The Effect of Foreign Aid on per capita Growth through
the Government Investment Channel
Stef van Vuuren (10578250) Bachelor Thesis Economics Faculty of Economics and Business University of Amsterdam Amsterdam, June 2018 Supervisor: Ms.N. J. Leefmans Word Count: 8026
Statement of originality
This document is written by Stef van Vuuren, 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 contents
1. Introduction 2. Literature review 2.1 Influential Studies Concerning the Effectiveness of Foreign Aid 2.2 Relevance of Research 3. Methodology 3.1 Per capita Growth Model 3.2 Estimation Method 3.3 Categorizing Variables 4. Data description 4.1 The Dataset 4.2 Summary Statistics 5. Results 5.1 Estimation Results 5.2 Validity and Robustness of Results 6. Conclusion and Discussion 7. References 8. Appendix 3 5 5 8 9 9 10 12 12 12 12 13 14 15 16 17 191. Introduction
There is a lot of discussion about the effectiveness of foreign aid. Politicians, opinion makers and even economists disagree on the subject. There are many studies trying to investigate whether foreign aid is effective, and by which variables the effectiveness if affected. The importance of finding answers to questions regarding the effectiveness of aid is more relevant than ever. The World Bank Development Indicators database shows that Official Development Assistance (ODA) worldwide increased from 39.47 billion constant 2015 US dollars in 1970 to 152.723 billion constant 2015 US dollars in 2015. A graphical presentation of the development of worldwide aid flows can be found in figure 1. The growth of worldwide ODA has been accelerating since 2000.
Figure 1: ODA received worldwide in billions of constant 2015 U.S. dollars
According to Easterly (2009) the increase in ODA can be explained by the United Nations’ Millennium Development Goals (MDGs). In 2000, the participating countries and development institutions agreed upon eight development goals, that had to be met in 2015. These goals include among other things clean drinking water for everyone, universal enrollment in primary school, reducing infant mortality, combatting diseases and reducing poverty by 50 percent of the level of poverty in 2000.
There is more than one explanation for the ongoing discussion. Easterly (2003) argues that development institutions do not have enough incentives to evaluate the effectiveness of projects. Without often reviewing the performance of past projects, there is no sound basis to justify the increasing foreign aid flows mentioned above. Another problem economists are facing when evaluating the effectiveness of aid is the specification of the models used. Hansen, Dalgaard & Tarp (2000) review the literature on the relationship between aid and economic growth extensively, and show both that this discussion has been ongoing for a long time and that there are a lot of different model specifications used to explain this relationship. The first generation of models, built in the early sixties, was based on the Harrod-Domar model, through which researchers tried to establish a relationship between aid and savings. Second generation models focus on the direct links between aid and economic growth and aid and investment. Third generation models are more extensive 0 20 40 60 80 100 120 140 160 180 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014
models, which include more control variables. These third generation models use panel data and a large dataset to estimate the effect of aid on economic growth. Moreover, these studies recognize the endogeneity of aid and the non-linear relationship between foreign aid and economic growth. Among others, the studies by Boone (1996) and Burnside & Dollar (2000) laid the groundwork for the way we evaluate the impact of foreign aid now. In next section, these studies will be reviewed.
There is little agreement about the effectiveness of foreign aid. For example, Boone (1996), Herzer & Grimm (2012) and Feeny & Fry (2014) find no positive relationship between foreign aid and economic growth, mostly because of the absence of incentives for the receiving country to invest. On the other hand, Burnside & Dollar (2000), Dalgaard, Hansen & Tarp (2004) and Chervin & Wijnbergen (2010) find a positive relationship between foreign aid and economic growth, when certain conditions are met. Interestingly enough, Roodman (2007) concludes that none of the important studies on the subject released up to 2007 establishes robust results that indicate a positive relationship. An important part of his conclusion is his explanation for the failure of the models; they probably do not pass the robustness tests because aid is no important factor in explaining economic growth.
One thing most research agrees on, it that aid will be more likely to affect economic growth if a larger part of it is invested instead of consumed. This thesis will therefore investigate the effect of foreign aid on growth, and aims to discover whether aid has a larger effect on economic growth when the receiving government invests a larger fraction of its total expenditure. The main question this thesis will attempt to provide an answer to is: does foreign aid have a larger effect on economic growth when the recipient government invests a larger part of its total budget? To my knowledge, there is no research on this exact question, although there is research on the effect of government investment on economic growth. Therefore, it is interesting to answer this question. This study uses more recent data than the majority of recent studies. To control for the endogeneity of the aid variable, which is correlation between the aid variable and the error term, the General Method of Moments (GMM) framework is used. Further explanations on this framework can be found in the methodology section.
2. Literature review
This section provides an overview of the most influential studies regarding foreign aid effectiveness in terms of the effect of aid on economic growth.
2.1 Influential studies concerning the effectiveness of foreign aid
The effect of foreign aid on economic growth has been subject to a lot of research. As mentioned in the introduction section, it is not clear which studies use well specified models, because the models often lack a theoretical backbone. The studies differ in model specification and data gathering, by using different sets of years, countries or variables, as well as in conclusions. This section provides an oversight of the most influential studies and is an attempt to give a clear description of the different perspectives.
