-6, using standardized criteria. Considerable care is taken to ensure that the ratings are comparable both within and between regions. While the scores include an irreducible element ofjudgement, they have a reasonable claim to being the best consistent and comprehensive policy data set. Although the rationale for collecting twenty different aspects of policy is that all are expected to be of some importance for development, potentially some are more important for growth than others. However, in the absence of good evidence to the contrary, we attached equal weights to each of the components. In an appendix we demonstrate that the results are not sensitive to reweighting of the components. Secondly, the Burnside-Dollar data set is for the period 1970-93, whereas the relationships between aid and policy may have changed since the end of the Cold War. Since our purpose is to compare poverty-efficient allocations with actual allocations, it is of little current interest to show that pre-1989 aid allocations were inefficient. We therefore switch from the Burnside and Dollar data set to data exclusively for the 1990s, covering the period 1990-96. The policy rating discussed above is only available for 1997, since for previous years criteria were different. However, policy and institutions in general only change slowly and differ considerably between countries, so that the rating is a reasonable proxy for the average policy stance prevailing during the 1990s. While the switch to the 1990s combined with the switch to a broader measure of policy has the advantage of making any results much more pertinent for current policy, it has the disadvantage of drastically reducing the sample size. Burmside and Dollar were able to 8 arrange their data into 272 growth-aid-policy episodes of four-years each, whereas having only a short period and a single observation of policy, we cannot similarly increase the sample size. Thirdly, the Burnside-Dollar study covered only 56 countries and so cannot provide comprehensive guidance on aid allocation. By switching to more comprehensive data sets, we are able to re-estimate the aid-growth relationship for 86 countries and ultimately to estimate poverty-efficient aid allocations for 109 countries. We now use the new data set to revisit the core Burnside-Dollar result. This is that the efficacy of aid in the growth process depends upon the policy environment: aid is more effective in raising growth the better is the policy environment. Thus, growth (G) is a function of exogenous conditions (X), the level of policy (P), the level of net receipts of aid relative to GDP (A), and the interaction of policy and aid:3 G =c + bX + b2.P + b3.A + b4.A.P. (1) To increase comparability with the original results and for statistical validity, we rebase both the new measure of policy and the Burnside and Dollar measure so that they have a mean of zero and a standard deviation of 1. Since the new measure is an ordinal index, its information content is unaffected by any such monotonic transformation. Table 1 column 1 presents the OLS results for the estimation of (1) on the new data set. Policy is directly significant in the growth process: a one standard deviation improvement in policy is associated with additional growth of 0.6 percentage points. As in Burnside and Dollar, policy is also significant via its interaction with aid. At the mean 9 of policy (=0), the marginal impact of aid is zero. At a policy level of 1, the marginal impact of 1 percent of real PPP GDP in assistance is additional growth of 0.3 percentage points. In column 2 we re-estimate to allow for the possibility that aid and the interaction of aid and policy are endogenous. We show that instrumenting for these variables does not alter the results: both variables remain significant and their coefficients are virtually unchanged.4 Hence, the core Burnside-Dollar result was not due to its particular choice of policy variables, nor was it due to a relationship which no longer holds in the post-Cold War era. These regressions establish the single most important component in the aid- growth relationship: that the efficacy of aid depends upon policy. However, they do not address a second necessary component, namely, diminishing returns to aid. While a given quantity of aid is more effective in a good policy environment, presumably, if aid programs are expanded, beyond a certain point the marginal efficacy of aid diminishes. To estimate diminishing returns to aid we need to establish the curvature of the aid- growth relationship, so that (1) is replaced by: G = c + b1.X + b2.P +b3.A + b4.A2 + b5.A.P (2) This is evidently more demanding of the data. In column 3 we report the Bumside-Dollar 2SLS results for (2) on their data set of 272 observations. They find that the coefficients on aid and Aid2 are both significant, the former positive and the latter negative: as expected, there are diminishing returns to aid. For a country with very good policy (=2), the return to aid reaches zero at about 5% of real PPP GDP. (Since PPP 10 measures of GDP are typically 3-4 times higher than GDP at nominal exchange rates, this corresponds to 15-20% of GDP evaluated at nominal exchange rates.) Column 4 repeats the regression on the new 86-observation data set. The addition of the Aid2 term does not alter the previous results: the efficacy of aid depends upon the policy environment, with both significance levels and coefficients essentially unaffected. However, we are unable to establish the curvature of the aid-growth relationship: both Aid and Aid2 are insignificant. Either this is because during the 1990s the world has changed in such a way that there are no longer diminishing returns to aid, or the sharp reduction in samnple size causes a loss of precision so that curvature cannot be estimated. Evidently, the latter is the more plausible interpretation: indeed, the presence of diminishing returns to aid is not controversial. While overall absorptive capacity might well be higher in the 1990s than previously owing to improvements in the policy environment, this would in no way contradict the notion that for a given policy enviromnent the marginal efficacy of aid diminishes as its volume is increased. The inclusion of diminishing returns to aid is critical for the establishment of a poverty-efficient allocation of aid. Without diminishing returns, the entire global aid budget would simply be allocated to whichever country happened currently to have the lowest marginal cost of poverty reduction. Not only would this be politically so unacceptable as to be uninteresting, it would be incredible. There are strong a priori grounds to suppose that there are diminishing returns: taking the point estimate of 0.3 percentage points of growth per 1% of GDP in aid, a country with good