Master Thesis International Economics and Business

(1)

International Economics and Business

Development aid from a sectoral perspective

Keywords: sectoral ODA, Dutch Disease, Medicine model.

Abstract:

The question whether or not development aid has a positive effect on growth is a hot topic. The latest research indicates that it could be negative, or only positive under certain conditions. In this research I will look at it from a sectoral perspective, trying to find evidence that the aggregated level loses its significance due to a distinctive effect of development aid at the sector level. I have found that development aid does positively influence growth at the aggregated level. This positive effect becomes less clear at the sector level, however there is some evidence that sectors are influenced differently by development aid.

Supervisor: dr. Robert Inklaar

Co-assessor: dr. Ger Lanjouw

Date: Wednesday 16-07-2012

Student: Remco Hoogma

Student number: 1554980

(2)

Introduction

A common way of charity is giving money to countries which are poorer. This is mostly done in the form of development aid. The question is whether this aid is beneficial for developing countries or not. Throughout history we can see an ever increasing flow of development aid, see figure 1. It all started with the Marschall plan, in which European countries where helped to overcome the damage of WW2. In this period the consensus was that development aid was necessary and effective. It the following decade’s researchers developed the theoretical ideas behind this relationship, at first it was seen in the context of the dual gap model. Basic gap models assert that the rate of economic growth is constrained by inadequate levels of foreign exchange and savings. In order to fill these gaps foreign aid is necessary to achieve a target rate of growth (White, 1992). This positive relationship between foreign capital flows and savings predicted by the Harrod–Domar model has not been observed. In fact, most studies generally find a negative association between the two (M. McGillivray et al. 2006).

Figure 1 World ODA (source: World Bank)

Up to the 1990, no consensus has been reached about the impact of foreign aid on economic growth. Some claimed that due to measurement issues it was difficult to measure the impact of aid on a macroeconomic level. For instance Mosley (1987) came up with the macro- micro- paradox, with this paradox it was meant that, while micro-, or project- related studies found a positive impact of aid, macro level studies did not found these positive relations. In 1998 the World Bank published the ‘Assessing Aid’ report which provided a new discussion on the

(4)

macroeconomic effectiveness of development aid. This study is one of the first to acknowledge that aid effectiveness depends on the specific circumstances of recipient countries. McGillivray et al. (2006) give a summary of all the critiques on the findings of the study of the World Bank, they show that in none of the other studies the interactive term of aid*policy index is found significant. Does this mean that we are at an unknowing state again? Maybe so, but many researchers continue to search for the relationship between development aid and economic growth. As Bourguignon and Sundberg (2007) mention in their paper, even though the cross-country evidence of the effectiveness of aid is fragile, this does not mean that the beneficial effect of aid is non-existing. They claim that in order to understand the effectiveness of aid, it requires that we open the so called “black box”. They say that there is a complex causality chain linking aid to final outcomes. Following this view this research will try to open up this “chain” and investigate one part of it.

One way of opening the black box is to not look at the flow of development aid in its aggregated form, but to take a sectoral view at its components. The effects of development aid could be very different for each sector. As Alfaro (2003) found that the relationship with Foreign Direct Investment (FDI) on economic growth is different for each sector, it is not hard to translate this to development aid. They are both inflows of capital into the country and are often used as investment. Some researchers claim that the insignificant result found in previous studies is caused by the fact that they use aggregated data and thereby they disregard the heterogeneity of aid (Clemens et al., 2004). The studies of Clemens et al. (2004), Asiedu and Nandwa (2007), and Rajan and Subramanian (2011) are some of the few who used aid data disaggregated by sector. Of these, the latter two studies only looked at one specific sector, making it impossible to value their results at the macroeconomic level. Dividing the economy into sectors in order to see if development aid has a distinctive effect on growth at a sectoral level is the novelty of this research. This brings me to my research question:

Does development aid have a significant effect on growth on a sectoral level?

Sub questions are:

To which sectors does development aid have a negative effect? To which sectors does development aid have a positive effect?

(5)

The importance of this research lies in the fact that donor countries doubt whether or not to continue or to reduce the amount of development aid. This is mainly caused due to the questionable results and effects of development aid found by different researches. If this research question is significant and we can determine to which sectors development aid has a positive or negative influence, we would be able to show that it matters to which sector you send it. Additionally, it could serve as an advice for organisations and policy makers who control the distribution of development aid.

The rest of this paper is organized in the following way: first there will be a discussion of the literature, followed by the theory and the model, after this the methodology and the data. The final parts consist of the robustness check and the conclusion.

Literature review

The aid effectiveness literature has mainly taken a perspective at a macro or micro level. In this perspective we encounter the micro- macro paradox of Mosley (1987). This paradox can be explained by the fact that at the micro level the success and failure of the projects can be seen directly, this in contrast to the macro level in which the aggregate is taken of all projects. This could mean that due to the fact that 50% of all aid projects fail (Cassen, 1994), the net effect on the macro perspective seems neutral. However, as Cassen mentions in his book even though only 50% of development projects work, the ones that fail cause very little harm. This would mean that you would still expect to see the positive effects of the micro level to show up at the macro level.

(6)

as well. So from 1995 on we can see a development to the conditional estimates in which the effect of aid on growth depends on a conditional variable.

One of these conditional mechanisms was developed by the World Bank. They developed the theory that national policies have a critical influence on the effectiveness of aid, for the same reasons that they affect economic growth (World Bank, 1998). The World Bank restored the image of the positive link between development aid and economic growth. According to this report, aid does help to increase growth, but only in countries with sound economic management or good governance. The claims of the Assessing Aid report with respect to the effectiveness of aid are mainly based on the studies of Burnside and Dollar (2000) and Collier and Dollar (2002)1. The Good Policy model of Burnside and Dollar (2000) claimed that aid only worked if the recipient country pursued good policies. This model has been investigated and reproduced by others and none of these studies found the positive relationship between development aid and growth, using the same conditional variable as Burnside and Dollar (Hansen, 2001, Hansen and Tarp, 2001, Lensink and White, 2001, Jensen and Paldam, 2006, Islam, 2002 and Ram, 2004).

This opened the debate again if there is an aid- growth relationship and if so, through which channels this relationship works. McGillivray et al. (2006) distinguished four main alternative views on the effectiveness of aid: aid has decreasing returns; aid effectiveness is influenced by external and climatic conditions; aid effectiveness is influenced by political conditions and aid effectiveness depends on institutional quality. Even with these new mechanisms, through which aid affects growth, the results are still ambiguous and the relationship is unclear. Minoiu and Reddy (2010) have disentangled the effects of aid in two kinds: in developmental and non-developmental and looked at the effect over longer periods2. They found that developmental aid promotes long run growth and hence is beneficial.

