The conditional effectiveness of aid in developing countries

(1)

The conditional effectiveness of Aid

in developing countries

Bachelor of Science Koen van der Ven

10599118

Supervised by N. Ciurila

June 2016

(2)

Statement of Authenticity

This document is written by Koen van der Ven who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is 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.

(3)

Introduction

The research report “Assessing Aid” by the World Bank, states that that 75 percent of their African agricultural projects were failures. The article “The Sorry Record of Foreign Aid in Africa” by James Peron mentions other aid debacles as well, like the funds of an “AIDS awareness” play in South Africa being squandered on a luxurious bus for the cast and crew. A newspaper article in The Express states among others that over the last five years around 526 million GBP were donated to Zimbabwe, falling into the hands of Robert Mugabe’s government that has spent the money on the oppression of the Zimbabwean people. In another article from Time Magazine, William Easterly, one of the leading economists in the field of economic development, compares Botswana with Ethiopia. In the latter president Meles Zenawi withheld aid-financed famine relief from everyone except members of the ruling party, while in the former the government successfully fought famine wherever drought struck. It is clear from these examples that foreign aid doesn’t always go to the people who need it most. But what is exactly meant by foreign aid?

Foreign aid is the voluntary transfer of resources from one country to another. The purpose of this transfer is not only for the noble cause of poverty reduction and stimulating development in poor countries, but also for other political reasons. It can be categorized in six clusters (Riddell, 2007):

(1) help address emergency needs

(2) assist recipients achieve their development goals (3) to show solidarity

(4) to further their own national political and strategic interest (5) to help promote donor-country commercial interests (6) historical ties

It depends on the type of foreign aid which of these purposes is the main motivation of the donation. For example, humanitarian aid, i.e. material aid in the case of immediate disasters such as an earthquake or a civil war, is provided to help address emergency needs, but not so much to help promote donor-country commercial types. Other types of foreign aid include military aid, development aid and project aid

(5)

among others. This paper focuses only on development aid. Therefore, the other types of foreign aid are not discussed in detail.

Development aid, or official development assistance (ODA), is financial aid given by governments and other agencies to support the economic, social, environmental and political development of developing countries. The members of the Development Assistance Committee1 donated about 135 billion dollars of ODA in the year 2013 to promote economic development and welfare of developing countries. The full definition of ODA is as follows:

Flows of official financing administered with the promotion of the economic development and welfare of developing countries as the main objective, and which are concessional in character with a grant element of at least 25 percent (using a fixed 10 percent rate of discount). By convention, ODA flows comprise contributions of donor government agencies, at all levels, to developing countries (“bilateral ODA”) and to multilateral institutions. ODA receipts comprise disbursements by bilateral donors and multilateral institutions.2

There is an ongoing debate on the effectiveness of aid. The proponents argue that aid can cause economic growth. Saving rates are low and financial assets are scarce in developing countries. The Solow growth model by Robert Solow (1956) remarks the importance of savings for economic growth. Dalgaard et al. (2004) also show in a two-period Diamond model augmented with aid influx the importance of savings and financial assets. Development aid is a way of providing the means to increase investment and therefore economic growth. A study by Clemens et al. (2004) shows that the relationship between aid and growth in real GDP per capita is positive.

The opponents however argue that the money is wasted due to corruption, bureaucracy and eventually ends up in the pockets of dictators. Therefore, the positive relationship between aid and growth is either conditional on “good” policy (Burnside & Dollar, 2000), or find insignificant evidence of a positive relationship (Rajan &

1

The Development Assistance Committee consists of 28 members consisting of the US, Canada, the EU, Japan, South Korea, Australia and New Zealand. In other words, the Western World

2

(6)

Subramanian, 2008; Boone, 1996; Landau, 1986). The meta-study of Doucouliagos and Paldam (2008) finds in 68 papers a positive average of the estimators of the effect of aid on growth, but it was insignificant.

A large number of articles are dedicated to this problem with mixed results. It is an important question, because the possibility of wasting taxpayers’ money is always a topic of discussion between politicians and civilians. Are there certain economic conditions that aid dependent countries need to fulfill in order for aid to encourage growth? Or is it a waste of money anyway? These important considerations lead to my research question:

Does foreign aid have a positive impact on growth conditional on good economic policies?

The research is based on the influential paper of Burnside and Dollar, which is discussed in more detail in the next chapter, extending the time frame of the original paper and adding additional variables to the model. Using different econometrical techniques such as OLS and 2SLS, the results obtained show support of a positive direct effect of aid on growth. The conditional effect of aid on growth is found to be negative.

The remainder of this thesis is structured as follows: Chapter I gives a review of the existing literature on the subject. Chapter II elaborates the methodology used, together with my hypotheses along with a description of the data. Chapter III presents the results, followed by a conclusion and a discussion of these results in the last chapter.

Chapter I: Literature review

The aid effectiveness debate is a diverse discourse. It is therefore convenient to distinguish two causal links. The first link focuses on the direct link between aid and growth, meaning that aid directly affects growth. The second link is the conditionality link. This link focuses on the possibility that aid affects growth, conditional on some other factor(s).

(7)

I.1. Direct effect of aid on growth

Aid is assumed to induce growth or fight poverty for it to be effective. As was mentioned in the introduction, the ongoing debate on the effectiveness of aid has proponents and opponents. The former show an unconditional positive impact of aid on growth; the latter are unable to find a positive impact of aid on growth.

The study of Dalgaard et al. (2004) uses different econometrical techniques, such as OLS and GMM, on a panel data set consisting of 56 aid receiving countries to answer whether aid is effective in causing growth. These countries are located in East Asia, Africa and Latin America. They find a robust and significant positive impact of aid on growth. They immediately weaken this result by saying that aid on itself is not the cure of poverty, meaning that aid does not cause poverty reduction but only a positive impact of aid on growth. They wonder why aid works better in some countries than in others. The most significant evidence is geography. Aid in countries located in tropical areas is less effective. The authors do not believe that geography is the cause of this ineffectiveness and call for further research on this topic.

On the other hand, the study of Boone (1996) finds insignificant evidence on the impact of aid on investment using OLS and IV panel regressions, but also insignificant evidence on improvements in infant mortality and primary schooling ratios. His data set consists of 97 countries that received aid (with the exclusion of OPEC countries and Israel). With little or no impact on growth via investment, and little or no impact on poverty reduction via infant mortality and education, Boone concludes that aid is ineffective.

Rajan and Subramanian (2008) also find no evidence of a positive relationship between aid and growth. They address a number of robustness checks. It turns out that their conclusion holds irrespective of time periods; time horizons; cross-section or panel data; good or bad policies and institutions; being in the tropics or outside among others. They argue that this ineffectiveness might come from too much noise in the data, making relationships hard to establish with growth regressions.

