IS AID PRO POOR?

(1)

IS AID PRO POOR?

Veronike Fikse

Abstract

Official development assistance is given by countries and donor institutions to increase economic and social development. Aid effectiveness studies have focused on measuring the effect of aid on economic growth. This research joins a new strand of literature by measuring the effect of aid on an income distribution in deciles. The effect on the poorest deciles answers the question whether aid is pro poor. Aid decreases the poorest income share by 2.1 percentage points and increases the income share of the rich. Therefore it is concluded that aid is relatively anti poor; its effect is smaller on the poor than on the non poor.

Keywords: Aid, pro poor and income distribution.

V.F. Fikse

Student number: 1271482 August, 2007

Supervisor: Dr. Ir. D.J. Bezemer

(2)

Preface

This thesis was written in order to finish my studies International Economics and Business at the University of Groningen. The topic of development economics gained my interest throughout the years. After having done research on Micro credits in India the subject of development aid attracted my attention. The main point of research on aid is finding out what the effects are and how to optimalise these effects. I have tried to add a new perspective to this the strand of literature.

(3)

Tables and figures

Figure 1. The three waves of models in the aid effectiveness literature... 6

Figure 2. Representation of the links in this research. ... 13

Figure 3. Lorenz curve. ... 23

Table 1. Income distribution with mean normalised average per capita income... 25

Figure 4. Mean normalised average per capita income per decile for three years... 26

Figure 5. Change between the subsequent deciles for three years... 27

Figure 6. Pie chart of income distribution in relative deciles (average 1988-1998)... 27

Figure 7. Countries receiving the most nominal Official Development Assistance... 29

Figure 8. Countries receiving the most Official Development Aid compared to GDP. ... 29

Table 2. Variance Inflation Factor analysis. ... 30

Table 3. Hausman test for Random Effects model ... 31

Table 4. Durbin Wu Hausman test for endogeneity ... 32

Table 5. Testing heteroscedasticity... 32

Table 6. Outcomes of aid in the regressions on relative deciles. ... 33

Figure 9. Change of income share due to inflow of aid in model 1 (ceteris paribus)... 34

Table7. Outcomes of the regressions on multiple deciles... 35

Table 8. Outcomes of the regressions of the Tobit model. ... 37

(5)

1. Introduction

“It is logical, that the United States should do whatever it is able to do to assist in the return of normal economic health to the world, without which there can be no political

stability and no assured peace. Our policy is not directed against any country, but against hunger, poverty, desperation and chaos. Any government that is willing to assist

in recovery will find full co-operation on the part of the USA." Secretary of State George Marshall, Harvard University on June 5, 1947.

The assistance mentioned above is the first form of what is now known as Official Development Assistance (ODA) and was given by the United States to support European recovery after World War II. Ever since aid is given its effectiveness has been studied. Several angles of measuring this effectiveness have been leading research strands, such as the focus on savings and investment, the influence of aid directly on economic growth and through conditions. New focus keep emerging and this research joins such a new strand of literature by measuring aid effectiveness on income levels of the poorest instead of economic growth according to the changing aims of the institutional donors. The aim of aid has shifted from increasing economic growth towards decreasing poverty (World Bank, 2000). Taking into account the difference between data and measurement overall research gives a positive conclusion on the effectiveness of aid (Dougouliagos and Paldam, 2005; McGillivray et al., 2006; Hansen and Tarp, 2000). Eradicating extreme poverty is the first Millennium Development Goal (MDG), but is aid effectively contributing to this aim, is aid pro poor? That is the main research question addressed here.

(6)

In total three models with three different aid variables are estimated with on the left hand side the relative deciles. These relative deciles gave the opportunity to compare the effect of aid on the income share of the poorest decile to the average income of the country. Also the difference in income share between poor and rich is calculated. The base model contains aid divided by GDP and some other explanatory variables, mostly taken from the World Development Indicators. In the second model the aid-policy condition is added to the base model. This condition tests the theory that countries with better economic policies provide better allocation of the inflowing aid. The policy condition seems to hold significantly; in countries with better policy rights the impact of aid on the poorest deciles is less negative relatively. The third model contains an aid² variable in order to measure a non linear relation between aid and income. No conclusions can be drawn from the outcomes of this model. The first two models estimated a statistically significant negative impact of aid on the three poorest income shares. Also the Tobit models, used to test the robustness of these outcomes, gave negative outcomes. The income share of the richest decile grew by the inflow of aid. Therefore this research concludes that aid is relatively anti poor.

(7)

2. Literature review

2.1 Aid on growth literature

After about 50 years of donating development aid Doucouliagos and Paldam (2005b) describe the history of aid effectiveness in literature. Their empirical research questions whether aid is effective by comparing the different methodologies of previous studies with meta-analysis. In their conclusion they state that the average result is small, but positive. Therefore they come to the suggestion that aid should be targeted more effectively in order to amplify the effect. They classify three waves of models in literature: the ‘accumulation’ model, the ‘direct growth’ model and the ‘conditional’ model. The links building these models are graphically displayed in the figure below.

Figure 1. The three waves of models in the aid effectiveness literature.

Source: Doucouliagos and Paldam 2005b

In the next paragraphs these models are consecutively described forming the framework of this section. In order to convey a more objective view on the literature this research compares Doucouliagos and Paldam (2005b) with two theoretical papers describing aid effectiveness literature. The first of these papers is the descriptive article by McGillivray et al. (2006). Their title already sums their conclusions on aid all in all; “It works; it

doesn’t; it can, but that depends.” The paper gives a clear overview of existing literature

(8)

into two groups of models. In figure 1 this is covered by the first two waves. Aid was supposed to create growth by increasing savings, therefore the link between aid and savings was extensively explored. McGillivray et al. do not address the second wave of direct growth literature in depth but emphasize another view: The strand of literature that takes into account the possible endogeneity of aid, which was initiated by Mosley (1980) and still forms an important estimation problem. McGillivray et al. start their second period at 1998 with the so called landmark article Assessing Aid by the World Bank (1998) and this lasts until the present. Their second period of time is related to the third wave that uses the conditional model.

The second article written by Hansen and Tarp (2000) has a very similar structure to the one used in this review. The same three movements in literature are distinguished; aid-savings-growth, aid-investment-growth and aid-growth. They intensively compare their own empirical results (Hansen and Tarp, 1999) with other work on the conditional aid-growth model. Their conclusion is positive i.e. aid positively affects aid-growth.

2.1.1 Accumulation model

The first wave of literature on the aid-growth relation is referred to as the accumulation model. This model is built on the Harrod Domar model1 which assumes that economic growth is constrained by capital and productivity of capital. Aid is the capital increase that leads to higher domestic savings. These savings lead to higher investments which instigate economic growth. In this theory countries would become self sufficient in creating economic development. These savings and investment are linked to economic growth in the national income identity stated in equation 1.1 (Doucouliagos and Paldam, 2005a).