policy would evidently not growth at 30% p.a. if provided with aid worth 100% of GDP. Hence, we adopt the Burnside-Dollar estimate of the diminishing returns to aid, namely that b3 is 11 2.21 and b4 is -0.29, and add these to the policy coefficient as estimated on our data set in regression 1. Specifically, the marginal impact of aid on growth is estimated to be: Ga = 2.21 + 0.29P - 0.58A (3) 3. The Poverty-EfficientAllocation ofAid The results above suggest that if donors wish to maximize the reduction in poverty, aid should be allocated to countries that have large amounts of poverty and good policy. The presence of large-scale poverty is obviously necessary if aid is to have a large effect on poverty reduction. The good policy ensures that aid has a positive impact. In the remainder of this paper we formalize this idea, examine the extent to which donors are already behaving optimally, and estimate the gains in poverty reduction that could be achieved through a more efficient allocation of existing aid volumes. One concern that arises immediately is whether in reality there are countries with good policies and mass poverty. It is possible that the two variables are negatively correlated to a high degree. However, in the 1990s poverty and policy are not highly correlated. Figure 1 shows the CPIA measure and the headcount poverty rate for 113 developing countries. There are 32 countries in the good policy, high poverty quadrant. More remarkably, a large majority of the world's poor lives in these 32 countries: 2 billion people out of a total of 2.7 billion poor people in the 113 countries included in the figure. Not surprisingly, most of the countries in quadrant I are ones that have reformed in the past 10 years (for example, India, Uganda, Ethiopia, and Mali). Our intuition is 12 that aid will have more impact on poverty reduction in this quadrant where there are two billion poor people and policy makes aid effective. In quadrant II, on the other hand, there are most of the remaining 0.7 billion poor people, but the effectiveness of aid in reducing poverty is hampered by poor policies. In quadrant IV, assistance is effective in promoting growth, but there is less work for it to do since these countries already have relatively low poverty. To formalize these ideas, we consider a world in which aid is given with the purpose of maximizing the reduction in poverty. That is, the objective function of donors is: Max Poverty Reduction = E G h' N' i Subject to T Aiy' N' = A, A' > 0 (4) where y is per capita income A is the total amount of aid h is a measure of poverty (for example, the headcount index) a is the elasticity of poverty reduction with respect to income N is population A denotes relative change, and the superscript "i" indexes countries. If the poverty measure is the headcount index, then this maximization has a simple interpretation: allocate aid so that the marginal aid cost of lifting someone above 13 the poverty line is the same in each country. For a more general poverty measure, such as the poverty gap, the objective is to equalize the marginal cost of reducing the gap. Considering for the moment only interior solutions (in which each country gets some aid), the first order conditions for a maximum are G a' h' N' =AY N' (5) Where A is the shadow value of aid. Using the estimate of Ga from (3) above, we can solve explicitly for each country's aid receipts as a function of its policy, poverty level, per capita income, and elasticity of poverty with respect to income: A =3.7+0.5p - 6u11 = (6) The basic properties of the equilibrium can be easily illustrated. Assume for simplicity that the elasticity of poverty reduction with respect to income is constant across countries. Then the equilibrium conditions define a set of relationships among aid, policy, and the poverty measure divided by per capita income - relationships that can be shown in two dimensions if we hold each variable constant in turn. For example, holding aid constant, we have the relationship between policy and poverty shown in figure 2a. Each isoquant shows combinations of policy and poverty that would justify a certain level of aid (given the shadow value of aid). The poorer a country, the lower is the policy quality required to justify a certain volume of aid. Intuitively, the aid will have less growth impact because of the weaker policies, but the poverty impact of a unit of growth 14 is higher. The isoquant for Aid=O is the dividing line between countries that receive aid in the efficient allocation and countries that receive none. We also show the isoquant for an aid level of two percent of GDP. Holding policy constant, the relationship between aid and poverty is upward- sloping, but with diminishing returns to aid (figure 2b). For a given poverty level, on the other hand, the optimal relationship between aid and policy is linear but kinked (figure 2c). There will be a threshold of policy below which even the first dollar of aid is not sufficiently productive in terms of poverty reduction. Above the threshold the poverty- efficient aid allocation is monotonic in policy and happens to be linear in our particular scaling of the ordinal policy rating. The reason for this is that, with poverty constant, the relationship shows combinations of aid and policy that maintain G,, at a constant level. This maps a linear relationship. The general point is that the optimal allocation of aid for a country depends on its level of poverty, the elasticity of poverty with respect to income, and the quality of its policies. In order to calculate a poverty-efficient allocation of aid we need to specify a measure of poverty and estimate the elasticity of poverty with respect to income for a large number of countries. We experiment with four different approaches in order to investigate the extent to which the choice of poverty measure and elasticity affects the allocation. For a given level of mean income, the aid-poverty mapping is affected by the distribution of income. In a highly unequal society there will be more poverty, but, unless aid changes the distribution, much of the aid-induced increase in income will accrue to people who are not poor. The former effect tends to make aid more effective in reducing 15 poverty in highly unequal societies, and the latter effect makes it less effective. Potentially, aid can change the distribution of income. Indeed, projects will normally attempt to target the poor. However, in aggregate aid tends to be fungible (Pack and Pack 1993; Feyzioglu et al. 1998) and so has distributional consequences which are similar to a general increase in public expenditure combined with a general decrease in taxation. Such evidence as there is on the distributional incidences of public expenditure and taxation in developing countries suggests