Clemens et al (2004) remark that previous research on aid and growth is imperfect, due to the fact that none of these studies distinguishes between categories of aid. When they use short-term aid, or aid that potentially impacts growth within four years like budget and balance of payment support, instead of aggregate aid, they find a large and positive effect on growth. It is labeled large due to the fact that it is two-to-three times larger than comparable studies using aggregate aid.

(8)

As displayed above there are numerous studies on the direct effect of aid on growth. Due to use of different techniques, different sample data or a different timeline, these studies differ in their conclusions. The meta-study of Doucouliagos and Paldam (2008) looks at 68 studies that investigate the direct effect of aid on growth. They mention that the majority of the authors agree upon a small positive effect on growth, but concluded something different. There seems to be too little evidence to accept that aid has a significantly positive effect on growth. This meta-study did not include conditional effects of aid on growth.

I.2. Conditional effect of aid on growth

The influential paper of Burnside and Dollar (2000) looks, with the use of panel growth regressions, at the impact of foreign aid on GDP per capita growth, conditioned on economic policies. These policies include fiscal policy (budget balance), monetary policy (price stability), and trade policy (degree of openness). The way they investigate this is by adding an interaction term of aid and policy. They find that the impact of aid is more positive in an environment of good policy (i.e. low inflation, positive budget balance, trade openness), while aid does not positively contribute to growth in the presence of “poor” policies.

The study of Durbarry et al. (1996) mentions similar results. With the use of different econometric techniques (cf. cross-country regression, panel regression) and different data samples, they find robust evidence that higher levels of aid have a positive effect on growth in least developed countries conditional on stable macro-economic policy (in their case they used trade openness and price stability). The difference with the study of Burnside and Dollar (2000) is that those authors use an interaction term between aid and policy, while Durbarry et al. do not include such an interaction term. They divide their sample into low and middle-income countries, and conclude that aid is more effective in raising growth rates in the middle-income countries, thus implying that the more favorable macroeconomic policy environments in the middle-income countries result in more effective aid. Another result from Durbarry et al. is that very high levels of aid to GDP lead to slower growth. This is evidence of diminishing returns of aid.

The study of Svensson (1999) concludes that aid has a positive impact on growth given that there are institutionalized checks on the governmental power. These

(9)

checks ensure civil and political liberties (for instance freedom of press). If these checks are properly institutionalized, i.e. well protected from within the country, governmental power is lower, which results in a more democratic society. Svensson also mentions that the data suggests that if those checks are not present, the aid will be squandered on the government’s own private goals. The study of Clemens et al. (2004) also mentions that the impact of aid on growth is greater for countries with strong institutions. More openness to trade, lower inflation and lower budget deficits are associated with higher growth rates ceteris paribus.

Burnside and Dollar revisit their original paper with a new data set focusing on the 1990s (2004). They arrive at similar conclusions, strengthening the robustness of their research. There is again evidence that aid impacts growth conditional on institutions using this new data set.

I.3 Criticism on conditional effect

The research of Burnside and Dollar (2000) fueled a number of critical studies. One of these studies is Hansen and Tarp (2001). They voice their criticism on the findings of Burnside and Dollar and other aid/growth regressions, like Durbarry et al. (1996). They argue that there is too much difference in research methods in the aid effectiveness debate, like the exclusion of poorly understood nonlinearities and critical methodological choices. In their paper they tackle these problems by adding them to their own model, essentially combining the critiqued models. The nonlinearity, which measures if aid has a diminishing effect, is excluded in the analysis in Burnside and Dollar (2000), because it was dependent on five large outliers. Dropping these outliers from the dataset resulted in insignificant outcomes for the diminishing effect. Excluding the diminishing effect in total is in the view of Hansen and Tarp wrong. On the subject of diminishing aid, Lensink and White (2001) investigate the existence of a Laffer curve of aid. A Laffer curve is the relationship between the level of tax and the tax revenue, which has a parabolic shape with a maximum (cf. Trabandt & Uhlig, 2011). Lensink and White’s empirical evidence suggests the existence of such a curve for aid, meaning that higher levels of aid do have a diminishing effect.

Durbarry et al. did not include an interaction term between aid and policy, which in the opinion of Hansen and Tarp is a poor methodological choice if the

(10)

conditional effect of policy on aid is investigated. So the Burnside and Dollar article, as well as the paper of Durbarry et al. omit an important aspect according to Hansen and Tarp. Their paper searches for regularities of the impact of aid among the studies and do show significant positive impact, irrespective of good or bad policies. Also, Easterly et al. (2003) conduct a follow-up study of the Burnside and Dollar (2000) article. They increase the data set and fill in the missing data for the original period. The conclusions of Burnside and Dollar appear to be less robust than is stated. Easterly et al. do not find significant evidence of a positive interaction between aid and policy, but also they do not find evidence for diminishing returns to aid.

The previously cited study of Dalgaard et al. (2004) mentions not only a positive impact of aid on growth, it also finds the evidence of a strong interaction between aid and policy to be insignificant. A study of Hudson and Mosley (2001) has similar criticism on the effect of good policy. Good policy appears to have a positive effect on growth, but it does not have a significant effect on aid effectiveness.

Again, the effectiveness of aid conditional on good policy is highly debated. Different data sets result in opposite conclusions. Doucouliagos and Paldam also investigate the conditional debate with the use of a meta-study (2010). In their study of 40 articles, they conclude that the interaction between aid and good policy does not differ significantly from zero. Good policies do spur growth, but their influence on aid effectiveness is marginal. They also mention that the reason why the Burnside and Dollar (2000) article got so much attention is that it was successful due to the evidence available at that time. Subsequent studies (cf. Easterly et al., 2003) increased the length of the data to fill in the missing gaps, thereby proving Burnside and Dollar’s conclusions not entirely correct.

I.4 Effects of other variables on growth

Burnside and Dollar’s growth model, which is discussed in more detail in the next chapter, employs several control variables besides the aid variables in order to estimate the effects on growth.