Y = C + I + G + X – M (1.1)

I – S = X - M = - H (1.2)

Where Y is national income, C stands for consumption, I for investment, G for government expenditure, X – M equals exports minus imports, S is savings and H is aid.

1

(9)

In equation 1.2 development aid (H) donated to a developing country bridges the difference between investment and savings as they should be equal in order to fulfil this equation. In turn investment is in positive relation with national income (equation 1.1). In case aid substitutes savings S falls and the effect of aid is crowded out. This leads to less growth. The theory above is the first gap of the so called gap model, which forms part of the accumulation model. This first ‘savings gap’ was empirically tested in the 1960s. Griffin (1970) and Griffin and Enos (1970) concluded that aid leads to lower domestic savings and that aid is used as government consumption which created dependency upon the donor countries instead of higher economic development. Israel is taken as an example where domestic savings decreased as a result of inflowing aid and total investment was paid by foreign capital.

An extension of the gap model is the two-gap model. This model describes, in addition to the savings gap, the lack of sufficient foreign exchange to import capital goods to invest (Chenery and Strout, 1966). Aid can fill this gap by increasing exports (X). Much later Bacha (1990) identifies a third fiscal gap for countries in debt. When these countries have low levels of inflation this fiscal constraint is important. High inflation leads to a more binding savings or foreign exchange constraint. All these models consider aid as an accumulation of income.

2.1.2 Direct growth model

(10)

measures their separate effects on economic growth. He reports that savings and foreign capital together account for a third of economic growth. McGillivray et al. (2006) emphasize the appearing of the book of Mosley (1980) in this period. He stated the endogeneity problem of aid in the endogenous growth model and made aid endogenous by lagging its value. The exogeneity assumption remains important for all following work.

2.1.3 Conditional growth model

The third stream of literature started with the work of Boone (1996). Focusing on political regime he analyses the effectiveness of aid and concludes that aid does increase government expenditure, but not economic growth. Hansen and Tarp (2000) consider this paper as an outlier as they come to the conclusion, through assessing several aid effectiveness papers, that aid improves economic performance. They refer to Assessing Aid (World Bank, 1998) as a turning point. Here it is claimed that aid is effective through the condition that aid is more effective in countries with good policies. In Doucouliagos and Paldam (2005b) this falls under the literature wave of the conditional model. Assessing Aid and Burnside and Dollar (1997 and 2000) built the theory that a government with sound policies is better able to allocate aid, as opposed to countries with poor policies. They differentiate between bilateral and multilateral aid and conclude that multilateral aid favours countries with sound policies. Overall they claim that aid does not form the impetus for good policy regime, but merely provides the needed capital to effectively create growth in case of good policies. This is referred to as the ‘Good policy’ model by Doucouliagas and Paldam, which takes good policy as a condition for effective aid.

(11)

datasets, based on the new growth theory, though the biggest innovation is the introduction of the non linear link between aid and growth. This functional form can give more precise estimation. McGillivray et al. (2006) put together a selection of found literature from the third wave and the results of their research (appendix table B5).

2.2 Pro poor aid

Over time, the aid-growth literature has changed thanks to new economic and econometric models used. The latest change of aid effectiveness literature is led by a change in the aim of development aid. The objective of development has shifted towards the objective of poverty reduction (World Bank, 2000). Institutions such as the OECD and UN have aimed their policies at decreasing poverty. Eradicating extreme poverty is the first of eight MDGs.2 After the three waves of aid effectiveness literature a new strand has emerged, in which the effectiveness of aid is measured by its impact on indicators of poverty. These studies are an extension of the growth regressions with panel data used in earlier years, but here the dependent variable is a Human Development Indicator (HDI) that gives a proxy of the level of social development. These can be the level of education, life expectancy, some health indicator or a combination of these. Countries that aim their policies towards increasing growth are most likely to sustain deterioration of the human development indicators, on the other hand policies implemented to raise social welfare also benefit from higher growth (White, 1999).

Boone (1996) was one of the first in estimating an aid regression in relation with HDIs, aid does not favour social development. In contrast to the results of Boone, Gomanee et al. (2003) and Gomanee et al. (2005) report that aid increases the HDI and lowers infant mortality. They distinguish between the direct and indirect effects of aid on the welfare of the poor. The direct effect is created by micro projects and for this indirect channel they have developed a Pro Poor Expenditure index consisting of public expenditure on housing, education and health. Mosley et al. (2004) use a slightly different version of this PPE and conclude that, together with inequality and corruption, it is a good means of aid’s poverty leverage.

2

(12)

Pro poor measures or policies increase the welfare level of the poor. This is a new phenomenon in development economics, and the discussion starts with its definition. There are two approaches: absolutely pro poor and relatively pro poor. Below both are explained and this research uses them as a means for measuring aid effectiveness.

2.2.1 Absolutely pro poor

Policies increasing the welfare of the poor are considered absolutely pro poor. Several researchers found that economic growth decreases the incidence of absolute poverty (Ravallion, 2001). Anderson and White (2001) discuss pro poor growth and claim that the income of the poor must increase compared to their current share, especially Eastern Europe and Central Asia have experienced strong pro poor growth. Dollar and Kraay (2002) find that the lowest quintile in income distribution gains one-on-one from economic growth. Deininger and Squire (1996) found a strong correlation between aggregate growth and growth of income of the poorest as well. Another method to measure absolute pro poor growth estimates the decline in a poverty measure of interest, such as the headcount ratio, the poverty gap and the squared poverty gap and the Watts index. (Kraay, 2004; Ravallion and Chen, 2003).3

Like most other aid effectiveness literature economic growth is used often to calculate pro poor growth. This research investigates the effectiveness of aid on level income. Lately donors have been diverting their capital from increasing economic growth to comply with the objective of decreasing poverty. This should increase the income and living standards of the poor. Here the efficiency of aid is tested on the poorest deciles directly and not through economic growth. The research question concerns whether aid is pro poor. The first hypothesis is derived from this definition concerning pro poorness.

Hypothesis 1: Aid is absolutely pro poor and increases the income of the poor.

3

(13)

2.2.2 Relatively pro poor

The next approach concerns relative pro poor policies, which positively affect the welfare of the poor relatively more than the welfare of the non-poor (Kakwani and Pernia, 2000). The non-poor can be the whole country taken together, this way the income of the poor is compared to the average income of the country. On the other hand, only a small rich group can also represent the non-poor. Comparing poor and rich groups in a country tends to measuring inequality. In fact relatively pro poor policies decrease inequality of an income distribution by bridging the difference between poor and rich.

(14)

Figure 2. Representation of the links in this research.

Source: own figure

The second hypothesis states the influence of aid on the income distribution; do the poor benefit relatively more from aid than the others? Does aid stimulate income equality?

