Development aid effects on economic growth:
Comparing loans and grants to total aid flows in IDA-eligible countries
Disaggregating foreign aid effects on economic growth in low-income countriesT. de Bree
Master Thesis International Economics
MSc Economics
August 2018
Author:
Tim de Bree, 11418915
tim.debree@students.uva.nl
Universiteit van Amsterdam School of Economics Supervisor: Ms. N.J. Leefmans Second Reader: Dr. D.J.M. Veestraeten Word count: 15.134
Statement of Originality
This document is written by Student Tim de Bree who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Content
I. Introduction 3
II. Literature Review 7
1. Literature on the effects of aid in general 7 2. Literature on the effects of disaggregate aid 11
III. Data and Variables Description 13
1. Variables 13
2. Countries 18
3. Periodization 19
4. Estimating Grants and Loans 20
IV. Model and Methodology 21
1. The timing and effect of aid 21
2. Early-impact Aid 22
3. Granger Causation 23
V. Results 24
1. All aid flows 25
2. Disaggregate aid flows 27
3. Discussion of Results 32
VI. Robustness Tests 33
1. Model Specifications 33
2. Mean Reversion and Reverse Causation 40
3. OLS assumptions 43
VII. Conclusions & Discussion 46
VIII. References 50
I.
Introduction
For a long time now, both economists as well as public officials have argued over the effects of foreign aid flows on developing countries. This debate has been at the center of attention ever since the foundation of the International Development Association (IDA) in 1960. As part of the World Bank Group, it specifically targets these countries with an extremely low GDP per capita. The current threshold being at an annual GNI of $1211,- per capita, there are at the moment worldwide 77 countries IDA-eligible1. Throughout 2015 the IDA alone had already committed $19 billion of development aid to these countries2. Multiple other institutions and governmental organizations donate similar numbers, amounting to a net total of over $58 billion flowing to IDA-eligible countries in 2015.3
The disbursement of this aid is meant to incentivize developing governments to invest in their economies and spur economic growth. Academic research has however to date not been able to reach a definite conclusion whether this aid, or aid in general, actually succeeds in increasing GDP growth in low-income countries. Clemens et al. (2011) made a first attempt to merge the findings from three seminal papers (Boone (1996), Burnside & Dollar (2001) and Rajan &
Subramanian (2008)) into one conclusion by taking an alternative approach to the datasets used
by the different authors. The paper’s methodology consists of three different steps and comes up with an alternative solution for the endogeneity problem based on econometric techniques and economic intuition. First, using once lagged aid as explanatory variable instead of current aid allows effects of aid to take time and resolves the problem of finding an adequate instrumental variable. Moreover, they control for any existing country-specific effects by using first differences. Thirdly, the authors focus on so-called ‘early-impact aid’, flows that are specifically directed to sectors that are likely to show short-term growth effects. They argue that aid directed towards for example humanitarian projects has a different aim than economic growth. Hence it would follow logically from this line of reasoning that such aid flows do not necessarily lead to growth effects. Limiting to early-impact aid would increase the probability of finding a significant effect through regression analysis. The resulting relation cannot be defined as purely causal, it is closer to Granger causation as first described in Granger (1969). Clemens and his
1
http://ida.worldbank.org/about/borrowing-countries
2 http://ida.worldbank.org/about/what-ida 3
colleagues however argue it is sufficient to know from a policy perspective that aid is likely to be followed by growth in the following years. Using their own methodology, the authors establish a common positive effect of aid on economic growth for all three papers. While Roodman (2014) raises some valid concerns on the methodology used, Clemens et al. (2011) concludes this method of lagging, first-differencing and limiting to early-impact aid is a more effective approach to establishing a significant effect of aid flows on economic growth.
Clemens and his colleagues are not the first ones to take a more in-depth look into differences between aid flows. Hence increasing amounts of attention have been drawn to the different flows underlying general aid. Authors such as Cordella & Ulku (2007) argue that one of the reasons that the academic community has not been able to agree on the impact of foreign aid in general, is that the effects of all types of aid differ hugely.The last two decades saw the introduction of new instruments such as debt cancellation and emphasized the importance of technical assistance, concessional loans and grants as well as combinations of these instruments. Amongst those different flows, the difference in incentives between grants and loans has increasingly generated attention in recent years4, both in policy circles as well as (to a lesser extent) in academic research. All these different types of aid makes one wonder if there is a difference between different types of aid flows and if so which instrument is the biggest contributor to economic growth. Is there a difference between a country’s performance after receiving a grant or after having been borrowed money on well below market-rates (ceteris paribus)? If so what are the factors that determine these differences? Knowing the answer to these questions would allow us to optimize the composition of aid flows and increase its effectiveness.
The public debate has seen many theoretical arguments both in favor of loans and in favor of grants. Opponents of grants point out that giving ‘free money’ to developing nations would remove incentives for the government to look for a sound way of investing in long term growth. Furthermore grants are argued to decrease tax collection efforts (Clist, 2016) and increase contemporary consumption instead of long-term (public) investment in economic growth. Proponents of grants however, have come up with theoretical insights to counter these arguments. In his seminal paper Krugman (1988), for instance, argued that loans remove the incentives for
4
Attention for grants significantly increased after the publication of the Meltzer Report (IFIAC, 2000), a report by an advisory committee from US Congress, that intended to improve international financial institutions and advocated, amongst others, the increased use of grants in foreign aid.
governments to stimulate economic activity, as all benefits arising would have to be transferred to creditors to repay the loans. Even though the debate on grants vs. loans has been ongoing for a while, few papers have taken an empirical approach to provide evidence for actual differences in effects on economic growth, although the effect on tax collection has received some attention. The aim of this paper is therefore twofold: First of all we aim to analyze the effects of total aid on the economic growth using the methodology from Clemens et al. (2011) in a more recent time period for IDA-eligible countries. Clemens and his colleagues conclude that the effects of aid on economic growth become positive and significant when following their methodology. However, the authors as well as Roodman (2007 & 2008) admit that results in the aid-growth literature are very dependent on the time period used. Replicating this methodology in a more recent time period therefore also allows to check the generalization and validity of the results as well as the methodology itself. If the methodology indeed generates positive and significant results in a more recent time period, this provides more robustness to the findings in the Clemens et al. (2011) paper. Moreover, , while the authors of Clemens et al. (2011) were forced to estimate the total aid disbursements per sector for the years up to 2002 based on commitments, the actual disbursements are now available. Furthermore, the further sophistication of aid packages as well as the increase of eligibility criteria for aid could have very well increased its efficiency.5 This would make it more plausible that a positive effect of aid on economic growth becomes apparent in recent data.
