Structural shifts in aid dependency and fiscal policy in developing countries

ABSTRACT

Foreign aid is a sizable source of government financing for several developing countries and its allocation matters for the conduct of fiscal policy. This article revisits the fiscal effects of shifts in aid dependency in 59 developing countries from 1960 to 2010. It identifies structural shifts in aid dependency and uses treatment effect methods to assess the fiscal effects of aid. It finds that shifts in aid dependency are frequent and have significant fiscal effects in developing countries. In addition to the traditional evidences of tax and investment displacement and ‘aid illusion,’ we show that upward shifts and downward shifts in aid dependency have asymmetric effects on fiscal accounts in developing countries. Large aid inflows undermine tax capacity and public investment while large reductions in aid inflows tend to keep recipients’ fiscal behaviour intact. Moreover, the tax displacement effect tends to be temporary while the impacts on expenditure items tend to last. Finally, we find that the undesirable fiscal effects of aid are more pronounced in countries with low governance score and low absorptive capacity.

KEYWORDS:

• Foreign aid
• fiscal policy

JEL CLASSIFICATION:

• E62
• F35
• H20
• E21
• E22

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ A distinct advantage of the model selection procedures based on hypothesis testing is that, unlike information criteria, they can directly take into account the possible presence of serial correlation in the errors and non-homogeneous variances across segments.

² In this article, the trimming $\epsilon$ = 0.10 is used to determine the minimal number of observations in each segment $\left(h=\left[\epsilon T\right]\right)$ expressed as a percentage of the number of observations, which constrains the minimum distance between consecutive shifts. Given our sample period of 1960–2010, each segment must contain the minimum number of 5 years. This is in line with the literature on growth accelerations (Hausmann, Pritchett and Rodrick 2005). In addition, we use the 0.10 significance level for the sequential testing.

³ We also identify 28 indeterminacies in which per capita aid increases but aid-to-GDP decreases.

⁴ See in Appendix A4 for the description of PSM model.

⁵ We also explore whether the holding of national elections, the plurality of political parties and the relative importance of the opposition represent decisive factors in aid allocation.

⁶ We applied Bai and Perron method both on aid-to-GDP ratio and aid per capita. This is because we combined both the two variables to define a shift in aid dependency. This means that, for each country, the two variables should work with Bai and Perron method. If aid-to-GDP ratio works with Bai and Perron method while aid per capita does not for a given country, we cannot include this country in the sample. Vice versa, if aid per capita works with Bai and Perron method while aid to GDP does not for another country, we cannot also include this country in the sample. The approach has been applied to all aid recipients and the sample size of 59 countries are those countries where both aid per capita and aid to GDP ratio work with Bai and Perron method.

⁷ Available at http://stats.oecd.org/qwids/#.

⁸ We follow Ravn and Uhlig (2002) who suggested a smoothing parameter of 6.25 for annual data.

⁹ Given the fact that the marginal impact of changing a variable is not constant in a probit model, we set all variables to their means to compute it.

¹⁰ Note that the covariates are balanced. Results for balance test are reported in appendix, Table A5.

¹¹ See in Appendix A3 for more detail of the index of absorptive capacity.

¹² We also use the mean score. The results remain broadly unchanged.

¹³ Note that we use ‘teffects’ command in stata that uses an estimation technique that implements both steps at once so that we do not need to correct the standard errors in the second step to reflect the uncertainty surrounding the predicted outcomes.

¹⁴ Noteworthy, that we present results for ATE and ATET because it is not currently possible to calculate ATET with AIPW estimator.