(15)

3. Methodology

The hypotheses are tested with an econometric model. The model estimates the influence of aid on income distribution divided into deciles. Ten regressions are estimated that give insight in the effect of aid on average income per decile. The data set combines 117 developing countries taken over three years (1988, 1993 and 1998). For this pooled time-series and cross-sectional data the pooled Ordinary Least Squares (OLS) method is used. The regression measures the level effect of aid on deciles and not the growth effect. It measures how much of the (level) income share is considered to be gained by development aid. The reason for investigating level effects over growth effects is to shed new light on the aid effectiveness literature, which until now mainly concerned growth data. This relation with level data creates a direct link between aid and income instead of growth. Henceforth, the changes on income of the poorest can be assessed. Another argument concerns the dataset, which was stated in level income data. Growth rates between these three time periods were calculated, but not used. This method reduced the number of observations to a number considered too small. Next, the basic econometric model with its variables is described. Thereafter, the method is explained for measuring whether aid is absolutely and relatively pro poor.

3.1 Econometric model

The model starts with the relation between aid flowing into a country and average income per decile. ) , ( jt jt ijt f A m y = (1)

Where y is annual $PPP income per capita in local national currency,

(16)

Average national per capita income (y) is taken in deciles (i). These deciles represent average per capita income of ten groups of the population. This dependent variable is transformed into relative income to make the data comparable as national income (y) is stated in local national currency. The variable is divided by the mean average income of the country (m), which makes the dependent variable the mean normalised average per capita income per decile. Formula 2 shows the function of the effect of aid (A) on the deciles. ) ( / jt jt ijt m f A y = (2)

In order to make a good fit the model is extended with explanatory variables that also influence income according to previous stated literature.

jt t j r t j a jt

ijt m A X ASIA EUR LAC e

y / =α +β ( −1)+β ( −1)+β11 +β12 +β13 + (3.1) Where

α

is a constant, βa are the parameters for the aid variables (A) and βr are all

the betas for X; the vector of other influencing variables. ASIA, Europe (EUR) and Latin America and Caribbean (LAC) are the regional dummy variables, for which β11, β12 and β13 respectively are the parameters and e is the error term.

In pooled OLS longitudinal and cross sectional data are estimated together in one model. The intercept and response parameters are able to vary across countries and time. Ways to account for this is to add a Fixed Effects estimator (FE). In the FE model the intercept parameter is able to vary across countries, but is fixed over time. For this dataset the FE estimator gave poor resultsaccording to Milanovic (2005)4, therefore the fixed effects estimator is not added but regional dummies are used. These account for regional differences instead of country differences. Adding regional dummies are also preferred for they have given significant conclusive results (Sala-i-Martin, 1997). Another model that is often used for pooling time series and cross sectional data is the random effects model (RE). This model also allows for different intercept parameters. Whether to use

4

(17)

this model is tested with the Hausman test. This compares the FE with the RE model, in case the coefficients are the same the RE model is consistent and efficient.

3.2 Explanatory variables

3.2.1 Official Development Assistance

The variable ODA5 is taken from the World Development Indicators (2006) and described as; Bilateral and multilateral aid flows consisting of a concessional loan and a grant part in current US dollars divided by GDP. Chang, Fernandez- Arias and Serven (1998) have derived a new measure of aid which only takes into account the grant part, Effective Development Assistance (EDA). The grant part is much bigger than the part of the aid that has to be repaid. As a result, correlation between the same dataset for ODA and EDA is 0.83. This research uses ODA because EDA data is only available until 1995 and contains few countries.

According to earlier studies the interaction variables aid-policy and aid² have shown significant influence on economic development. The first variable is established by Burnside and Dollar (1997, 2000) who advocate that countries with good institutional policies are better able to allocated aid. They state: “Aid affects economic growth but it is

conditional on the same policies that affect growth”.6 The difference here is that this variable is tested on level data. Level income increases with aid when this is seen as an income transfer. Previous research stated a positive influence, which is translated into a negative effect here due to the form of the policy variable; higher values represent lower levels of policy freedom. More on this variable is explained in the next section.

Hadjimichael et al. (1995) are the first to establish a non linear relation between aid and growth. This non linear relation is assumed due to the low absorptive capacities of the poor countries and the ‘Dutch Disease’.7 Together with Durbarry et al. (1998) and

5

When referring to ODA and aid both are taken as a percentage of GDP. 6

Burnside and Dollar, 2000 p. 847 7

(18)

Hansen and Tarp (1999) they include aid² in their aid-growth regressions. This gives the opportunity to calculate a turning point i.e. the amount of aid that creates the most effect.

3.2.2 Other explanatory variables

Other explanatory variables are captured in the vector X. All data is taken from the World Development Indicators (2006) except for Policy rights.

POL = Policy Rights are taken from the Freedom House8 which measure the level of political rights and civil liberties in a country, from 0 being free to 7 being not free. FDI: The inflow of Foreign Direct Investment in current US dollars relative to GDP. GDP: Gross Domestic Product in current US dollars converted from domestic local currencies divided by a million.

GOVEXP: General government final consumption expenditure includes all government current expenditures for purchases of goods and services as a percentage of GDP.

INFL: Inflation is measured as the annual percentage change in consumer prices.

LEXP: Life expectancy at birth indicates the number of years a newborn infant would live if prevailing patterns of mortality at the time of its birth were to stay the same throughout its life.

LL: Liquid Liabilities are all the money and money equivalents held by the residents as part of GDP.

TRADE: Exports and imports of goods and services measured as a share of GDP.

URB: Urbanization as a percentage of urban population on total population reported to the United Nations.

These are included in the regression to get a higher fit of the model as are found to be associated with income in previous research (Barro, 1991; Sala-i-Martin, 1997).

Milanovic (2005) is the only work referred to where the regression contains level income data. All the others take growth variables, but this should not be a problem as growth is determined by the difference in level income.

8

(19)

The variables FDI, GDP, trade and urbanization are also taken by Anderson and White (2001). They found a very small but significant relation between FDI and income, considering the fact that FDI is relatively small in developing countries. Though, FDI is expected to have no effect on the income of the poorest and a positive effect on the rich, therefore increasing inequality. Initial GDP is taken as a proxy of country size. It was found significantly negative for every income quintile and for economic growth overall, but GDP did not influence the income distribution. Trade is taken as openness (Milanovic, 2005; Anderson and White, 2001) and they found the opposite effect compared to GDP. Openness is positive for all income groups. They also found robust evidence that increased urbanization is positive for the poor. Urban settings are created by the attraction of industrialization, making urbanization a proxy for industrial development. This research expects the same outcomes for these variables.

Inflation is found to be significantly negatively related with income growth by several researchers and is therefore included in this regression (Hadjimichael et al., 1995; Burnside and Dollar, 1997; Hansen and Tarp, 1999).