Secondly, this thesis aims to look in more detail at the effects on economic growth of various instruments, by further dividing aid flows into loans and grants. The increased availability of data in recent years enables us to look at these types of aid separately. Hence the World Bank and Credit Reporting System6 now make a distinction between grants and loans in aid flows. Not only does it allow us to get a more detailed understanding of development aid mechanisms, it could also be useful in designing future development aid packages.
This thesis hence analyzes aid flows (following the official definition of Official Development Assistance) to 70 IDA-eligible countries over the time period 1996 – 2015. By conducting the research this thesis aims to answer the following research question:
5
Cordella & Ulku (2007) indeed hint that aid has become more efficient over time for their panelset.
6 The CRS is a database from the OECD on aid information. Data are collected on individual projects and
‘What are the differences in the effects of aid in general, loans and grants on economic growth for IDA-eligible countries in the years 1996 -2015?’
In order to answer this research question we divide it into three different sub-questions:
- ‘What is the effect of aid in general on economic growth for IDA-eligible countries in
the years 1996 – 2015?’
` - ‘What is the effect of those parts of development aid defined as loans on economic growth for IDA-eligible countries in the years 1996 - 2015?’
- ‘What is the effect of those parts of development aid defined as grants on economic growth for IDA-eligible countries in the years 1996 - 2015?’
The methodology used follows the approach from Clemens et al. (2011) and draws on recent papers to develop a model including the most often used control variables. Following Clemens and his colleagues, we fix traditional endogeneity issues using lags and first-differencing and limit aid flows to so called ‘early-impact aid’7. The resulting relation does not follow the strict definition of scientific causality, but is more close to Granger causality. Most importantly however, this strategy does not require instrumental variables, as proper instruments for aid are not easily found in the literature. The methodology from Clemens et al. (2011) subsequently applies various robustness tests to exclude any other possible sources of the established effects of aid on growth. We add others from Roodman (2014) to further test the strength of the findings. This thesis finds that total aid flows have a significant, positive effect on average economic growth in IDA-eligible countries over the period 1996 – 2015. Interestingly, opposite to the findings in Clemens et al. (2011), limiting aid flows to early-impact sectors does not generate significant results. The results are however strongly dependent on model specifications. On the disaggregate level, grants do not appear to have any significant effect on growth, neither when limiting to early-impact sectors only. Loans however, do indeed appear to show a somehow significant positive effect of aid flows on economic growth, albeit the link is fragile to some robustness tests and mean reversion.
The remainder of this thesis is structured as follows. In Section II we provide a literature review on the effects of development aid on economic growth in developing nations. Its aim is to provide a critical analysis of the current literature on this topic and emphasize the contribution of this
7 Clemens et al. define early-impact sectors as those sectors where aid can be expected to affect growth within the
thesis to the existing research. Section III shows the data used and outlines the choices made concerning control variables and periodization. Section IV describes the model and methodology used and goes through the different steps of the approach by Clemens et al. (2011). In Section V the results of the regression analyses are presented. First it looks at the effects of overall aid flows on economic growth in IDA-eligible. Secondly, we assess the differences with respect to the impact of grants and loans on economic growth. Section VI tests the results for robustness and finally Section VII presents the main conclusions to be drawn from the analysis, discusses possible policy implications and provides directions for future work.
II.
Literature Review
While there exists an extensive academic literature on the effects of development aid on economic growth, with hugely diverging results and conclusions, the literature on the effects of disaggregate aid flows is minimal. Multiple theoretical arguments can be made (and have been made) concerning the aid-growth relation, but the following section will merely focus on those papers with an empirical perspective since this paper will take a similar approach. The literature on the effects of aid flows on economic growth is generally divided in four generations starting from the 1970s up to the present8. The following section 2.1 describes the evolution of the general aid literature, its most seminal papers and draws attention to the problem of endogeneity within the aid-growth literature, which creates uncertainty on the direction of causality. We end with a literature review of the small literature on disaggregate aid flows in section 2.2.
2.1 Literature on the effects of general aid flows on economic growth
First generation models
The first generation of research was based on very intuitive and simple economic assumptions. Most such models were based on the Harrod-Domar model, where growth results from the level of savings and productivity of an economy. The underlying assumption in these papers is that all aid is consequently invested by the recipient government. One of the first researchers to use a multivariate model to test this was Papanek (1972, 1973). He established, in contrast to most of his contemporaries, a positive effect of aid on growth using the following regression:
ẏ/y = α + β*Aid + χ*η + ε
8
Where ẏ/y shows growth rates in income per capita, η is a vector of constants and χ represents a vector of country characteristics such as the saving rate, imports and education. The analysis includes 51 countries in the period 1950 – 1965. Due to the lack of solutions for endogeneity issues and the small datasets used, the (mostly negative) results of these first generation models are rarely taken into account nowadays. The endogeneity problem is one of the most discussed issues in the aid-growth literature and describes how aid flows might be biased towards countries that are doing well (poorly) and are in a state of growth (recession), hence influencing the perceived relation between the two variables. If indeed it is the case that aid is biased towards countries with specific growth rates (this can theoretically be either negative, low or high growth rates), the effect on growth rates cannot be defined as causal, as economic growth theory dictates that economic growth is correlated with past rates. Hence similar future growth rates in countries receiving aid would not be caused by (the increase in) foreign aid itself, but merely by the natural development of economic growth over time.