This unclarity at the macroeconomic level is also present when we look at the influence of foreign direct investment on economic growth. Here we see the phenomenon that some studies find a positive impact of FDI on economic growth and others find a negative relation between

1

These paperswhere originally circulated prior to their publication date.

2

(7)

these two variables (Li and Liu, 2005). In the study of Alfaro (2003) she claims that the effect per sector can neutralize the effect at the macroeconomic level. She decomposes the macro environment in three sectors and she found that these sectors are all differently affected by the inflow of FDI. This would mean that the results at the macro level could be ambiguous because the positive effects of flows to one sector are neutralized by the negative effects of another sector. This neutralizing effect might explain the absence of a positive or negative effect of development aid on the macro economic situation of countries. This makes it necessary to investigate whether or not this relation becomes clearer on the sectoral level.

Until now not many researches have shifted to a sectoral perspective. Most studies have looked at the aggregated effect of aid on economic growth. The studies of Clemens et al. (2004), Rajan and Subramanian (2011) are one of the few who used aid data disaggregated by sector. Clemens et al. divided aid into three different components: emergency and humanitarian aid; short-impact aid, which is defined as aid that stimulates growth within 4 years (this includes budget support, infrastructure, banking, agricultural, and industry); and long-impact aid, includes technical assistance, democracy, environment, health, education. In their research they focused on short term aid, which was found to have a positive and significant impact on growth. This division can be seen as one of the first steps in looking at the effect of development aid on different parts of the economy. However, Rajan and Subramanian (2005) claim that by looking at the short-term aid as well as long term, economic, social, and food aid on growth that none of these types of aid show a robust effect on growth. Although the measurement of development aid has been disaggregated, both studies still combine the disaggregated development aid flows into two or three aggregated sectors, making the independent variables of development again in a big pool of aggregated development aid.

(8)

effect of aid on the manufacturing sector. They found a negative relationship, which could be explained by what they termed ‘The Dutch Disease’.

Theory and model

In this theoretical part I will explain why the positive effect of development aid on economic growth could be expected, I will also look at the mechanisms through which this effect can turn negative or perform in a non-optimal way. Thirdly, I will discuss why each sector can be influenced differently.

Positive effect of aid

A theoretical impact of aid on growth can be derived by using the Solow-Swan Growth model. This model assumes that a fraction of aid goes toward financing public investment, which has an impact on long-run growth via capital accumulation. By using this model Rajan and Subramanian (2005) calculated that a “1 percentage point increase in the ratio of aid to GDP

should at most raise the long-run growth rate by 0.16 percent”. This number is derived by

making use of: the capital share in income; the fraction of aid that is invested and the output capital ratio. They assumed that all aid is invested, the capital share of 0.35 was used of Bosworth and Collins (2003) and the output capital ratio of 0.45 was derived by using their own data. Rajan and Subramanian (2005) add one comment to this number: due to the IT revolution the upper limit for the impact of aid on growth has gone up. This shows that it is not a fixed number, but an estimation depending on the circumstances. Even though this number is interesting, the message is that from a pure textbook point of view an increase in investments or flows to a sector should positively influence the growth level of an economy or a sector. That is why I assume in an optimal situation, in which there are no constraints on the absorptive capacity, that is all the development aid is invested in beneficial projects, the effect of development aid will have a positive influence on growth at the sectoral level and at the macro level. This perspective will be the starting point of the research, so the first hypothesis will be:

Hypothesis 1: Development aid distributed to a sector will positively influences the growth level of that sector.

(9)

which the positive effect of development aid is moderated or even could harm a sector. This will be discussed in the following section.

Negative effect or non-optimal situation of aid

There are a number of mechanisms which can affect the effectiveness of aid, some are at the country level while others are more sector specific. The effectiveness of aid and how a country is able to use it is sometimes called the absorptive capacity. Many factors can influence this absorptive capacity, for instance at an environmental level governance issues like corruption, political instability and lack of accountability can all negatively influence the absorptive capabilities of a country (De Renzio, 2005). Also the quality of institutions and even social and cultural factors can influence the effectiveness of aid.

Two other factors are the sensitivity to large external shocks and the type of aid. Countries which are dependent on aid are more sensitive to large external shocks. They are not able to deal with shocks, due to their liquidity constraints and the lack of effective countercyclical policy tools (Bulíř and Hamann, 2008). This means that volatile, unpredictable and procyclical aid can decrease the macroeconomic stability. Aid is found to be highly volatile in developing countries, it turns out that shortfalls in aid and domestic revenue tend to coincide (Bulíř and Hamann, 2003). According to Gemmell and McGillivray (1998) shortfalls in aid are frequently followed by reductions in government spending and sometimes by increases in taxes. This indicates that a typical aid receiving/dependent country is unable to offset an unexpected non-disbursement of aid by borrowing and has to resort to other fiscal adjustments. This uncertainty is detrimental to growth according to Lensink and Morrissey (2000). Minoiu and Reddy (2010) found that the type of aid also influences the effect on growth. They make a distinction between two kinds: developmental and non-developmental. They even found that the type of donor influences the potential growth.

(10)

Dutch Disease effect

The Dutch Disease concept of van Wijnbergen was introduced in 1984. It referred to the adverse impact on Dutch manufacturing industry due to the increase in income associated with the discovery of natural gas in the Netherlands in the 1960’s. This led to the appreciation of the Dutch real exchange rate, and as a consequence, the shrinkage of the tradable sector. This mechanism can be translated to development aid in the following way: in the case of a boom caused by large aid flows, it is seen that the public sector crowds out the private sector through its increased spending associated with aid (Nkusu, 2004). In turn a real appreciation may cause a reallocation of labour to the non-traded goods sector, which will result in a rise of real wages in terms of the price of tradables. This could have a deterioration effect in competitiveness and lead to a decline in export performance and a negative effect on growth (Agénor et al., 2008). Some claim that if there is learning by doing, this negative effect could be ambiguous on the long run (Torvik, 2001; Adam and Bevan, 2003). However, for this research I will look if the exporting sector is negatively influenced by development aid. This gives the second hypothesis:

Hypothesis 2: Sectors or countries which are more export intensive will show detrimental results to development aid.

The medicine model

(11)

Figure 2 Medicine model (source: Jensen and Paldman, 2006)

The interval in which the effect of aid is positive is not set in stone and is less clear. For instance the study of Lensink and White (2001) found that the turning point of aid to GNP ratio is about 50 per cent. Hadjimichael et al. (1995) found this turning point twice as low at a level of 25 per cent. Collier and Dollar (1999) found it at an even lower point of 18,5 per cent. Even though the exact number is unclear, several studies did find evidence for the Laffer curve. In this study the emphasis does not lie on figuring out where this turning point lies, but to show that there are some negative effects and that these can be different per sector. This gives us our third hypothesis:

Hypothesis 3: Large percentages of aid to GDP or sector size will result in diminishing effects on growth.