Government expenditure is sometimes added to extend the base model (Milanovic, 2005; Burnside and Dollar, 2000). No robust evidence was found of its relation with growth. This variable is still included as high level of government expenditure is expected to be effective for all residents and therefore positive on income. Life expectancy is included by Lensink and White (2001) as a proxy for initial stock of human capital (Sala-i-Martin, 1997). It is expected to be positively related to income.

Liquid liability is taken as financial development indicator and found to have a significant positive effect on all income groups and thus no influence on income distribution (Milanovic, 2005).

3.2.3 Regional dummies

(20)

distinguish between the regions, as regions go through similar experiences such as crisis and war, but they also relate in climate conditions. Countries trade intensively with their region, but regional trade can also cause contagious effects. This division of continents is implemented in the regression to take into account the effects of country differences. Not all regions can be included, to avoid extreme multicollinearity. The dummy variable for Africa is excluded from the sample. Africa contains a wide diversity of countries according to their development level, besides Africa lacks a lot of observations of the year 1988.

The included dummies for the geographical areas in this regression are Asia, Latin America and Eastern Europe. Eastern European countries found themselves in transition economies. This region had negative growth rates, but quite equal income distributions (Milanovic, 1998). Aid flowing into this part of the world is expected to be allocated appropriately and support the poorest. Asia is more equal in its income distribution due to the fact that they experienced a land-reform as opposed to Latin America and Africa (Higgins and Williamson, 1999). During the time period of the sample especially the Asian Tigers were already developing quickly and experiencing high levels of economic growth9, this has contributed to good growth rates for the region. Therefore a positive relation is expected between aid and income of the poor also compared to the non poor. The income distribution in Latin America is more unequally divided and negatively related to growth (Sala-i-Martin, 1997). A negative relation is expected to hold between inflowing aid and income of the poorest absolutely and relatively.

3.3 Specifications of the model

In order to build the right model several effects should be taken into account; i) multicollinearity ii) endogeneity iii) heteroscedasticity. The first is a so-called data problem while the latter two are estimation problems. Collinearity arises when variables move in the same systematic way. Multicollinearity is the existence of collinearity between several explanatory variables. This can lead to incorrect estimation of the parameters, such as the incorrect sign or value. The collinearity between the independent

9

(21)

variables is checked by the correlation matrix and the multicollinearity is checked with the Variance Inflation Factor (VIF) analysis. In the VIF method one of the independent variables is taken as dependent variable. In case of high R² the model fits well which indicates a high relation between the variables. Solution is then to model for this multicollinearity or delete one of the correlating variables from the regression.

The right hand side variables are supposed to be exogenous and influencing the dependent variable. Endogeneity of one of the explanatory variables can be a problem which results in correlated error terms. Burnside and Dollar (1997) advise to instrument the ODA/GDP ratio in case of endogeneity of aid and use the Two Stages Least Squares (TSLS) method. They average their data over four year periods. Milanovic (2005) encounters the same problem and takes the five year average of the explanatory variables and uses the Generalised Method of Moments (GMM) approach which has more efficient properties. The downside of these models is the requirement of instrumental variables for aid, which are hard to define and could bias the results. Lensink and White (2001) come to the conclusion that they do not have to instrument their aid variables. They measured this by testing substitutes of aid against aid as the dependent variable. Here the Durbin-Wu-Hausman test is performed. The endogeneity of aid is measured by taking the possible endogenous explanatory variable as the dependent variable. The explanatory variables remain the same, but the lagged value for aid and mean normalised decile are added as explanatory variables. Secondly, this error term is added to the base regression. In case of significant coefficients of the error term the aid variable is considered endogenous.

(22)

Another specification is the right functional algebraic form for the regression which is chosen by trial-and-error. Six basic options are commonly used (Carter Hill et al., 2001): (the first is the form of the dependent and the second is the form of the independent variable) i) linear ii) log iii) log-linear vi) log-log v) log-inverse vi) linear-reciprocal. This last model is not used, because the reciprocal term creates very small outcomes and even deletes many observations. Also logging the dependent variable gives very small outcomes. As a result only the first two models are considered good options.

3.4 Absolute pro poor aid

The econometric model described above uses a relative measure of income. The dependent variable is an index of average income per decile divided by the average income of the country, both in local currencies. Absolute per capita income levels per decile are taken as dependent variable to measure the absolute effect of aid on the deciles. The average per capita income per decile is divided by the PPP exchange rate with the dollar in the years of their recording. These exchange rates were given by the same database (Milanovic, 2005) used throughout this paper. Special interest goes out to the poorest deciles in order to see whether aid is pro poor and lifts these averages.

3.5 Relative pro poor aid

Relative pro poor aid concerns the comparison between the effect of aid on the poor and non poor. Therefore the average per capita income per decile is divided by the mean average per capita income of the country. Changes in this decile ratio can be derived from changes in average decile income and/or changes in total average income of a country. Meaning that if there is a positive effect of aid on the poorest deciles the average income of the poorest has increased more from aid than the country overall. This is relatively pro poor aid. Thanks to the extensive information of the database the effects can also be compared over the whole income distribution. Ten level regressions are run with the same explanatory variables.

(23)

dependent variable. One decile has to be excluded to avoid extreme multicollinearity. The effect of aid on the poorest decile is compared to the effect of aid on an income group including the effect of decile 1.

jt t j r t j a jt i e LAC EUR ASIA X A m i y + + + + + + =       ∑ − −1) ( 1) 11 12 13 ( 1 *

α

β

(3.2) Where i is 1, 2, .., 9.

In order to test whether aid is even more pro poor in countries with better policy rights an interaction variable of aid with policy rights is included. The policy index of this policy rights variable is scaled from zero to seven, with seven having low policy rights. Aid interacting with policy rights is pro poor, when the regressions provide a negative relation with income of the poorest. Aid² has an exponential relation with the dependent variable. If aid is pro poor there needs to be a negative relation between aid² and the poorest deciles.

In estimating the difference of aid effects on the poor and non poor, also the richest part of society is taken as a comparison. Comparing rich and poor gives insight in the level of inequality in a nation. Inequality decreases (increases) if the level of income of the poor increases (remains the same or decreases) and the level of income of the rich decreases or remains the same (increases) i.e. income of the poor increases relatively more, or decreases less, than the non poor. From this line of reasoning it follows that relative pro poor aid should also decrease inequality.

(24)

Figure 3. Lorenz curve.

Source: Wikipedia10

The dependent variables used previously are now transformed into a gini index. This research takes the richest 20 % of the income distribution (deciles 9 and 10) divided by the poorest 20 % of the income distribution (deciles 1 and 2). Mathematically a lower gini value comes from either an increase of income of the poor and/or a decrease of income of the rich, equalizing society. The drawback of this estimation is the negligence of the changes in middle incomes shares.