Second generation models
Second generation models continued to build on the impact of aid on growth through investment. Their point of view however encompassed a political economy dynamic in which the effectiveness of aid depends on fundamentals of the recipient. Hence these models incorporate variables such as policy environment, political stability, inflation levels and geographic location into their models. Most studies found a positive effect of aid in general on growth. One of the most influential papers of this period, is Mosley (1980) which is one of the first papers to include an instrumental variable to isolate the causal part of the effect of aid and growth. Instrumental variables are an often used technique to correct for endogeneity problems and take the following form:
ẏᵢ,ₜ/yᵢ,ₜ = α + βDᵢ,ₜ + Xᵢ,ₜη + εᵢ,ₜ
Dᵢ,ₜ = Zᵢ,ₜξ + vᵢ,ₜ
Where Xᵢ,ₜ represent control variables,, Zᵢ,ₜ is a vector of exogenous instruments that are both uncorrelated with the error term as well as correlated with the explanatory variable, and ξ is a vector of constants. Similar to the first generation, these second generation studies provide mixed
evidence on the aid-growth relation. All studies from this generation are however perceived to be biased by short time periods and small samples of countries. (Gupta and Islam, 1983)
Third generation models
Around the turn of the millennium, third generation models evolved along various new dimensions, due to the availability of bigger panel data sets. Boone (1996) was one of the first to use an elaborate dataset, covering 96 countries with yearly data between 1971 and 1990. In this seminal paper he finds no significant effect of aid. Another paper worth mentioning here is the contribution of Burnside & Dollar (2000) for their influence on policy. They analyze 56 countries from 1970 to 1993 for the effects of aid on economic growth conditional on institutional and political conditions. They conclude that: “aid has a positive impact on growth in developing countries with good fiscal, monetary and trade policies … [but] … in the presence of poor policies, aid has no positive effect on growth.” Hansen & Tarp (2000) were among the first researchers to include a quadratic term for aid to account for any possible non-linear returns to aid. Overall Roodman (2007) finds that the results of these third generation models are extremely sensitive to methodological choices and generate different results when changing for example the period covered. Furthermore he shows that the instruments used by both Hansen & Tarp (2000) and Burnside & Dollar (2000) are likely to be biased by country-specific characteristics. He concludes “aid is probably not a fundamentally decisive factor for development.”
Recent models
The most recent generation of models can be characterized by widely differing assumptions, econometric techniques and consequently diverging results. These models all look at the effects of aid flows directly on economic growth. However, they differ in their variable choice and country characteristics, whereas some also incorporate time fixed effects. Rajan & Subramanian (2008) is one of the most cited papers from this generation. They rely on a combined instrumental variable based on population size and donor country characteristics such as colonial ties. In their research covering 93 countries they find no evidence of aid having any effect on growth over different periods of time, neither conditional on better policy or geographical environments nor depending on the type of donor. As a possible explanation for this result they state the data could contain large amounts of noise, which would make it difficult to find the effect on economic growth and/or establish a significant relation. Alternatively, they argue the effects of aid on
growth could be significantly smaller than theory suggests, making the effects on a general level hard to discern with current models. Sumner et al. (2015) provide a very concise meta-analysis based on the most recent papers. They conclude aid works, however only in specific situations, depending on, for example, the level of aid, domestic political institutions, the composition of aid as well as the volatility and fragmentation of aid. In a very influential paper, Clemens et al. (2011) review and remodel the methodology of the most seminal recent papers, that is Boone (1996), Burnside & Dollar (2000) and Rajan & Subramanian (2008), maintaining the original datasets and variables. The authors give three important reasons for both the difference in results between past papers and the failure to distinguish a significant effect on economic growth. First of all they question the assumption that current aid directly impacts economic growth, without allowing for the flows to take time to generate economic effects. They hence use (a) lagged variable(s) for aid to allow aid to affect growth with a time lag. Secondly the authors argue the literature contains no valid instrumental variables. They show for example that most of the explanatory power of the IV used by Rajan & Subramanian (2006), came from the population part of the variable, which has been proven to influence economic growth by Bazzi & Clemens (2009) and therefore violates the required exogeneity assumption. Therefore Clemens et al. (2011) focus on alternative solutions and choose to, besides lagging the aid variable, also include first differences in the regression, simultaneously controlling for fixed country characteristics. Lastly, the authors restrict the aid variable to those aid flows directed to so-called early-impact sectors. They argue aid directed to these sectors aims to generate short-term growth and hence effects should be more apparent. Early-impact aid covers sectors such as infrastructure or agriculture, whereas aid for social sector investments or administrative costs is excluded. Following most papers in the aid-growth literature, Clemens et al. (2011) use a multivariate regression for their methodology defined as:
ẏ/y = β0 + β1*Aid + χ*η + ε
As explained above, Clemens et al. (2011) do not come up with new datasets or (control) variables, but merely use those from the papers they intend to improve and subsequently apply their own methodology. Hence depending on the underlying paper ẏ/y shows either GDP or GNI growth rates, χ represents a vector of country characteristics such as budget balance, policy quality or population and η is a vector of estimators with ε being the error term. Clemens and his co-authors find that the conclusions from the three papers change drastically when they apply
their own methodology. Boone (1996) and Rajan & Subramanian (2008) originally did not find aid to have a significant or positive effect on economic growth and Burnside & Dollar (2000) found it to be conditional on good policy. Clemens et al. (2011), however, find a significant, persistent and positive effect of aid on economic growth for all three papers even though the respective country selections, time periods and variable selections differ. Interestingly, Roodman (2014) reproduces the Clemens et al. (2011) paper, reviews its methodology, proposes improvements and finds limiting to early-impact aid does not always generate a positive, significant result. The methodology will be explained in more detail in section IV.
2.2 Literature on the effects of disaggregate aid
Clemens and his colleagues are not the first researchers to disaggregate aid flows to try to define a clearer picture of the effect of foreign aid on economic growth. Mosley (1987) was one of the earliest to look at aid flows on different levels and established the so-called micro-macro paradox. He finds a difference in the effectiveness of aid, where foreign aid generates economic growth on a micro-level but does not appear to have a significant effect on the macro-level. Therefore this paper is one of the first to look at aid flows on different levels. Mosley considers three possible explanations for this hypothesis: (i) errors in the data; (ii) switching of expenditure within the public sector to more economic growth-enhancing spending; (iii) indirect effects of aid on the private sector. The paper finds some evidence of the second and third type of effect. More recently, Minoiu & Reddy (2009) use a panel data approach and divide total flows in aid for developmental and non-developmental purposes. With this clear distinction they find a robust, positive effect for developmental aid on growth. Rajan & Subramanian (2008) also distinguish between different aid flows, albeit on the donor side (multilateral, bilateral etc.), when researching the effects of aid on economic growth. They, however do not find any significant effect on GDP growth after disaggregating aid based on donor side specifications.