In line with previous studies I expect to find this turning point between the range of 18% and 50%.

Why difference per sector?

(12)

exhibit stronger negative effects of aid volatility; the type of aid will influence the effectiveness of aid and hence the growth level per sector. In figure 3 a graphical representation is presented.

This research will concentrate on the Dutch Disease effect and the Medicine Model. How this will investigated will be explained in the next section.

Methodology

To investigate the relationship between development aid and growth a simple cross sectional pooled OLS regression technique is an appropriate method. This has also been used by Rajan and Subramanian (2005), Minoiu and Reddy (2010), and Jensen and Paldman (2006). An advantage in comparison to a single cross-section is the increase in sample size, some disadvantages are that this estimation technique is very sensitive for omitted variable bias and fails to differentiate between aid flows and other foreign capital inflows (M. McGillivray et al. 2006). The first could be a problem which needs to be kept in mind and will be checked for with an ovtest, the latter is more a data collection issue which can be solved by collecting the appropriate data.

The pooled OLS regression is simply an OLS regression applied to the whole dataset, which does not take into consideration the countries across time periods. Prior to the regression the data will be checked for normality by plotting it and by using a swilk r test. The swilk r test performs the Shapiro-Wilk W test for normality, the p-value is based on the assumption that the distribution is normal. Presence of autocorrelation will be checked by making use the Wooldridge test and the Breusch-Pagan/Cook-Weisberg test for heteroskedasticity will indicate if there is evidence of heteroskedasticity.

(13)

In order to test the hypotheses and to find out if development aid at a sectoral level has a different influence on growth, I will conduct several regressions. The setup will be divided into three streams: the first will look at the aggregated data and the effect on country growth, the second will look at sector growth with aggregated ODA and the third will look at sector growth with sector ODA. In each stream I will look at three models: the general model in which ODA affects growth directly; the Dutch Disease model which will be investigated in two ways: one in which growth is affected by the export intensity of a country and a second by making use of a dummy variable for the most export intensive sectors; and the third model will be the Medicine model in which growth is affected according to the Laffer curve. In the following section the regressions for each stream are written down, in addition to those regressions I will also conduct a regression for each stream with the three models together. This regression will not be written out, but it is simply the combination of the three models.

After each regression I will check if the regression is subjected to omitted variable bias with an ovtest. This test creates new variables based on the predictors and refits the model using those new variables to see if any of them would be significant. I will check for multicollinearity with a Variance Inflation Factor (VIF) test. The rule of thumb is that if a variable whose VIF values are greater than 10 needs further investigation.

Stream A

This stream looks at the aggregated form of growth and ODA, it looks on a country level. The regressions are given underneath, where i is the country, j is the sector, t is the year, X is a term for the export intensity.

General model:

= α + β ODA

GDP + γcontrols + ε Dutch Disease model:

(14)

Stream B

This stream looks at how the sector growth is affected by the aggregated flow of development aid to that country. The regressions are:

General model:

= α + β ODA

= + + ODA GDP + ∗ ODA GDP + + = + + ODA GDP + ∗ ODA GDP + + Medicine model: = α + + + δcontrols + ε Stream C

In this stream I will look at the effect of disaggregated ODA on sectoral growth. The regressions are:

General model:

= α + β ODA

= + + ODA GDP + ∗ ODA GDP + + = + + ODA GDP + ∗ ODA GDP + + Medicine model : = α + β ODA GDP + γ ODA GDP + δcontrols + ε

Data

(15)

Independent variable development aid

The independent variable will be measured in the form of Official Development Assistance (ODA). This is defined by the OECD in the following way: “those flows to developing countries

and multilateral institutions provided by official agencies, including state and local governments, or by their executive agencies, each transaction of which meets the following tests: i) it is administered with the promotion of the economic development and welfare of developing countries as its main objective; and ii) it is concessional in character and conveys a grant element of at least 25 per cent.” For the data of ODA flows to the sectors the Credit Reporting

System (CRS) of the OECD will be used. The CRS reports aid commitments disaggregated by 233 distinct purposes, for all donors and recipients annually. They have made a distinction between aid commitments and disbursements by donors. Because I am interested in the actual effect of aid I will use the disbursement aid instead of the commitments number. The data for disbursements is available for the years 2002 until 2010. The OECD reports that it is not recommended to use flows prior to 2002, because the annual coverage is below 60%, while from 2002 on it is around and over 90% and continuously improving. An aspect that should be taken into consideration is the fact that some countries receive a small amount of ODA. Since I want to look at the effect of ODA on growth it is necessary that the ODA level is not too small, otherwise it will be almost impossible to detect an effect of it on growth. Therefore I will only include the countries that at least receive a minimum of 1% of ODA/GDP. Similar is done by Rajan and Subramanian (2011).

Dependent variable growth

The dependent variable is the growth of the sector or country, which will be measured in value added by economic activity, annual average rate of growth in percentages. The data will come from the UN National Accounts Main Aggregates Database, this database presents a series of analytical national accounts tables from 1970 onwards for more than 200 countries and areas of the world. Due to the limitation of the independent variable I will use the data from 2002 until 2010. The data is allocated to a sectoral perspective with the International Standard Industrial Classification of All Economic Activities, Rev.3.1.

Control variables

(16)

necessary to select a few of them which seem most suitable. In this paper I will include five control variables which all affect growth, these are the same as the control variables used by Asiedu and Nandwa (2007), when they looked at the effect of ODA in the education sector and the same as Hansen and Tarp (1999) used in their research. These are:

• Rule of law, this is from the Kaufmann World Governance Indicators (WGI) it captures the extent to which agents have confidence in the rules of society, contract enforcement, property rights, the courts, and the likelihood of crime and violence in a country. The maintenance of the rule of law is found to positively influence growth (Barro, 1996).

• Fixed investment/GDP (%), this comes from the World Development Indicators (WDI) of the World Bank, it represents the gross fixed capital formation. High rates of fixed investment are one of the sources for economic growth (Blomstrom et al., 1996).

• Log (initial GDP per capita), this is also from the WDI World Bank. This variable captures the conditional convergence effects and has been used in many empirical growth studies (Hansen and Tarp, 2001).

• Inflation, this comes from the WDI and is the annual percentage change in the cost of acquiring a basket of goods and services. Extreme numbers of inflation or deflation have a negative influence on growth.

• Government consumption/GDP (%), is the expense for operating activities of the government in providing goods and services. This number also comes from the WDI. Growth rates are affected by government consumption because they are one of the biggest consumers in a country.

Sector classification

(17)

• Agriculture, hunting, forestry, fishing (ISIC A-B)

• Mining, Utilities (ISIC C&E)

• Manufacturing (ISIC D)

• Construction (ISIC F)

• Transport, storage and communication (ISIC I)

• Other Activities (ISIC J-P) (service sector)

In the following pie chart the disbursement of ODA on the sector level is presented. What we can see is that a big part cannot be attributed to any specific sector. This part accounts for 40% of the total ODA.