As mentioned briefly before, one of the first influential studies that uses modern theory as background and modern estimation methods to address the effectiveness of foreign aid was constructed by Boone in 1996. Boone wanted to address this issue, because he was convinced that the growing flows of development assistance were not supported by
economic theory. In earlier studies, an often used argument for foreign aid was the absence of access to capital markets in developing countries. These studies argue that there are investment opportunities in developing countries, which are not exploited, because of a shortage of savings to finance them. Boone estimates the effect of foreign aid on macroeconomic performance indicators by performing an OLS-regression, using official development assistance (ODA) data from the OECD database. This research uses 5-year periods where the data is averaged for 96 countries. Boone is not able to find a significant relationship between aid and total investment, only between aid and total consumption. Only for very small countries, he finds a significant impact on total investment. Focusing on the government, his conclusion is that aid increases government consumption, not investment. For long-term sustainable growth, an effect on investment is necessary. Boone compares the effect on investment and consumption for different regimes, but concludes that aid primarily benefits the elite, because aid is given without much conditionality. Dalgaard, Hansen and Tarp (2004) indicate that nowadays, donor countries take the performance of recipient countries on different measures of policy into account when allocating their aid flows.
Easterly (2007) mentions a comparable concern: although there are large aid flows, the donor countries do not know what to do to give the recipient countries the proper incentives to use the aid for good purposes. According to Easterly, the reality matches this statement. Despite the large flows of aid, the top quarter of recipient countries experienced almost no economic growth between 1965 and 2007. Easterly criticizes research that found a significant impact between aid and growth, because those estimation methods do not pass robustness tests. Easterly mentions a Paradox from Bauer (1976): when a country faces incentives to invest, aid is not necessary. This implies that all countries that need aid, will not use it for investment purposes, simply because they lack the incentives to do so. As mentioned by Easterly, it is hard to find out what actions work because the donors do not have incentives to review their projects. The reviewing process is hard, because they get little feedback from the recipients, which makes well evaluated projects even less likely. Many of the recipient countries are no democracies, so the poor are not able to communicate their satisfaction and their needs through voting. In normal markets, people signal their satisfaction with the good or service provided by buying or not buying. In the case of aid, the only way to signal this level of satisfaction is voting, but in absence of democracy, this channel does not work either. It is hard to hold anyone accountable for the failure of the projects, if recognized, because many organizations work together to achieve the same goal. It is also mentioned that there are informational problems, because governments have bureaucratic structures, which lack market forces to get enough incentives to gather the information needed to become aware which actions work and which do not.
The accountability and incentive problems are not only of concern to Easterly. Williamson (2010) explains carefully that aid is not distributed in the most efficient way. This occurs mostly because there are a lot of stakeholders, that all want to serve their own interest. Political decisions in donor countries are blurred because they need the stakeholders, like special interest groups, to support them in upcoming elections. The agencies that are expected to use the aid have mismanagement problems, and are competing for larger budgets, while as mentioned before, it is hard to attribute the results of the aid sponsored programs to a particular agency.
One of the most influential studies on the effectiveness on foreign aid is Burnside and Dollar (2000). The first part of their conclusion is in line with the work published by Boone (1996). When the recipient country has poor policies, aid primarily affects government consumption, not investment. Burnside and Dollar find that up to the 1990s, there is no correlation between the amount of aid received and the economic policy in the recipient countries. Their sample consists of 56 countries, where they define low-income countries as countries where the real GDP per capita was under 1900 constant 1985 U.S. dollars in 1970. They use a time-period of 24 years, up to 1993. Their data is retrieved from the World Bank database. The most important variables in their models are aid, policy and their interaction term. Of the two, aid is defined as Effective Development Assistance (EDA). EDA only consists of grants and the grant part of loans, while ODA also includes the total amount of all loans that include a grant part of at least 25 percent and excludes al loans in which grants account for less than 25 percent of the amount. Burnside and Dollar created a proxy for policy, which includes the inflation rate, the budget surplus and an openness dummy. In this study, multiple regressions are performed, both OLS and IV. The main conclusion is that aid has a limited impact on growth, but that the effect becomes more positive when the recipient country has good policies. For low-income countries, the policies matter even more for the effectiveness of foreign aid. Contrary to the results of Boone (1996), this study suggests that aid effectiveness does depend on the policies in the recipient country. The Burnside and Dollar study was of big support to the pro-aid campaigns, because policies in low-income countries are becoming better over time. This finding in combination with the statement that the effectiveness of aid increases when there are good policies in the recipient country, suggests that aid might be more effective in the future.
Hansen, Dalgaard & Tarp (2001) use the same data as Burnside and Dollar, but they use an alternative growth model. They add squared terms for all variables that include aid. This study concludes that aid has a diminishing effect on economic growth, because the squared term of aid turns out to be more important than the interaction term of policy and aid. The squared term of aid is found to be significant at the 5% level. In this paper, Hansen & Tarp test the Burnside-Dollar estimations for robustness, and the estimations turn out to be very sensitive to model specification and changes in the dataset.
Dalgaard, Hansen and Tarp (2004) also use the same specification as Burnside and Dollar, but they add a variable: Fraction in tropics. Their study uses the same time-period and the same countries, but includes other instruments. It seems that donor countries tend to differentiate countries in their aid allocation based on their policy. This is no surprise, since the Burnside and Dollar study influenced thinking about the allocation of aid in the donor countries. Dalgaard, Hansen and Tarp conclude that aid does stimulate convergence of GDP/capita across countries, but that aid is barely effective in tropical areas.