The disaggregation of aid leads us to the discussion concerning grants and loans and the possible different effects on economic growth. Disaggregating aid flows based on the underlying financial conditions seems a natural extension to the same line of reasoning as dividing foreign aid into micro and macro levels. Authors such as Odedokun (2004), and Clements (2004) have looked at the effects of loans and grants on government investment and tax collection. The literature on the direct effect of grants and loans on economic growth is however limited. In a theoretical
experiment using a simple two-period mathematical model, Cohen et al. (2007) find that a combination of loans with debt forgiveness generates the highest effects on growth. Using another approach Dovern & Nunnenkamp (2007) study 124 countries over the period 1960 – 1994. They perform an analysis of aid both in general, as well as on disaggregate levels. These disaggregate levels distinguish between different aid categories to account for heterogeneity of aid and divide short-impact aid from long-impact aid as well as separating grants from (concessional) loans. The control variables included are economic reforms, openness to trade, the political environment, geographic location and the mortality rate of infants. The authors find consequently that neither loans nor grants on average increase the probability of a country seeing its growth rate accelerate, both for contemporaneous as well as for lagged aid data. The results are however highly sensitive to the definition of the probability of growth acceleration. Cordella
& Ulku (2007) are amongst the few authors to have generated a regression analysis of the impact
on GDP growth, studying an unbalanced panel of 62 low-income countries over four-year periods in the timeframe 1976 - 1995. The paper tests the extent of concessionality of loans and its effects on economic growth. They find that a higher degree of concessionality is associated with higher growth. Control variables included are similar to those in the traditional aid literature and include economic development, policy quality, institutional level, terms of trade, life expectancy, levels of foreign debt and a civil war dummy. Overall Sumner et al. (2015) conclude in their meta-literature study on the loans vs grants debate that past results are simply too divergent to reach a final conclusion concerning the effects on economic growth and further work is required.
It is therefore clear from the above overview that an elaborate, regression-based analysis, combining the past findings in the literature on the effects of aid on economic growth with further research on disaggregate aid flows on the level of loans and grants, both contributes to the development of current academic literature and has its clear relevance for modern policy choices.
III.
Data and Variables Description
The following chapter describes the variables, trends and datasets used in the empirical part of this study. We analyze 70 IDA-eligible countries over a time period of 20 years from 1996 to 2015 both on an general aid level and on level of loans and grants. Multiple control variables are used, based on their occurrence in previous papers and relevance. Moreover, we generate estimates for the data on grants and loans for those years where data is missing.
3.1 Variables
Dependent & independent variables
The dependent variable used in this thesis to proxy for economic growth is the growth rate of GDP per capita level, as defined by the World Bank’s World Development Indicators. While some studies use the GNI per capita ratio or look directly at investment levels, Clemens et al. (2011) show that the choice of the dependent variable only influences the level of the estimator, not its significance or sign. We are, however, mostly interested in establishing the direction of the relation instead of finding the exact size of the effect., The ratio of ODA (Official Development Assistance as defined by the DAC) over GDP will be used as proxy for development aid9, conforming to the approach from Clemens and his colleagues and the majority of previous papers. ODA comprises all development aid from multilateral and bilateral sources. The disaggregate variables are made up of those parts of ODA that the DAC characterizes as grants and loans as a percentage of GDP.10 As explained above the segmentation into early-impact aid and non-early-impact aid follows the definition of Clemens et al. (2011). Hence aid flows that might not be expected to cause growth in the short-term are excluded from the early-impact limitation. A more precise explanation of the definition of early-impact sectors can be found in section 4.2.
Control variables
Previous papers throughout the academic literature have used a multitude of different control variables. According to Roodman (2007), this is one of the explanations for the diverging results throughout the years. This thesis takes a two-sided approach to reach both comparable as well as robust results. Hence on the one hand we compare the control variables included in the three studies used by Clemens et al. (2011), as shown in Table 1. On the other hand we test for robustness of the results by rerunning the analysis with different control variables. As it turns out the direction, standard errors and robustness of the results do not vary significantly with the inclusion of other variables. Clemens and his colleagues graciously provided their dataset,
9
It should be noted that all regressions that use early-impact ODA also include Repayments/GDP and (Repayments/GDP)², since early-impact aid is a gross flow while total ODA is a net flow.
10
The total of loans and grants does not add up to the total of development aid as total ODA contains corrections and includes equity investments that comprise direct financing of enterprises in a developing country that do not (as opposed to direct investment) imply a lasting interest in the enterprise. See Appendix 9.3 for the exact definitions.
sources, regressions and a technical appendix to the public11, allowing this thesis to recreate and use the same variables.
Table 1: Covariates used in aid studies
Source: Clemens et al. (2011)
Overall there exists quit some overlap between the variables included in each of the three studies from Clemens et al. (2011), or at least the channel they aim to proxy for. Hence the ‘budget balance/GDP’ variable in Rajan & Subramanian (2008) and ‘Policy’ variable in Burnside &
Dollar (2000) both aim to describe the governmental economic policy quality. Other covariates
such as ethnic fractionalization or geographic location are neither significant in any of the above studies nor do they turn out to have a significant impact when added to the regression used in this thesis. Previous studies from academic literature had found similar results that differences in geography do not have explanatory power. Notwithstanding, we control for static country characteristics using first differences in order to get rid of any existing country fixed effects. The assassinations variable is part of a protected database and therefore we proxy for this by adding the ‘No Violence’ Indicator from the World Governance Indicators instead. Compared to the above, we change one variable: While most authors create some synthetic variable for institutional quality based on different indicators (rule of law, corruption control and bureaucracy quality), we include them separately to avoid discussions on assigned weights for each indicator. While this could create multicollinearity problems, a test shows this is not the case; the returning VIF values are all around 1. Based on the above selection process the main OLS regressions for this study takes the following form:
11 All data used in the paper by Clemens et al. (2011) can be downloaded at aiddata.org. Hence this thesis used the
General aid regression:
ΔGRGDPCi,t = α1ΔAIDi,t-1 + α2ΔAIDi,t-1² + α4ΔPOPi,t + α5ΔBURi,t + α6ΔRULEi,t +
α7ΔCORRi,t + α8ΔBUDBALi,t + α9ΔMSUPi,t + α10ΔINFLi,t + α11ΔNOVIOi,t + Δεi,t
Disaggregate aid regressions on loans and grants:
ΔGRGDPCi,t = α1ΔLOANSi,t-1 + α2ΔLOANSi,t-1² + α4ΔPOPi,t + α5ΔBURi,t + α6ΔRULEi,t +
α7ΔCORRi,t + α8ΔBUDBALi,t + α9ΔMSUPi,t + α10ΔINFLi,t + α11ΔNOVIOi,t + Δεi,t
ΔGRGDPCi,t = α1ΔGRANTSi,t-1 + α2Δα1ΔGRANTSi,t-1² + α4ΔPOPi,t + α5ΔBURi,t +
α6ΔRULEi,t + α7ΔCORRi,t + α8ΔBUDBALi,t + α9ΔMSUPi,t + α10ΔINFLi,t + α11ΔNOVIOi,t + Δεi,t
Where GRGDPC is the dependent variable, AID/LOANS/GRANTS represent the independent variables of interest (in percent of GDP), the remaining variables are covariates and ε represents the error term. The covariates used are the following: POP = population, BUR = bureaucracy quality, RULE = Rule of Law, CORR = control of corruption, BUDBAL = budget balance (as a percentage of GDP), MSUP = money supply (as a percentage of GDP), INFL = inflation, NOVIO = no violence. The exact definitions as well as their sources can be found in Appendix 9.3.