) / /( ) / (y9jt y10jt mjt y1jt y2jt mjy G= + + (4)

Again the right hand side variables are the same as in the basic model for measuring the effect of aid (regression 3.1). Aid is expected to be relatively pro poor by having a positive effect on income of the poor, relatively more than on the rich. This has an equalizing effect and thus lowers the gini coefficient. The coefficient of this regression is expected to be negative.

10

(25)

4. Data description

4.1 Dependent variable

The dependent variables in this model consist of an income distribution in deciles. The last regressions are assembled by a combination of deciles. The data is taken from Milanovic (2005), who compiled a unique dataset with income distributions of 129 countries and regions from 321 surveys using the Deininger and Squire Database and World Income Distribution. Here only the developing countries are included, this amounts to a total of 117. The income distribution is divided into deciles11 made up by average income of ten percent of individuals. The absolute income levels are calculated by dividing these local currencies into the PPP dollar exchange rate of the three sample years and can be found in the appendix (graph B1). The regressions with these data did not show any conclusive results and the outcomes are therefore left out. It is concluded that when using the 1$ per day poverty line the first decile is considered poor and when using the 2$ per day poverty line the first two deciles are considered poor.

The average per capita income of a decile is divided by mean average per capita income in order to make a relative comparison between income of the decile and the average income of the country on a whole. Besides, the data is taken in local currencies and could not be estimated in the same regression with the explanatory variables. This gives ten income shares per country and ten regressions per model. Table 1 gives a clear overview of the relative deciles.12 The dataset used in this research contains secondary data taken for three years; 1988, 1993 and 1998. Not all data is measured exactly in these years, though most is taken close to these ‘benchmark years’.

Not all countries are present in every time period, which makes the sample unbalanced. The first period has observations of 80 countries, the second 98 countries and the last

11

Part of the dataset consisted of deciles with their average GDP per capita, another part was made up of certain GDP per capita amounts and the respective percentage of population falling under this decile. The second group is extrapolated into the first form. For example quintiles are split in two deciles with the same average per capita income and in distributions of five percent the weighted averages are added together. 12

(26)

time period consists of 101 countries. Only PPP household data is used, mostly income data, otherwise expenditure. India, China, Bangladesh, Egypt and Indonesia are admitted twice in an urban and rural part because of the big difference in average income.

The uniqueness of this new dataset lies in the fact that Milanovic divided the income distribution into deciles. Income distribution taken in deciles is quite a new approach as most other research takes quintiles, for which the poverty literature is interested in the poorest twenty percent of the population. Taking deciles gives more specific information as the groups are smaller and thus divides the poorest twenty percent in two groups of ten percent. Also the estimation of average income will be more precise because the range of the income group is smaller.

Table 1. Income distribution with mean normalised average per capita income.

1988 1993 1998 Deciles Relative ∆Subsequent deciles Relative ∆Subsequent deciles Relative ∆Subsequent deciles 1 0,3278 0,2297 0,2254 2 0,4375 10,97% 0,3632 13,35% 0,3576 13,22% 3 0,5379 10,04% 0,4620 9,88% 0,4623 10,50% 4 0,6171 7,92% 0,5514 8,94% 0,6025 13,99% 5 0,7183 10,12% 0,6609 10,95% 0,658 5,56% 6 0,8430 12,46% 0,7814 12,05% 0,834 17,61% 7 0,9798 13,69% 0,9384 15,70% 0,9502 11,60% 8 1,1691 18,93% 1,1636 22,52% 1,1488 19,86% 9 1,5347 36,56% 1,5556 39,19% 1,5467 39,79% 10 2,7872 125,24% 3,2914 173,58% 3,2140 166,72%

Source: own compilation of dataset Milanovic (2005)

(27)

Figure 4. Mean normalised average per capita income per decile for three years. 0 0,5 1 1,5 2 2,5 3 3,5 1 2 3 4 5 6 7 8 9 10 Deciles M e a n n o rm a li s e d p e r c a p it a in c o m e 1988 1993 1998

(28)

Figure 5. Change between the subsequent deciles for three years. 0,00% 20,00% 40,00% 60,00% 80,00% 100,00% 120,00% 140,00% 160,00% 180,00% 200,00% 1 2 3 4 5 6 7 8 9 10

Mean normalised deciles

1988 1993 1998

The total average income can be seen as a pie with the deciles as shares of this pie. Below the percentages are the average income shares over the total sample period.

Figure 6. Pie chart of income distribution in relative deciles (average 1988-1998).

In the appendix table B2 gives an overview of the descriptive statistics of the mean normalised average per capita income in deciles. All deciles have 274 observations. In

(29)

coherence with figure 4 table B2 shows that the richest decile has a much higher average income compared to mean income. Logically, the range is also much higher than the other groups. Deciles 2, 3 and 4 are normally distributed according to a significance level of 5% by the Jarque Bera test13.

4.2 Explanatory variables

The explanatory variables are also known as the right hand side variables. As this data is not always exactly gathered in these three benchmark years the explanatory variables are taken as averages of five years; from 1984-1988, 1989-1993 and 1994-1998. Another reason for taking averages is that the variables aid and inflation show high yearly variation for developing countries, taking averages is a good means to mediate extreme effects. Third reason for taking the average value of the explanatory variables is to include lags in the model. The effect of these variables on the dependent variable now has time to develop.

The variable aid is represented by ODA as a percentage of GDP. The two figures below present the five countries receiving most aid and the five countries receiving most aid compared to their national income, for the three sample years. Countries big in population and surface seem to receive more aid, but relatively more aid flows to poorer countries.14 These are the countries with the highest percentage of aid in their national income. Descriptive statistics on the ten independent variables can be found in the appendix in table B3.

13

Jarque Bera measures the normality of a variable. _

     ₋ + = 4 ) 3 ( * 6 2 2 K S n JB , where n is number of

observations, S skwewness and K kurtosis. 14

(30)

Figure 7. Countries receiving the most nominal Official Development Assistance. Indi a Egy pt Ban glad esh Chi na Indo nesi a Egy pt Chi na Indi a Indo nesi a Ban glad esh Chi na Egy pt Indi a Pol and Rus sia 0 500 1.000 1.500 2.000 2.500 3.000 3.500 4.000

Year 1988 Year 1993 Year 1998

O d a ( m il li o n s )

Figure 8. Countries receiving the most Official Development Aid compared to GDP.

