APPENDIX 1. Country Experiences with Scaled-Up Aid
- Kevin Fletcher, Sanjeev Gupta, Duncan Last, Gerd Schwartz, Shamsuddin Tareq, Richard Allen, and Isabell Adenauer
- Published Date:
- April 2008
In the past, aid recipients often experienced sharp swings in aid flows. Net aid flows to Pakistan, for example, increased by a factor of 2.5 between 1997 and 2004, and nearly tripled to Ethiopia during the same period.51 Analyzing country experiences around such aid spurts can be useful, both for understanding the transmission mechanism of scaled-up aid to various fiscal variables and for drawing lessons regarding appropriate institutional arrangements for facilitating aid management and absorption.
Some Statistical Properties of Aid Flows
Many low-income countries already receive more funds in the form of aid than they collect in the form of own revenues (Appendix Table A1.1).52 This is particularly true for African countries, which account for almost 60 percent of the sample. African countries received, on average, 16 percent of GDP in aid flows, substantially more than the Latin American or Asian countries in the sample. In contrast, the average revenue-to-GDP ratio in African countries was less than 10 percent. Breaking down the sample into five-year intervals shows that aid levels, expressed as a share of GDP, have declined in many countries.
|Number||Mean||Median||Standard deviation||Mean||Median||Standard deviation||Relative Variance1|
Ratio of variances between the aid and revenue variables, as in Bulíř and Hamann (2006).
Also, at least for Africa, aid flows have remained substantially more volatile than revenues (Table A1.1). Although the absolute volatility of both aid and revenues has declined, aid flows remain more volatile than revenues, a finding that is similar to the findings of other researchers.53 Volatility of aid is higher in African countries than for the sample as a whole, reflecting the quantitative importance of aid (both grants and loans). Conversely, relative aid volatility, which is measured as a ratio of the variances of aid and revenues, has worsened in recent years. Volatility of aid has contributed to additional fiscal uncertainties in aid recipient countries.
Among the main components of aid, grants are much more volatile than loans (Table A1.2). The fairly large standard deviation around the mean for grants underscores that spending financed by external grants faces larger uncertainty than spending financed by loans. Statistically, this simply reflects the fact that grants are usually substantially larger than loans, but for actual fiscal management, absolute volatility is more relevant than relative volatility (that is, a normalized measure of volatility such as the coefficient of variation).
|Mean||Standard deviation||Mean||Standard deviation||Mean||Standard deviation|
Past aid surges have been relatively short-lived. Achieving the Millennium Development Goals (MDGs) would require countries to manage and execute ambitious social and infrastructure projects that often have long gestation periods. Aid inflows for financing such projects would have to be much smoother and more sustained in the coming years than what appears to have been the norm in the past.
Aid Flows, Government Spending, and Fiscal Institutions
In general, aid flows have remained difficult to predict while past aid surges have been short-lived. A set of panel regressions of aid shows that only revenues and lagged values of aid consistently explain aid flows, and even then with relatively weak explanatory power (Table A1.3).54 The negative relationship between aid and revenues conforms to the findings of other researchers (for example, Gupta and others, 2004). There is also some indication that aid flows rise with growth and behave countercyclically with respect to the output gap and revenues (that is, as the output gap widens and revenues increase, aid flows decline). For the most part, however, and despite trying out a wide range of explanatory variables, regression residuals remained large. The significantly smaller-than-unity coefficient of the lagged dependent variable suggests that aid is mean reverting, meaning a large aid spurt seldom persists. Various event studies carried out to probe deeper into the issue of aid flow volatility confirm that large increases in aid have consistently been followed by a tapering off of aid (Figure A1.1).55
Figure A1.1.Event Study: Aid Flows After an Aid Spurt
Sources: OECD, Development Assistance Committee (DAC) database; and IMF staff estimates.