Figure 4 Sector ODA

The biggest receiver of ODA is ISIC J-P which is the service sector. This sector gets approximately 40% of all the ODA. To make a good judgement we should compare it to the sector size. This however is tricky because some countries which do receive a little bit of ODA are well developed and hence have large developed sectors. This causes that the average percentage of ODA to sector is highly influenced by these countries. For example China still receives development aid, but the sizes of the sectors are very big. In the following figure 5 ODA size to sector we can see how much ODA a sector on average gets compared to its size. I have only included the countries which receive on average more than 1% of ODA/GDP.

What we can see by looking at both of these figures is that the service sector not only gets the biggest number of development aid, but that on average it also gets the biggest part compared to the size of its sector. The construction sector is the sector which barely receives 1% of ODA compared to its size.

ODA ISIC A+B _{ODA ISIC C+E}

ODA ISIC D ODA ISIC F ODA ISIC I ODA ISIC J-P Other

(18)

Figure 5 ODA % in sector

Dummy variable export

In order to measure the Dutch Disease effect, a dummy variable needs to be introduced which signifies the exportability of the specific sector. This dummy variable for export is made in a similar way as Rajan and Subramanian (2010), they developed an exportability index for the different subsectors of the manufacturing sector. This index was constructed by looking at a few countries and then generalizing it. This means that every country gets the same score for its sectors. I have formed this dummy variable by looking at the WITS World Bank data of export per sector and the OECD trade data per sector. The first has also been used by Rajan and Subramanian (2010), to create their exportability index. Taking this together we see that the following sectors are the most export intensive of the six sectors:

• Agriculture, hunting, forestry, fishing (ISIC A-B)

• Mining, Utilities (ISIC C&E)

• Manufacturing (ISIC D)

These sectors will get the value of 1 in the dummy variable.

0 2 4 6 8 10 12 A+B C+E D F I J-P Perce nt age Sector

(19)

Empirical analysis

In this part the empirical analysis will be conducted. First a summary table is given with all the data. After this overview there will be devoted a subsection to each stream describing the data in more detail, the problems with the data, the regression techniques and the results of the regressions. The control variable government spending will be left out of the regression because of the limited availability.

Table 1 Summary Statistics

Stream A

Variable Obs Mean Std De. Min Max

GDP growth rate UN 1263 0.049 0.046 -0.140 0.350

Total ODA/GDP 785 0.093 0.129 0.010 1.515

Total export % GDP 1120 0.343 0.216 0.000 0.999

Fixed investments % GDP 1269 0.182 0.111 0.000 0.600

Log GDP per capita 1232 0.742 1.204 4.466 10.233

Inflation 1106 0.074 0.079 -0.100 1.090

Government spending 564 0.216 0.085 0.000 0.510

Rule of law 1255 -0.506 0.712 -2.677 1.731

Stream B

Variable obs Mean Std De. Min Max

Sector growth rate 7614 0.055 0.110 -0.760 2.080

Total ODA/GDP 4710 0.093 0.129 0.010 1.515

Dummy export 7614 0.500 0.500 0.000 1.000

Total export % GDP 6720 0.343 0.216 0.000 0.999

fixed investments % GDP 7614 0.182 0.111 0.000 0.602

Inflation 6636 0.074 0.079 -0.100 1.090

Rule of law 7530 -0.506 0.711 -2.677 1.731

Stream C

Variable obs Mean Std De. Min Max

Sector growth rate 7614 0.055 0.110 -0.760 2.080

Sector ODA/value added sector 2545 0.160 0.651 0.010 2.490

Dummy export 7614 0.500 0.500 0.000 1.000

Total export % GDP 6720 0.343 0.216 0.000 0.999

fixed investments % GDP 7614 0.182 0.111 0.000 0.602

Inflation 6636 0.074 0.079 -0.100 1.090

(20)

Stream A

Stream A looks at the influence of the aggregated ODA on a country level. The first step is to investigate the data by plotting it and looking for values that could be seen as outliers. If we look for outliers in the growth data we see a few countries which have remarkable numbers. Liberia in 2003 has a growth number of -31% and Iraq also has a negative number of -33%. In 2004 Iraq shows a growth number of +54%, Chad in 2004 +34% and Afghanistan +65% in 2002, all these numbers can be seen as extreme and not normal. These numbers are caused by extreme situation like wars or social instability in that specific year or in the years prior to it, so these are excluded from the regressions. I will discuss the effect of this exclusion in the section robustness.

If we look at the ODA/GDP we also encounter a few extreme numbers. The Marshall Islands, Micronesia, Timor-Leste and Tuvalu get on average more than 40% of ODA/GDP. Liberia also gets on average more than 30% of ODA. In 2008 it got 151% of ODA and 2010 81% compared to their GDP. In 2005 Iraq got 121% of ODA/GDP and Congo in 2003 91%. Even though these numbers are much higher than the average of 9.3% they could be important to find evidence of the Medicine model, so they are not excluded. The inflation data also shows some extreme numbers, in 2003 in Haiti 39% and 23% in 2004. Myamar has several years with very high inflation of 30% or higher. And Zimbabwe shows hyperinflation with inflation number of over a 100%. After examining the data further we find that it is normally distributed there is no evidence of autocorrelation in the panel and that it is heteroskedastic. Because of the heterosketasticity I will relax the assumption of zero error correlation over time for the same individual. I will cluster by countries, this will cause that the standard errors are slightly higher compared to the uncorrected regression. The output can be found in table 2 stream A.

Discussion stream A

(21)

Table 2 Stream A Stream A GM DD MM TM Aid variable Total ODA/GDP 0.039*** 0.079 0.083*** 0.114 (0.003) (0.173) (0.000) (0.108) (Total ODA/GDP)^2 -0.048*** -0.087* (0.001) (0.085) Exp 0.014 0.011 (0.331) (0.241) Exp* (ODA/GDP) -0.132 -0.040 (0.456) (0.803) Control Variables Rule of law 0.004 0.005 0.003 0.004 (0.325) (0.232) (0.345) (0.308) Fixed investments % of GDP 0.091*** 0.108*** 0.094*** 0.109*** (0.000) (0.000) (0.000) (0.000) Inflation 0.062*** 0.059*** 0.062*** 0.058*** (0.000) (0.002) (0.000) (0.002)

Log GDP per capita -0.006** -0.007*** -0.005** -0.006**

(0.010) (0.004) (0.028) (0.012) Constant 0.067*** 0.068*** 0.057*** 0.061*** (0.000) (0.000) (0.000) (0.001) Number of observations 601 550 601 535 R squared 0.088 0.105 0.093 0.113 F statistic 8.590 5.940 7.780 9.060 Prob>F 0.000 0.000 0.000 0.000 VIF meana 1.230 2.000 2.260 2.770 Ovtestb 0.517 0.992 0.413 0.989

Note P-values in parentheses. *,**,*** denote significance at the 10, 5, and 1 percent level, respectively.

a

value below 10 no evidence of multicollinearity,

b

(22)

All the aid variables in the DD model have the expected sign. The interaction variable between aid and export is negative, which indicates that there is a negative interaction that affects growth. However, all the aid variables are insignificant making the support for the DD very weak. For the MM we see strong evidence of this phenomenon: the two aid variables have opposite values and are highly significant. The turning point however lies at a value of 86%. This would indicate that only under extreme circumstances aid will negatively influence growth. In the total model most numbers turn insignificant indicating that the model does not improve if we put it together. What we can conclude from the first stream is that aid does seem to have a direct effect on growth, that on an aggregated level there is no significant evidence of the Dutch Disease model and that the Medicine model only plays a role in very high numbers of ODA to GDP.