Furthermore all variables are eventually first-differenced and the independent variables are both lagged and squared. The argumentation for these changes can be found in the next section. The results section will present both the regression output of these final regressions as well as the output from intermediate steps in the methodology of Clemens et al. (2011), (i.e. lagging, first-differencing, squaring and limiting to ‘early-impact aid), to enable a discussion of the added value and impact of each intermediate step.
Description of main variables
This section describes the main variables and their development over time12. We look at the variables GDP per capita growth, total ODA, (early-impact) grants and (early-impact) loans. Both the loans as well as the total aid variables show signs of a trend. However, first-differencing should remove any negative effects arising from non-stationarity. Moreover an Im-Pesaran-Shin test for unit roots (Im–Pesaran–Shin, 2003) finds evidence of stationarity in all main variables.
12
GDP per capita growth: Figure 1 shows the development of GDP per capita growth on a yearly
basis for the period 1996 – 2015. Pictured are the averages per continent as well as the average for all countries. Europe and Asia show the highest growth rates, average growth rates ranging between 0 – and 5%. The spike in GDP-growth in Europe around 2001 is explained by the fact that the dataset only includes 2 countries from Europe (Kosovo and Moldova).
Interestingly, (continental) averages do not fluctuate as much as the growth rates on an individual country’s levels. The main breaks in the years 2008/2009 and 2014/2015 can be explained by the aftermath of the Global Financial Crisis and the European Sovereign Debt Crisis.
Figure 2A Figure 2B
Total ODA: Figures 2A & 2B portray the development of total ODA flows as a percentage of
GDP over time. Relative foreign aid flows decrease significantly in the 1990s after which remains a light declining trend in the following years. Countries in Oceania receive relatively most aid, as most of these countries included in the dataset are small island states with low
GDP-levels. Furthermore we see a huge spike in early-impact ODA around the year 2006, before the Global Financial Crisis hit the developed world. 13
Figure 3A Figure 3B
(Early-impact) Loans: The amount of (early-impact) loans as a percentage of GDP decreases
steadily throughout the period, as shown in Figure 3. While relative flows are comparable between continents, Asia and Oceania receive relatively fewer loans than other continents. The flow of early-impact loans follows a similar development to that of overall loans, it seems however somewhat more volatile.
Figure 4A Figure 4B
(Early-impact) Grants: Figures 4A & 4B shows the relative amount of grants received by
countries as a percentage of GDP and the percentage of early-impact grants. Similar to figure 2 displaying total ODA, Oceania receives relatively more grants than other continents due to their
13
It is important to note here that Total ODA, loans and grants are net flows, whereas all early-impact flows are gross flows. In some years this causes early-impact flows to be higher than total flows due to higher repayments in that specific year. Unfortunately, the CRS and OECD do not allow to yet to filter gross flows and repayments on a sectoral level. As we are not interested in the size of the effect but the significance, this is of minor importance.
small size. Most striking is the steady increase in % Grants after the year 2000 (supposedly following the Meltzer Report) and the spike in 2006 for Latin-America and Africa before the Global Financial Crisis hit the world economy.
3.2 Countries
This study focuses on 70 out of the 77 countries countries eligible for IDA development assistance14, as development aid in general is mostly oriented towards these low-income countries. For the seven remaining IDA-eligible countries, insufficient data was available. The three papers studied in the Clemens paper use respectively 96 (Boone (1996)), 56 (Burnside &
Dollar (2000)) and 93 countries (Rajan & Subramanian (2008)) in their sample. The authors do
not give any explanation in their papers for the specific country selection; it is likely to be related to data availability for those early years. The majority of the samples overlap with our selection and all countries under investigation can be characterized as developing economies. The risk of selection bias should therefore be sufficiently small. Additionally, Clemens et al. (2011) argue in their paper that their results hold, irrespective of the country selection under investigation. Investigating the effectiveness of foreign aid in IDA-eligible countries is extremely important as the inflows represent a big proportion of the available budget to these countries, which makes the effectiveness of these flows crucial (e.g. foreign aid for Liberia amounted in 2008 and 2010 to more than 100% of the country’s GDP). Unfortunately insufficient data was available Uzbekistan, Somalia, Sao Tome and Principe, Kiribati, Afghanistan, Tuvalu and the Maldives to include these IDA-eligible countries in the analysis. Countries included are dispersed in their geographic location (e.g. Papua New Guinea in South-East Asia and Moldova in the Balkan) and have different sizes (ranging from Nigeria to island states such as Vanuatu). Hence the sample should be balanced enough to allow for some generalization of the results.
3.3 Periodization
The two main questions regarding periodization concern the time frame to study and the frequency of variable measurement. The question of when to test for growth is mentioned throughout the entire growth literature. Short periods decrease the bias from omitted variables that change slowly over time and allow for first-differencing or fixed-effects (Islam, 1995).
14
However, according to Barro (1997) the shorter the time periods, the higher the probability of the estimators to be biased by measurement error or misspecifications. Longer periods however are more prone to be influenced by omitted variable bias and reverse causation, while allowing for long term growth effects to appear (Clemens et al., 2011). Hence Durlauf and Quah (1999) warn against short periods. Islam (2003) agrees with Temple (1999) that the use of panels is often the best way forward’. As both periodization approaches seem to have clear advantages and disadvantages we choose to stick to the time period where the best data is available, while keeping above pitfalls in mind. The OECD and CRS started collecting and distinguishing between aid data on sectoral and concessional level from 1996 onwards. Hence this naturally generates a scope of 1996-2015. Regarding the frequency of data, the underlying aid variables are usually measured on a yearly basis. Most papers within the aid-growth literature, as well as
Clemens et al. (2011), however aggregate these yearly measurements to create periodical
averages of either 4, 5 or 8 years. The underlying reasoning is that this differences away any effects on growth due to the business cycle of an economy. Analysis of our dataset however shows that there is no sign of a constant business cycle effect in the data that could bias the results as illustrated in Figure 5.
