The effect of a policy-aid interaction
term on the growth of a country
Florian van der Peet 10284966 University of Amsterdam Bachelor thesis: first version Supervisor: Rob van Hemert Instructor: Maurice Bun December 24, 2014
2
Table of contents
1. Introduction ... 3
2. Background literature... 5
2.1 Development aid ... 5
2.2 The policy index ... 6
2.3 Previous studies on a policy-aid interaction term ... 8
3. Model ... 13
4. Results and analyses ... 15
4.1 New instruments ... 15
4.2 Ordinary least squares ... 18
4.3 New policy index ... 19
5. Conclusion ... 24 References ... 25 Appendix I ... 26 Appendix II ... 27 Appendix III ... 28 Appendix IV ... 29 Appendix V ... 31
3
1. Introduction
The development aid to Third World countries climbed to a new record in 2013. This new record contained a total amount of 134.8 billion dollars1. However the effect of development aid on the growth of a country is still controversial even after a lot of studies. Doucouliagos & Paldam (2006) analyzed 68 papers about the effect of development aid on growth. Their results indicate, based on a meta-analysis, that the effect of development aid on growth is not significantly larger than zero. However their results also show that a majority of authors agree that development aid has a small positive effect on growth.
In contrast to the conclusion of Doucouliagos & Paldam (2006), do the results of Collier and Dollar (2002) indicate that since the mid 1990’s more and more researchers conclude that development aid has got a positive effect on the growth of a country. However the magnitude of this effect differs per country. The study of Dalgaard, Hansen & Tarp (2004) states that it is the key to find what causes these differences between countries. The study of Burnside & Dollar (2000) finds that the effectiveness of development aid depends on the policy of a country. Their study mentions that a corrupt, incompetent government is not going to use aid wisely.
However the study of Rajan & Subramanian (2008) concludes that the effect of a policy-aid interaction term is never significant. However their results show relative large standard errors. Their study exploited an IV-regression to estimate the coefficient of a policy-aid interaction term. However it is hard to find relevant instruments in the growth literature, since there is a lot of correlation. Therefore did the study of Rajan & Subramanian (2008) construct a new instrument, called fitted aid. Fitted aid is based on why countries receive development aid. However the validity of this instrument seems questionable and is reviewed in this study.
A solution is given by the study of Bun & Harrison (2014) in the form of OLS. Their study shows that even in the case of endogeneity an OLS-estimator is consistent and asymptotically normal distributed for the coefficient of the interaction term. Additionally they show that the further conditions regarding higher order dependencies in the data, OLS interference on the coefficient of the interaction term is warranted. Additionally is the validity of the OLS-method on the interaction term between development aid and policy tested with a Hausman test.
1
http://www.oecd.org/newsroom/aid-to-developing-countries-rebounds-in-2013-to-reach-an-all-time-high.htm
4
The study of Rajan & Subramanian (2008) uses a measure of openness designed by Sachs & Werner (1995) as a policy index. However it is questionable if this openness indicator effects the growth of a country, according to the study of Rodriquez & Rodrik (2001). Therefor I have constructed a different policy index. This policy index is an augmentation of the policy index used in the study of Burnside & Dollar (2000) and is designed in the study of Easterly (2003). Both the IV-regression from the study of Rajan & Subramanian (2008) and the OLS-regression, as mentioned before, will be redone with this new policy index instead of the measure for openness. I expect that the new policy index reflects the policy of a country better compared to the openness indicator.
The data used in this study is from the same two databases used in the study of Rajan & Subramanian (2008)2. The first datasets contains information about the relations between 22 donor countries and 181 development countries over the period 1960-2000. The second database contains information on 74 development countries3, for example how much development aid the countries received4.
The structure of the paper is as follows. The next section examines the theoretical effect of development aid and policies set by a government on growth. Section 3 specifies the interaction models that are used for this analysis. Section 4 contains the results of the analyses and tests the validity of the OLS-estimator with a Hausman test, followed by a conclusion.
2
The dataset is kindly given by Rajan & Subramanian
3
A list of the countries in the two databases is in appendix V
5
2. Background literature
Since the 1990’s development aid and growth studies started focusing on the relations between aid, policies and growth, according to the study of Hansen & Tarp (2001). This study will also focus on the relation between an aid-policy interaction term and growth in GDP per capita. The first paragraph of this section explains how development aid is used. The second paragraph examines the effect of the policy of a country on growth. Finally different methods to estimate models with heterogeneity are compared and there is a short analysis of the study of Rajan & Subramanian (2008).