Note: t denotes years, and the dotted lines denote 1 standard deviation error bands.
|Total aid/GDP||Revenue (minus grants)/GDP||–––Capital spending/GDP–––||–––Current spending/GDP–––||Education spending/GDP||Health spending/GDP|
|Total aid/GDP squared||–0.01**||–0.01**||–0.01||–0.01||–0.01||–0.01**|
|Lagged total aid/GDP||0.53**|
|Revenue (minus grants)/GDP||–0.42**|
|Number of countries||28||28||28||30||30||20||20||30||30||20||20||26||26||26||26|
Own revenues are correlated positively with loans and negatively with grants.56 The contemporaneous correlation findings do not necessarily imply that grants induce reduced tax effort; rather, the finding could well be associated with the fact that donors give more grants to less-developed, fiscally constrained countries. This argument is also advanced in a recent paper by Morrissey (2006). Adding an indicator of political risk as an explanatory variable yields positive and statistically significant coefficients, indicating that countries with better political institutions and lower risk tend to be associated with higher revenue collection.
The impact of aid on spending was analyzed with four regressions that use different spending aggregates as the dependent variable (Table A1.3). The main results follow:
Capital spending rises with total aid, although the result is more robust with increases in grants as opposed to loans. However, capital spending does not increase proportionately with more aid, as shown by a negative but small coefficient in the squared aid-to-GDP term.
Current spending also increases with grants and, overall, tends to rise with aid flows by more than capital spending does.
Social spending (that is, health and education) is fairly unaffected by aid flows. Health spending is positively correlated with grants (but not with loans), although the parameter is very small. There is no statistically significant effect of different aid aggregates on education spending. In general, countries with better political risk ratings are also associated with higher levels of health and education spending. The lack of responsiveness of health and education spending to aid flows may reflect government attempts to maintain such spending even when funding is volatile and uncertain. Indeed, countries use various mechanisms to protect certain spending items in these sectors from allocation shortfalls.
Although data on the quality of fiscal institutions are scarce, countries with better fiscal institutions would also seem to experience less aid volatility (Figure A1.2). Scatter plots of the standard deviations of aid flows with total HIPC-AAP (Heavily Indebted Poor Countries Assessments and Action Plans) scores—or any of the components (that is, budget formulation, execution, and reporting)—suggest that a higher institutional quality score goes hand in hand with lower aid volatility. Similar results hold when the HIPC-AAP scores are replaced by the fiscal portion of the World Bank Country Policy and Institutional Assessment (CPIA) ratings for a larger group of countries.
Figure A1.2.Aid Volatility and Fiscal Institutional Quality
Sources: IMF country documents and staff estimates.
Note: AAP = Assessments and Action Plans.
Countries that improved their ratings for budget execution also tended to reduce current spending while increasing capital spending (Figures A1.3 and A1.4). The two HIPC-AAP surveys, done a few years apart, allow analyzing the impact of improvements in fiscal institutions on budgetary activities. The data suggest that five out of seven countries with a deterioration in budget execution ratings during 2001–04 increased current spending relative to GDP; similar results were found for other components of the HIPC-AAP scores. Conversely, countries that improved their budget execution ratings during 2001–04 also increased their capital spending, on average, although only slightly.
Figure A1.3.Changes in Current Spending and Institutional Quality
Sources: IMF country documents, PEFA secretariat, and IMF staff estimates.
Figure A1.4.Changes in Capital Spending and Institutional Quality
Sources: IMF country documents, PEFA secretariat, and IMF staff estimates.
See Mattina (2006) for a detailed discussion.
The analysis presented here is based on panel data from 51 PRGF-eligible countries during 1990–2004. Data on aid flows are taken from the OECD’s Development Assistance Committee (DAC) database, which captures the majority of (but not all) aid flows to the sample countries. The rest of the information is obtained from the IMF’s World Economic Outlook (WEO) and Monitoring of Fund Arrangements (MONA) databases.
Bulíř and Hamann (2006) find that the average volatility of a country’s aid share in GDP is about 40 times higher than that of its revenue share in GDP.
The regressions use a fairly large number of explanatory variables, including economic growth, outcome gaps, commodity prices, political risk, revenues, and past values of aid. To test for robustness, the regressions were run in various permutations, using levels and changes of the variables, and with different estimation techniques. Some of the selected regression results are presented in Table A1.3.
Aid spurts were defined as periods when a country’s aid flows were notably higher than its average aid flows (by 0.75, 1, or 1.5 standard deviation), and then the average aid flows before and after these events were plotted.
Only selected regression results are reported. The initial set of regressions was run with data from the 51-country sample for the period 1990–2004. Adding the political risk variable in the specification significantly reduces the number of observations. The core results discussed in this section, however, hold across both the larger and smaller samples. Overall, the results are strong and survive a battery of controls and robustness tests.