Easterly, Levine and Roodman (2004) wrote a comment with important criticism on the Burnside and Dollar study. This work uses the same dataset as Burnside and Dollar, but it adds more countries and more years to the dataset. This paper uses exactly the same methodology, to check whether the results of Burnside and Dollar are robust to the expansion of the dataset. It turns out that the interaction term between aid and policy becomes insignificant, it even switches sign. A possible explanation for this result: the model built by Burnside and Dollar is built without a solid theoretical base. There are many ways to build a model on the subject, and it is hard to determine which one is theoretically valid. Rajan and Subramanian (2008) attempt to find the impact of aid on economic growth. This study contains a time period from the 1960s to the 1990s. To address
endogeneity problems, GMM-estimation is used. Rajan and Subramanian find no significant positive relationship between aid and economic growth, and no evidence that other aforementioned factors such as policies in the recipient countries, geographical settings or different measures of aid have a significant impact on the effectiveness of aid. In their study about the effect of volatile aid, Chervin & Wijnbergen (2010) focus on the growth effect of aid when the amount of aid is volatile and when it is not. The amount of aid given is often volatile, and this study concludes that aid has a positive effect on growth when the amount is not volatile, but no significant effect when it is. However, a more recent study concludes that the volatility of aid it not worth worrying about, because a shock in the amount of aid needs to be 9 times as large as a shock in aggregate production to result in the same change in GDP growth (Annen, Batu & Kosempel, 2016). In this thesis, it is therefore assumed that volatility of aid is not a big problem. Roodman (2007) performs robustness tests on influential cross-country studies. This study concludes that even the most influential studies, such as Burnside and Dollar (2000) do not pass the robustness tests. There are no robust findings that indicate a significant effect from aid on economic growth when years or countries are added to the dataset. This finding might explain the very different results across studies. Roodman does not state that foreign aid therefore is ineffective, but he indicates that it is hard to find a significant impact. He argues that aid is probably not one of the most important factors in explaining economic growth, and that there are kinds of aid, like food aid, that do not directly influence economic growth.
The study of Clemens, Radelet, Bhavnani and Bazzi (2012) also explains the very different conclusions of the aforementioned articles. According to Clemens et al., there are two major problems that cause the wide range of conclusions. First of all, the influential studies mentioned before focus on the direct effect of aid on economic growth. Some projects may directly affect economic growth, but for instance health campaigns can take decades to have a significant impact on economic growth. Second, these studies use instrumental variables to distinguish between correlation and causation, but the instruments often lack strength and validity. Clemens et al. address the first problem by introducing a time lag, which allows aid to affect growth a period later. Moreover, only the fraction of aid that is expected to affect economic growth within the next few years is included. The second problem is addressed with first-differencing, which removes the omitted variable bias caused by country-specific fixed effects. This study uses the original data of the study of Burnside and Dollar (2000) and the Rajan and Subramanian (2008) study. For the Boone (1996) study, some additional data gathering is needed, but the dataset used is close enough to the original to treat them as equal. The results indicate a much smaller difference in conclusions. On average, an increase of aid as fraction of GDP of one percent point is followed by an increase in total investment of 0.3-0.5 percent point and an increase of per capita growth of 0.1-0.2 percent point. 2.2 Relevance of research Although there are many studies around that address this subject, there is still no agreement on the actual effect of aid on economic growth and the channels through which this possible effect is established. This thesis aims to find whether aid fuels economic growth, and more in particular, whether aid has a larger effect on growth when the recipient government invests a larger part of its budget. To my knowledge, there are no studies available that answer exactly this question. Because of the earlier mentioned incentives problem, this
seems an interesting question. Unlike most influential studies on the aid-growth relationship, which use OLS or IV regression, here the System GMM framework will be applied to eliminate possible bias from endogenous regressors.
3. Methodology
This section contains the model used in this analysis, an explanation of the GMM-framework and the way of using this framework specified to this model. 3.1 Per capita growth model This thesis aims at looking whether foreign aid has a significant impact on per capita growth, especially through the public investment channel. The model contains variables on aid, investment and some control variables. The regression model presented below is the most comprehensive version of the model, and has ten independent variables. 𝑔𝑑𝑝𝑐𝑎𝑝𝑔𝑟𝑜𝑤𝑡ℎ+, = 𝛽0 + 𝛽1 ∗ 𝑔𝑑𝑝𝑐𝑎𝑝𝑔𝑟𝑜𝑤𝑡ℎ,34+ 𝛽2 ∗ 𝑠𝑐ℎ𝑜𝑜𝑙𝑖𝑛𝑔+,+ 𝛽3 ∗ 𝑝𝑜𝑝𝑔𝑟𝑜𝑤𝑡ℎ+, + 𝛽4 ∗ 𝑓𝑟𝑎𝑐𝑔𝑖+,+ 𝛽5 ∗ 𝑝𝑟𝑖𝑣𝑖𝑛𝑣+,+ 𝛽6 ∗ 𝑜𝑝𝑒𝑛𝑛𝑒𝑠𝑠+,+ 𝛽7 ∗ 𝑖𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛+, + 𝛽8 ∗ 𝑎𝑖𝑑𝑔𝑑𝑝+,+ 𝛽9 ∗ 𝑎𝑖𝑑2+,+ 𝛽10 ∗ 𝑓𝑟𝑎𝑐𝑔𝑖𝑎𝑖𝑑𝑔𝑑𝑝+,+ 𝜇+, In addition to the above presented model, some variations of this model will be tested. In this variations, some independent variables are excluded.
The choice of independent variables is based on the research question and on the model specification in some influential studies.
The dependent variable is GDP growth per capita, measured in annual percentages. The one period lag of per capita growth is included as independent variable, to account for the fact that economic growth is a dynamic process. This variable can be an indication of the persistence of growth.