The previously discussed studies of Dalgaard et al. (2004), Hansen and Tarp (2001), Hudson and Mosley (2001), Easterly et al. (2003) and Burnside and Dollar (2000, 2004) all include the same three policy elements. The first is budget balance. A positive budget balance (i.e. a surplus) has a positive impact on growth; a negative

(11)

budget balance (i.e. a deficit) has a negative impact on growth. There are several channels through which this can be explained. One of these channels is the signaling channel (Fisher, 1993). A surplus signals that a government is doing a good job, a deficit therefore does the opposite and this increases uncertainty. Increased uncertainty makes it less attractive to invest, which harms growth. Another channel is the tax channel. Easterly and Rebelo (1993) mention that higher taxes like income tax, lowers the net rate of return on private investment. This makes investment opportunities less attractive. When a government has a budget deficit, the chance of an increase in taxes is higher than with a budget surplus. Both the studies of Fisher (1993) and Easterly and Rebelo (1993) show a negative relationship between budget deficit and growth, i.e. a positive relationship between growth and budget surplus, using cross-sectional historical data.

An alternative for budget balance is government consumption, as proposed by Easterly and Rebelo (1993). Barro (2001) argues that government consumption measures expenditure of government that does not directly enhance productivity. Increasing government consumption therefore decreases productivity, hence growth. Several studies investigate the effect of government consumption on growth and is found negative (cf. Barro, 1989; Easterly & Rebelo, 1993; Barro & Lee, 1994; Burnside and Dollar, 2000; Barro, 2001; Burnside and Dollar 2004). Burnside and Dollar use this variable in their growth regression as a robustness check.

The second policy element is inflation. Fisher (1993) argues that inflation and growth are negatively associated, and presents the evidence to back it up using a panel growth regression. The main reason for the negative sign is uncertainty: higher levels of inflation increase uncertainty, which in turn affects likeliness of investment and therefore growth. Fisher’s panel and cross-sectional growth regressions confirm this negative relationship, which is in line with outcomes of other studies (cf. Burnside & Dollar, 2000; Hansen & Tarp, 2001; Dalgaard et al., 2004).

The last policy element is trade openness. Standard economic textbooks on international trade (cf. Krugman et al., 2015) explain that trade between two countries is beneficial because of comparative advantages. Both countries can produce the product at which they have a comparative advantage. The resulting trade increases welfare in both countries. Harrison (1996) argues that openness to trade has an impact on technological change, which in turn has an impact on long-run growth. Openness provides access to imported inputs, and therefore new technology. She finds a

(12)

positive association using several different measures of trade openness with cross-country and panel data, which is again in line with the other studies (cf. Burnside & Dollar, 2000; Hansen & Tarp, 2001; Dalgaard et al., 2004).

Most of the studies (cf. Burnside & Dollar, 2000; Dalgaard et al., 2004; Hansen & Tarp, 2001; Barro, 2001) add other variables to their growth regressions, namely a measure of convergence and a measure of political stability. One implication of the Solow growth model is that countries with a lower GDP per capita grow faster than countries with a higher GDP per capita due to higher marginal product of capital, thus these countries display convergence. Parente and Prescott (1993) did not find evidence for convergence, i.e. the distance between the richest and the poorest countries in their dataset remained almost the same. On the other hand, Barro and Lee (1994) find convergence to be one of the main determinants of growth. The evidence for convergence in studies such as Burnside and Dollar (2000) and Dalgaard et al. (2004) is mixed. Not all studies conclude convergence due to insignificant results. The political stability has a positive impact on growth, because of reduced uncertainty and favorable business environment (cf. Barro, 1989).

All in all, the studies discussed above use similar variables in their growth regressions, and find similar results. Any deviation of the outcomes of these variables in this paper is to be analyzed further.

This paper will also investigate if the conclusions of Burnside and Dollar are correct. It will increase the number of data points by looking at not only 20th century data, but also more recent figures. Besides that, the hypothesized diminishing effect of aid will be investigated. Furthermore, the outcomes of other key variables will be discussed.

Chapter II: Methodology

In order to analyze the effectiveness of aid on growth, this chapter introduces the model and variables used by Burnside and Dollar along with the elements added by this paper. The chapter begins with an explanation of the dataset; at the end the hypotheses are discussed. The applied econometric techniques are also considered.

(13)

II.1 Data set

The countries that are included in the dataset are the same set of countries used by Burnside and Dollar. It consists of 26 African countries, 21 countries from Latin America and the Caribbean, and 9 countries from Asia, summing up to 56 countries. The only difference with Burnside and Dollar is the exclusion of Korea, due to the fact that nowadays it is a highly developed high-income country and the focus of this paper is also on more recent data making Korea an outlier. For a full list of the countries that are in the data set, refer to the appendix. The sources of the data are the

World Development Indicator database of the World Bank and the World Economic Outlook Database of the IMF.

There are eight periods, each spanning four years. The first period is from 1983 till 1986; the last period is from 2011 till 2014. The values of the variables for a period are given by the average values within these periods. Some countries do not have data-points for all eight periods. Thus the dataset is unbalanced.

II.2 The model

The model used in this paper is based on the model constructed by Burnside and Dollar (2000). This model consists of the following elements:

𝐺𝐺𝑖𝑖𝑖𝑖 = 𝑐𝑐 + 𝑏𝑏1𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖+ 𝑏𝑏2𝐴𝐴𝑖𝑖𝑖𝑖+ 𝐵𝐵′3𝑃𝑃𝑖𝑖𝑖𝑖+ 𝑏𝑏4𝐴𝐴𝑃𝑃�𝑖𝑖𝑖𝑖+ 𝐵𝐵′5𝑋𝑋𝑖𝑖𝑖𝑖+ 𝛾𝛾𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (E1)

The dependent variable is 𝐺𝐺_{𝑖𝑖𝑖𝑖}, which is the growth rate of real GDP per capita in country 𝑖𝑖 at year 𝑡𝑡. The first independent variable is 𝑙𝑙𝑙𝑙𝑙𝑙_{𝑖𝑖𝑖𝑖}: the log of real GDP per capita of country 𝑖𝑖 at the beginning of period 𝑡𝑡 . This variable examines the convergence of countries. A negative sign would indicate convergence. Although the evidence for convergence is mixed, as is discussed in the previous chapter, the expectations are that there is some convergence.

Next is 𝐴𝐴_{𝑖𝑖𝑖𝑖}, which represents the level of official development aid as a percentage of GDP of country 𝑖𝑖 at period 𝑡𝑡. The expected sign is positive. Referring to the previous chapter, previous empirics on this is mixed, with aid having a positive but sometimes insignificant effect on growth.

𝑃𝑃𝑖𝑖𝑖𝑖 is a vector of policy variables, which include budget balance, level of

(14)

consider the budget balance variable in their analysis as fiscal policy variable, they also use government consumption as a percentage of GDP as a robustness check as proposed by Easterly and Rebelo (1993). There is more data on government consumption in the World Development Indicator database, therefore government consumption is used primarily in this paper instead of budget balance. The expectations of the policy variables have been discussed already.