Stream B

The distinction between stream B and A is that stream B looks at the growth number of the sector. It does however still take the aggregated flow of ODA to the country as the independent variable. In other words, it looks how a specific sector is influenced by the total flow of ODA. The setup is the same as in the previous stream, only the check for the Dutch Disease model will be done in two ways. The first is the same as in stream A where I multiply the ODA/GDP of a country with the export intensity of that country. The second method is by making use of the dummy variable as an exportability index.

(23)

Table 3 Stream B Stream B GM DD1 DD2 MM TM Aid variable Total ODA/GDP 0.035** 0.139* 0.021 0.155*** 0.042 (0.019) (0.053) (0.412) (0.000) (0.470) (Total ODA/GDP)*Exp -0.311 0.125 (0.176) (0.126) Exp 0.035** 0.011 (0.027) (0.289)

(Total ODA/GDP)*Dummy Var. 0.023 0.043

(0.475) (0.580) Dummy Var. -0.025*** -0.028*** (0.000) (0.000) (Total ODA/GDP)^2 -0.129*** 0.032 (0.000) 0.685 Control Variables Rule of law -0.004 0.001 -0.004 -0.005 0.002 (0.418) (0.739) (0.471) (0.349) (0.583) Fixed investments % of GDP 0.083*** 0.103*** 0.081*** 0.092*** 0.118*** (0.005) (0.002) (0.008) (0.001) (0.001) Inflation 0.099** 0.034 0.100** 0.100** 0.038 (0.016) (0.196) (0.016) (0.017) (0.136)

Log GDP per capita -0.008*** -0.012*** -0.008*** -0.006** -0.010***

(0.008) (0.000) (0.008) (0.081) (0.000) Constant 0.084*** 0.103*** 0.096*** 0.059** 0.105*** (0.000) (0.000) (0.000) (0.021) (0.000) Number of observations 3618 3306 3588 3618 3096 R squared 0.017 0.025 0.027 0.022 0.048 F statistic 8.870 8.950 13.54 9.170 12.24 Prob>F 0.000 0.000 0.000 0.000 0.000 VIF meana 1.230 3.970 1.590 2.250 2.260 Ovtestb 0.000 0.212 0.003 0.002 0.031

a

b

(24)

Discussion stream B

In this stream we can see that the general model again shows a positive influence of aid on growth. The first regression of the DD model shows that the effect of ODA/GDP is insignificant at the 5% level, combined with the negative number of the interaction term this could indicate that ODA affects growth through the interaction with the export variable. However, this interaction term is again not significant. This is also the case in the second regression for the DD model. We see that the export dummy is negative and significant, this indicates that there is a difference between the sectors. The Medicine Model regression again shows the expected signs and the aid variables are significant at a p value of 0.001. The turning point or the top of the laffer curve lies higher than Lensink & White (2001), Hadjimichael et al. (1995) and Collier & Dollar (1999) found, in stream B it lies at 60%. This number as in stream A is again really high, making it a rare situation in which this phenomenon would affect growth. In the total model we see that only the dummy variable is significant of the aid variables. This again shows that the total model does not show a better fit than the models separately.

One thing of concern is that in four of the five regressions there seems to be a model specification error. This could be caused by one or more relevant variables are omitted from the model or one or more irrelevant variables are included in the model.

Stream C

(25)

sectors appear to receive more than a 100% of the value added in the sector. There will be no upper limit on the ODA percentage to the sector, because I am also looking to find evidence of the Medicine Model. The Breusch-Pagan / Cook-Weisberg test for heteroskedasticity again shows that the data is heteroskedastic. Also the Wooldridge test for autocorrelation in panel data shows that the data suffers from autocorrelation. I will control for this in the same way as in stream A and B. The regression output can be found in table 4 Stream C.

Discussion stream C

In this stream I have looked at the effects of sectoral ODA on sectoral growth. What we encounter is that the significant effect of the aid variables of stream A and B is lost if we take it down to the lowest level. The DD model however becomes significant. The first regression of the DD model shows that all the aid variables are significant at the 5% level. In this regression we encounter for the first time a direct negative effect of ODA on growth. This is not according to expectations, also the interaction term does not show the expected sign. If the Dutch Disease effect would work, you would expect to see the positive and negative signs the other way around. In the second regression for the DD model we see again that the dummy variable is negative and highly significant, which is an indication that the sectors affect growth differently. In this stream the signs are as expected but not significant.

The aid variables of the MM are significant at a p value of 0.10. The value of the turning point of the MM lies at 150% (sector ODA/added value sector). This is a really high number which only happens in very rare occasions. Interesting is the fact that this lies higher than the aggregated level that was found in stream B. Maybe this is an indication that the saturation point of the sector to absorb ODA lies higher than the aggregated level.

(26)

Table 4 Stream C

Stream C

GM DD1 DD2 MM TM

Aid variable

Total ODA sector / value added sector 0.009 -0.051** 0.030 0.048* 0.003

(0.387) (0.047) (0.463) (0.065) (0.955)

(Total ODA sector /value added sector)*Exp

0.245** 0.224**

(0.013) (0.026)

Exp. 0.030** 0.019

(0.019) (0.128)

(Total ODA sector /value added sector)* Dum Var.