The schooling variable, defined as gross primary school enrollment, is included because investments in human capital are expected to fuel economic growth. Colclough (1982) was able to conclude that especially primary schooling increases labor productivity. Enrollment in primary school is therefore expected to have a positive effect on per capita growth.
According to theory, population growth can have both a positive and a negative effect on per capita growth. Population growth can contribute to economic growth because it expands the labor force, but if the resources available are not able to keep up with the population growth, it might have a negative effect on economic growth per capita.
The fracgi variable is defined as the fraction of the government budget that is invested. The government budget is, because of a lack of data, approximated by the sum of the government investment and the government consumption. Because investment tends to have a positive effect on economic growth, it is expected that the coefficient of this variable has a positive sign. As mentioned before, there is a lot of literature on the incentive problem: governments tend to consume a large part of the received aid, which hampers long-term growth. Private investment is also expected to have a positive effect on per capita growth. This variable is included because previous studies indicate that excluding this
variable leads to misspecification (Herzer & Morrisey, 2013). Private investment is measured as fraction of total GDP.
Openness is measured as the sum of imports and exports as percentage of GDP. Just like most authors, Brueckner & Lederman (2015) find a significant positive effect of trade openness on economic growth. Therefore, the sign of the coefficient of openness is expected to be positive. The inflation variable is present in the model as a proxy for macroeconomic stability. Inflation is measured as the annual GDP-deflator in percentages. The coefficient of this variable is expected to have a negative sign, since a low and stable inflation is known to have a positive effect on the economy.
The aidgdp variable is measured as ODA as percentage of total GDP. ODA is official development assistance, which includes grants, soft loans and technical assistance, but excludes any transfers or loans that are related to the military (World Bank). The squared term of aid is included because of the often assumed diminishing returns to aid (Dalgaard, Hansen & Tarp, 2004). The purpose of this thesis is to investigate whether foreign aid is more effective when the government invests a larger part of its budget. Therefore, the interaction term of the aid variable with government investments is added to the regression. Burnside & Dollar (2000) used an interaction term with respect to aid and policy, while Dalgaard, Hansen and Tarp (2004) used an interaction term of aid and fraction in the tropics. Following the design of the variables of those studies, the aforementioned interaction terms regarding aid and government investment are included. 3.2 Estimation Method Even the most influential studies encounter estimation problems using a dynamic panel data model. As explained by Leefmans (2017), estimation results are biased when a model has both fixed effects and a lagged dependent variable, because the lagged variable will be correlated with country-specific fixed effects that are not modeled. If a variable in the model correlates with another not modeled variable, this creates an omitted variable bias, which causes the variable in the model to be correlated with the error term. In his publication in the Stata Journal, Roodman (2009) describes which convenient assumptions are incorporated in the Generalized Method of Moments (GMM) estimators. It can be used for dynamic processes, even when there are endogenous regressors and fixed country-specific effects. This is relevant for this thesis, because the variable aid is known to cause endogeneity problems (Chervin & Wijnbergen, 2010). Countries with low economic growth might receive more aid, but the causality also works the other way, because countries that receive larger amounts of aid might experience higher economic growth. The first influential paper on GMM was Arellano-Bond (1991). Although this was not the first paper on a comparable estimation method, it is widely seen as the basis of GMM estimation (Roodman, 2009). The method used in Arellano-Bond (1991) is called difference GMM. In difference GMM, values of exogenous and predetermined variables in the model are used to build instruments for the endogenous variables in the model, by first differencing the variables. The latter eliminates the constant out of the equation and removes the fixed effects that were included in the error term. It uses lagged levels as instruments. It is no longer necessary for the variables that are used to create instruments to be strictly exogenous, but the regressors have to be predetermined. This method can be used when the assumption holds that there is no serial correlation in the errors. This approach can be used to remove the fixed effects with the use of first-differencing (Leefmans, 2011).
Unfortunately, first differencing is not sufficient to eliminate the estimation bias when lagged values are present in the model. Leefmans explains that the correlation between the lagged values and the error term causes a bias. To solve this problem of endogeneity, it is necessary to instrument the lagged variable and other possible endogenous variables with instruments that are not correlated to any of the fixed effects (Roodman, 2009). Blundell & Bond (1998) difference the instruments instead of the regressors, to make them exogenous. This way the untransformed lagged variables can be used when we instrument them with their differenced lag. According to Roodman (2009), these instruments are not only exogenous, but also more relevant, because past differences are more closely related to present levels than past levels. He states that in xtabond2, the Stata command for this kind of system GMM analysis, it is best to use all appropriate lags as instruments, in order to get a more efficient estimation method. This is valid when the differences in the instrumental variables are not correlated to the part of the error term that contains the fixed effects. Which lags are valid, depends on the kind of variable. Roodman distinguishes between three kinds of variables. For endogenous variables, only lags from 2 periods back and more are valid, because they can be correlated with both present and past error terms. For predetermined variables, lags of one period back and older are usable, because they can only be correlated with past error terms that go at least 2 periods back. In the third category are the exogenous variables, which are not correlated with any error terms. Just as in Arellano-Bond estimation, autocorrelation in errors must still be absent. In order to be as efficient as possible, all valid lags should be used. It is however also possible that GMM creates too many instruments. In that case, the instrumental variables overfit the endogenous variables. According to Windmeijer (2005) it is important to deal with the overfitting problem, because reducing the number of instruments can dramatically decrease the bias in the estimations.