Other exogenous variables are included in the 𝑋𝑋_{𝑖𝑖𝑖𝑖} vector. It consists of level of M2 monetary aggregates, which controls for the effect of financial markets on growth (King & Levine, 1993), and dummy variables for Sub-Saharan nations and Latin-American & Caribbean nations. The expected sign for the M2 variable is negative, which is a proxy for financial market distortions. King and Levine (1993) argue that these distortions increase the costs of investment, making it harder to mobilize funds and find the appropriate investment opportunities. Sub-Saharan nations and Latin-American nations tend to have lower growth rates. The expected signs of these dummy variables are therefore negative. A political stability variable is not added due to lack of availability. The variable used by Burnside and Dollar and others, namely the number of assassinations, is not available for the time period considered. Several other variables used by Burnside and Dollar are also not employed due to lack of availability3.

The term 𝛾𝛾_𝑖𝑖 indicates a period dummy to account for the world business cycle. 𝑐𝑐 is the constant term, 𝑏𝑏𝑗𝑗 is the coefficient of the corresponding independent variable

(note that 𝐵𝐵₃ and 𝐵𝐵₅ are vectors of coefficients), and 𝜀𝜀_{𝑖𝑖𝑖𝑖} is the mean zero error term. 𝑃𝑃�𝑖𝑖𝑖𝑖 is defined as follows:

𝑃𝑃�𝑖𝑖𝑖𝑖 = 𝑎𝑎0+ 𝐴𝐴′1𝑃𝑃𝑖𝑖𝑖𝑖 (E2)

This creates a policy index from the policy variables used in a linear way. This way a single interaction term is created between aid and policy (𝐴𝐴𝑃𝑃�_{𝑖𝑖𝑖𝑖}), making it easier to analyze the interaction. The calculations on 𝑃𝑃�_{𝑖𝑖𝑖𝑖} are elaborated in the next chapter.

3

These variables include the institutional quality measure of Knack and Keefer (1995), the ethnolinguistic fractionalization variable by Easterly and Levine (1997) and the interaction term between ethnolinguistic fractionalization and the

(15)

As was stated at the end of the literature review, the point of criticism that is addressed in this paper is the diminishing effect of aid. This is done by adding an aid-squared term to the model:

𝐺𝐺𝑖𝑖𝑖𝑖 = 𝑐𝑐 + 𝑏𝑏1𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖+ 𝑏𝑏2𝐴𝐴𝑖𝑖𝑖𝑖+ 𝑏𝑏3𝐴𝐴𝑖𝑖𝑖𝑖2 + 𝐵𝐵′4𝑃𝑃𝑖𝑖𝑖𝑖+ 𝑏𝑏5𝐴𝐴𝑃𝑃�𝑖𝑖𝑖𝑖+ 𝑏𝑏6𝐴𝐴2𝑃𝑃�𝑖𝑖𝑖𝑖+ 𝐵𝐵′7𝑋𝑋𝑖𝑖𝑖𝑖

+ 𝑎𝑎𝑖𝑖+ 𝛾𝛾𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖

(E3)

Notice that the interaction term between aid squared and the policy index is also added, due to the possible diminishing effect of aid. The model incorporates individual effects (𝑎𝑎_𝑖𝑖), time effects (𝛾𝛾_𝑖𝑖) and time-invariant dummies (for Sub-Saharan Africa and Latin America & Caribbean). When a fixed effect regression is estimated, the time-invariant dummies are excluded, due to the fact that the effect of these dummies cannot be estimated in a fixed effects regression. Standard OLS regression techniques are used to estimate the model. The model therefore assumes that all the variables are exogenous.

(16)

Table T1. Summary of variables Variable Description Used by B&D Remark Expected impact 𝑮𝑮𝒊𝒊𝒊𝒊

Real per capita GDP

growth yes -

𝒍𝒍𝒍𝒍𝒍𝒍𝒊𝒊𝒊𝒊

Real per capita GDP at

beginning of period t yes < 0

𝑨𝑨𝒊𝒊𝒊𝒊

Net ODA as a percentage

of GDP yes > 0 𝑨𝑨_{𝒊𝒊𝒊𝒊}𝟐𝟐 _{The square of}_𝐴𝐴 𝑖𝑖𝑖𝑖 no Hypothesis < 0 𝑷𝑷𝒊𝒊𝒊𝒊 Government consumption as a percentage of GDP yes Alternative: Government net borrowing or lending as a percentage of GDP < 0 (alt. > 0)

Inflation (based on CPI) yes < 0

Trade openness index: sum of exports and import

as share of GDP

no

Burnside and Dollar use a different measure for trade openness

> 0

𝑷𝑷�𝒊𝒊𝒊𝒊 Policy index based on 𝑃𝑃𝑖𝑖𝑖𝑖 yes -

𝑨𝑨𝑷𝑷�𝒊𝒊𝒊𝒊

Interaction term between 𝐴𝐴𝑖𝑖𝑖𝑖 and 𝑃𝑃�𝑖𝑖𝑖𝑖

yes Hypothesis > 0

𝑨𝑨𝟐𝟐_𝑷𝑷� 𝒊𝒊𝒊𝒊

Interaction term between 𝐴𝐴𝑖𝑖𝑖𝑖2 and 𝑃𝑃�𝑖𝑖𝑖𝑖

no < 0

𝑿𝑿𝒊𝒊𝒊𝒊

Money and quasi money (M2) as a percentage of GDP (lagged one period)

yes < 0

Regional dummies yes

SSA is Sub-Saharan Africa, LAC is Latin America & Caribbean

< 0

(17)

II.3 Endogeneity problems

It is, however, not clear that all variables are actually exogenous. Other researchers mention this problematic assumption frequently (cf. Boone, 1996; Burnside & Dollar, 2000; Dalgaard & Hansen, 2001; Hansen & Tarp, 2000). If this assumption is relaxed for aid, meaning that aid can affect growth, but growth can also affect aid, the problem of simultaneous causality occurs. In order to address this problem Stock and Watson mention that a 2SLS regression, in which aid is instrumented, can mitigate this problem (2015). The instruments proposed by Burnside and Dollar (2000) are population, infant mortality rate and proxies for donor’s strategic interest (i.e. dummies for countries being in the Franc Zone in Africa and Egypt). The same donor’s strategic interest dummies are used to stick as close as possible to the paper of Burnside and Dollar. The interaction terms 𝐴𝐴𝑃𝑃�_{𝑖𝑖𝑖𝑖} and 𝐴𝐴2𝑃𝑃�_{𝑖𝑖𝑖𝑖} are instrumented by interaction terms between population and policy, and infant mortality and policy. Dalgaard and Hansen propose a different set of instruments (2001). A summary of the instruments is given in table T2.