that on average such changes will not be distributionally progressive to any great extent. The incidence of public spending in developing countries is mildly progressive (van de Walle 1995; Devarajan and Hossain 1998). However, the tax reduction effect of aid is likely to be regressive. Furthermore, there is evidence that the distribution of income is fairly stable over time in a majority of countries (Li, Squire, and Zou 1998). We will therefore assume that the net effect of aid is distributionally neutral. This should be understood, however, not as an empirically well-grounded result, but rather as a neutral assumption pending evidence which would enable distribution to be endogenized with respect to aid. Thus, at present the aid-poverty mapping is being endogenized with respect to income distribution, but that distribution is not endogenous with respect to aid. High-quality infornation on the distribution of income is now available for more than 60 developing countries (Deininger and Squire 1996). For these countries we have estimates of the headcount index of poverty for a $1 per day poverty line (hi), the poverty gap corresponding to that line (pgl), the headcount index for a $2 per day poverty line (h2), and the poverty gap corresponding to that line (pg2). For a given 16 distribution of income, there is a remarkably simple formula for the elasticity of the poverty gap with respect to mean income (Datt and Ravallion 1993): apg= (pg-hypg. (7) Thus, the poverty gap and the headcount contain all of the information that we need about the distribution in order to know the elasticity ofpg with respect to income. For the poverty gap measure of poverty, which is conceptually superior to the headcount, we will calculate a country-specific elasticity based on the formula above. The elasticity of the headcount, on the other hand, is far more complicated and requires full knowledge of the density function. However, Ravallion and Chen (1997) have estimated the elasticity of headcount poverty with respect to mean income for a large sample of countries. The mean elasticity for this sample is two, and we will adopt this as our measure for the headcount. So, we have two different concepts of poverty, poverty gap and headcount, with two different poverty lines ($1 and $2 per day). We now bring together the aid-growth mapping and the growth-poverty mappings, and generate poverty-efficient allocations of aid, initially using 58 countries for which we have complete data. Our first result is that if aid allocation is not politically constrained, even despite diminishing returns, aid budgets would be allocated overwhelmingly to India. Because India has reasonable policies, very high poverty and a very large population, its capacity to absorb aid is enormous. While this result is of interest, telling us that under any politically realistic aid allocation India will be under- funded on the criterion of poverty reduction, it does not provide a good basis for 17 discrimination between other environments and so does not guide marginal improvements in aid allocation. We therefore constrain both India and China to their actual levels of aid, and investigate poverty-efficient allocation among remaining countries.5 The results are reported in Table 2. (The total amount is the $26 billion in aid that these countries actually received in 1996.) Our second result addresses the considerable conceptual attention which has been given to refining the measurement of poverty from the simple headcount to the more sophisticated poverty gap. We find that in allocating aid, it makes virtually no difference whether one uses the headcount index with a constant elasticity or the poverty gap with a country-specific elasticity. For the $1 poverty line the correlation between the allocations resulting from the different approaches is .998, while for the $2 poverty line it is .967. It makes a small difference which poverty line is used, but still the correlation between the allocation based on the $1 headcount and that based on the $2 headcount is .90. It is reassuring that the different approaches yield similar allocations. By any measure countries such as Uganda or India have a high incidence of poverty, whereas a country such as Chile, while it may have good policy, does not have a comparable poverty problem and receives no aid in any of the four allocations. The final step in this section is to bring more countries into the analysis. The number of countries in Table 2 is constrained by the availability of information on the distribution of income. Now that we have evidence that using a simple headcount measure of poverty and a constant elasticity of 2 produces results very similar to those from a more sophisticated approach, we can use that technique to arrive at an optimal allocation of aid for more than 100 countries. As before, we constrain China and India to 18 their actual level of aid, and otherwise allocate aid to maxinize the reduction in poverty based on the $2 per day headcount. The marginal cost of poverty reduction, given the current allocation of aid, varies enormously. For around twenty countries additional aid would not reduce poverty (and indeed may even increase it). At the other end of the spectrum, the marginal cost of poverty reduction is only $333 per person in Bangladesh, and is almost as low in India and Ethiopia. In these countries an additional $1 million of aid would permanently lift between 2,400 and 3,000 people out of poverty. The poverty-efficient allocation of aid (table 3) gives each country an amount such that the marginal dollar is equally effective in reducing poverty. Given the present total aid budget this marginal cost of poverty reduction is $1,502. Countries such as Chile have a marginal cost of poverty reduction which is above this level and so receive a zero allocation. It is worth noting that in the unconstrained optimum (in which India gets the bulk of all aid) the marginal cost is $665. Not surprisingly, the poverty-efficient allocation of aid is more sharply targeted to high-poverty countries than is the actual allocation. The correlation between the poverty rate and the poverty-efficient allocation is .79, compared to .52 for the actual allocation. Within the high-poverty group of countries, the poverty-efficient allocation gives particularly large amounts to good-policy countries such as Ethiopia or Uganda. Thus, it is possible to make the allocation of aid more sharply targeted to poverty and more sharply targeted to good policy, simultaneously. 