System GMM, as developed by Blundell & Bond combines both methods. The first method, as introduced by Arellano & Bond, is used for the differenced equation and the second method, as introduced by Blundell & Bond, is used for estimation with the untransformed equation. Together these two equations are used for estimation in this system of equations. (Leefmans, 2017).
In order to check the validity of the results, it is necessary to check whether the assumptions are met. The most critical assumption is the assumption of no autocorrelation in errors. Arrelano and Bond (1991) constructed a test to make sure this assumption holds. This test gives two statistics. AR(1) will be significant if there is order-one autocorrelation. According to Roodman (2009), the presence of autocorrelation of order one is no reason for concern. Because of mathematical relations, autocorrelation of order one is expected in differences. It is important that we check for serial correlation in the levels. This can be done by testing for second-order autocorrelation in differences. Therefore, a significant value for AR(2) in the Arrelano-Bond test is problematic. The Arrelano-Bond test is useful to test for autocorrelation in the GMM-framework, as long as there are only pre-determined, exogenous and endogenous regressors. If there are post-determined regressors, this test is not valid. Since there are no post-determined regressors in the model used in this thesis, this is no problem.
The second required assumption in order to get valid results is that the instruments used are strictly exogenous. This can be tested using the Sargan/Hansen test of overidentifying restrictions. This test can only be done when the estimation is overidentified, which means that there are more instruments than endogenous regressors. In GMM-output,
Stata reports the value of this test-statistic on default. If there is a significant value, there is a problem with the exogeneity of the instruments. Both the abovementioned Arellano-Bond test for autocorrelation and the Sargan-Hansen test for overidentifying restrictions are performed for this thesis. The results can be found in section 5.2.
4.3 Categorizing variables
As mentioned before, it is important to distinguish exogenous, pre-determined and endogenous regressors. In the model used in this thesis, only inflation and openness are categorized as exogenous, because they are assumed not to be influenced by other variables in the model. Schooling, population growth and the initial value of GDP are treated as pre-determined, because their current values are determined in the past. Therefore, these variables can be instrumented with lags one period back and further. The investment variables and the aid variables are considered to be endogenous regressors, because there is a reverse causality problem with the dependent variable. For example, countries tend to receive more development assistance when their economic performance is poor, while at the same time, aid can boost economic performance. This reverse causality creates a bias when the variables are not properly treated as endogenous. The endogenous variables can only be instrumented with lags two periods back and more.
4. Data description
This part contains a brief description of the dataset, as well as an explanation of some extreme datapoints.
4.1 The dataset
The dataset contains 58 countries. All countries with enough data availability that are in the Burnside-Dollar study are included, and some extra countries with low per-capita GDP in 1970. Burnside and Dollar (2000) define low-income countries as countries with a per capita GDP lower than 1900 U.S. dollars in constant 1985 dollars. An overview of the included countries can be found in the appendix. The time-period used starts in 1970 and ends in 2015. This is the longest possible period with good data availability and is more recent than the data used in the majority of recent studies. Almost all data is retrieved from the World Bank Development Indicators database. The data on government investment, GDP and private investment comes from the IMF Investment and Capital Stock dataset 1960-2015. Table 1 provides a summary of the statistics for all variables used.
4.2 Summary statistics
As visible in the minima and maxima, there are large differences between countries and years. For example, the minimum economic growth value was found in Rwanda, during the 1994 genocide. There are countries with periods of deflation, like Congo in 2015, responsible for the minimum value of inflation. There is also an extreme maximum value for inflation, which corresponds to Nicaragua in 1988, with severe hyperinflation. Schooling is measured as primariy enrollment rates. It might seem strange to see primary enrollment rates above 100 percent. This occurs because this is a gross variable, which means that the base of 100 percent are the children that should be in primary school according to their age, but the actual value contains also late enrollers, repetitors and early enrollers. The Aid
variable contains negative values, because some net recipient countries were net donor countries in other time periods. The extreme value where aid is 24,38% of GDP, refers also to the genocide in Rwanda in 1994, and in the same year the country experienced a sharp decline in GDP. In this year, the population shrank by almost 6%. The extreme positive value of the GDP growth per capita is in Zimbabwe in 2009 when there was one central government installed and the own currency, which was suspect of severe hyperinflation, was abandoned. All these extreme values might indicate the presence of outliers, but when plotted against a normal distribution, there is no reason to believe these outliers have a big influence on the regression results. The plot can be found in the part about robustness in the results section, which is section 5.
Variable Obs Mean SD Min Max
gdpcapgrowth 2608 0.016 0.051 -0.463 0.527 inflation 2569 42.149 435.405 -29.691 13611.635 openness 2472 0.602 0.301 0.063 2.204 schooling 2376 92.924 26.439 11.714 165.645 popgrowth 2609 0.023 0.009 -0.060 0.082 fracgi 2444 0.472 0.190 0.055 0.953 aidgdp 2662 1.335 1.810 -0.156 24.377 aid2 2662 5.055 17.843 0.000 594.248 privinv 2483 0.121 0.071 0.001 0.653 fracgiaidgdp 2439 0.627 0.936 -0.102 6.862 Table 1: Summary of statistics of all variables in the model
There are some trends visible when we evaluate the descriptive statistics. In the period investigated, the amount of aid as fraction of total GDP increases, as expected. The statistics of the fraction of the total budget that the government invests shows some convergence over time, because the differences between countries diminish. Countries report less extreme values of inflation in more recent years, which indicates increasing stability. One of the primary targets of the development agenda is the enrollment rate. The statistics of primary school enrollment shows a large positive trend. Population growth is almost always positive, and the statistics shows no clear change in population growth rates over time. Countries tend to become more open to trade as time passes. The private investment statistics display less extreme values in more recent years, and the trend shows a small increase. A graphical representation of the trends can be found in figures (3) to (11) in the appendix.