-0.012 0.009

(0.734) (0.821)

Dummy Var. -0.023*** -0.027***

(0.000) (0.000)

(Total ODA sector /value added sector)^2 -0.016** -0.025*** (0.041) (0.009) Control Variables Rule of law -0.016*** -0.010* -0.012** -0.016*** -0.005 (0.006) (0.056) (0.028) (0.005) (0.246) Fixed investments % of GDP 0.072*** 0.085*** 0.071*** 0.076*** 0.085*** (0.004) (0.005) (0.003) (0.002) (0.004) Inflation 0.028 0.004 0.035 0.028 0.016 (0.462) (0.878) (0.365) (0.453) (0.517)

Log GDP per capita -0.003 -0.007** -0.006** -0.003 -0.009***

(0.304) (0.018) (0.046) (0.421) (0.001) Constant 0.054** 0.072*** 0.089*** 0.045 0.104*** (0.047) (0.009) (0.000) (0.109) (0.000) Number of observations 2121 1879 1934 2121 1705 R squared 0.019 0.041 0.034 0.023 0.071 F statistic 6.340 6.860 8.990 5.500 8.690 Prob>F 0.000 0.000 0.000 0.000 0.000 VIF meana 1.170 2.240 2.080 2.190 2.800 Ovtestb _0.340 _0.842 _0.193 _0.004 _0.935

a

(27)

General discussion

What we can see in the process of going from stream A until C is that the effect of ODA on growth becomes less clear. Were we found a significant and direct effect of ODA on growth in the first two streams, this significant effect was lost at the lowest stream. Also the evidence of the MM gets weaker and the saturation point was found at an extreme high number. In all the regressions we have a low variance inflation factor (vif) indicating that there is no direct evidence of multicollinearity. The rule of thumb is that a variable whose VIF values are greater than 10 may merit further investigation. The explanatory power of the models also decreases as we go from an aggregated level to a sectoral level. In stream A we started with an R-square of almost 0.10, this decreased until barely 0.02 in stream C. One of the reasons for this decrease could be the fact that the control variables are at a national level.

If we look at the control variables we also see some interesting results. We can see that rule of law is not significant in stream A and B and negative in stream C. The initial GDP level per capita is almost always significant and negative, this could indicate that a lower GDP per capita is beneficial for growth. This is surprising to find and interesting to investigate further, however this does not fall within the borders of this research. The control variable ´fixed investment as a percentage of GDP´ is most of the time significant and always positive, which is in line with expectations. In stream B and C the dummy variable for exporting sectors was introduced. This variable was highly significant in both streams. On the contrary, the interaction variable which would give evidence of the Dutch Disease model, was not found significant. Only in one occasion was it significant, but it did not show the expected sign. Making the support for the Dutch Disease model very weak. The Medicine model was found significant in all three streams, only the saturation point is not in line with previous studies.

Robustness

(28)

Dummy variable

In order to check the correctness of the dummy variable I have done the general regression for each sector, see appendix 2 dummy variable. The results of these regressions where that indeed sector 1(agriculture, hunting, forestry and fishing) and sector 3(manufacturing) are negatively affected by ODA. Sector 2(mining and utilities) which was also included in the dummy as a sector that is export intensive, does not show this negative effect, however the positive effect is small. Sectors 5(transport, storage and communication) and sector 6(service sector) both show positive effects of ODA on growth. However, of all the regressions only the regression of sector 6 was significant. These results give evidence that the effects of ODA on growth at a sector level are different for each sector. Additionally, the classification of the sectors in the dummy variable seems to be acceptable, which give some evidence in the direction of the Dutch Disease model.

Dummy model

A disadvantage of the pooled OLS is that it does not distinguish between two different individuals and the same individual at two points in time. This can be a drawback when differences exist between cross-sectional individuals. To filter out individual, sector and year effects a dummy model can be used. This method is called the least squares dummy variable estimator. The regression results are presented in appendix 3 dummy model. What we can see in the regression is that the results do not improve compared with the pooled OLS regression. Most of the variables turned insignificant in the dummy model. Additionally, what we could see is that most years are insignificant in the model indicating that it is not necessary to include a time effect in the regression. The dummy variable of the sectors was significant and different per sector for the majority of the regression. This supports the hypothesis that the effect of ODA is different per sector.

Data cleaning impact

(29)

significant anymore. This clearly shows the necessity of cleaning the data in the way I did. I have not included the regression tables of stream B and C but the results are similar.

Conclusion

The starting point of this research was the ambiguous effect of development aid on growth at the macroeconomic level found in many papers. From there I developed the research question, which looks at the effect of development aid at a sectoral level. Interesting to find was that development aid had a positive and significant effect at the macroeconomic level, but an insignificant effect if we looked at the sectoral data. So from an empirical point of view I need to negatively answer the research question. Development aid did not show a significant effect on growth at the sectoral level. A few reasons could cause this: the data is limited and only covers 9 years; the control variables were kept at the national level instead of at the sectoral level and not all ODA could be attributed to a sector. This caused that when we looked at the sectoral level 40% of the ODA was not allocated and hence it could not be meassured. The positive effect that I found is however hopeful and might be the beginning of a turn around on the view of development aid. The difference that might have caused that I found this positive effect is that I, contrary to some other researches, applied a minimum level of ODA and did not look at commitments but at disbursements by donors.

(30)

Because of the autocorrelation and the heteroskedasticity in the majority of the data it could be useful to use a statistical model which uses instrumental variables. This could also be a solution for the endogeneity issue, because the question remains if ODA is send to countries which have good growth numbers or if it is send to countries because of the low growth numbers? In this research I applied a lag of one year to tackle this endogeneity issue. However, this is far from perfect, so for future research it could be interesting to use instrumental variables and to use an estimation technique that incorporates this. One of them is the Generalized Method of Moments (GMM) (Bun and Kleibergen, 2010), which accounts for a short time span of a large cross section. The GMM technique provides asymptotically efficient inference assuming a minimal set of statistical assumptions (Bun and Kleibergen, 2010). The GMM can be used to obtain robust panel standard errors with data that suffers from heteroskedasticity, autocorrelation and outliers. This could be a topic for future research, to find instrument variables so the GMM could be applied.

A few of the limitations in this research are that some of the development aid is not attributable to a specific sector. This means that when I started to look at the effect on the sectoral level, 40% of the ODA was lost. Additionally, the ODA which was allocable to a specific sector might not be spent in that particular sector. Many of the countries which receive ODA have low levels of institutional development, making it more difficult to check if the ODA is spent in the sector to which it is allocated. Also the fact that I did not use control variables at the sectoral level could have influenced the estimations on the sector level. The time span used in this research can be seen as short, I have looked at the disbursement of ODA which is only available for nine years. This limits the data and only makes it possible to look at the effect over the last 9 years. It could be interesting for future research to also look at commitments of ODA. This could be interesting for two reasons one is to see if there is a significant difference between disbursements and commitments and the second is to see how the sectors are affected over a longer time span.

(31)

Literature

Adam C., D. Bevan (2003), Aid, public expenditure and Dutch disease Working Paper, vol. 2003-02Centre for the Study of African Economies, Oxford University February

Agenor, P., N. Bayraktar and K. El Aynaoui (2008), .Roads out of poverty? Assessing the links between aid, public investment, growth, and poverty reduction..Journal of Development

Economics 86, 277-295.