19 4. Reallocating Aidfor Poverty Reduction We now consider in more detail how the actual allocation of aid compares to the allocation that maximizes poverty reduction and quantify the gains from moving from the current allocation to the allocation derived in the previous section. In our model of efficient aid, what a country receives relative to GDP should be a monotonic but non-linear increasing function of the headcount index divided by per capita income (which we will denote POV). It should be a monotonic increasing function of policy, and linear in our transformation of the ordinal policy measure (CPIAS). The actual allocation of aid in 1996 across 106 countries has a broadly appropriate relationship to poverty (Table 4, regression 1). However, there is no significant linear relationship to policy. One possible reason might be that in practice countries with small populations receive higher per capita aid and this may be disguising the true aid-policy relationship. When population is added to the equation it is indeed highly significant, but there is still no significant linear relationship between policy and aid (regression 2). Nor is the absence of a significant relationship caused by outliers. Omitting three outliers with aid to GDP above 20% does not change this result (regression 3).6 Nevertheless, there is a significant relationship between policy and aid, but it is non-monotonic. Regression 4 shows that policy is significant once it is included as a quadratic. The shape of the relationship is shown in figure 3, evaluated at the mean level of POV. For a given level of poverty, in the range between bad policies and mediocre policies aid is positively related to policy. However, still controlling for poverty, in the range between mediocre policies and good policies aid sharply declines. Thus, just as 20 policy moves into the realm in which aid becomes effective in reducing poverty, aid starts to be phased out. Whereas the efficient aid-poverty mapping would require that aid should taper in with policy reform, actual donor behavior is for aid to taper out with reform. Evidently, if policy is treated as exogenous to aid, this represents a large misallocation of aid on the criterion of poverty reduction. Clearly, the rationale for the present aid allocation rule is that aid is being used to induce policy change. This is why it is concentrated over the range of policy in which there is the most scope for improvement. Were policy highly responsive to aid then this assignment might be poverty-efficient even though, for given policy environments, it is highly inefficient. However, Burmside and Dollar (1997) find that increases in the amount of aid do not typically result in betfer policy. Although there are undoubtedly particular instances where aid does induce reform, there are also cases where it delays reform, and these results are consistent with much other literature (Collier 1997; Killick 1991; Rodrik 1996; Williamson 1994). Unfortunately, this ineffective use of aid to induce policy change has had a very high opportunity cost. As can be seen from Figure 3, the optimal relationship between aid and policy is the opposite of the actual. While aid has not proved effective at inducing sustained policy change in poor policy environments, it is proving highly effective in reducing poverty in the newly reformed environments. We now turn from the overall allocation of aid to the allocation of World Bank concessional lending (IDA). The policy rating used in our study is already a factor in the allocation of IDA and so it might be expected that IDA allocations are more poverty- efficient than for aid in general. We test this by re-estimating the aid allocation equations for IDA. Using the maximum number of observations there is a significant, positive 21 relationship between IDA and policy, after controlling for poverty (Table 5, regression 1). The relationship is stronger if population is added to the equation (regression 2). In the case of IDA, about half of the countries in the data-set receive none at all. Essentially, there is a poverty threshold that countries have to cross before they become eligible. If the model is run for only the observations with positive amounts of IDA (53 countries), the relationship between IDA and policy is stronger (regression 3), though the coefficient declines somewhat if three large outliers are dropped (regression 4). The mean level of IDA receipts for these 50 countries is 0.47% of PPP GDP. The coefficient in regression 4 indicates that a one standard deviation improvement in policy results in an increase in IDA of about 25%. The actual allocation of IDA in 1996 is compared to the ideal allocation of aid in Figure 4, and they have remarkably similar slopes. (To adjust for the fact that IDA is only a fraction of total aid we have scaled IDA receipts up by a factor of five so that mean receipts are about the same as for total aid.) At first glance it may seem that the IDA line corresponds to a larger volume of aid. However, the IDA relationship is drawn for a given population, while the poverty-efficient allocation does not depend on population. Thus, relative to the efficient allocation, IDA gives too much to small countries and too little to large ones. (An increase in population leads to a parallel shift of the IDA relationship to the right. At a high level of population, the IDA line would be to the right of the poverty-efficient allocation.) Even with its present allocation, aid achieves much in terms of poverty reduction. We estimate that without aid each year there would be an additional 30 million poor people. However, at present aid overall is not allocated very efficiently with respect to 22 poverty reduction. To return to our poverty-policy quadrants, a large majority of poor people (74%) live in Quadrant I: countries with high poverty and satisfactory policies. On our analysis, a poverty-efficient allocation of aid would assign a larger share of aid to these countries than is accounted for by their share of global poverty because aid is more effective in these environments. We have already noted that one rationale for the current allocation of aid is a reluctance to increase aid to the very populous countries - India and. China. But even in the constrained optimum in which these countries are held at their actual levels of aid, 69% of total aid should go to the countries in Quadrant I. In fact, as shown in figure 5, the share of aid going to Quadrant I countries (55%) is considerably less than this and far less than the quadrant's share of poor people. T'he allocation in table 3 would lead to an additional 27 million people per year rising out of poverty. Recall that this is a constrained optimum in which India and China are held at their actual aid levels. The figure would be even larger (51 million) if the allocation were unconstrained. To put this in perspective: it would take a four-fold increase in the total volume of aid holding allocations constant to achieve the same poverty reduction. Restated, the average cost of permanent poverty reduction would fall from its current level of $1,205 per person to only $445 at the unconstrained optimum, Finally, suppose that the world raised an extra $10 billion in assistance, what would be the differential poverty impact of allocating it across-the-board versus allocating the increment efficiently (that is, allocating the increment so that the marginal impact is equalized across countries receiving part of the increment). We estimate that an across-the-board increase totaling $10 billion would lift 7 million people out of poverty. An efficient increase, on the other hand, would raise about 25 million out of poverty. 