5. Results
In this section, the results of the analysis are presented and interpreted. At the end of the section, a robustness check is done to check for the influence of outliers on the estimation results. The results of the GMM-estimation can be found in table 2.
5.1 Estimation results (1) (2) (3) (4) (5) Variables gdpcapgro wth gdpcapgro wth gdpcapgro wth gdpcapgro wth gdpcapgro wth gdpgrowthlag 0.237*** 0.195*** 0.201*** 0.212*** 0.206*** (0.063) (0.070) (0.077) (0.064) (0.057) schooling -0.000 -0.001 -0.001 -0.001 -0.001* (0.000) (0.000) (0.001) (0.000) (0.000) popgrowth 0.333 0.263 -0.069 -0.380 -0.693 (0.902) (0.713) (0.740) (0.720) (0.723) fracgi 0.075 -0.003 0.076 0.018 0.071 (0.050) (0.071) (0.089) (0.059) (0.065) privinv 0.264*** 0.246** 0.343*** 0.237*** 0.305*** (0.087) (0.108) (0.095) (0.088) (0.096) openness -0.008 -0.003 -0.016 -0.005 -0.012 (0.006) (0.007) (0.010) (0.011) (0.011) inflation -0.000*** -0.000*** -0.000*** -0.000*** -0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) aidgdp 0.007 0.038** 0.037* 0.051*** (0.006) (0.017) (0.021) (0.013) aid2 -0.005 -0.005** (0.004) (0.002) fracgiaidgdp -0.066** -0.035 (0.030) (0.025) Constant -0.049 0.024 -0.003 0.036 0.014 (0.060) (0.056) (0.071) (0.056) (0.057) Number of instruments AR(2) AR(2) p-value Hansen statistic Hansen p-value 13 0.180 0.860 2.85 0.723 15 -0.13 0.898 6.61 0.358 17 -0.08 0.939 7.63 0.366 17 -0.18 0.855 4.29 0.746 19 -0.19 0.850 3.32 0.913 Observations 2,085 2,085 2,085 2,085 2,085 Number of countries 58 58 58 58 58
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 2: GMM-estimation results The first regression presented in the column(1) of the above table, contains no aid variables. It only contains other factors that affect economic growth, in order to have a basic model specification to which we will add aid variables in columns (2) to (6). Most important for answering the research question is the effect of the aid variables and the interaction term of aid and government investment on economic growth. In regression (2), the aid variable is added to the variables that are in regression (1). In regression (2), the amount of foreign aid has no significant effect on per capita growth, but in regression (4) the coefficient is significant at the 10%-level, in regression (3) at the 5%-level and in regression (5) even at the 1-% level. As expected, aid seems to have a significant
impact on per capita growth, which is an important results for answering the research question of this thesis. The quadratic aid-term is only significant in regression (5), so this analysis is no strong support for the statement of diminishing returns to aid made by Dalgaard, Hansen & Tarp (2004). The research question of this thesis can be answered by interpreting the coefficients of the interaction terms. The interaction term of aid and the fraction of government budget that is invested only appears to be significant at the 5%-level in regression (3), and surprisingly enough, it is negative. These findings mean that there is no strong support for the hypothesis that aid is more effective when the recipient government invests a larger part of its budget.
The gross primary school enrollment rate was hypothesized to have a positive coefficient, because of the assumption that education contributes to the growth of human capital, which can fuel economic growth. Nevertheless, the coefficient on schooling is negative, but only significant in regression (5), and only at a 10%-level. The coefficient of population growth is insignificant across the entire analysis. In earlier research, the effect of population growth was ambiguous. Some researchers predict a positive effect (Berry, 2010), others a negative effect (Klasen & Lawson, 2007). Considering these ambiguous findings, the results in this analysis are not surprising. Also important for answering the research question is the effect of the investment variables on per capita growth. Private investment contributes, in line with the hypothesis, positively to economic growth. The positive effect of private investment is significant at the 1%-level throughout the whole analysis. However, the fraction of government budget that is invested has no significant effect on the per capita growth whatsoever. This is quite surprising, since investments are assumed to fuel economic growth. Easterly (2009) suggests the absence of a positive significant relationship is present because of incentive problems and institutional problems. He states that the evaluation of investment projects by the recipient governments is often inadequate, the projects lack accountability and the recipient government might not know what actions are able to contribute to economic growth. The exogenous variables in this analysis are inflation and openness. The openness is expected to have a positive coefficient, but it has a negative insignificant coefficient. Apparently, the amount of openness is not an important factor in explaining per capita growth for the countries and years in this dataset. Inflation is expected to have a negative effect on growth, because a low and stable inflation rate usually is a good approximation for macroeconomic stability, and a stable economic environment has a positive effect on growth. In this analysis, inflation indeed has a significant negative effect on per capita growth. This holds at a 1%-level throughout all regressions.