2.1 Development aid
Development aid started shortly after the second world-war and has become a great cash flow from rich countries to poor countries. The purpose of this cash flow is to reduce poverty and inequality and to increase growth and building capacity5. Aid effectiveness measures how well development aid has reached its purpose. In this study I will only analyze the effect of development aid on the growth in GDP per capita. An advantage of the GDP is available for (almost) every country in the world. The development aid regressor is used as a percentage of development aid of the GDP:
%𝑑𝑒𝑣𝑒𝑙𝑜𝑝𝑚𝑒𝑛𝑡 𝑎𝑖𝑑 = 𝑑𝑒𝑣𝑒𝑙𝑜𝑝𝑚𝑒𝑛𝑡 𝑎𝑖𝑑 𝐺𝐷𝑃
The advantage of using %𝑑𝑒𝑣𝑒𝑙𝑜𝑝𝑚𝑒𝑛𝑡 𝑎𝑖𝑑 is that countries with different GDP’s are better comparable.
The OECD keeps track of the amount development aid receipt by development countries. The criteria set by the OECD to be development aid are:6
i. Provided by official agencies, including state and local governments, or by their executive agencies; and
ii. Each transaction of which:
a) Is administered with the promotion of the economic development and welfare of developing countries as its main objective; and
b) Is concessional in character and conveys a grant element of at least 25 per cent (calculated at a rate of discount of 10 per cent).”
The three most important cash flow that are included in this definition are the following. Firstly, bilateral aid. This is the aid that is given by a government directly to the government
5
http://data.worldbank.org/topic/aid-effectiveness
6
of the development country. Additionally multilateral aid is included; this is aid that comes from private organisations, who received their money from governments. Finally non-government aid is included, this is aid given by charity organisations like Unicef.
The three most important cash flow that are not included in this definition are the following. Firstly investment cash flows. Secondly military aid is excluded, which is used to defend a country. Finally and most importantly the money that is sent from migrant workers to their family that is living in their home land. In 2008 foreign workers sent over twice as much money to family compared with the official aid flows from OECD members, according to the study of Ratha, Mohapatra & Silwa (2009). However not all of this money goes to poor development countries, for example Poland and Mexico also receive a lot of money.
This study analyses the effect of an aid-policy interaction term between 1960 and 2000. Unfortunately the three cash flows mentioned above are not available for this time series. The most reliable and consistent data that is available is from the OECD. This study uses therefore the same definition for development aid as the OECD does.
However he study of Hansen & Tarp (2001) makes a difference between development aid given by governments and development aid given by private investors. I assume, however, that both the government and private investors have the same purpose, which is making the development aid effectiveness and letting the recipient country grow.
2.2 The policy index
The way a country is ruled, effects the growth of that country. The study of Tavares & Wacziarg (2001) gives the following reasons for this effect.
First of all there is political stability. Political instability leads to uncertainty about the future state of properties and policies for companies and laborers. A democracy for example stimulates transparency. Laborers and companies are aware of the plans of the government and the government tries to do what is best for the country. Additionally is the population stimulated to join an open discussion about what is best for the country. However the downside of a democracy is that every election can lead to a completely different policy, giving it the possibility to be more political instable compared to a dictatorship. A side note is that foreign investors prefer political stable countries, which possibly causes correlation between development aid and the policy index. However in this study, we assume that the policy index is exogenous.
There is also a causal link between the quality of a government and the growth of a country. An autocratic government will tend to set up a policy, where only a small group
7
benefits and not the entire population. A democracy with a good competition system, that is enough different representatives, will tend to set up a policy for the entire population; otherwise the government will not be reelected. For example democracies tend to set inequality reducing policies. This leads to a more fair distribution of national income between capital and labor. In addition is education stimulated by a democratic. The effect of education on growth is significant larger than zero, according to the results of Makiw, Romer & Weil (1992).
Finally the size of the government matters. Theoretically there are reasons for a positive and a negative effect. Negatively, a big government costs a lot of money. Positively, a big government is able to harmonize in conflicts and secure a socially optimal direction for growth and development. The results of Ram (1986) conclude that the size of the government positively effects the economic growth of a country.
The study of Burnside and Dollar (2003) states that including a dummy for democracy is not enough to analyze the effect of a good policy. Because a democracy may mean many things and not all democracies work on the same way. For example can a democracy be affected by corruption, or can there be a lack of representatives. As a solution for this problem the study Burnside and Dollar (2003) suggests to use a policy index, which measures how good the policy of a country is. There are multiple methods to construct a policy index. In this study we will use and discuss two different approaches to measure the policy of a country.