Gui nea Bisa u Gam bia Com oros Djib uti Leso tho Gui nea Bisa u Nic arag ua Gam bia Guy ana Djib uti Gui nea Bisa u Rw anda Bos nia Zam bia Mal awi 0% 10% 20% 30% 40% 50% 60%

Year 1988 Year 1993 Year 1998

O d a /G D P

(31)

5. Empirical results

5.1 Model specification tests

The three aid variables aid as a percentage of GDP, aid interacting with policy and aid² all include ODA, therefore there is a high chance of collinearity. In the appendix the correlation matrix (table B4) shows the correlation between all explanatory variables including the three kinds of variables representing aid. The correlation between these variables is over 80%, which is quite high. Therefore the Variance Inflation Factor (VIF) analysis was conductedVIF=1/(1-R²). In table 2 the outcome of six regressions can be found in which the interaction variables aid interacting with policy rights and aid² are taken as dependent variable. The other variables present in the regression are marked in the table and all the other explanatory variables remain the same. There are no hard measures for accepting some level of multicollinearity, but the VIF higher than 10 causes multicollinearity problems and asks for removal of one of the variables. The VIFs are below the critical value and at least ten percent is thus explained by other explanatory variables.

Table 2. Variance Inflation Factor analysis. VIF analysis Dependent variables

ODA ODA ODA ODAPOL ODAPOL ODAPOL ODAQ ODAQ ODAQ

ODA X X X X

ODAPOL X X X X

ODAQ X X X X

R² 0.884160 0.885231 0.927224 0.893240 0.833643 0.893240 0.833812 0.737136 0.835880

VIF 8.33 7.14 12.5 9.09 5.88 9.09 5.88 3.85 5.88

This gives three equations for which the outcomes can be found in the appendix. They are named model 1, 2 and 3.

(32)

jt t j t j t j t j t j t j t j t j t j t j t j jt ijt e LAC EEU ASIA URB TRADE LL LEXP INFL GOVEXP GDP LN FDI POLR ODA ODA m y + + + + + + + + + + + + + + + = − − − − − − − − − − − 14 13 12 ) 1 ( 11 ) 1 ( 10 ) 1 ( 9 ) 1 ( 8 ) 1 ( 7 ) 1 ( 6 ) 2 ( 5 ) 1 ( 4 ) 1 ( 3 ) 1 ( 2 ) 1 ( 1 ) ( ² / β β β β β β β β β β β β β β α (3.4)

GDP is lagged two periods (t-2) instead of one (t-1) like the other right hand side variables. This takes away the direct relation with the dependent variable as they both represent income of a country. Lagging explanatory variables and capturing the dependent variable in time t makes it a distributed lag model.

The Hausman test is performed to check whether the random effects estimator must be added. The first decile is taken as dependent variable for the three equations respectively 3.2, 3.3 and 3.4. The outcomes of this test can be found in table 3. The hypothesis of accepting the random effects model over the fixed effects model is not rejected and the RE for cross sections is added to the regressions. The RE model is estimated by the Generalised Least Squares (GLS).15

Table 3. Hausman test for Random Effects model Dependent variable: decile 1

Model (1) (2) (3) Chi-Sq stat 10.5496 10.5131 12.5135 Chi-Sq. d.f. 10 10 11 Prob 0.3037 0.3967 0.3263

The endogeneity of the aid variables could pose a problem in estimating the right model. Like other research the explanatory variables are taken as five year averages. Again the tests for endogeneity are measured on the first decile. Table 4 shows the outcomes of the error terms added to the regressions. They are significantly different from zero, concluding that the aid variables are not endogenous and this makes the GLS model still appropriate.

15

(33)

Table 4. Durbin Wu Hausman test for endogeneity Dependent variable: decile 1

Model (1) (2) (3)

Error 0.1274 -0.0483 0.0233

(0.3082) (0.1021) (0.6643)

R² 0.5760 0.5819 0.5716

Heteroscedasticity is measured by difference in error terms between the time series or cross sections. The null hypothesis of the Bartlett test assumes that the subgroup variances are equal. These two tests conclude that the variances are equal in time series and not between countries (table 5).

Table 5. Testing heteroscedasticity Test for equality of variances between

Countries Time

Barlett 1.228.489 0.7210

(0.0455)** (0.6973)

The Panel Corrected Standard Error (PCSE) Seemingly Unrelated Regression (SUR) for cross sections is added to account for this heteroscedasticity. This corrects for heteroscedasticity between countries. This method is preferred for panel data as a variant of the commonly used White estimator (Beck and Katz 1995).

The standard errors can not be correlated in time. Otherwise we speak of autocorrelation which is not discussed here, because this panel dataset contains only three time series.

5.2 Outcomes relative pro poor aid

(34)

Table 6. Outcomes of aid in the regressions on relative deciles.

Between parentheses the standard error is displayed.

Note: *is significant at a level of 10 percent, ** is significant at a level of 5 percent and *** is significant at a level of 1 percent.

The first regression (column 1) solely estimates aid as the aid variable. This effect on the first decile is -0.21, meaning that if the ratio of ODA per GDP increases by 1 percentage points the income share of the first decile decreases by 2.1 percentage points. This is significant at a ten percent significance level. For example, in a country where total average income per capita is $1000 the first decile has an average income of (see figure 6; decile 1 has 3%) $300. Consider an increase of the ODA to GDP ratio of 1 percentage point in the ten years of this sample. In case total average income did not change (ceteris paribus), this results in a decrease of 2.1 and average income of decile 1 decreases from 3% to 0.9% and from $300 to $90.

In figure 9 the effect of aid on the income shares is put together. The coefficients seem to fluctuate throughout the distribution. Aid is very positive for the last decile.

Model (1) (2) (3)

Variable ODA ODA ODA*Pol Pol ODA ODA²

(35)

Figure 9. Change of income share due to inflow of aid in model 1 (ceteris paribus). -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 1 2 3 4 5 6 7 8 9 10 Deciles C o e ff ic ie n t o f A id Model 1

The second model estimates aid and aid interacting with policy rights on every decile. The effect of aid on the first three deciles is significantly decreasing income shares ranging from -0.21 to -0.33. The effect of an interacting variable is calculated as follows:

pol ODA y * 2 1 β β + = ∂ ∂ ₍₅₎

Where the change of income ( y∂ ) is divided by the change of aid ( ODA∂ ).

β

1 is the coefficient of aid and β2*pol the coefficient of the interaction variable times

average of policy rates.

The calculation of aid on the first decile is -0.05*4.11 = -0.21, β1 is insignificant and not

(36)

The policy variable ranges from 0 to 7. The limits of the effect of this policy variable are calculated with the same formula only the average of policy is substituted by 0 and 7. The policy effect on the first decile can range from -0.5 to -3.5 percentage point for countries with lower levels of policy rights. The negative sign of aid*policy gives evidence for the theory that countries with good economic policies are better able to allocate aid, even though aid still negatively affects the income shares of the poor.