Figure 5
The trends and breaks in figure 5 can be clearly explained by the global financial crisis from 2007/2008 and the European Sovereign Debt Crisis from 201415 and show no obvious signs of a business cycle. Considering the limited data points available, averaging over multiple years would seriously decrease the degrees of freedom for our regression analysis. Hence this thesis
15
Including a dummy variable for the crisis years 1997/1998 (Asia Crisis), 2007/2008 (Financial Crisis) and 2014 (European Sovereign Debt Crisis) does not alter the results significantly. Neither is the dummy variable itself significant in the regression.
will rely on yearly data and therefore use a slightly different approach to periodization compared to Clemens et al. (2011).
3.4 Estimating Grants and Loans Disbursements
One challenge arising from above choices is that for the years 1996 - 2001 the OECD and the Creditor Reporting System do not report for loans and grants disaggregate disbursements on a sectoral level but only disaggregate commitments. Luckily, Clemens and his colleagues encounter a similar problem in their research and we can therefore learn from their approach to solve this issue. Hence at their time of writing, there existed a gap in the OECD data on an overall aid level for disaggregate disbursements by purpose. Since then the database has been improved and the gap is limited to the grants/loans purpose specific level while overall aid disbursements are now included. Their approach for filling this gap is using information obtained from purpose-disaggregate commitments to estimate disbursements on a pro rato basis. The following example serves as illustration: To calculate the early-impact loans disbursed to Zambia in 1999, we take the early-impact loans committed to Zambia in 1999. This number is divided by the total amount of loans committed to get the ratio of early-impact loans commitments to total loans commitments for Zambia in 1999. This number is then multiplied with the total number of loans disbursed to Zambia in 1999 to give an estimate of the early-impact loans disbursements to Zambia for that year. This procedure is repeated for every country and every year for both loans and grants to complete the dataset. As explained in the Clemens et al. (2011) paper, this method is attractive from a theoretical perspective as it is reasonable to assume that the share of aid committed is representative for the share of aid actually disbursed, even within a purpose or sector. Furthermore, in the periods where the actual disbursements are known, these are highly correlated with the disbursements levels estimated through the above method.
IV.
Model and Methodology
This thesis is based on the methodology and logic used in the Clemens et al. (2011) paper. The same methodology is used to investigate the effects of aid both on an general level and on the level of loans and grants. A sound methodology is of utmost importance as previous research16 hinted that the lack of clear, significant results is likely to be due to noise in the model or the
16
effect being relatively small and difficult to distinguish. As mentioned earlier, Clemens et al. (2011) graciously provided a concise technical appendix to complement their research, allowing us to be fairly confident about recreating the same methodology in this thesis. The following chapter introduces the different steps within the methodology as applied in this thesis. We start with the explanation of the timing and effect of aid. Consequently, we describe what Clemens and his colleagues coin ‘early-impact aid’ and finish with the definition of Granger causation.
4.1 The timing and effect of aid
Clemens et al. (2011) argue that a possible explanation why previous research has not found a
clear positive effect of general aid on growth is that ‘some aid is aimed at activities whose growth impact has weak theoretical basis within the time periods used in the panel’. Hence the effect of certain flows on economic growth, if spurring growth at all, might appear only long after actually receiving foreign aid (e.g. on education projects). Allowing the aid variable to take on lagged form(s) can solve these issues. The question of how to solve the bias from omitted variables and country fixed effects has already been mentioned before. In order to answer this question, it must be taken into consideration that we use a (relatively) short period for our analysis, namely 1996 – 2015. Nickell (1981) describes the disadvantages of using fixed-effects in autoregressive models with on the one hand a limited amount of years and on the other hand both a large number of control variables and countries, causing the estimators to be biased. Hence given the small timeframe, Clemens and his colleagues choose the more straightforward approach of first-differencing. It should be kept in mind however, that first-differencing decreases the number of time-periods, possibly impacting the significance of results. In summary, the approach of
Clemens et al.(2011) consists of using first-differencing to allow country fixed effects to be
differenced away. Furthermore they allow for the possibility for the growth effect of foreign aid to surface with a time lag by including (multiple) lags in the regression. It is argued this also removes any possible impact of reverse causation from higher (or lower) growth leading to higher (or lower) levels of foreign aid. This combined approach therefore offers an alternative solution to the endogeneity problem, while avoiding poor-quality instrumental variables. Following Hansen & Tarp (2000), the model also includes a quadratic term to allow for non-linear, diminishing returns. This leads to the final regressions in this thesis taking the form:
General aid regression:
Δẏt/yt = Δβ1*Aidt-1 + Δβ2*Aidt-1² Δχ*ηt + ε
Disaggregate aid regressions on loans and grants:
Δẏt/yt = Δβ1*Loanst-1 + Δβ2*Loanst-1² Δχ*ηt + ε
Δẏt/yt = Δβ1*Grantst-1 + Δβ2*Grantst-1² Δχ*ηt + ε
Where Δẏ/y shows changes in growth rates, χ represents a vector of estimators and η is a vector of control variables with ε being the error term.
4.2 Early-impact Aid
As mentioned in prior sections, different aid flows can have varying impacts according to their purposes and area of use. Hence not all foreign aid carries the objective to promote economic growth. Clemens et al. (2011) therefore argue that the aid variable should be restricted to those portions of aid that can be expected to cause economic growth during the time period under study, calling this type of aid ‘early-impact aid’. Limiting aid flows to early-impact aid would therefore allow the effects to become more apparent and lead to significant results. The challenge hence arises to limit the aid variable to this ‘early-impact aid’. The database from the Creditor Reporting System17 allows for this distinction through the possibility to distinguish between different types of aid based on the recipient sector. Following Clemens et al. (2011) early-impact aid hence includes project aid disbursed for real sector investments such as infrastructure or to directly support production in transportation (including roads), communications, energy, banking, agriculture and industry. It excludes any aid flow that clearly and primarily funds an activity whose growth effect might arrive far in the future or not at all, such as for example most social sector investments, all humanitarian aid and donors’ administrative/overhead costs and
17
Both the World Bank and Creditor Reporting System have significantly improved and enlarged their database, since the publication of the Clemens paper. While the authors of the initial paper were forced to estimate the disbursements per sector, the actual disbursements are now available. Therefore we can use the actually disbursed amounts of aid for both the aggregate as for (most of) the disaggregate flows.
expenditures on promotion of development awareness.18 If there is any ambiguity within the data on what is being funded we leave it in ‘early-impact aid’ as it might affect economic growth.