Firstly an openness index designed by Sachs & Werner (1995) and updated by Wacziarg and Welch (2003). In this index ‘1’ represents an open economy and ‘0 represents a closed economy. This index is based on the following five dummy criteria:
1. Average tariff rates of 40% of more (TAR)
2. Nontariff barriers covering 40% or more of trade (NTB)
3. A black market exchange rate that is depreciated by 20% or more relative to the official exchange rate, on average, during the 1970s or 1980s (BMP)
4. A state monopoly on major exports (XMB)
5. A socialist economic system (as defined by Kornai, 1992) (SOC)
However there is some criticism on this openness index. Firstly, in 1970-89 there are 31 open economies and 74 closed economies and in the 1990s there are 78 open economies and 27 closed economies, which is a huge contrast. According to the study of Wacziarg and Welch (2003), the openness index can no longer effectively partition fast growing from slow growing countries. In addition does the study of Rodriquez & Rodrik (2001) conclude that there is little evidence between the dummy criteria used to construct the openness index and
8
the economic growth of a country. Furthermore they mention that a closed economy may mean many different things.
Secondly, the study of Easterly (2003) gives a different way to calculate the policy index of a country. Their policy index is based on five variables. Firstly the black market premium is included to measure the openness of a country. Also financial depth is included with the ratio of M2 to GDP. In addition is the ratio of trade to GDP included to measure the integration with global trade. Finally the budget surplus and the inflation are included.
Two steps are needed to calculate the policy index. Firstly the weights of the variables in the policy index are determined by the following regression. This model is designed in the study of Easterly (2003).
𝑦𝑖 = 𝛽1+ 𝛽2𝐵𝑀𝑃𝑖+ 𝛽3𝑇𝑅𝐴𝐷𝐸𝑖 + 𝛽4𝐼𝑁𝐹𝑖 + 𝛽5𝐵𝑆𝑖 + 𝛽6𝑀2𝑖 + 𝑢𝑖 (2.1) The variable to explain 𝑦𝑖 is the real growth in GDP per capita. 𝐵𝑀𝑃𝑖 Represents the black
market premium and the ratio trade of GDP is given by trade. In addition is the inflation given by INF and the budget surplus by BS. Finally M2 reflect the ratio M2 of GDP.
Secondly, the coefficients obtained from estimation (2.1) are used to calculate the policy index with equation (2.2).
𝑝𝑜𝑙𝑖𝑐𝑦 𝑖𝑛𝑑𝑒𝑥𝑖 = 𝛽1+ 𝛽2𝐵𝑀𝑃𝑖 + 𝛽3𝑇𝑅𝐴𝐷𝐸𝑖 + 𝛽4𝐼𝑁𝐹𝑖+ 𝛽5𝐵𝑆𝑖+ 𝛽6𝑀2𝑖 (2.2) A high policy index score reflects a good policy and a low or even negative policy index score represents a bad policy. A disadvantage of this method is that the different variables are forced into a prespecified relationship, according to the study of Rajan & Subramanian (2008).
2.3 Previous studies on a policy-aid interaction term
The growth literature gives many possibilities to estimate models with parameter heterogeneity. In this study we will discuss three different approaches to estimate data with heterogeneity.
Firstly, a quantile regression as used in the study of Canarella & Pollard (2004). Their study created groups based on the dependent variable. A quantile regression is a statistical technique designed to estimate and conduct inferences about condition quantile functions, according to the study of Canarella & Pollard (2004). Their study also mentions that when data are heteroskedastic, the set of slope coefficients of the quantile regressions will differ from each other as well as from the IV slope parameters. However the study of Buchinsky (1998) observes that these differences can also be caused as a result of systematic differences between the observations.
9
Secondly, the study of Durlauf & Johnson (1995) exploited a cluster method. This method creates groups based on a regressor. Their study divided countries into different groups based on the output and the literacy rate of a country. The main problem with dividing into groups is that it is unknown how many breaks there are and where these breaks are. The study of Durlauf & Johnson (1995) used an algorithm that minimizes the residual sum of squares to divide the groups.
A different approach is given by the study of Burnside & Dollar (2000). Their study exploits an interaction term between development aid and a policy index. By adding this interaction term, the model allows interaction between development aid and the policy index.However this method only solves the heterogeneity of one variable.
A problem with the quantile regression method and the cluster method, is that countries are divided into self-created groups. In the case of this study, how high does the policy index have to be in order for a country to have a good policy? How much growth must a country have to be part of the top quantile? Therefor this study chooses for the interaction term method. An advantage of this method, is that the policy of a country is not divided into self-created groups.
The study of Rajan and Subramanian (2008) also exploits the interaction term method. Their study does not find a significant relationship between an aid-policy interaction term and the growth of a country. The openness indicator designed in the study of Sachs & Werner (1995) is used as a policy index in the study of Rajan and Subramanian (2008). Their study analyzes the effect of an interaction term between this openness index and development aid with an IV-estimator. The instruments that are used are fitted aid and fitted aid * openness. This is consistent with the results of the study of Bun & Harrison (2014), since their study concludes that the instruments need to be interacted with the exogenous part of the interaction to achieve identification7.