It is possible that other variables, that are assumed to be exogenous, actually are endogenous. Inflation is the rise in the price level due to changes in the money supply. This money supply is managed by the local central bank. Any changes to the money supply is decided by the central bank, therefore the assumption that inflation is exogenous is a reasonable one. Exogeneity of the M2 variable is ensured, due to the fact that the variable is lagged.

The assumption that trade variable used in this paper is exogenous is probably a weak assumption. The level of imports and exports are indeed influenced by growth. A classic example is China, which is exporting and importing more because of its growth in comparison with 50 years ago. Dollar and Kraay (2004) propose to use the trade openness variable lagged one period as an instrument for trade. Another instrument is a dummy for being landlocked, as proposed by Frankel and Romer (1999). They argue that not being connected to the sea has an impact on trade.

(18)

Table T2. Summary of instruments

Instruments for aid Instruments for trade Instruments Burnside and

Dollar

Instruments Dalgaard and Hansen

Log(population) Log2(population)

Infant Mort Infant Mort2 Policy x infant mort Policy x log(population)

Dummy Egypt Dummy Franc Zone

Aid/GDP lagged Aid/GDP2 lagged Policy x aid/GDP lagged Policy x log (initial GDP) Policy x log2(initial GDP) Policy x log(population)

Dummy Franc Zone

Trade/GDP lagged Dummy landlocked

II.4 Hypotheses

To test if aid has a positive impact on growth conditional on good policy environment the estimated coefficient of 𝑏𝑏₅ of equation (E3) is considered. A positive impact means that the estimated coefficient of 𝑏𝑏₅ is positive. This results in the following null and alternative hypothesis:

𝐻𝐻0: 𝛽𝛽5 = 0

𝐻𝐻1: 𝛽𝛽5 > 0

To test whether aid has indeed a diminishing effect, the estimated coefficient of 𝑏𝑏₃ must be negative. The resulting null and alternative hypothesis are:

𝐻𝐻0: 𝛽𝛽3 = 0

𝐻𝐻1: 𝛽𝛽3 < 0

II.5 Assumptions

When using OLS and 2SLS techniques on panel data some assumptions are made (Stock & Watson, 2015), namely:

(19)

(A2) All variables and errors are independent identically distributes (i.i.d.) draws from their joint distribution

(A3) Large outliers are unlikely

(A4) There is no perfect multicollinearity

The first assumption can be analyzed by the graphs presented in Appendix B. Graph (G1) shows no clear pattern between the fitted values and the residuals. Graphs (G2) to (G4) compare the distribution of the residuals to the normal distribution. These graphs indicate that the residuals are indeed approximately normally distributed with mean zero. Thus assumption (A1) holds. The standard errors of fixed effects regression models are robust to both heteroscedasticity and to correlation over time. This means that assumptions (A2) and (A4) hold for fixed effects models. However, not all models make use of fixed effects, but the most important ones do. Therefore, assumptions (A2) and (A4) are maintained. The validity of assumption (A3) is discussed later.

In order to answer the research question, this chapter outlines the methods used for this purpose. Equation (E3) is estimated using standard OLS panel regression. Due to a simultaneous causality between aid and growth (and trade and growth) these estimates might be biased, therefore also a 2SLS panel regression, in which aid (and trade) is instrumented, is estimated.

Chapter III: Results

This chapter displays the results from the various regressions along with an analysis of these results. First a summary of the main variables is displayed, followed by the OLS results. After that the IV regressions are discussed.

III.1 OLS results

Table T3 gives a summary of the main variables used in the analyses. Most of the variables are in percentages (except 𝑙𝑙𝑙𝑙𝑙𝑙_{𝑖𝑖𝑖𝑖}, population, and infant mortality). To give an example the average level of per capita GDP growth is 1.46%, the lowest level of trade to GDP is 12.98%, and the highest level of aid received is 35.86% of GDP.

(20)

Notice that the mean value of inflation is 101.9% which is large. The standard deviation of inflation is also large, indicating that inflation is volatile in this data set. This can be attributed to very high levels of inflation (hyperinflation) in countries such as Zimbabwe. These observations might violate assumption (A3). This problem is discussed in more detail in the next chapter.

Table T4 displays the estimated coefficients and corresponding test statistic in brackets of standard OLS panel regression. Three stars indicates that the coefficient is significantly different from zero at a 1% significance level; two stars indicates significance at a 5% significance level; one star indicates significance at a 10% significance level. Regression (R1) is used to calculate the policy index 𝑃𝑃�_{𝑖𝑖𝑖𝑖} from equation (E2). The constant term 𝛼𝛼₀ of equation (E2) is the sum of the estimated coefficients of regression (R1) (excluding inflation, trade, and government consumption) multiplied by their corresponding mean value. The values of 𝐴𝐴₁of equation (E2) are taken directly from the estimated coefficients of regression (R1). The resulting policy index is follows:

𝑃𝑃�𝑖𝑖𝑖𝑖 = 0.0122 − 0.0008 ∗ 𝐼𝐼𝑙𝑙𝐼𝐼𝑙𝑙. +0.0201 ∗ 𝑇𝑇𝑇𝑇𝑎𝑎𝑇𝑇𝑇𝑇 − 0.0754 ∗ 𝐺𝐺𝐺𝐺𝐺𝐺. 𝐶𝐶𝐺𝐺𝑙𝑙𝐶𝐶. (E4)

Table T3. Summary of Statistics

Variable Observations Mean Minimum Maximum Std. Dev

𝐺𝐺𝑖𝑖𝑖𝑖 447 0.0146 -0.1251 0.1103 0.029 𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 443 7.221 4.809 9.576 1.077 𝐴𝐴𝑖𝑖𝑖𝑖 443 0.0522 -0.0019 0.3586 0.067 Inflation 448 1.019 -0.0293 122.1 7.771 Trade 431 0.6434 0.1298 2.480 0.319 Gov. Cons. 426 0.1307 0.0100 0.3447 0.045 M2 (lagged) 444 0.3758 0.0246 1.316 0.220 Budget Bal. 326 -0.0248 -0.1489 0.0943 0.034 Population 448 49.3m 0.72m 1270m 140m Infant Mort. 448 53.54 6.48 161.03 34.358

(21)

Regression (R2) to (R5) use OLS to determine the effect of aid on growth, and the conditional effect of aid on growth. Aid has a positive impact on growth, being significant in all regressions except (R2). The hypothesis of diminishing aid is supported only little in these regressions, due to the negative sign of the aid-squared variable being only significant in one of the regressions. Furthermore, the hypothesis of aid having a bigger impact in a good policy environment is not supported in these regressions. The sign of the interaction term between aid and policy is negative, indicating that the null hypothesis cannot be rejected in favor of the alternative hypothesis. The significant negative sign is counterintuitive: it indicates that in countries which do well with respect to the three policy components, aid has less effect (or even a negative effect) on growth. Complementary to this is that in countries that conduct bad policy, aid has a constructive effect on growth. This will be discussed further in Chapter IV.