23 5. Conclusion Although aid is allocated coherently, it is allocated inefficiently with respect to poverty reduction. At present, aid is allocated primarily as an inducement to policy reform. This produces a pattern in which aid is targeted on weak policy environments. Often these are environments in which poverty is less severe than in better policy environments: aid is being diverted from high-poverty countries. Further, even among countries with similar poverty problems, aid is being diverted from countries in which the poverty problem is soluble, to those in which, given policy, it is insoluble. The policies which matter for aid to be effective in poverty reduction are not narrowly macroeconomic, but include both distributional policies and the provision of social safety nets. This diversion of aid from poverty reduction to policy improvement would only be justifiable if there was clear evidence that the offer of finance was effective in inducing policy improvement. However, there is now reasonable evidence to the contrary: finance is ineffective because it impairs government ownership of the process of policy reforn. A large majority of poor people now live in policy environments in which aid is effective is reducing poverty. Indeed, far more poor people live in such environments than in the poor policy environments on which aid is currently disproportionately targeted. We estimate that with the present allocation aid lifts around 30 million people permanently out of poverty each year. With a poverty-efficient allocation this would increase to around 80 million per year. Even with political constraints which kept India and China at their present allocations, there would be an increase to around 60 million per year. Hence, the attempt over the past decade to buy policy reform has been hugely expensive for the world's poor. 24 We have argued that the choice between finance for poverty reduction and.finance for policy reform is relatively straightforward: finance is effective for the former but not for the latter. However, there are other choices which are more problematic. There is some evidence that aid is effective in reducing the risk of conflict, both pre-conflict and post-conflict (Collier 1998). Indeed, some of the largest deviations from poverty-efficient allocations favor post-conflict countries such as Rwanda and Bosnia. Similarly, newly reformed countries often experience a phase of low private investment as investors wait for uncertainties to be reduced. There is some evidence that in these environments aid is effective in raising private investment (Dollar and Easterly 1998). Hence, even with poverty-efficiency as a benchmark, donors will continue to face genuinely hard choices between poverty alleviation and other effective uses of aid. We plan further work on these alternative priorities for aid. 25 References Alesina, Alberto, and David Dollar, 1998, "Who Gives Aid to Whom and Why?" NBER Working Paper, no. 6612. Boone, Peter. 1994. "The Impact of Foreign Aid on Savings and Growth." London School of Economics, mimeo. Burnside, C. and D. Dollar, 1997, "Aid, Policies, and Growth," Policy Research Working Pper, no. 1777, The World Bank. Cassen, Robert, 1994, Does Aid Work? Oxford, Clarendon Press. Collier, Paul, 1997, "The Failure of Conditionality," in C. Gwin and J. Nelson, eds., Perspectives on Aid and Development, Washington, DC: Overseas Development Council. Collier, Paul, 1998, "The Economics of Civil War," World Bank mimeo. Datt, Gaurav, and Martin Ravallion, 1993, "Regional Disparities, Targeting, and Poverty in India," in M. Lipton and J. van der Gaag, eds., Including the Poor, World Bank. Devarajan, Shantayanan, and Shaikh I. Hossain, 1998, "The Combined Incidence of Taxes and Public Expenditures in the Philippines," World Development. Dollar, David, and William Easterly, 1998, "The Search for the Key: Aid, Investment, and Policies in Africa," World Bank mimeo. Feyzioglu, Tarhan, Vinaya Swaroop, and Min Zhu, 1998, "A Panel Data Analysis of the Fungibility of Foreign Aid," World Bank Economic Review 12(1): 29-58. Killick, Tony, 1991, "The Developmental Effectiveness of Aid to Africa," World Bank Working Paper No. 646. Kruger, Anne 0., Constantine Michalopoulos, and Vernon Ruttan, 1989, Aid and Development, Johns Hopkins University Press: Baltimore and London. Levy, Victor, 1987, "Does Concessionary Aid Lead to Higher Investment Rates in Low- Income Countries?" Review of Economics and Statistics Li, Hongyi, Lyn Squire, and Heng-fu Zou, 1998, "Explaining International and Intertemporal Variations in Income Inequality, " The Economic Journal, 108: 1-18. 26 Mosley, P., 1987, Overseas Aid. Its Defence and Reform, Brighton: Wheatsheaf Books. Pack, Howard and Janet Rothenberg Pack, 1993, "Foreign aid and the Question of Fungibility," Review of Economics and Statistics, Ravallion, Martin and Shaohua Chen, 1997, "What Can New Survey Data Tell Us about Recent Changes in Distribution and Poverty?, " The World Bank Economic Review, Vol. 11, No. 2, 357-82. Rodrilc, D., 1996, Understanding Economic Policy Reform, Journal of Economic Literature, XXXIV. Summers, R. and A. Heston, 1991, "The Penm World Table, Version V," Quarterly Journal ofEconomics, 106: 1-45. United Kingdom Secretary of State for International Development, 1997, Eliminating World Poverty: A Challenge for the 21st Century, White Paper on International Development. van de Walle, Dominique, 1995, "The Distribution of Subsidies Through Public Health Services in Indonesia, 1978-87," in van de Walle and Nead, eds., Public Spending and the Poor, Johns Hopkins U. Press for the World Bank. van de Walle, Nicholas, and Timothy Johnston, 1996, Improving Aid to Africa, Washington, DC: Overseas Development Council. White, H., 1992, "The Macroeconomic Analysis of Aid Impact," Journal of Development Studies, 28. Williamson, J. (ed.), 1994, The political economy of policy reform, Institute for International Economics, Washington, DC. 27 End Notes 'New York Times, August 28"' 1998, p.A4, 'Norway's Awesome Nature Awesomely Overcome'. 