4.3 Granger Causation
After the aforementioned steps we are left with an effect of aid on growth that cannot undoubtedly be described as causal. Most papers in the aid-growth literature focus on establishing a truly causal effect on economic growth. Instead Clemens and his colleagues introduce the notion of Granger causation, named after the seminal paper by Granger (1969). Granger causation occurs when an independent variable helps to (partly) predict the future values of a dependent value (Toda and Phillips, 1994). Even though this type of causation does not meet the strict requirements of scientific causation, it can be very valuable to a donor country. Hence from a policy’s perspective it is sufficient to know whether aid receipts are likely to be followed within several years by any change in economic growth. Furthermore proving Granger causation is generally easier compared to scientific causation. While it is possible for non-causal mechanisms to lead to Granger causality, Clemens et al. (2011) argue that it is very unlikely that these mechanisms occur and we test for it in through robustness checks in section 6.2.
V.
Results
The following section describes the results of the different regressions analyzing the effects of foreign aid on economic growth in IDA-eligible countries for the period 1996 – 2015. The main objective is to establish the direction of the effect of both aid in general and/or loans and grants on economic growth. The actual size of the effect is considered to be of secondary importance as it is likely to be influenced by the model specifications, making it difficult to compare in detail the level of the estimates to other findings within the literature. Similarly to Clemens (2011), the results are presented gradually to show each intermediate step. The first column shows the results of a regression without lagging, first-differencing, or restricting to early-impact aid. Each subsequent column adds one of the advocated changes to the regression.. The last column therefore always shows the results of the regression including all three methodological steps of lagging, first-differencing and restricting to early-impact sectors. For a clear overview of the estimated aid coefficients variables the tables below do not include the estimated coefficient for
18
control variables. All control variables presented in section 3.1 where however included in the regressions.19As proposed by Hansen & Tarp (2000), we also include the results of running the regressions including a squared term. We start with the results for all aid flows, both on a total level as well as restricted to early-impact aid, and compare them with the findings from the paper of Clemens and his colleagues. Consequently we analyze the differences in the findings with respect to loans and grants. The findings suggest, that aid in general does, on average, lead to economic growth in the period 1996 – 2015, for our selection of countries. Interestingly, total aid, after limiting to early-impact aid, does not seem to have a significant effect on average economic growth, contrary to the results from Clemens et al. (2011) for the period 1960 - 2005,. Furthermore, we find no evidence of (early-impact) grants having an effect on GDP growth for the period. Loans, on the other hand, do seem to be followed by higher economic growth, albeit with diminishing returns. The section ends with a brief discussion of the results.
5.1 All aid flows
Table 3 below shows the results of the regressions for the effects of all aid flows to on GDP-growth. The first column includes the results of a simple regression of GDP growth on foreign aid and the chosen control variables. In parentheses are the standard errors. The null hypothesis in this case posits the estimator for aid, β1, to be equal to 0. Hence if we can reject this hypothesis it
would indicate foreign aid indeed is followed by economic growth in the broadest sense. Following the logic of Clemens et al. (2011), columns (1) & (5) are likely to be biased by endogeneity, due to the lack of controls for country fixed effects. These columns are however included to indicate the starting point for each different step of the methodology and allow for a comparison throughout the stepwise approach. Column (2) show the results of the regression after replacing the aid variable by the lagged aid term, in column (3) the entire regression is first differenced whereas in column (4) the aid variable is restricted to early-impact aid. Columns (5) to (8) replicate the regressions and additionally introduce a squared term to allow for non-linear effects on economic growth. Therefore column (8) shows the results of the final regression as presented in section 3.1.
19 An entire overview of the regression results including the coefficients of the control variables can be found
Table 3A: All aid flows linear regression results (1996 – 2015)
Table 3B: Quadratic regression results
Notes: Standard errors in parentheses. Dependent variable is GDP per
capita growth. The control variables and sample size are held constant throughout each step. The only change throughout each step is made to the dependent variable. * p<0.1, ** p<0.05, *** p<0.01.
The results show a significant estimator for total aid on economic growth after controlling for endogeneity through lagging and first-differencing in column (3) and column (7). This suggests aid does have a positive impact on economic growth for the period and countries studied. The squared term in column (7) is significant and negative, hinting at decreasing returns to aid. Moreover, the coefficient for aid increases throughout columns (1) to (3) and columns (5) to (7). Hence it seems the effect does indeed become more apparent by conducting the methodological steps and removing possible endogeneity or time-invariant country characteristics. However, ultimately, neither in the linear (column (4)) nor in the quadratic regression (column (8)) does the effect become significant. As the null hypothesis for the estimator of foreign aid cannot be
rejected, this suggests early-impact aid is not followed by economic growth. The coefficient for aid in column (7) of 0.114 is within the same order of magnitude as the estimators in the Clemens
et al. (2011) study. Given the significant and negative quadratic estimator, the turning point for
the effectiveness of total aid following from this regression would be approximately 1.01%. This turning point indicates which amount of aid generates the highest impact on GDP-growth. This represents the extremum20 of the parabola given by the coefficients on AID and AID² and can be calculated as -b/2a or -(βAid)/ 2*(βAid²). As the negative estimate indicates decreasing returns to
scale, the positive impact on GDP-growth subsequently decreases after this turning point and eventually turns negative for extremely large portions of aid. Roodman (2014) additionally suggests to look at the maximum impact of total aid, which is given by the y-coordinate of the parabola extremum and can be calculated as -b²/4a or -(βAid)²/ 4*(βAid²). We find aid flows are
on average followed by a maximum additional 0,057% in GDP-growth. A Wald test on the significance of the turning point returns significant results. In summary, we find total aid on average does seem to have a positive and significant effect on economic growth in the period 1996 – 2015 for our selection of countries. Furthermore, the significant squared term hints at decreasing returns to scale. Surprisingly the three-step methodological approach does not lead to a significant effect. In this case, restricting to early-impact aid does not generate significant results in the time period 1996 – 2015 for IDA-eligible countries. This contradicts conclusion from Clemens et al. (2011) that growth effects are more likely to arise when considering ‘a subset of aid that does not include aid flows whose growth effect is most likely to arrive decades in the future, or never.’