Regression (R6) uses the budget balance variable instead of the government consumption variable as a proxy for sound fiscal policy. As is mentioned in the methodology, Burnside and Dollar primarily use this budget balance variable in their analysis and use the government consumption variable as a robustness check. This paper does it the other way around, due to the fact that the government consumption variable is more complete. Referring to table T3, the number of observations for government consumption is 426 (95% coverage), while the number of observations for budget balance is 326 (73% coverage). Nevertheless, regression (R6) yields the same results as regression (R5), in spite of the coefficients being less significant.

Regression (R2) is a random effects regression, whereas regression (R3) to (R6) are fixed effects regressions. The Hausman test checks whether it is appropriate to use fixed effects in favor of random effects. Under the null hypothesis the random effects model is preferred. A significant Hausman test statistic thus means that the fixed effect model is preferred over the random effect model, and therefore the fixed effects model is used for that particular regression. The corresponding Hausman 𝜒𝜒2 test statistics are given below in table T4.

The other variables in the regressions (R2) to (R6) are robust and significant. The three policy variables plus budget balance have the expected effect on growth. The M2 variable and the regional dummies also have the expected signs. There is, however, no evidence for convergence, due to the positive but insignificant sign of 𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖.

(22)

Table T4. OLS regression outcomes (R1) RE (R2) RE (R3) FE (R4) FE (R5) FE (R6) FE Variable 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} Constant 0.0077 (0.54) 0.0001 (0.00) 0.0070 (0.13) -0.0121 (0.23) -0.0071 (0.13) -0.0604 (0.77) 𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 0.0018 (0.96) 0.0031 (1.27) 0.0014 (0.18) 0.0033 (0.44) 0.0028 (0.36) 0.0079 (0.71) Inflation -0.0008*** (5.55) -0.0008*** (5.48) -0.0011*** (5.16) -0.0011*** (5.09) -0.0017*** (5.13) -0.0002 (0.29) Trade 0.0201*** (3.74) 0.0192*** (3.42) 0.0460*** (4.66) 0.0462*** (4.70) 0.0498*** (5.03) 0.0240* (1.80) Gov. Cons. -0.0754** (2.31) -0.0828** (2.47) -0.1838*** (4.14) -0.1752*** (3.93) -0.1973*** (4.35) - M2 (lagged) -0.0246** (2.61) -0.0247** (2.60) -0.0649*** (4.89) -0.0652*** (4.92) -0.0668*** (5.07) -0.0347** (2.17) SSA -0.0252*** (4.77) -0.0259*** (4.71) - - - - LAC -0.0160*** (3.39) -0.0170*** (3.43) - - - - 𝐴𝐴𝑖𝑖𝑖𝑖 - 0.0229 (0.71) 0.1370*** (2.62) 0.2674*** (2.71) 0.50668** (3.60) 0.4728* (1.77) 𝐴𝐴_{𝑖𝑖𝑖𝑖}2 _- _- _- -0.4362 (1.56) -1.432*** (2.84) -1.605 (1.35) 𝐴𝐴𝑃𝑃�𝑖𝑖𝑖𝑖 - - -5.337** (2.42) -5.074** (2.30) -20.51*** (2.98) -9.372 (0.63) 𝐴𝐴2_𝑃𝑃� 𝑖𝑖𝑖𝑖 - - - - 63.57** (2.36) 32.56 (0.48) Bal. Budg. - - - 0.1574*** (3.40) Obs. 422 419 419 419 419 308 R2 0.3137 0.3178 0.3552 0.3597 0.3698 0.2956 Hausman 10.9 7.25 39.3*** 42.4*** 43.8*** 24.1*** *** p<0.01; ** p<0.05; * p<0.1

(23)

III.2 2SLS results

As is mentioned in Chapter 2, there is a problem of simultaneous causality between aid and growth and between trade and growth. This causes the estimated coefficients of standard OLS regression to be biased. To mitigate this problem, a two stage least squares regression has to be considered.

Regressions (R7) is a 2SLS regressions where only aid is instrumented using the instruments of Burnside and Dollar. This means that the variables 𝐴𝐴_{𝑖𝑖𝑖𝑖}, 𝐴𝐴2_{𝑖𝑖𝑖𝑖}, 𝐴𝐴𝑃𝑃�𝑖𝑖𝑖𝑖, 𝐴𝐴2𝑃𝑃�𝑖𝑖𝑖𝑖 are all instrumented. The signs of the estimated coefficient are the same

as the OLS regression results. Regression (R8) instruments both aid and trade using the instruments of Burnside and Dollar and the two trade instruments. Regressions (R9) and (R10) use the instruments of Dalgaard and Hansen4 instead of Burnside and Dollar’s. Although the signs of the coefficients remain consistent over these four regressions, there is more difference in the size, making the 2SLS results less robust than the OLS results. The evidence of a positive effect of aid on growth remains, as well as the evidence for the diminishing effect of aid (which is now stronger). The counterintuitive negative sign of the aid policy interaction term is not very robust, due to the fact it is not always significant and differs in size. Nevertheless, there is no evidence of a positive effect of policy on aid, making it impossible to reject the null hypothesis.

To test whether the instruments used are valid, the Sargan test or overidentifying restrictions test is employed. Under the null hypothesis the instruments are exogenous; rejecting the null hypothesis means the instruments are endogenous and therefore not valid. The Sargan test statistic is not significant in all four regressions, meaning we do not reject the null hypothesis. This indicates that the over-identifying instruments are valid.

To test if the OLS results are valid in favor of the 2SLS results, the Durbin-Wu-Hausman test for endogeneity is conducted. This tests whether the results obtained from OLS are consistent or not. Under the null hypothesis of this test OLS is consistent. When aid is instrumented by the set of instruments of Burnside and Dollar, the corresponding Durbin-Wu-Hausman test statistic is 2.65 with a p-value of 0.10. The null hypothesis is not rejected, so in this case OLS is consistent. When aid is

4

The instrument of Policy x log (initial GDP) is excluded, decreasing the Sargan statistics to acceptable levels

(24)

instrumented by the set of instruments of Dalgaard and Hansen, the corresponding Durbin-Wu-Hausman test statistic is 8.72 with a p-value of 0.003. In this case the null hypothesis is rejected, making OLS is not consistent. Instrumenting trade results in a Durbin-Wu-Hausman test statistics of 9.31 with corresponding p-value of 0.002, resulting again in rejecting the null hypothesis and therefore OLS is inconsistent. Hence, OLS consistency cannot be concluded from the Durbin-Wu-Hausman tests.