5.2 Validity and robustness of results
As mentioned before, it is important to check the validity of the results using the Arellano-Bond test for autocorrelation and the Sargan-Hansen test of over-identifying restrictions. The p-value of the AR(2) is in all regressions well above the 10% level, so it seems that there are no problems regarding autocorrelation in this analysis because all test statistics are insignificant. The p-values of the Sargan-Hansen test are also all above the 10%-level, so it can be assumed that the instruments used are exogenous. On the basis of these results, there is no reason to question the validity of the estimates. In the data description section, the presence of outliers is confirmed. To investigate whether the outliers have an important effect on the estimates, it is necessary to predict the residuals and plot them
against the standard normal distribution. This plot can be found in figure (2) in the appendix. It appears that the residuals fit the distribution quite well, except of the two observations in the bottom-left corner. These values correspond to Rwanda in 1994 and 1995. This is the period of the genocide, which is mentioned before in the data description section.
Dropping these outliers is no good practice if there is no theoretical justification. For robustness purposes, it is interesting to run the regressions again when the outliers are dropped. Regressions (3) and (5) are evaluated, because those regressions are most important to answer the research question of this thesis.
The results indicate that the analysis is quite robust, since there are no large differences in significance and signs when the outliers are dropped. Table 3 contains the results of the regression without the outliers.
(3) (5)
Variables gdpcapgrowth gdpcapgrowth
gdpgrowthlag 0.198** 0.192*** (0.078) (0.055) schooling -0.001 -0.001** (0.001) (0.000) popgrowth -0.135 -0.773 (0.737) (0.719) fracgi 0.080 0.075 (0.090) (0.070) privinv 0.350*** 0.316*** (0.095) (0.099) openness -0.017* -0.010 (0.009) (0.012) inflation -0.000*** -0.000*** (0.000) (0.000) aidgdp 0.036** 0.045*** (0.016) (0.016) aid2 -0.004* (0.002) fracgiaidgdp -0.064** -0.032 (0.030) (0.030) Constant -0.003 0.023 (0.071) (0.053) Observations 2,085 2,085 Number of countries 58 58
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 3: Regression results with outliers dropped
6. Conclusion and discussion
As stated in the introduction, the aim of this study was to investigate whether aid positively affects economic growth and whether this effect is larger when the recipient government invests a larger part of its budget. Previous studies focus on the incentives of the recipient governments and the effectiveness of aid conditional on institutional quality or macroeconomic policy. In previous literature, it is mentioned that governments consume a
large part of the received aid, while it would be better to invest a larger part. This study was not able to find a significant relationship between aid, the fraction of government budget that is invested and economic growth throughout the analysis. There is some evidence that foreign aid can boost per capita growth, although not necessarily through the government investment channel. The interaction variable of the fraction invested by the government and the received aid even had a coefficient with a negative sign. This study is of support to the pro-aid campaigns, because of the mentioned significant positive effect of aid on economic growth. Private investment on the other hand, has a significant positive effect on per capita growth. It could be interesting to build a model with which it is possible to find out which channels are preferred to establish a growth effect of foreign aid. However, that can be difficult, because there is a lack of data from several developing countries. From the results in this analysis, it seems that the government investment channel is not preferred, because no significant result is found. For the political discussion mentioned in the introduction section, the finding that foreign aid has a significant positive effect on economic growth is important, but the way make foreign aid most effective still remains unclear.
This study has some limitations that need to be mentioned. The most important one is the aforementioned lack of data. There is some data missing for the countries that are used in the dataset, which might influence the estimates, but there are also countries excluded from the dataset, because there was no data available. If the countries that do not provide data frequently are significantly different from the countries that do, the results from this study are not generalizable, which affects the external validity of the analysis.
7. References
Annen, K., Batu, M., & Kosempel, S. (2016). Macroeconomic Effects of Foreign Aid and Remittances: Implications for Aid Effectiveness Studies. Journal of Policy Modeling, 38(6), 1136-1146.
Arellano, S., & Bond, S. (1991). Some Tests of Specification for Panel Data. Review of Economic Studies, 58, 277-297.
Berry, C. (2014). The Relationship Between Economic Growth and Population Growth.
Retrieved from http://www.academia.edu/8762205/The_relationship_between_economic_growth_ and_population_growth Blundell, R., & Bond, S. (1998). Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of Econometrics, 87, 11-143. Boone, P. (1996). Politics and the Effectiveness of Foreign Aid. European Economic Review, 40(2), 289-329. Brueckner, M., & Lederman, D. (2015). Trade Openness and Economic Growth: Panel Data Evidence from Sub-Saharan Africa. Economica, 82, 1302-1323. Burnside, C., & Dollar, D. (2000). Aid, Policies and Growth. American Economic Review, 90(4), 847-868.
Chervin, M., & Van Wijnbergen, S. (2010). Economic Growth and the Volatility of Foreign Aid (Discussion paper Tinbergen Institute). Retrieved from
https://ideas.repec.org/p/tin/wpaper/20100002.html
Clemens, M. A., Radelet, S., Bhavnani, R. R., & Bazzi, S. (2012). Counting Chickens when they Hatch: Timing and the Effects of Aid on Growth. Economic Journal, 122(561), 590-617. Colclough, C. (1982). The Impact of Primary Schooling on Economic Development: A Review of the Evidence. World Development, 10, 167-185.
Dalgaard, C., Hansen, H., & Tarp, F. (2004). On the Empirics of Foreign Aid and Growth. Economic Journal, 114(496), 191-216.
Dalgaard, C. J., Hansen, H., & Tarp, F. (2001). On Aid, Growth and Good Policies. The Journal of Development Studies, 37(6), 17-41.
Dowrick, S., & Rogers, M. (2002). Classical and Technological Convergence: Beyond the SolowSwan Model. Oxford Economic Papers, 54(3), 369-385.
Easterly, W. (2003). Can Foreign Aid Buy Growth? Journal of Economic Perspectives, 17(3), 23-48.