Alfaro, L. (2003), “Foreign direct investment and growth: does the sector matter?”. In mimeo, Harvard Business School

Arellano,M. and S. Bond. (1991). “Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations.” Review of Economic Studies, 58, 277-297.

Arellano, M. and O. Bover. (1995). “Another look at the instrumental variable estimation of error-components models.” Journal of Econometrics, 68, 29-52.

Asiedu, E. and B. Nandwa. (2007), “On the Impact of Foreign Aid in Education and Growth: How Relevant Is the Heterogeneity of Aid Flows and the Heterogeneity of Aid Recipients? Review of World Economics 143(4):631–49.

Barro, R. (1996), “Democracy and growth.” Journal of Economic Growth 1, 1-27.

Blomstrom, M., R. Lipsey, and M. Zejan, (1996)"Is Fixed Investment the Key to Economic Growth," Quarterly Journal of Economics 111, 269–276.

Blundell, R. and S. Bond. (1998). “Initial conditions and moment restrictions in dynamic panel data models.” Journal of Econometrics, 87(1), 115-143.

Blundell, R. and S. Bond. (2000). “GMM estimation with persistent panel data: an application to production functions.” Econometric Reviews, 19(3), 321-340.

Blundell, R., S. Bond and F. Windmeijer. (2000). “Estimation in dynamic panel data models: improving on the performance of the standard GMM estimator.” In B. Baltagi (ed.),

Nonstationary Panels, Panel Cointegration and Dynamic Panels, Elsevier Science.

Bulíř A., J. Hamann (2003), Aid volatility: An empirical assessment IMF Staff Papers, 50 (1) pp. 64–89

(32)

Bun, M.J. and F. Kleibergen(2010), Identification pathologies and their effects on GMM test statistics for dynamic panel data models.

Burnside C, Dollar D. (2000), Aid, policies and growth. American Economic Review 90(4): 847-868.

Cassen, R. and associates. (1994), Does Aid Work? Oxford: Oxford University Press.

Clemens M. A., S. Radelet, and R. Bhavnani (2004). Counting Chickens When They Hatch: The Short-Term Effect of Aid on Growth. Center for Global Development Working Paper 44.

Washington, D.C.

Collier P, Dollar D. (2002), Aid allocation and poverty reduction. European Economic Review 46(8):1475–1500.

De Renzio, P. (2005), ‘Increased Aid vs Absorptive Capacity: Challenges and Opportunities Towards 2015’, IDS Bulletin 36 (3): September.

Doucouliagos H., M. Paldam (2009), The aid effectiveness literature. The sad results of 40 years of research. Journal of Economic Surveys. Vol. 23, No. 3, pp. 433–461

Gemmell, N., and M. McGillivray, (1998), "Aid and Tax Instability and the Government Budget Constraints in Developing Countries," CREDIT Working Paper No. 98/1 (Nottingham, England: University of Nottingham Center for Research in Economic Development and International Trade).

Hadjimichael, M.T., Ghura, D., Mu¨hleisen, M., Nord, R., Ucer, E.M. (1995), Sub- Saharan Africa: growth, savings, and investment, 1986–93. IMF Occasional Paper, No. 118

Hansen H, Tarp F. (1999). The Effectiveness of Foreign Aid. University of Copenhagen: Development Economics Research Group, Institute of Economics.

Hansen H, Tarp F. (2000), Aid effectiveness disputed. Journal of International Development 12(3): 375–398.

Hansen H, Tarp F. (2001), Aid and growth regressions. Journal of Development Economics 64(2): 547– 570.

Islam, M. (2005). Regime Changes, Economic Policies and the Effect of Aid on Growth. Journal

of Development Studies, 41(8), pp. 1467–1492.

Jensen, P. S., & Paldam, M. (2006). Can the two new aid-growth models be replicated? Public

(33)

Judson, R. and Owen, A. 1999. Estimating dynamic panel data models: a guide for macroeconomists.Economics Letters 65, 9-15.

Lensink, R., and O. Morrissey (2000), "Aid Instability as a Measure of Uncertainty and the Positive Impact of Aid on Growth," Journal of Development Studies, Vol. 36, No. 3, pp. 31-49.

Lensink R. and H. White (2001), Are There Negative Returns to Aid?, Journal of Development

Studies, 37:6, 42-65

Li, X. and X. Liu (2005), ‘Foreign Direct Investment and Economic Growth: An Increasingly Endogenous Relationship’, World Development, 33, 3, 393–407.

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.

Minoiu, Camelia, and Sanjay G. Reddy. (2010), ‘‘Development Aid and Economic Growth: A Positive Long-Run Relation.’’ The Quarterly Review of Economics and Finance 50 (2010) 27– 39

Mosley P. (1987), Overseas Development Aid: Its Defence and Reform. Wheatsheaf: Brighton

Nkusu, M., (2004), Aid and the Dutch disease in low-income countries Working Paper, vol. 04/49International Monetary Fund

Rajan, R., and A. Subramanian (2005), “Aid and Growth: What Does the Cross- Country Evidence Really Show?” IMF Working Paper WP/05/127. International Monetary Fund, Washington, D.C.

Rajan, R. and A. Subramanian (2011), "Aid, Dutch Disease, and manufacturing growth." Journal

of Development Economics 94, 106-118.

Ram R. (2004), Recipient country’s ‘policies’ and the effect of foreign aid on economic growth in developing countries: additional evidence. Journal of International Development 16(2): 201– 211.

Torvik R. (2001), Learning by doing and the Dutch disease European Economic Review, 45 pp. 285–306 (February)

White H. (1992), The macroeconomic impact of development aid: a critical survey. Journal of

Development Studies 28(2): 163–240.

(34)

Appendix 1 country list

Country or area

Afghanistan Djibouti Liberia Serbia

Albania Dominica Libyan Arab Jamahiriya Seychelles

Algeria Dominican Republic Madagascar Sierra Leone

Angola Ecuador Malawi Solomon Islands

Antigua and Barbuda Egypt Malaysia Somalia

Argentina El Salvador Maldives South Africa

Armenia Equatorial Guinea Mali Sri Lanka

Azerbaijan Eritrea Marshall Islands Sudan

Bahrain Ethiopia Mauritania Suriname

Bangladesh Fiji Mauritius Swaziland

Barbados Gabon Mexico Syrian Arab Republic

Belarus Gambia

Micronesia (Federated States

of) Tajikistan

Belize Georgia Moldova Tanzania

Benin Ghana Mongolia Thailand

Bhutan Grenada Montenegro Timor-Leste

Bolivia Guatemala Morocco Togo

Bosnia and Herzegovina Guinea Mozambique Tonga

Botswana Guinea-Bissau Myanmar Trinidad and Tobago

Brazil Guyana Namibia Tunisia

Burkina Faso Haiti Nepal Turkey

Burundi Honduras Nicaragua Turkmenistan

Cambodia India Niger Turks and Caicos Islands

Cameroon Indonesia Nigeria Tuvalu

Cape Verde Iran, Islamic Republic of Oman Uganda

Central African Republic Iraq Pakistan Ukraine

Chad Jamaica Palau Uruguay

Chile Jordan Panama Uzbekistan

China, People's Republic

of Kazakhstan Papua New Guinea Vanuatu

Colombia Kenya Paraguay Venezuela (Bolivarian Republic of)

Comoros Kiribati Peru Viet Nam

Congo, Dem, Rep, Korea, Rep, Philippines Yemen

Congo Kosovo Rwanda Zambia

Costa Rica Kyrgyzstan Samoa Zimbabwe

Côte d'Ivoire

Lao People's Democratic

Republic Sao Tome and Principe

Croatia Lebanon Saudi Arabia

(35)

Appendix 2 Check of export dummy variable

In the table of the export dummy a simple pooled OLS model regression has been done for each of the individual sectors. All the regressions are controlled for autocorrelation and heteroskedasticity.