The third model containing aid and aid² is never statistically significant; none of the deciles have an exponential relation with aid. The first 3 deciles and the richest decile have normally distributed errors. The signs show a negative relation between aid² and the poorest seven income shares and a positive relation with the richest three relative deciles. This negative relation shows the existence of an optimal point of how much aid would be most efficient.

Total outcome of the regressions, including the other variables, are displayed in the appendix (table B6). Striking are the statistically significant outcomes of the dummy variables. The first 8 deciles of Eastern Europe and Asia are positively correlated with aid and the two richest deciles are negatively related to aid. These results are found in all three models. Aid is relatively pro poor in these two regions. In Latin America, on the other hand, the first decile is negatively related to aid. Out of the other explanatory variables FDI has a statistically negative impact on income share of the poorest and increases income share of the rich. Although the effects are close to zero. The variable trade also significantly decreases relative income of the poor. Again the effect is near to zero. FDI and Trade are relatively anti-poor.

Table 7. Outcomes of the regressions on multiple deciles.

Model (1) (2) (3)

Variables ODA ODA ODA*Pol ODA ODA²

(37)

Besides the parsimonious aid regressions this research estimates these three models with different dependent variables in order to compare the effects of aid on the poor to the non poor in another way. The regression estimates the effect of aid on an average of multiple deciles. The average income of several deciles includes the first decile, which makes a good comparison possible (table 7). The first row is the estimation of the aid regression on an average of the two poorest deciles. This results in an effect of -0.25 = -0.06*4.11. The same is done for average of the first three, four and five deciles. The average of decile 1 until 5 results is the last to give significant results and can be found in the second row. In the second model aid has a statistically negative effect of -0.03 (= 0.42+ (-0.11*4.11)), which is the identical to the first model. The average income share of the poorest fifty percent of the income distribution is negatively affected by an increase of aid compared to total average income. The negative impact decreases after subsequently adding a weighted average decile to the dependent variable. Including higher income deciles gives insignificant results. The negative relation of the aid*policy variable suggests that countries with better policy rights gain more from aid.

The effect of aid on the inequality index is only statistically significant for the second model. In case of a percentage point increase of ODA to GDP the gini index will increase by 172.8 percentage points in the first model. This means that the income of share of the richest twenty percent increases by 172.8 percentage points more than the income share of the poorest twenty percent. These results should be interpreted with care since the residuals of this regression are not normally distributed.

5.3 Robustness

(38)

would include these and therefore the probability of their coefficients can be incorrect. The Tobit I model, which is used here, only takes into the regression the non-zero values (Amemiya, 1984). The Tobit model is taken to compare the GLS outcomes. Both techniques include all the observations, because there are no observations of zero average income or below.

Table 8. Outcomes of the regressions of the Tobit model.

The outcomes of the three models using the Tobit model are presented in table 8. Similar to the GLS model the regressions with the first three and the last decile contain normally distributed residuals. The first model with only aid and other explanatory variables shows a statistically significant negative effect of aid on average income of the poorest three relative deciles. Income of the poorest decile decreases with 2.6 percentage points compared to mean average income after one percentage point increase of ODA per GDP. Deciles 2 and 3 relatively decrease with 3.1 percentage points and average income share of the richest decile increases more due to aid.

Model (1) (2) (3)

Variable ODA ODA ODA*Pol ODA ODA²

(39)

The second model with the aid-policy variable also finds that the effect of aid on the first three relative deciles is negatively increasing and that the relative average income of the richest decile increases. The same happens in model 3, although these results are not statistically significant.

The outcomes of the three regressions in the Tobit model give about the same outcome as the GLS model. The effect of aid on the poorest three relative deciles is negative and decreases their income share. The tenth decile is positively affected by aid compared to mean average income.

Besides checking the robustness of the results on aid, another regression is estimated to test the robustness of the outcomes of the aid-policy variable. The policy rights variable of the Freedom House is replaced by the law and order (LawOr) variable from the International Country Risk Guide (ICRG). This variable contains the level of the legal system in a country. Low rating of 1 is given to countries with high criminality and no measures to sanction this behavior. The highest rating is 6, given to countries with good functioning judicial systems.

Below table 9 shows the outcomes of the ICRG data in the aid-policy model, previously referred to as the second model. Many coefficients are statistically significant and deciles 1 until 3 and the last decile have normally distributed residuals. The effect on the first three relative deciles is negative16, implying that aid has a slightly smaller effect on the income of the poorest inhabitants than on mean average income of the country as a whole. The upper ten percent relatively gain 1.28 (=3.8 – (0.8*3.15)), which is a 12.8 percentage point increase after a percentage point increase of aid.

16

(40)

Table 9. Outcomes of the regressions of the IRCG data.

Even though the effects of these last outcomes are smaller, both the Freedom House and ICRG data give negative outcomes on the first three relative deciles. Also the positive effect of aid on the tenth decile is confirmed, now statistically.

Model ICRG

Variable ODA ODA*LawOr

(41)

6. Conclusion

The aid effectiveness literature has changed over the past fifty years from measuring the effect of aid on savings, directly on growth, with a policy condition and lately on poverty. Since the aim for decreasing poverty has gained importance for donor institutions (World Bank, 2000) the effect of aid on the well being of the poor has become the prevalent means for measuring aid effectiveness. This research joins this strand of literature by questioning whether aid is pro poor. The effect of inflowing aid per country GDP is measured on income shares by three GLS models. The base model assumes ODA per GDP as explanatory variable and other variables that have proved to be of influence on income were added to increase the fit of the model. This model was extended by aid interacting with policy level according to the empirical results that countries with good economic policies are better able to allocate aid, obtained by Burnside and Dollar (1997, 2000). The third regression was extended with aid² revealing a non linear relation.

The uniqueness of the database is the separation of the income distribution of 117 developing countries in deciles (Milanovic, 2005). The data was exchanged from local currency into dollars. The regressions that measure the aid impact on absolute income levels did not have normally distributed standard residuals. Therefore the hypothesis on the effect of aid on absolute income levels cannot be confirmed with statically significant evidence.

(42)

the poorest is compared to its effect on the rich. The coefficients of the regressions with decile 10 are positive, but not statistically significant. Aid increases the income share of the richest decile. All in all it is concluded that hypothesis 2 is rejected and aid is relatively anti-poor.

The negative relation of the variable aid interacting with policies relates to the fact that income share of the first deciles decrease less in countries with more political freedom. This finding is in coherence with the theory stated by Burnside and Dollar (2000). Inflowing aid has a relative negative effect on the poorest decile even in countries with good economic policies. The negative signs in the third model are an indication of negative marginal returns of aid which is shown by Durbarry et al. (1998), Hadjimichael et al. (1995) and Lensink and White (2001). Furthermore their results are not matched because the outcomes were insignificant and only measured returns per decile instead of the whole country.