Combining the outcomes of the Sargan tests and the Durbin-Wu-Hausman tests, the regressions (R9) and (R10) are the most appropriate regressions. OLS turns out to be inconsistent in these regressions, and the instruments used in these regressions are valid according to the Sargan test. Regressions (R7) and (R8) indicate that OLS is consistent according to the insignificant Durbin-Wu-Hausman statistic, which contradicts the conclusion of the Durbin-Wu-Hausman tests of the other regressions.

The estimated coefficients of the policy variables and other exogenous variables are as expected. In regressions (R9) and (R10) the variables are more significant than in regressions (R7) and (R8), increasing the reliability of the (R9) and (R10) over (R7) and (R8).

(25)

5

The output of regression (R10) did not give a R2 number

Table T5. 2SLS regression outcomes

(R7) FE (R8) RE (R9) FE (R10) FE Variable 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} 𝐺𝐺_{𝑖𝑖𝑖𝑖} Constant -0.1124 (0.87) 0.0781 (1.12) -0.0901 (1.37) -0.1913* (1.71) 𝑙𝑙𝑙𝑙𝑙𝑙𝑖𝑖𝑖𝑖 0.0132 (0.82) -0.0089 (0.86) 0.0128 (1.36) 0.0257* (1.69) Inflation -0.0018 (1.11) -0.0046** (2.36) -0.0013** (2.30) -0.0065** (2.45) Trade 0.0391** (2.22) 0.0266 (1.17) 0.0390*** (2.98) 0.0735*** (2.60) Gov. Cons. -0.1129 (1.58) -0.0168 (0.19) -0.1725*** (2.72) -0.4345*** (2.92) M2 (lagged) -0.0645*** (4.22) -0.0324* (1.81) -0.0684*** (4.83) -0.0874*** (3.76) SSA - -0.0201* (1.67) - - LAC - -0.0015 (0.12) - - 𝐴𝐴𝑖𝑖𝑖𝑖 1.431** (2.30) 1.545* (1.67) 0.8209*** (2.56) 3.570** (2.38) 𝐴𝐴𝑖𝑖𝑖𝑖2 -5.122* (1.76) -9.560** (2.03) -2.159* (1.65) -11.58** (2.60) 𝐴𝐴𝑃𝑃�𝑖𝑖𝑖𝑖 -30.87 (0.70) -133.9** (2.32) -13.06 (0.89) -135.8** (2.18) 𝐴𝐴2_𝑃𝑃� 𝑖𝑖𝑖𝑖 144.7 (0.74) 720.4** (2.36) 41.17 (0.65) 490.5** (2.14) Obs. 419 415 417 414 R2 0.1916 0.0633 0.3276 5 Hausman 𝜒𝜒2 11.7** 0.91 41.1*** 14.8** Sargan test 1.66 1.77 2.23 0.23 *** p<0.01; ** p<0.05; * p<0.1

(26)

This chapter outlines the results obtained from several OLS and 2SLS regression. The OLS results are robust and do not support the first hypothesis of a positive impact of aid on growth conditional on policy; there is some evidence for the diminishing effect of aid. The 2SLS results are less robust. They also do not support the first hypothesis, but the evidence for the diminishing effect of aid is more significant. The effects of other variables are in line with expectations.

Chapter IV: Conclusion

This chapter answers the research question posed at the beginning of the thesis, along with other conclusion from the obtained results. Afterwards the conclusions are discussed and suggestions for further research are given.

IV.1 Conclusions

This thesis looks at the effect of aid on growth in a sample of 56 developing countries. Using different econometrical techniques, there is enough evidence to conclude that aid has a positive impact on growth: the estimated coefficients of aid is positive in all nine regressions, and significant in eight of them. The effect of aid on growth conditional on good economic policy, is found negative and significant in five out of eight regressions. This gives a disconfirming answer to the research question posed at the beginning of this thesis It also contradicts the findings of the research conducted by Burnside and Dollar, on which this thesis is based.

Furthermore, the results show a robust negative effect of the estimated aid-squared coefficient that is significant in five out of seven regressions. Therefore, the null hypothesis of the second hypothesis is rejected in favor of the alternative. This indicates that aid has a diminishing effect on growth.

IV.2 Discussion

The results obtained in chapter III do not support the hypothesis of a positive impact of aid on growth in a good policy environment. In fact, the evidence points in the opposite direction: a bad policy environment is more advantageous for the effectiveness of aid when compared to a good policy environment. This is a counterintuitive result. The cause of this result might be one of the following.

(27)

First, some of the variables used by Burnside and Dollar are not used in this paper due to lack of availability. These variables include number of assassinations as a proxy for country stability, ethnic fractionalization and institutional quality measures to capture long-term characteristics of countries, and an interaction term between assassinations and ethnic fractionalization. Excluding these variables might have resulted in the omitted variable bias, therefore the results can be biased. Further research is needed to estimate the effect of the exclusion of these variables, and check the robustness of the results obtained in this thesis.

Second, to calculate the policy index 𝑃𝑃�_{𝑖𝑖𝑖𝑖}Burnside and Dollar use three components, namely budget balance, inflation, and a trade openness dummy. This paper uses government consumption instead of budget balance and another measure of trade openness. The data on government consumption is more complete, and the trade dummy used by the other authors is not available for the whole time period considered here. These differences in calculating the policy index might be the second cause of the surprising result. Redoing the analysis of this paper with the policy index more closely aligned with the one used by Burnside and Dollar, i.e. updating the data set with the same variables as Burnside and Dollar, would be a suggestion for follow-up research.

Third, the assumption of unlikely outliers (A3) might be violated by some observations on inflation. As is discussed earlier some countries enjoyed hyperinflation during the time period considered here, which might have an influence on the outcomes. Dropping these observations from the data set might influence the outcomes of the analysis. However, to establish which large observations are actually outliers, and which are just large, is something to be investigated more thoroughly.

(28)

Reference list

Barro, R. J. (1989). Economic growth in a cross section of countries (No. w3120). National Bureau of Economic Research.

Barro, R. J. (2001). Human capital and growth. The American Economic Review, 91(2), 12-17.

Barro, R. J., & Lee, J. W. (1994, June). Sources of economic growth. In Carnegie-Rochester conference series on public policy (Vol. 40, pp. 1-46). North-Holland.

