VoxEU Column Development Poverty and Income Inequality

Poorer without aid

If there seems to be near unanimity among policymakers about the positive role of aid, the academic community has not found any robust evidence that aid contributes to development. This column presents a new empirical strategy that isolates the causal effect of aid on growth. The effect is found to be larger than previously estimated. The average developing country citizen would be about 15% poorer today had aid never been disbursed.

The academic debate about aid effectiveness is a long and still unsettled one. Economists have struggled to find any robust evidence that aid contributes to development, so many have come to the conclusion that, at best, aid has on average no effect on GDP growth. This is somehow at odds with the field experience that underlines the positive impact of some aid-funded projects. There seems to be a paradox inherent to aid. Its micro effects can be evaluated and appear to be positive in many studies, but its macro effects are elusive. David Roodman, a senior fellow at the Center for Global Development who scrutinised many aid-growth studies, wrote in its macro aid effectiveness guide for the perplexed that the effects of aid on growth probably cannot be detected (Roodman 2007).

The prominent reason behind this statement is that aid, by its very nature, is to a large extent allocated to low-performing countries. Hence growth and aid tend to be negatively correlated. This simple remark makes the causal link from aid to GDP growth impossible to establish by looking at simple correlations between these two variables. To observe the causal effect of aid, researchers resort to instrumental variables. Such variables must be exogenous to growth and explain aid flows well. In other words, one must find variables whose only effect on growth occurs through aid, and that are good predictors of aid. There are unfortunately no obvious candidates and researchers tended to use variables that hardly qualify as instruments. For this reason, the whole aid efficiency literature has been subject to some fundamental criticisms. For instance, Deaton (2010) criticised it strongly and argued that instruments used in these studies cannot plausibly be exogenous. Bazzi and Clemens (2009) showed that most results in the literature hinged in practice on population being a valid instrument, which is highly questionable.

A new instrumental variable

Our recent working paper (Frot and Perrotta 2012) takes these criticisms seriously and proposes a new instrument, arguing that it is a significant improvement relative to past approaches. Our strategy stems from the observation that the temporal order in which donor-recipient partnerships are established matters for aid quantities.

This is shown in Figure 1 below, where we group recipients into six cohorts based on entry dates. Each donor-recipient partnership is characterised by its order relative to the years of activity of the donor. For example, partnerships created in the first year of activity of the donor belong to cohort 1, and so on. The figure presents the average normalised aid portfolio share received by recipients in each cohort over the years. It represents whether the average donor portfolio is biased towards some cohorts.

Figure 1. Average aid share in deviation from equal sharing, by recipient cohort

As shown by the figure, early entrants into donors' portfolios on average have received larger aid shares since 1960. Of course, it is likely that donors created partnerships that prioritised poor and heavily populated countries, and that such countries received larger aid shares because of these characteristics, and not because of their entry dates. We show in our working paper that this is indeed the case but that, even after controlling for these determinants, entry dates still matter. Developing countries with earlier entry dates have been receiving more aid than those with later entry dates for the last 40 years. We thus construct an instrument variable that quantifies for each country its expected aid receipts based only on its various partnerships’ entry dates.

Unlike actual aid, predicted aid is not influenced by shocks to economic performance in the recipient country, hence it does not suffer from reverse causality. To be a valid instrument, predicted aid must also affect growth only through aid, and not through any other channel. For example, it might be the case that early entrants receive not only more aid but also larger trade flows, which in turn affect growth. It must also be the case that donors did not sequentially pick recipients according to unobserved variables that were themselves correlated with subsequent growth rates. We also need to ensure that multiplying predicted aid shares with donors’ aid budgets in order to obtain aid quantities does not re-introduce endogeneity. That would be the case if, for instance, higher aid budgets were correlated with other transfers, such as FDI, technology transfers, lower trade barriers, etc. Finally, the correlation between aid and our instrument must be high enough.


We find a statistically significant and positive effect of aid on growth. It implies that a 1% increase in aid disbursements would increase GDP per capita by 0.03%. Put differently, increasing the aid-to-GDP ratio by one percentage point from the 8% sample mean, which requires a 12.5% increase in aid disbursements, would lead to an associated growth rate in per-capita GDP of 0.375%.

This effect may sound small, but it is actually almost three times larger than previously estimated in the literature. Taking our estimates at face value, it is possible to compute the growth rate that would have prevailed, had no aid ever been disbursed. This forecasting exercise is quite demanding for our empirical model, but we still see it as a useful exercise to understand the effects at play.

Figure 2 presents the estimated density of growth rates under the actual and the counterfactual scenarios. The shift in the distribution is non-negligible. Many more countries, in fact most of them, would have had negative growth rates over the period 1963-2007 without aid. The average GDP-per-capita growth rate in our data is 1%. Without aid, it would have been -0.01%. In order to estimate the impact of aid from 40 years of disbursement, we compute the ratio of the counterfactual 2007 GDP to its actual counterpart.

Figure 2. GDP-per-capita growth density

If aid had not been disbursed since 1963, GDP per capita in the average developing country would be today about 30% lower. While it is interesting to think in terms of countries, large countries tend to receive relatively little aid compared to most small countries and giving equal weight to China and Cape Verde to compute the effects of aid on poverty may be questionable. An alternative is to compute the change in the income of the average developing country citizen. We find she would be about 15% poorer, while the median citizen would be 7% poorer.


The question on aid effectiveness investigates the effects of aid in its generality: the raw sum of everything that enters the flow of money, goods and services classified as development assistance. This approach has been criticised, on the point that, being different components of aid so heterogeneous in nature, aim and function, it is not only difficult but also misleading to search for an aggregate effect. More potential, is argued, lies in the question about specific interventions that more easily meet the requirements for rigorous scientific evaluation with the method of the randomised trial.

The more general question on aid still begs an answer, though. Foreign assistance has been disbursed for decades and is still seen as a major tool of development policy. A hot debate about how the aid system can be improved, or whether any improvement is feasible, is ongoing. In this article, we voluntarily shy away from this debate. More modestly, our goal is to uncover what can be said about aid effectiveness in the past 40 years. Our answer is clear and robust across specifications. The effect of aid has been, on average, positive. Whether the size of the effect – a 15% increase in the income of the average developing country citizen – was worth the while is out of the scope of this article.


Bazzi, S and M Clemens (2009), "Blunt instruments: On establishing the causes of economic growth", Working Paper 171, Center for Global Development.

Deaton, A (2010), "Instruments, randomization, and learning about development", Journal of Economic Literature 48, 424-455.

Roodman, D (2007), "Macro Aid Effectiveness Research: A Guide for the Perplexed", Working Paper 135, Center for Global Development.

Frot, Emmanuel and Maria Perrotta (2012), "Aid Effectiveness: New Instrument, New Results?", mimeo.

2,310 Reads