Export Dummy Variable sector* Isic A+ B Agriculture hunting forestry fishing Isic C+E Mining, utilities Isic D Manufac- turing Isic I Transport storage communication Isic J-P Service sector Aid variable

Total ODA sector / value added sector -0.055 0.002 -0.011 0.002 0.131

(0.420) (0.902) (0.851) (0.973) (0.044) Control Variables Rule of law -0.008 -0.011 -0.047** -0.004 -0.017* (0.188) (0.276) (0.040) (0.719) (0.093) Fixed investments % of GDP -0.002 0.108* 0.261*** 0.046 0.104*** (0.939) (0.065) (0.003) (0.412) (0.003) Inflation -0.003 -0.022 -0.025 -0.010 0.063 (0.976) (0.716) (0.798) (0.849) (0.371)

Log GDP per capita -0.009** -0.015** 0.003*** -0.011* 0.007

(0.028) (0.012) (0.819) (0.051) (0.239) Constant 0.093*** 0.128*** -0.061 0.149*** -0.047 (0.004) (0.007) (0.427) (0.000) (0.405) Number of observations 307 495 74 359 682 R squared 0.031 0.047 0.074 0.016 0.054 F statistic 3.430 2.390 3.070 1.120 2.940 Prob>F 0.008 0.044 0.032 0.358 0.016 VIF mean 1.240 1.140 1.470 1.180 1.210

ovtest Ho: model has no omitted variables 0.449 0.000 0.523 0.538 0.000

*sector 4 has too few observation with the restrictions that have been placed on the data

(36)

To tal OD A/GD P (To ta l O D A/GDP )*Exp E xp _(Tota l O D A/GDP )* D u mm y ex p Du m my E xp (To ta l O D A/GDP )^ 2 C o ns tant R -squ are N u m be r of o bs Stream A GM 0.022* 0.026 0.4022 601 DD1 0.002 0.027 0.008 0.040 0.463 535 MD 0.028 -0.005 0.4022 601 TM 0.084 -0.007 0.008 -0.098 0.013 0.4642 535 Stream B GM -0.015 0.195*** 0.1201 3603 DD1 0.140* -0.036 0.019* 0.088** 0.1451 3201 DD2 -0.034 0.042 -0.011** 0.193*** 0.1211 3588 MD 0.088 -0.090** 0.164*** 0.1238 3603 TM 0.390** -0.173 0.020* -0.056 -0.051*** -0.246** 0.140*** 0.1462 3186 Stream Ca GM -0.012 0.169 0.1527 2121 DD1 -0.048* 0.127 0.016 0.132 0.22 1879 DD2 -0.115 0.104 -0.012 0.206* 0.1539 2102 MD 0.026 -0.014 0.160 0.1549 2121 TM 0.004 0.106 0.016 0.011 -0.034** -0.031 0.179 0.224 1860

∗, ∗∗, ∗∗∗ denote significance at the 10, 5, and 1 percent level,

a _{total oda/gdp is at sector level for stream C (Total ODA sector / value added sector)}

(37)

autocorrelation and heteroskedasticity. Stream A not significant or omitted.

GM all years, DD 2003 2004, MM only 2009, TM 2003 2004 Stream B Not significant or omitted of sector or year

GM Sector 3, Year 2004 2005 2007 2008, DD1 Year 2003 2008 2010, DD2 Sector 3, year 2003 2005 2006 2008 2010, MM Sector 3, year 2004 2005 2008, TM Sector 4 and 5, year 2003 2004 2005 2006 2007 Stream C Not significant or omitted of sector or year

(38)

Appendix 4 Data cleaning impact

Stream A GM GM* DD DD* MM MM* Aid variable Total ODA/GDP 0.039*** 0.036* 0.079 0.066 0.083*** 0.088 (-0.003) (0.050) (-0.173) (0.245) (0.000) (0.244) (Total ODA/GDP)^2 -0.048*** -0.134 (-0.001) (0.420) Exp 0.014 0.028 (-0.331) (0.025) Exp* (ODA/GDP) -0.132 -0.098 (-0.456) (0.529) Control Variables Rule of law 0.004 -0.005 0.005 -0.005 0.003 -0.006 (-0.325) (0.283) (-0.232) (0.253) (-0.345) (0.249) Fixed investments % of GDP 0.091*** 0.097*** 0.108*** 0.101*** 0.094*** 0.096*** (0.000) (0.001) (0.000) (0.002) (0.000) (0.001) Inflation 0.062*** 0.036 0.059*** 0.039 0.062*** 0.035 (0.000) (0.135) (-0.002) (0.123) (0.000) (0.143) Log GDP per capita -0.006** -0.002 -0.007*** -0.003 -0.005** -0.001 (-0.010) (0.529) (-0.004) (0.254) (-0.028) (0.826) Constant 0.067*** 0.037 0.068*** 0.035 0.057*** 0.028 (0.000) (0.101) (0.000) (0.128) (0.000) (0.277) Number of observations 601 943 550 869 601 943 R squared 0.088 0.053 0.105 0.069 0.093 0.053 F statistic 8.590 4.580 5.940 3.770 7.780 3.890 Prob>F 0.000 0.001 0.000 0.001 0.000 0.001

Note P-values in parentheses. *,**,*** denote significance at the 10, 5, and 1 percent level,respectively.

Master Thesis International Economics and Business

International Economics and Business

Development aid from a sectoral perspective

Supervisor: dr. Robert Inklaar

Co-assessor: dr. Ger Lanjouw

Date: Wednesday 16-07-2012

Student: Remco Hoogma

Student number: 1554980

Table of Contents

Introduction

Literature review

Theory and model

Methodology

Data

Empirical analysis

Robustness

Conclusion

Literature

Appendix 1 country list

Appendix 2 Check of export dummy variable

Appendix 4 Data cleaning impact