2 Earlier literature generally did not find any robust effect of aid on investment or growth (Boone 1994; Levy 1987; White 1992). The Bumside-Dollar result is consistent with this earlier literature in that during the 1970s and 1980s their estimate of the effect of aid on growth in a country with mean policy level is not significantly different from zero. 3Burnside and Dollar consider the possibility that policy is endogenous and in particular is influenced by the level of aid, but they find no significant effect of the amount of aid on policy. Though we have fewer control variables, we also estimated an equation for the policy index, included aid, and instrumented for it. There was no relationship. Our specification for growth makes use of this information, that the policy measure is not affected by the level of aid and can be taken as independent of it. 4As instruments we use population and donor interest variables such as membership in the franc zone, and also interact these with policy. 5In the unconstrained optimum, China does not get much aid. But if India is constrained, then China would receive the bulk of all aid. 6Our objective here is not to estimate a full behavioral model of aid allocation, but-simply to look at whether the allocation of aid meets the efficiency condition that we have established. Alesina and Dollar (1998) show that much of the allocation of aid can be explained by strategic-political variables such as colonial past or UN voting patterns. Figure 1. Poverty and Policy, 113 Developing Countries, 1996 f f: C; of 0f L0 f° ; t 0Ctf, fiSI :> lf7 0 II~~~~~~~~~2 Figure 2a _~~~~~~~~ M.~~~~~~~02 -3 -2-1 01 23 4 Policy Figure 2b 3.5 3 . ,., .. ... . ' '' '-" '''"" '"'''' ' '_~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~. . . .. . O 2.5 02 .L 1 . . ._. . . *~0.5 0 ~ ~ O0440.0$ O.04 2 -0.5 Pov Figure 2c L Poicy 29 a 2. o ii .~~~~~~~~, 2 .i ,3 o 2~~~~~~~~~~2 Figure 3. Actual Aid, Poverty-Efficient Aid, and Policy -4 -3 -2 -1 0 1 2 3 4 Policy 30 Figure 4. IDA and Poverty-Efficient Aid . .... . ..7777777~a AA A~~~~-7 : . .. 7777 AAA AA~~~~~~~~~~~~~~~~~~!, Y 1, ~ ~ ~ ~ A A ~A4~AA A A b ~~A ' A ~ A ~ ~ A~ AAAAA~~ ~ A ~ 3 ,~ A A A A~~ ,... A AAI. AA4n. ~ A . ...... .. .. ... . .... .. ..~ ~ ~ ~ ~ ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~A *AA l S 'A ~~~~~~~~~&&A~~~~~~~AA ~ ~ ~ ~ ~ ~ ~~........ ........ . ... ... . ....... A A A~~~~~~A A'~~~~~~w-' AA~~~~~~~~~ A5 5. A~~~~~~~~~~~~~AA~~~~A~~~ AA~U , . .......... A ~ ~ ~~~~~~~ I~~~~~ 'A .. . ... . . . . .. .....~~~~~~~~ ~ ~ ~ ~ ~ ~ ~ ~ . ......... in ~ ~ 'AAAAˇSA o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~. . ... OA~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~k AA 5~~~~~~~~~~~~~~~~~~~~~~ Figure 5. Poverty, Aid, and Poverty-Efficient Aid SHARE OF POOR PEOPLE SHARE OF POVERTY-EFFICIENT AID (UNCONSTRAINED) SHARE OF POVERTY-EFFICIENT SHARE OF ACTUAL AID AID (CONSTRAINED)SHROFATLAI I: (GOOD POLICY, HIGH POVERTY) lI:(POOR POLICY, HIGH POVERTY) Wii:(POOR POLICY, LOW POVERTY) IV:(GOOD POLICY, LOW POVERTY) 32 Table 1 Dependent variable: Growth rate of per capita GNP (1) (2) (3) (4) Method OLS TSLS TSLS OLS Cross Section Cross Section Panel Cross Section Time Period 1990-96 1990-96 1970-93 1990-96 Initial Income 0.62 0.52 -.80 0.59 (1.82) (1.46) (1.22) (1.84) Policy 0.63** 0.72** 0.19 0.59 (2.18) (2.19) (0.28) (1.91) Aid/GDP -0.01 -0.06 2.21 -0.02 (0.13) (0.41) (1.89) (0.64) (Aid/GDP) x Policy 0.27* 0.24* 0.65** 0.29* (3.31) (2.63) (1.96) (3.38) (Aid/GDP)2 -- -- -0.29** 0.02 ______________________ ____________ ____________ (2.06) (0.60) N 86 86 272 86 R2 _ 0.59 0.59 *Significant at the 1 percent level. **Significant at the 5 percent level. Note: t-statistics in parentheses. Regional dummies included. Regression (3) includes all of the control variables used in Burnside and Dollar (1997). 33 Table 2 Poverty Elasticities Reallocated Aid based on: Pop below -Pov gap at Pop below Pov gap at Elasticity: Pop Elasticity: Pov Elasticity: Pop Elastlcity: Pov gap Pop blow S1/ Pov gap at SI/ Pop below $21 Pov gap at$2I $1 i day S1 I day $2 1 day $2 I day below $1 day gap at $11 day below $2 day at $2 1 day day ($ mn) day (S mn) day (S mnn) day IS mn) CHN CHINA 22.2 6.9 57.8 24.1 2 2.22 2 1A0 2,617 2.617 2,617 2.617 ION INDONESIA 11.8 1.8 58.7 19.3 2 5.56 2 2.04 0 0 450 1,677 MYS MALAYSIA 5.6 0.9 26.6 8.5 2 5.22 2 2.13 0 0 0 0 PHL PHILIPPINES 28.6 7.7 64.5 28.2 2 2.71 2 1.29 6.267 6,454 4.094 3,283 THA THAILAND I .. 23.5 5.4 2 .. 2 3.35 a .. 0 0 BLR BELARUS I .. 6.4 0.8 2 .. 2 7.00 0 .. 0 0 BGR BULGARIA 2.6 0.8 23.5 6 2 2.25 2 2.92 0 0 0 0 CZE CZECH REP. 3.1 0.4 55.1 14 2 6.75 2 2.94 0 0 0 0 EST ESTONIA 6 1.6 32.5 10 2 2.75 2 2.25 0 0 0 0 HUN HUNGARY I .. 10.7 2.1 2 .. 2 4.10 0 .. 0 0 KAZ KAZAKHSTAN I .. 12.1 2.5 2 .. 2 3.84 0 .. 0 0 KGZ KYRGYZ REP. 18.9 5 55.3 21.4 2 2.78 2 1.58 275 283 258 251 LTU LITHUANIA 1 .. 18.9 4.1 2 . 2 3.61 0 .. 0 0 MDA MOLDOVA 6.8 1.2 30.6 9.7 2 4.67 2 2.15 0 0 21 34 POL POLAND 6.8 4.7 15.1 7.7 2 .45 2 0.96 0 0 0 0 ROM ROMANIA 17.7 4.2 70.9 24.7 2 3.21 2 1.87 0 0 0 0 RUS RUSSIA I .. 10.9 2.3 2 .. 2 3.74 0 .. 0 0 SVK SLOVAK REPUB 12.8 2.2 85.1 27.5 2 4.82 2 2.09 0 0 0 2 TKM TURKMENISTAN 4.9 0.5 25.8 7.6 2 8.80 2 2.39 0 0 0 0 BRA BRAZIL 23.6 10.7 43.5 22A 2 1.21 2 0.94 0 0 0 0 CHL CHILE 15 4.9 38.5 16 2 2.06 2 1A1 0 0 0 0 COL COLOMBIA 7A 2.3 21.7 8.4 2 2.22 2 1.58 0 0 0 0 CRI COSTA RICA 18.9 7.2 43.8 19.4 2 1.63 2 1.26 0 0 0 0 ECU ECUADOR 30.4 9.1 65.8 29.8 2 2.34 2 1.22 0 0 0 0 GTM GUATEMALA 53.3 28.5 76.8 47.6 2 0.87 2 0.61 1.173 910 524 0 HND HONDURAS 46.9 20.4 75.7 41.9 2 1.30 2 0.81 429 405 329 248 JAM JAMAICA 4.3 0.5 24.9 7.5 2 7.60 2 2.32 0 0 0 0 MEX MEXICO 14.9 3.8 40 15.9 2 2.92 2 1.52 0 0 0 0 NIC NICARAGUA 43.8 18 74.5 39.7 2 1A3 2 0.88 284 270 217 16S PAN PA..MA 25.6 12.6 46.2 24.5 2 1.03 2 0.89 71 0 0 0 VEN VENEZUELA 11.8 3.1 32.2 12.2 2 2.81 2 1.64 0 0 0 0 OZA ALGERIA I .. 17.6 4A 2 ,. 2 3.00 0 , 0 0 EGY EGYPT 7.6 1.1 51.9 15.3 2 5.91 2 2.39 0 0 1,806 2,365 JOR JORDAN 2.5 0.5 23.5 6.3 2 4.00 2 2.73 0 0 0 0 MAR MOROCCO I .. 19.6 4.6 2 .. 2 3.26 0 .. 0 0 TUN TUNISIA 3.9 0.9 22.7 6.8 2 3.33 2 2.34 0 0 0 0 IND INDIA 52.5 15.6 88.8 45.8 2 2.37 2 0.94 1.936 1.936 1,936 1,936 NPL NEPAL 50.3 16.2 86.7 44.6 2 2.10 2 0.94 547 54 472 425 PAK PAKISTAN 11.6 2.6 57 18.6 2 3.46 2 2.06 3,340 3.805 5,011 5,237 LKA SRI LANKA 4 0.7 41.2 11 2 4.71 2 2.75 0 0 501 689 BWA BOTSWA.. 