Boone, P. (1996). Politics and the effectiveness of foreign aid. European economic

review, 40(2), 289-329.

Burnside, A. C., & Dollar, D. (2000). Aid, policies, and growth. American Economic

Review, 90(4), 847-868.

Burnside, A. C., & Dollar, D. (2004). Aid, policies, and growth: revisiting the evidence. World Bank Policy Research Working Paper, (3251).

Clemens, M. A., Radelet, S., & Bhavnani, R. R. (2004). Counting chickens when they hatch: The short term effect of aid on growth. Center for Global Development

working paper, (44).

Dalgaard, C. J., & Hansen, H. (2001). On aid, growth and good policies. Journal of

development Studies, 37(6), 17-41.

Dalgaard, C. J., Hansen, H., & Tarp, F. (2004). On the empirics of foreign aid and growth. The Economic Journal, 114(496), F191-F216.

Dollar, D., & Kraay, A. (2004). Trade, growth, and poverty*. The Economic Journal, 114(493), F22-F49.

Dollar, D., Pritchett, L. (1998). Assessing aid - what works, what doesn't, and why. World Bank policy research report. Washington, D.C. : The World Bank. Doucouliagos, H., & Paldam, M. (2008). Aid effectiveness on growth: A meta study.

European journal of political economy, 24(1), 1-24.

Doucouliagos, H., & Paldam, M. (2010). Conditional aid effectiveness: A meta‐ study. Journal of International Development, 22(4), 391-410.

Durbarry, R., Gemmell, N., & Greenaway, D. (1998). New evidence on the impact of

foreign aid on economic growth. Nottingham: Centre for Research in

(29)

Easterly, W. (2014, March 13). Stop Sending Aid to Dictators. Time. Retrieved from http://time.com/23075/william-easterly-stop-sending-aid-to-dictators/

Easterly, W., & Levine, R. (1997). Africa's growth tragedy: policies and ethnic divisions. The Quarterly Journal of Economics, 1203-1250.

Easterly, W., Levine, R., & Roodman, D. (2003). New data, new doubts: A comment

on Burnside and Dollar's" aid, policies, and growth"(2000) (No. w9846).

National Bureau of Economic Research.

Easterly, W., & Rebelo, S. (1993). Fiscal policy and economic growth. Journal of monetary economics, 32(3), 417-458.

Fischer, S. (1993). The role of macroeconomic factors in growth. Journal of monetary economics, 32(3), 485-512.

Frankel, J. A., & Romer, D. (1999). Does trade cause growth?. American economic review, 379-399.

Gutteridge, N. (2015, December 15). Foreign aid scandal: Iran, China, Zimbabwe.. why is YOUR cash given to foreign dictators? The Express. Retrieved from http://www.express.co.uk/news/politics/626700/UK-foreign-aid-India-China-Zimbabwe-foreign-dictators-Tory-Government-cuts-Cameron

Hansen, H., & Tarp, F. (2000). Aid effectiveness disputed. Foreign Aid and Development: Lessons Learnt and Directions for the Future, 103-128.

Hansen, H., & Tarp, F. (2001). Aid and growth regressions. Journal of development

Economics, 64(2), 547-570.

Harrison, A. (1996). Openness and growth: A time-series, cross-country analysis for developing countries. Journal of development Economics, 48(2), 419-447. Hudson, J., & Mosley, P. (2001). Aid policies and growth: In search of the Holy

Grail. Journal of International development, 13(7), 1023-1038.

IMF. “World Economic Outlook Database”. The International Monetary Fund, Washington D.C., US. (retrieved June 2016)

King, R. G., & Levine, R. (1993). Finance, entrepreneurship and growth. Journal of Monetary economics, 32(3), 513-542.

Knack, S., & Keefer, P. (1995). Institutions and economic performance: cross‐country tests using alternative institutional measures. Economics & Politics, 7(3), 207-227.

(30)

Krugman, P. R., Obstfeld, M., & Melitz, M. (2015). International trade: theory and

policy. Harlow, England: Pearson Education Limited.

Landau, D. (1986). Government and economic growth in the less developed countries: an empirical study for 1960-1980. Economic Development and

Cultural Change, 35(1), 35-75.

Lensink, R., & White, H. (2001). Are there negative returns to aid?. Journal of

development Studies, 37(6), 42-65.

Parente, S. L., & Prescott, E. C. (1993). Changes in the Wealth of Nations. Federal Reserve Bank of Minneapolis. Quarterly Review-Federal Reserve Bank of Minneapolis, 17(2), 3.

Peron, J. (2001, August 1). The Sorry Record of Foreign Aid in Africa: African Governments Are Destroying Their Countries with Aid from the West.

Foundation for Economic Education. Retrieved from https://fee.org/articles/the-sorry-record-of-foreign-aid-in-africa/

Rajan, R. G., & Subramanian, A. (2008). Aid and growth: What does the cross-country evidence really show?. The Review of economics and Statistics, 90(4), 643-665.

Riddell, R. C. (2007). Does foreign aid really work?. Oxford: Oxford University Press

Solow, R. M. (1956). A contribution to the theory of economic growth. The quarterly

journal of economics, 65-94.

Stock, J. H., Watson, M. W. (2015). Introduction to Econometrics. Harlow, England: Pearson Education Limited.

Svensson, J. (1999). Aid, growth and democracy. Economics & Politics, 11(3), 275-297.

Trabandt, M., & Uhlig, H. (2011). The Laffer curve revisited. Journal of Monetary

Economics, 58(4), 305-327.

The World Bank. “World Development Indicators Database”. The World Bank, Washington D.C., USA. (retrieved June 2016)

(31)

Appendix

A. List of countries included in data set.

Africa Latin America and the

Caribbean Asia

Algeria Madagascar Togo* Argentina Haiti India Botswana Malawi Tunisia Bolivia Honduras Indonesia

Cameroon* Mali* Zambia Brazil Jamaica Malaysia

Congo, DR Morocco Zimbabwe Chile Mexico Pakistan Egypt Niger* (= 26) Colombia Nicaragua Philippines

Ethiopia Nigeria Costa Rica Paraguay Sri Lanka

Gabon* Senegal* Dominican

Republic Peru Syria

Gambia Sierra Leone Ecuador Trinidad &

Tobago Thailand

Ghana South Africa El Salvador Uruguay Turkey

Ivory

Coast* Sudan Guatemala Venezuela (=9)

Kenya Tanzania Guyana (= 21)

(32)

B. Graphs

Graph (G1): Fitted values

(33)

Graph (G3): Standardized normal probability plot

The conditional effectiveness of aid in developing countries