Several shortcomings are also encountered in this conclusion. Do the results make sense? Aid seems to increase economic growth, which in turn increases income of the poor (Dollar and Kraay, 2002), but acts anti poor and stimulates inequality (Boone, 1995). Widespread research already has been done, but overall lack of robustness is the unavoidable result of fragile datasets and weak methodologies (Roodman, 2007). Further investigation and better data can give insight in the robustness of these results.

(43)

References

Articles:

Amemiya T. 1984. Tobit Models; a survey. Journal of Econometrics 24, 3-61

Anderson E. and White H. 2001. Growth and Distribution. Does the pattern of growth matter? Development Policy Review, Volume 19, Number 3, pp. 267-289

Atkinson A. 1987. On the measurement of poverty. Econometrica Vol. 55 No. 4, 749-765 Bacha EL, 1990. A three gaps model of foreign transfers and GDP growth rate in

developing countries. Journal of Development Economics 32(2): 279-296

Barro, R.J. (1991), Economic Growth in a Cross Section of Countries. Quarterly Journal

of Economics, 106, 2, 407-443

Beck N. and Katz J. 1995. What to do (and not to do) with Time-Series Cross-Section Data. The American Political Science Review, Vol. 89, No. 3. pp. 634-647

Burnside C. and Dollar D. 1997. Aid Policies and Growth. Policy Research Working

Paper World Bank

Burnside C. and Dollar D. 2000. Aid, Policies and Growth. American Economic Review

90(4):847-868

Borensztein E., de Gregorio J. and Lee J. 1998. How does Foreign Direct Investment affect economic growth? Journal of International Economics 45 115-135

Collier P. and Dollar D. 2000. Can the World Cut Poverty in Half? How Policy Reform and Effective Aid Can Meet the International Development Goals. Development

Research Group, World Bank

Chenery H. and Strout A. 1966. Foreign Assistance and Economic development.

American Economic review 56:144-51

Chang C., Fernandez – Arias E. Serven L. 1998. Measuring Aid flows; a new approach.

Inter American Development Bank, Working paper 387

Deininger K. and Squire L. 1996. A new data set measuring Income Inequality. The

World Bank Economic Review Vol 10. No. 3 565-91

Dollar D. and Kraay A. 2002. Growth is good for the poor. Journal of Economic Growth,

7, 195-225

(44)

Doucouliagos H. and Paldam M. 2005b. The Aid Effectiveness Literature, sad result of 40 years research. Working paper 15 University of Aarhus

Durbarry, R., Gemmell, N. and Greenaway, D. 1998. New Evidence on the Impact of Foreign Aid on Economic Growth. CREDIT Research Paper 98/8, Centre for Research

in Economic Development and International Trade, University of Nottingham

Griffin K.B. and Enos J.L. 1970. Foreign Assistance: Objectives and consequences.

Economic Development and Cultural Change, Vol. 18, No. 3. pp. 313-327

Gupta, K.L. and Islam, M.A. 1983. Foreign Capital, Savings and Growth - An International Cross-Section Study, Dordrecht: Reidel Publishing Company

Hadjimichael M.T., Ghura D. Mühleisen M., Nord R. and Uçer E.M. 1995. Sub-Saharan Africa: Growth, Savings, and Investment, 1986-93. Occasional Papers 118, Washington

D.C.: International Monetary Fund

Hansen, H. and Tarp F. 1999. The Effectiveness of Foreign Aid, Development

Economics Research Group, Institute of Economics, University of Copenhagen.

Hansen H. and Tarp F. 2000. Aid effectiveness disputed. Journal of International

Development 12 375 -398

Higgins M. and Williamson J. 1999. Explaining Inequality the World Round: Cohort Size Kuznets Curve, and Openness. Federal Reserve Bank of New York Working Paper 79 Kakwani N. and Pernia E. 2000. What is pro poor growth? Asian Development Review,

18, 1-16

Kraay A., 2004. When is growth pro poor? Evidence from a panel of countries. The

World Bank, Policy Research Working Paper No. 3225

Lensink R. and White H. 1999. Is there an Aid Laffer curve? Credit paper 99/6.

University of Nottingham. Centre for research in economic development and international trade

Lensink R. White H. 2001. Are there negative returns to aid? Journal of Development

Studies 37, 42-65

McGillivray M., Feeny S., Hermes N. and Lensink R., 2006. Controversies over the impact of development aid: it works, it doesn’t: it can, but that depends… Journal of

International Development 18: 1031 – 1050

Mosley P. 1980. Aid, savings and growth revisited. Oxford Bulletin of Economics and

(45)

Mosley P. Hudson J. and Verschoor J. 2004. Aid Poverty reduction and the ‘New Conditionality’ The Economic Journal 114 F217-F234

Milanovic B. 1997. A simple way to calculate the Gini coefficient and some implications.

Economic Letter 57, 45 - 49

Milanovic B. 2005. Can we discern the effects of globalization on income distribution? Evidence from Household surveys. World Bank Economic Review Vol 19, no. 1, pp.

21-44

Papanek G. 1973. Aid, foreign private investment, savings and growth in less developed countries. Economic Journal 81(1): 120-130

Pen, J. 1971. Income Distribution: Facts, Theories, Policies. , Praeger, New York Ramirez A, Ranis G, Stewart F. 1997. Economic growth and human development (mimeo)

Ravallion M. and Chen S. 2003. Measuring pro poor growth. Economic Letters 78, 93-99 Ravallion M. 2001. The debate on globalization, inequality and poverty: Looking beyond averages. World Development 29: 11

Ravallion M. 2003. The debate on globalization, poverty and inequality: Why measurement matters. International affairs 79:4, 739-753

Roodman, 2007. The anarchy of numbers Aid development and cross country empirics.

World Bank Economic Review, Volume 21, number 2

Sala-i-Martin X. 1997. I just ran 2 million regressions. American Economic Review 87, 2,

178 – 183

White H. 1999a. Global poverty reduction: are we heading in the right direction? Journal

of International Development 11: 503 - 519.

World Bank 1998. Assessing aid. What works, what doesn’t and why? World Bank policy

research report.

(46)

Books

Amemiya T. 1985. Advanced Econometrics. Cambridge MA: Harvard University Press Carter Hill, R. Griffiths, W. E. and Judge, G. G. 2001. Undergraduate econometrics. New

York Wiley & Sons

Mankiw G. 2002. Macroeconomics. Palgrave Mc Millan

Milanovic B. 1998. Income, Inequality and poverty during the transition from planned to market economy. World Bank Publications

Mosley P. 1987. Overseas Development Aid: Its Defence and Reform. Wheatsheaf:

Brighton.

Websites

IS AID PRO POOR?

IS AID PRO POOR?

Preface

Table of Contents

Tables and figures

1. Introduction

2. Literature review

3. Methodology

α

α

β

β

β

β

β

4. Data description

5. Empirical results

β

6. Conclusion

References