33 12A 61 30.4 2 1.66 2 1.01 161 106 a 0 CIV COTE D'IVOIRE 17.7 4.3 54.8 20.4 2 3.12 2 1.69 601 630 581 581 ETH ETHIOPIA 46 12A 89 42.7 2 2.71 2 1.08 1.170 1,172 1,133 1.114 GIN GUINEA 26.3 12.4 50.2 25.6 2 1.12 2 0.96 267 221 170 104 GNB GUINEA-BISSAU 88.2 59.5 96.7 76.6 2 0.48 2 0.26 34 31 29 16 KEN KENYA 50.2 22.2 78.1 44.4 2 1.26 2 0.76 845 813 710 593 LSO LESOTHO 48.8 23.8 74.1 43.5 2 1.05 2 0.70 128 121 105 82 MOo MADAGASCAR 72.3 33.2 93.2 59.6 2 1.18 2 0.56 363 3S4 320 268 MRT MAURITANIA 31.4 15.2 68A 33 2 1.07 2 1.07 130 114 108 94 NER NIGER 61.5 22.2 92 51.8 2 1.77 2 0.78 250 247 224 203 NGA NIGERIA 31.1 12.9 59.9 29.8 2 1A1 2 1.01 2,394 2,295 2,054 1,647 RWA RWANDA 45.7 11.3 88.7 42.3 2 3.04 2 1.10 125 126 118 115 SEN SENEGAL 54 25.5 79.6 47.2 2 1.12 2 0.69 505 480 414 323 ZAF SOUTH AFRICA 23.7 6.6 50.2 22.5 2 2.59 2 1.23 0 0 0 0 TZA TANZANIA 10.5 2.1 45.5 15.3 2 4.00 2 1.97 547 568 591 598 UGA UGANDA 69.3 29.1 92.2 56.6 2 1.38 2 0.63 919 908 845 766 ZMB ZAMBIA 84.6 53.8 98.1 73.4 2 0.57 2 0.34 307 294 282 222 ZWE ZIMBABWE 41 14.3 68.2 35.5 2 1.87 2 0.92 632 613 400 254 23 26.316 26316 26310 34 Table 3. Actual and Poverty-Efftcient Allocations of Aid Population below $2 Population below $2 Actual Aid (per Poveny-Efficient Allocation Country a day 1%) a day (people) Real GDP) (per Real GDP) ETHIOPIA 89.00% 48,S36,477 2.90% 3.75% UGANDA 92.20% 16,625,S29 3.34% 3.68% MOZAMBIQUE 100.00% 15.858,402 9.21% 3.56% MALAWI 96.04% 8,882,111 7.09% 3.49% ZAMBiA 98.10% 8,339,111 7.63% 3.46% MALI 92.79% 8,544,451 6.95% 3.44% BANGLADESH 87.63% 101,726,385 1.02% 3.38% RWANDA 88.70% 6,126,779 15.75% 3.22% BURiNA FASO 86.43% 8,488,849 4.11% 3.09% NIGER 92.00% 7,789,197 2.97% 3.03% SIERRA LEONE 77.48% 3,333,213 8.11% 3.01% GUINEA-BiSSAU 96.70% 991,558 1S.67% 3.00% MADAGASCAR 93.20% 11,763,329 2.84% 3.00% CHAD 85.40% S,240,924 5.07% 3.00% BENIN 79.94% 4,140,478 4.15% 2.99% LAO, PDR 83.44% 3,647,992 5.73% 2.80% LESOTHO 74.10% 1,407,570 3.09% 2.79% SENEGAL 79.60% 6,292,971 4.03% 2.78% BURUNDI 88.03% 6,241,706 5.31% 2.77% KENYA 78.10% 19,799,948 1.91% 2.72% NEPAL 86.70% 17,647,835 1.70% 2.70% VIETNAM 80.00% 66,715,974 0.78% 2.64% NIGERIA 59.90% 62,964.795 0.19% 2.62% HAMT 68.28% 4,697,680 4.51% 2.52% GHANA 68.30% 11,063.068 2.04% 2.49% HONDURAS 75.70% 4,232,519 2.82% 2.39% MAURITANIA 68.40% 1,480,059 6.15% 2.36% NICARAGUA 74.50% 3,071,999 10.21% 2.31% PAiJSTAN 57.00% 70,045.184 0.41% 2.31% TAJIKISTAN 47.80% 2,696,884 2.12% 2.25% TOGO 65.26% 2,527,496 2.33% 2.23% COTE DiVOIRE 54.80% 7,227,401 3.91% 2.20% CAPEVERDE 56.61% 205,931 15.49% 2.17% KYRGYZ REP. 55.30% 2,479,662 2.45% 2.16% MONGOLIA 57.29% 1,364,219 4.34% 2.13% CENT.AFR.REP. 70.24% 2,204,091 3.41% 2.07% CONGO, OEM. REP. 70.71% 29,163,044 0A1% 1.98% CAMEROON SSA9% 7,343,886 1.57% 1.86% GUYANA 60.16% 490,507 6.96% 1.81% ZIMBABWE 68.20% 7,170,704 1.45% 1.77% CONGO, REP. 64.82% 1,614,468 8.86% 1.73% GUINEA 50.20% 3,142,361 2.45% 1.64% COMOROS 63.95% 299,132 4.49% 1.64% AZERBAiJAN 36.12% 2,667,190 0.93% 1.33% EL SALVADOR 52.43% 2.830,701 1.94% - 1.24% PHIWPPINES 64.50% 43,320,003 0.36% ' 1.17% GUATEiUALA 76.80% 7,713,746 0.51% 1.06% ANGOLA 68.17% 6,916,587 2.45% 0.93% MALDiVEs S6.58% 133,084 3.77% 0.88% EGYPT 51.90% 29,010,796 1.31% 0.83% SRI LANKA 41.20% 7,271,972 1.16% 0.83% SOLOMON ISL. 54.24% 192,639 4.79% 0.83% MOLDOVA 30.60% 1,330,958 0.59% 0.75% PAPUA NEW GUIN. 58.46% 2,404.443 2.87% 0.55% VANUATU 51.53% 82,428 6.36% 0.48% EQUATORIAL GUIN. 78.06% 297,013 2.39% 0.38% INDONESIA S8.70% 110,169,378 0.16% 0.26% SWAZILAND 55.92% 473,597 0.99% 0.25% INDIA 88.80% 797,248,087 0.13% - CHINA 57.80% 680,302,826 0.06% _ TURKMENISTAN 25.80% 1,079,247 0.26% 0.00% BELARUS 6.40% 659,969 0.16% 0.00% UZBEKISTAN 43.44% 9,511,108 0.15% 0.00% ECUADOR 65.80% 7,223,496 0.44% 0.00% PARAGUAY 40.74% 1,864,685 0.56% 0.00% UKRAINE 30.52% 15,801.298 0.33% 0.00% BULGARIA 23.50% 1,999,285 0.46% 0.00% RUSSIA 10.90% 16,169,747 0.00% 0.00% ROMANIA 70.90% 16,201,271 021% 0.00% FIJI 37.46% 288,538 1.33% 0.00% 35 Table 3. Actual and Poverly-Efficient Allocations of Aid Population below $2 Population below $2 Actual Ald per Poverty-Efficient Allocation Country a day (%I a day (people) Real GDP (per Real GDP) VENEZUELA 3220% 6,732,427 0.02% 0.00% ALGERIA 17.60% 4,728,311 022% 0.00% BELIZE 45.31% 93,095 1.87% 0.00% JAMAICA 24.90% 615,849 0.66% 0.00% GABON 54.41% 568.089 1.51% 0.00% DOMNICAN REP. 47.71% 3,597,500 0.29% 0.00% KAZAKHSTAN 12.10% 2,030,282 0.23% 0.00% MEXICO 40.00% 35,348,244 0.04% 0.00% MOROCCO 19.60% 5,003,306 0.70% 0.00% TURKEY 47.91% 28,485,601 0.06% 0.00% JORDAN 23.60% 903,797 3.26% 0.00% THAILAND 23.50% 13,616,464 0.20% 0.00% BRAZIL 43.60% 67,340,389 0.04% 0.00% SLOVAK REPUB 65.10% 4,524,904 0.35% 0.00% NAMBIA 60.32% 738,253 2.27% 0.00% LITHUANiA 18.90% 704,189 0.54% 0.00% ST. IQTTS & NEV 36.20% 14,991 2.19% 0.00% KOREA, REP. 30.36% 13,393,366 -0.02% 0.00% COSTARICA 43.80% 1,416,364 -0.03% 0.00% MALAYSIA 26.60% 5,111,283 -0.20% 0.00% PERU 50.09% 11,462,327 0.37% 0.00% ST. LUCIA 34.12% 52,188 4.62% 0.00% SOUTH AFRICA 50.20% 17,961,207 0.13% 0.00% TRINIDAD & TOB. 32.31% 408,928 0.19% 0.00% COLOMBIA 21.70% 7,708.792 0.10% 0.00% LATVIA 30.27% 782,870 0.86% 0.00% URUGUAY 34.12% 1,074,147 0.20% 0.00% CZECH REP. 55.10% 5,691,302 0.11% 0.00% MAURITIUS 33.61% 368,290 0.19% 0.00% PANAMA 46.20% 1,172,560 0.46% 0.00% ARGENTINA 35.98% 12,184,026 0.08% 0.00% BOTSWANA 61.00% 843,912 0.71% 0.00% TUNISIA 22.70% 1,968,131 0.29% 0.00% ESTONIA 32.50% 494,276 0.91% 0.00% POLAND 15.10% 5,801,369 0.36% 0.00% HUNGARY 10.70% 1,100,773 026% 0.00% CHILE 38.50% 5,299,686 0.12% 0.00% 36 Table 4 Dependent variable: ODA as a percent of GDP -- 1996 (1) (2) (3) (4) Constant .37 13.1 10.3 10.8 ______ __ (.80) (6.58) (7.38) (7.73) POV 78.2* 81.9* 61.9* 57.0* (3.90) (4.85) (5.21) (4.78) Pov2 -217.2 -213.8** -143.2 -122.3 (1.73) (2.02) (1.93) (1.66) CPIAS 0.17 0.28 0.13 0.03 (0.58) (1.13) (0.75) (0.16) CPIAS2 -0.25** (2.02) Ln (POP) -- -0.008* -0.006* -0.006* = ___________________ __________ _ 1(6.52) (7.10) (7.29) N 106 106 103 103 lR2 0.35 0.54 0.60 0.62 *Significant at the 1 percent level. *"Significant at the 5 percent level. Note: t-statistics in parentheses. 37 Table 5 Dependent variable: IDA as a percent of GDP -- 1996 (1) (2) (3) (4) Constant -0.07 0.74 1.50 1.35 (1.26) (2.82) (3.26) (4.01) POV 11.6* 11.9* 10.5** 14.2* (5.02) (5.36) (2.29) (3.56) poV2 -20.9 -20.9 -14.6 -56.0** .______________ (1.45) (1.51) (0.59) (2.43) CPIAS 0.06 0.07** 0.15** 0.11** (1.81) (2.10) (2.12) (2.07) Ln (POP) -0.05* -0.09* -0.09* (3.15) (3.09) (3.97) N 104 104 53 50 R2 0.57 0.61 0.48 0.42 *Significant at the 1 percent level. *Significant at the 5 percent level. Note: t-statistics in parentheses. 