5.2 Disaggregate aid flows
The next sections dig deeper into foreign aid flows and aim to answer the question whether grants and/or loans do have a significant effect on economic growth and whether there are any differences in these effects. The existing literature21 suggests aid in the form of grants or loans could result in different economic behaviors. If this also affects the levels of consumption and/or investment of aid differently, we should logically see a difference in the effects on economic growth between the two forms of aid. Indeed any existing difference in effects could potentially
20
Following Roodman (2014) we define the coordinates of the extremum of the parabola y = ax² + by as (-b/2a, -b²/4a).
21
explain why the academic community has not been able to reach a definite conclusion on the effects of aid in general on economic growth. We find that grants on average do not have any significant effect on growth whereas loans do seem to have a positively effect, both in general as for early-impact sectors.
Grants
Table 4 presents the effects of limiting foreign aid flows to those aid flows that the DAC considers to be grants.22 Similar to the above tables, it shows the intermediate results from the stepwise methodology used in this thesis and omits the estimators of the control variables for readability. Columns (3) and (7) show the results for the lagged and first differenced total grants, whereas columns (4) and (8) present the results of limiting these flows to early-impact aid. The null hypothesis tested is whether the estimator for grants, β1, is significantly different from zero.
Hence, once again, rejecting this hypothesis would indicate grants are likely to generate economic growth. Columns (1) to (4) show the results of the linear regression and columns (5) to (8) replicate these regressions including a squared term.
Table 4A: Grants disaggregate flows linear regression results (1996 – 2015)
22
Table 4B: Quadratic regression results
Notes: Standard errors in parentheses. Dependent variable is GDP per
capita growth. The control variables and sample size are held constant throughout each step. Only change throughout each step is made to the dependent grants variable. * p<0.1, ** p<0.05, *** p<0.01.
First of all, most importantly, none of the estimated coefficients of interest appear to be significant at the 10% level, neither in the linear nor in the quadratic regression. We find a significant result (10% significance) in column (2), however, this result is likely to be biased by endogeneity due to not controlling for country fixed effects. This would imply that grants, on average, do not have any significant effect on the economic growth of IDA-eligible countries for the chosen timeframe. Furthermore, interestingly, the coefficient on grants does not significantly increase (it decreases even in the linear regression) after first-differencing.. Simultaneous causation by country fixed effects therefore does not seem to be an important determinant of the aid coefficient for grants .23 Most striking however, are the extremely low coefficients for grants when limiting the flows to impact aid. In conclusion, neither grants in general nor early-impact grants do seem to have an effect on economic growth.
Loans
Table 5 presents the results of regressing loans on economic growth for IDA-eligible countries in 1996 - 2015.24 Once again the null hypothesis tested is whether the estimator for loans is
23
Clemens et al. (2011) argue that an increase in the estimator between column (2) & (3) shows signs of country-specifics influencing the former regression.
24 It is important to note here that grants and loans do not add up to total foreign aid as the OECD & DAC use
significantly different from zero and hence whether we can distinguish any (positive) effect of foreign loans on economic development. Columns (1) to (4) display the linear relation and columns (5) up to and including (8) show the quadratic form.
Table 5A: Loans disaggregate flows linear regression results (1996 – 2015)
Table 5B: Quadratic regression results
Notes: Standard errors in parentheses. Dependent variable is GDP per
capita growth. The control variables and sample size are held constant throughout each step. Only change throughout each step is made to the dependent loans variable. * p<0.1, ** p<0.05, *** p<0.01.
Above tables outline an interesting image of the effect of loans on economic growth. Contrary to grants, (early-impact) loans do appear to lead to an increase in economic growth. Hence, according to the last step in column (8), an increase of 1% in early-impact loans as a percentage of GDP would be followed by a 0.345 percentage point increase in GDP growth. This result is direct financing of enterprises in a developing country that do not (as opposed to direct investment) imply a lasting interest in the enterprise. See Appendix 9.3 for exact definitions.
broadly speaking within the same order of magnitude as found in previous studies on general aid. However, the analysis seems to require a deeper look. Hence, first of all, the estimator for loans changes significantly between the different columns, both throughout the different steps of the methodology as well as in comparison between the two tables. According to Clemens and his colleagues such differences between steps in the methodology indicate that the initial estimators are likely to be affected by both reverse causation between aid and growth as well as simultaneous causation by country specific, time-invariant characteristics. Secondly, the columns (1) up to (8) show three significant results for the loans dependent variable above the 95% confidence interval. The estimator in column (3) shows a confidence level of 99% and is the result of the lagged and first-differenced regression using all concessional loans before limiting to early-impact sectors. The coefficient in column (4) does not differ much in order of magnitude compared to the previous step, it only decreases in significance level. Hence limiting to early-impact aid does not necessarily make the aid-growth relation more apparent in this case. The third significant estimator for loans, at a 95% confidence level in column (8), is the result of the regression when we not only include a one-year lag and first-differencing, but also allow for a squared term and restrict to early-impact loans. Interestingly the quadratic relation does only become significant in this final methodological step. Therefore this would imply that restricting the aid variable to early-impact aid does indeed sometimes add value. Furthermore, the squared term in the last step, as shown in column (8), is also significant, albeit at a 90% confidence interval. 25 As it has a negative coefficient, this would imply diminishing returns to loans, similar to previous findings in the aid-growth literature and theoretical arguments. 26 As explained in the section on all aid, the resulting turning point is given by the extremum of the parabola and defines the point of maximum impact of early-impact loans on economic growth. The significance can be tested through another Wald test. This Wald test shows a joint significance level of more than 99%. The extremum indicates foreign loans would be efficient in generating economic growth up to a level of 0.226% of GDP, which seems relatively low given the average levels of past development aid. The suggested maximum impact of aid resulting from the results in column (8), would be an increase of 0,423% in GDP growth. Another Wald test on the
25 Clemens et al. (2011) consider coefficients with a statistically significance level of 10% to be relevant as well. 26