Easterly, W. (2007). Was Development Assistance a Mistake? American Economic Review, 97(2), 328-332.
Easterly, W. (2009). Can the West Save Africa? Journal of Economic Literature, 47(2), 373-447.
Easterly, W., Levine, R., & Roodman, D. (2004). Aid, Policies and Growth: Comment. American Economic Review, 94(3), 774-780. Feeny, S., & Fry, T. (2014). How Sustainable is the Macroeconomic Impact of Foreign Aid? Journal of Policy Modeling, 36(6), 1066-1081. Hansen, H., & Tarp, F. (2000). Aid Effectiveness Disputed. Journal of International Development, 12(3), 375-398.
Herzer, D., & Grimm, M. (2012). Does Foreign Aid Increase Private Investment. Applied Economics, 44(20), 2537-2550. Herzer, D., & Morrissey, O. (2013). Foreign Aid and Domestic Output in the Long Run. Review of World Economics, 149(4), 723-748. Klasen, S., & Lawson, D. (2007). The Effect of Population Growth on Economic Growth and Poverty Reduction in Uganda (Working Paper No.133). Retrieved from
http://www.academia.edu/8762205/The_relationship_between_economic_growth_ and_population_growth
Leefmans (2011), draft version PhD dissertation, University of Amsterdam.
Roodman, D. (2007). The Anarchy of Numbers: Aid, Development and Cross-Country Empirics. The World Bank Economic Review, 21(2), 255-277.
Roodman, D. (2009). How To Do Xtabond2. Stata Journal, 9(1), 86-136.
Williamson, C. (2010). Exploring the Failure of Foreign Aid: The Role of Incentives and Information. The Review of Austrian Economics, 23(1), 17-33. Windmeijer, F. (2005). A Finite Sample Correction for the Variance of Linear-Efficient Two-Step GMM-estimators. Journal of Econometrics, 126(1), 25-51.
8. Appendix
Figure 2: Residuals plotted against normal distribution for the entire sample -1 0 -5 0 5 R e si du al s -2 -1 0 1 2 Inverse NormalFigure (3): GDP per capita growth between 1970 and 2015, in decimals. Figure (4): The fraction of total government budget that is invested between 1970 and 2015. -0,4 -0,3 -0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 1970 1980 1990 2000 2010 0 0,2 0,4 0,6 0,8 1 1,2 1970 1980 1990 2000 2010
Figure (5): The share of aid in GDP in percentages between 1970 and 2015.
Figure (6): The spread of the values of the interaction term of aid and the fraction of the government budget that is invested between 1970 and 2015. -2 0 2 4 6 8 10 1970 1980 1990 2000 2010 -1 0 1 2 3 4 5 6 7 1970 1980 1990 2000 2010
Figure (7): The primary enrollment rate between 1970 and 2015. Figure (8): Population growth between 1970 and 2015 in declimals. 0 20 40 60 80 100 120 140 160 180 1970 1980 1990 2000 2010 -0,06 -0,04 -0,02 0 0,02 0,04 0,06 0,08 0,1 1970 1980 1990 2000 2010
Figure (9): Inflation, in annual percentages between 1970 and 2015..
Figure (10): Openness, measured as the share of the total trade volume of total GDP, between 1970 and 2015. -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 100 1970 1980 1990 2000 2010 0 0,5 1 1,5 2 2,5 1970 1980 1990 2000 2010
Figure (11): Private investment as share of total GDP between 1970 and 2015.
Algeria Costa Rica Guatamala Mauritania Peru Trinidad and Tobago Argentina Côte d’Ivoire Haiti Mali Phillipines Tunisia Bolivia Dominican
Republic Honduras Mexico Rwanda Turkey
Botswana Ecuador India Morocco Senegal Uganda
Brazil Egypt Indonesia Nepal Sierra Leone Uruguay
Burkina Faso El Salvador Kenya Nicaragua Sri Lanka Venezuela
Cameroon Ethiopia Lesotho Niger Syria Zambia
Chile Gabon Madagascar Nigeria Tanzania Zimbabwe
Colombia Gambia Malawi Pakistan Thailand
Congo Ghana Malaysia Paraguay Togo
Table 4: All countries in the dataset 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 1970 1980 1990 2000 2010
Correlation Coefficients
gdpcapgrowth inflation openness schooling popgrowth fracgi aidgdp aid2 privinv fracgiaidgdp
gdpcapgrowth -0.032 0.064 0.076 -0.172 0.080 -0.114 -0.112 0.236 -0.087 inflation -0.098 -0.163 0.067 -0.002 0.123 -0.138 -0.141 -0.165 -0.106 openness 0.069 0.038 0.087 0.021 0.014 0.177 0.182 0.219 0.166 schooling 0.094 0.011 0.123 -0.368 0.222 -0.310 0.310 0.176 -0.334 popgrowth -0.053 0.021 -0.001 -0.321 0.203 0.510 0.509 -0.237 0.526 fracgi 0.070 0.041 0.079 -0.161 0.186 0.032 0.031 -0.038 0.253 aidgdp -0.068 0.003 0.097 -0.257 0.222 0.055 0.999 -0.200 0.969 aid2 -0.097 0.002 0.035 -0.172 -0.018 0.013 0.777 -0.197 0.968 privinv 0.186 0.028 0.196 0.157 -0.152 0.043 0.173 0.098 -0.201 fracgiaidgdp -0.029 0.004 0.076 -0.218 0.270 0.300 0.900 0.614 0.111 Table 5: Correlations