38 Appendix: Reweighting the Components of the CPLA The Country Policy and Institutional Assessment has 20 different components, receiving equal weight in the overall index. The components are grouped into four categories: macroeconomic policies, structural issues (trade, financial sector, property rights), poverty and safety net policies, and public sector efficiency and accountability (including absence of corruption). A natural question to explore is the sensitivity of results to reweighting the components of the index. One version of reweighting is to put all of the weight on one of the four categories, tying each in turn. Taking each category as the measure of policy, the basic results are broadly the same: the coefficient on policy ranges from .38 to .58, while the coefficient on the interactive term, aid times policy, ranges from .22 to .29 (and is always significant at the 2% level) (appendix table 1). This finding should not be too surprising since the correlations among the components are quite high (appendix table 2). The lowest correlation is .72 between the structural policies and the poverty/safety net measures. Nevertheless the two perform nearly identically in the growth regressions. A more extreme version of reweighting is to put all of the weight on each of the twenty components in turn. In this case, 10 of the 20 measures of policy produce a significant coefficient on the interactive term (the 10 are starred in appendix table 3). A simple average of these 10 is correlated .97 with the overall CPIA. This suggests that these 10 contain almost all of the information in the 20. In the growth regression, the average of the 10 produces nearly identical results to the CPIA - except that the t-statistic on both the policy term and the interactive term is larger when all 20 components are used. The basic finding that aid has more impact in a good institutional-policy environment comes through with different measures of policy focusing on macroeconomic management, the business environment, social policy, or public services and accountability. This should increase confidence in the result. On the other hand, there is no basis for excluding any of these components from the measure of policy, and we prefer to work with a simple average of all 20 components. 39 Appendix Table 1 Use of Alternative Policy Measures Coefficient on Policy Coefficient on Aid x Policy CPIAS 0.63** 0.27* (2.18) (3.31) Macro Components 0.38 0.22* (1.01) (2.37) Structural Components 0.58 0.26** (1.47) (2.37) Poverty/Safety Nets 0.55 0.22** (1.37) (2.34) Public Sector Management 0.56 0.29** __________________ __ _ (1.58) (2.60) Big Oa 0.72 0.33* (1.52) (2.95) Note: t-statistic in parentheses. aSimple average of the 10 components starred in Appendix Table 3. 40 Appendix Table 2 Correlations among Different Measures of Policy Big 10 Macro Structural Poverty Public Sector CPLAS 0.97 0.94 0.95 0.86 0.93 Big 10 0.93 0.87 0.92 0.93 Macro 0.84 0.76 0.84 Structural 0.72 0.81 Poverty 0.85 Public Sector 41 Appendix Table 3 Components of the CPIA Measure A. Macroeconomic Management and Sustainability of Reforms 1. General Macroeconomic Performance* 2. Fiscal Policy* 3. Management of External Debt* 4. -Macroeconomic Management Capacity 5. Sustainability of Structural Reforms* A. Structural Policies for Sustainable and Equitable Growth 1. Trade Policy 2. Foreign Exchange Regime 3. Financial Stability and Depth 4. Banking Sector Efficiency and Resource Mobilization* 5. Property Rights and Rule-based Governance 6. Competitive Environment for the Private Sector 7. Factor and Product Markets 8. Environmental Policies and Regulations A. Policies for Reducing Inequalities 1. Poverty Monitoring and Analysis* 2. Pro-poor Targeting and Programs* 3. Safety Nets* A. Public Sector Management 1. Quality of Budget and Public Investment Process* 2. Efficiency and Equity of Revenue Mobilization 3. Efficiency and Equity of Public Expenditures 4. Accountability of the Public Service* * Signifies that the t-statistic on Aid x Policy is 1.90 or greater if this component alone is taken as the measure of policy. 42 Policy Research Working Paper Series Contact Title Author Date for paper WPS2019 The Effects on Developing Countries A. Denny Ellerman December 1998 T. Tourougui of the Kyoto Protocol and Carbon Henry D. Jacoby 87431 Dioxide Emissions Trading Annelene Decaux WPS2020 Trade Liberalization, Fiscal Delfin S. Go December 1998 C. Bernardo Adjustment, and Exchange Rate Pradeep Mitra 31148 Policy in India WPS2021 Balance, Accountability, and Anwar Shah December 1998 A. Shah Responsiveness: Lessons about 37687 Decentralization WPS2022 The Implications of Foreign Aid Shantayanan Devarajan December 1998 C. Bernardo Fungibility for Development Vinaya Swaroop 31148 Assistance WPS2023 Capital Inflow Reversals, Banking Samuel Talley December 1998 M. Geller Stability, and Prudential Regulation Marcelo M. Giugale 85155 in Central and Eastern Europe Rossana Polastri WPS2024 Re-Engineering Insurance Lawrie Savage December 1998 H. Arbi Supervision 34663 WPS2025 Dynamic Capital Mobility, Capital Hong G. Min December 1998 K. Labrie Market Risk, and Exchange Rate 31001 Misalignment: Evidence from Seven Asian Countries WPS2026 Circuit Theory of Finance and the Biagio Bossone December 1998 B. Bossone Role of Incentives in Financial 33021 Sector Reform WPS2027 Child Labor: Cause, Consequence, Kaushik Basu December 1998 M. Mason and Cure, with Remarks on 30809 International Labor Standards WPS2028 Access to Markets and the Benefits Hanan G. Jacoby December 1998 M. Fernandez of Rural Roads 33766 WPS2029 Small Manufacturing Plants, Susmita Dasgupta December 1998 Y. D'Souza Pollution, and Poverty: New Robert E. B. Lucas 31449 Evidence from Brazil and Mexico David Wheeler WPS2030 The Political Economy of Financial Cevdet Denizer December 1998 T. Hailu Repression in Transition Economies Raj M. Desai 44396 Nikolay Gueorguiev WPS2031 Addressing the Education Puzzle: Ram6n L6pez December 1998 T. Shiel The Distribution of Education and Vinod Thomas 36317 Economic Reform Yan Wang Policy Research Working Paper Series Contact Title Author Date for paper WPS2032 Institutional Investors and Securities Dimitri Vittas December 1998 A. Yaptenco Markets: Which Comes First? 38526 WPS2033 Diversification and Efficiency of Stijn Claessens December 1998 R. Vo Investment by East Asian Simeon Djankov 33722 Corporations Joseph P. H. Fan Harry H. P. Lang WPS2034 Information, Accounting, and the Phil Burns December 1998 G. Chenet-Smith Regulation of Concessioned Antonio Estache 36370 Infrastructure Monopolies WPS2035 Macroeconomic Uncertainty and Luis Serven December 1998 H. Vargas Private Investment in Developing 38546 Countries: An Empirical Investigation WPS2036 Vehicles, Roads, and Road Use: Gregory K. Ingram December 1998 J. Ponchamni Alternative Empirical Specifications Zhi Liu 31022 WPS2037 Financial Regulation and James R. Barth January 1999 A. Yaptenco Performance: Cross-Country Gerard Caprio, Jr. 38526 Evidence Ross Levine WPS2038 Good Governance and Trade Policy: Francis Ng January 1999 L. Tabada Are They the Keys to Africa's Global Alexander Yeats 36896 Integration and Growth? WPS2039 Reforming Institutions for Service Navin Girishankar January 1999 B. Casely-Hayford Delivery: A Framework for Development 34672 Assistance with an Application to the health, Nutrition, and Population Portfolio WPS2040 Making Negotiated Land Reform Klaus Deininger January 1999 M. Fernandez Work: Initial Experience from 33766 Brazil, Colombia, and South Africa