Aid, policies, and growth in developing countries: a new look at the empirics.

1. Introduction

The link between foreign aid and growth in receiving countries has been controversial for many years; specifically, the effectiveness of aid in promoting growth remains highly contested. A recent paper by Burnside and Dollar (2000), which reported that good policy increases the effect of aid on growth has ratcheted up that debate. This particular policy view has also had an enormous impact on donor policies (Easterly 2003). Given limited resources and the implication that aid should be directed to countries with good policies, many developing poor countries with questionable policy records would be especially vulnerable. Not surprisingly, a large number of empirical papers followed to re-examine the aid-growth relationship, in general, and the aforementioned policy view, in particular (for example, Collier and Dehn 2001; Dalgaard and Hansen 2001; Hansen and Tarp 2001; Collier and Dollar 2002; Burnside and Dollar 2004; Clemens, Radelet, and Bhavnani 2004; Dalgaard, Hansen, and Tarp 2004; Easterly, Levin, and Roodman 2004; and Rajan and Subramanium 2005, to name a few). Although there is some corroboration, several papers dispute this policy view. Some papers argue that aid works by itself but with diminishing returns, and the controversy continues.

A number of issues have been raised in the existing empirical literature, of which nonlinearity is a major one. Previous papers have tried to address nonlinearity in the aid variable mainly by adding quadratic terms in the linear regressions. However, it is almost impossible to pinpoint the exact form of nonlinearity in the aid-policy-growth relationship that would be appropriate for all data sets and for all data ranges. By using widely referenced aid and policy variables (which have been applied by authors in both camps in the debate), we revisit the issue from a different intuitive and econometric angle that has not been explored before. We do not superimpose any a priori functional form on the relationship, so that our analysis remains free from any possible functional form misspecification bias. We then raise an important question: Do various aid-policy combinations (ranges) yield different returns to growth, and, more importantly, can we discern any meaningful patterns in the underlying aid-policy-growth relationship? By pursuing a nonparametric estimation framework, we are able to derive specification-free, data-driven, point-wise estimates that capture the varying effects on growth at different levels of aid and policy. This is particularly helpful because we can focus on specific values (ranges) of aid and policy in assessing when they are growth enhancing, neutral, or possibly growth detracting. If they happen to accommodate all three possibilities, perhaps in different data segments, linear estimation would possibly tend to average out the varying effects and could possibly "tilt" the results in either direction, depending on the sample of countries and periods included.

To keep the main contributions of our paper crisp and focused, we use the traditional measures of aid and policy--effective development assistance (purchasing power parity [ppp] adjusted) and a macroeconomic policy index a la Burnside and Dollar (2000). This allows us to abstract from several other issues, including the use of a measure of aid that is adjusted for emergency relief and other political and long-impact noneconomic motivations "(for example, democracy, health, and education; see Clemens, Radelet, and Bhavnani 2004; and Rajan and Subramanium 2005) and broader measures of policy that incorporate institutions and rule of law (as in Burnside and Dollar 2004). Following some of the important papers in this debate (including Burnside and Dollar 2000 and Easterly, Levin, and Roodman 2004), we keep the institutional variable as a control variable in the regression. Also, in keeping with tradition, we abstract from the issue of policy endogeneity in a growth regression, as recently pointed out by Rodrik (2005). (1) It also allows us to examine the aid-policy-growth nexus in various ranges of aid and policy in a data-driven way while staying with the conventional measures of the variables.

As is common in the literature (for example, Burnside and Dollar 2000; Clemens, Radelet, and Bhavnani 2004; Dalgaard, Hansen, and Tarp 2004; and Rajan and Subramanium 2005, to name a few), we use standard parametric estimations as benchmarks, although our conclusions are solely based on the semiparametric framework. We apply several statistical tests to validate the semiparametric estimation strategy. The main findings are the following: policy is mostly an important determinant of growth; there is partial corroboration of the Burnside-Dollar policy view; and we find some evidence of diminishing returns to aid, though at very high levels of aid.

The paper is organized as follows. Section 2 reviews the relevant literature. Section 3 discusses the econometric model and tests. The data and empirical results are discussed in section 4, and section 5 concludes with a discussion.

2. Related Literature

Throughout the 1970s and 1980s, the two-gap model of Chenery and Strout (1966) provided the main framework to conceptualize the links between aid and growth, and this model was subsequently extended by Bacha (1990) to incorporate the fiscal behavior of government. Aid was perceived to fill gaps that were the most pressing. The currently popular view is that developing countries suffer more from an "institutions gap" and a "policy gap" than a financing gap. In this regard, the usual empirical framework for testing the aid-growth relationship augments aid and aid-policy interaction with some measure of institutions, policy, and/or governance and political stability (see, for example, Burnside and Dollar 2000; Dalgaard, Hansen, and Tarp 2004; and Easterly, Levin, and Roodman 2004).

Although the earlier generation of papers had a different focus, assessing whether aid-induced saving fosters growth or if aid partly substitutes investment, (2) the current generation addresses the role of macropolicies, influence of new growth theory, potential endogeneity in aid and policy, and nonlinearity in the aid-growth relationship. Of particular relevance to our paper, Burnside and Dollar (2000) investigated whether macropolicy matters for aid effectiveness, and they found that aid has a positive impact on growth only in the presence of good economic policy. They observed that it is not aid but the interaction term (aid x policy) that is consistently positive and significant, where the policy index is constructed by taking a linear combination of three well-documented policy variables in the growth literature (budget deficit, openness, and inflation). Using different specification, measures, and/or data, Collier and Dehn (2001), Collier and Dollar (2002), and Burnside and Dollar (2004) all support Burnside and Dollar's original conclusion. However, Easterly, Levin, and Roodman (2004), using the exact same specification and methodology as Burnside and Dollar (2000), failed to support Burnside and Dollar's conclusion when they expanded the data set to include a few more countries and more years. (3) Hansen and Tarp (2001) addressed the issue of nonlinearity in aid by including an aid-square (aid (2)) term, where they found that squared aid drove out the significance of the aid x policy term, and they concluded that aid, on average, works, although with diminishing returns. Ghura et al. (1995) and Durbarry, Gemmell, and Greenaway (1998) also supported the conclusion of Hansen and Tarp (2001). Roodman (2004) performed a series of robustness checks and concluded that most of his findings supported Hansen and Tarp (2001) rather than Burnside and Dollar (2000).

To summarize, the debate in the existing empirical literature is centered on two major views, which are independently and often jointly contested.

(i) Policy view: Aid works only in the presence of good policy (i.e., the interaction term [aid x policy] has a positive and significant effect on growth).

(ii) Diminishing Returns view: Aid works, but with diminishing returns, irrespective of good policy (i.e., the aid-squared term drives away the significance of the interaction term).

3. Econometric Model and Tests

One of the key issues in this debate is the functional form specification of the aid-policy-growth relationship. It is well known that any parametric linear or nonlinear model may lead to misspecification bias if the functional form imposed is not consistent with the true functional relation. Nonparametric modeling, which does not impose any functional form restrictions, is particularly useful in addressing this issue. However, as is well known, a nonparametric model suffers from the dimensionality problem; (4) therefore, it requires a large sample, where the size of the sample required increases rapidly with the number of regressors that are treated nonparametrically. Given our sample size, which is a potential limitation, this is overcome by using the partially linear model. Hence, we use a semiparametric framework where we model only the primary variables of interest nonparametrically and leave the other covariates linear. The semiparametric approach we use, specifically, the partially linear model, is as follows:

[Y.sub.it] = [beta][X.sub.it] + m([Z.sub.it]) + [u.sub.it]. (1)