Fitted aid is based on the assumption that donor countries are more likely to give aid to a recipient country if the donor country expects that their money buys influence over the recipient country8. The instrument fitted aid can be considered as a strong instrument, since it has a correlation coefficient of 0.70 with aid itself. Fitted aid is constructed based on thirteen
7 For more details and proof see Bun & Harrison (2014) 8
An example from Rajan & Subramanian (2008) to help to understand the idea behind fitted aid: the United Kingdom should be willing to give more aid per capita to Uganda than to India; but it will be more willing to give aid to Uganda than to a similar-sized country in Africa, say Senegal, because France is likely to have a significant aid presence in the latter, thus diluting any possibility of British influence.
10
variables9. Model (2.3) is estimated, using the thirteen variables as regressors and the percentage development aid of the GDP as variable to explain. Additionally, the subtext d does refer to donor countries and the subtext r to recipient countries. Finally equation (2.4) is calculated, fa is in this formula fitted aid.
𝑑𝑎𝑑,𝑟 = 𝛽1′𝑌
𝑑,𝑟+ 𝑣𝑑,𝑟 (2.3)
𝑓𝑎𝑑,𝑟 = ∑ 𝛽1′𝑌 𝑑,𝑟
𝑑 (2.4)
This method has its disadvantages and raises questions regarding its actual implementation. Firstly, the exogeneity of fitted aid cannot be tested with a Sargan test, because the model is exactly identified. Secondly, the 13 variables witch were used to create fitted aid cannot be tested on multicollinearity or exogeneity, since they are not used in the IV-regression.
It is also possible to use an OLS-estimator according to the study of Bun & Harrison (2014). Their study has shown that an OLS-estimator of the coefficient of an endogenous interaction term can be consistent and asymptotically normally distributed10. However this only applies for the coefficient of the interaction term, thus the coefficient of the regressors cannot be analyzed with this method, however in this study I only want to analyze the effect of a policy-aid interaction term. A Hausman test will be used to test if the OLS-estimator is indeed consistent and asymptotically normally distributed.
The data used in this study is from the study of Rajan & Subramanian (2008). Their data consists of two databases. From here an on referred to as database A and database B. Both databases contain information for four periods. These are 1960-2000, 1970-2000, 1980-2000 and 1990-1980-2000. Database A was used in the study of Rajan & Subramanian to construct the instrument fitted aid. Database A contains information on the relation between donor countries and receiving countries, for example colonial history, common language and difference in population size11. There are 22 donor countries and 181 developing countries for every period and thus a total of 3842 observations.
Database B contains characteristics of group of recipient countries. The most important characteristics that are included in the database are real growth GDP per capita, percentage development aid of GDP, fitted aid and the measure for openness12. Row 1 of Table 1 shows the number of development countries for every period.
9 List of variables used to construct fitted aid in appendix IV 10
Theoretical proof for the use of this OLS-estimator with the presence of an endogenous interaction term is given in the study by Bun & Harrison (2014).
11
Full list of variables used in database A is in appendix IV
11
Database A and database B have been merged into database C to augment the IV-regression of Rajan & Subramanian with the thirteen variables that have been used to construct fitted aid. A downside of this merger is the loss of data. Database A contains data on 181 recipient countries and database B 70-78 recipient countries. Which means that more than 100 observations from database A are lost. Table 1 shows the number of development countries that are left in database C. The number of recipient countries is given in row 1 and the number of donor countries in row 2.
Table 1. – Number of countries in database C
Period 1960-2000 1970-2000 1980-2000 1990-2000 (1) Number of recipientcountries 74 78 75 70 (2) Number of donor countries 22 22 22 22 Notes: In database C are observations from every donor country to every development country. The total number of observations is there for equal to the product of the two.
In database C every donor country is listed 22 times. This means that every 22 observations contain the exact same information for all the variables from database B. Thus for example the growth rate of real GDP per capita: 𝑦1 = 𝑦2 = ⋯ = 𝑦22≠ 𝑦23.
However the data from database A differs per observation. For example the development aid is now unique for every observation, since every donor country gives a different amount of aid to each recipient country: 𝑑𝑎1 ≠ 𝑑𝑎2 ≠ ⋯ ≠ 𝑑𝑎22.
Table 2. – Colonial dummy variables
Variable Dummy colony United Kingdom Dummy colony France Dummy colony Spain Dummy colony Portugal Sum in database A 82 54 44 35 Sum in database C 24 16 14 1
Notes: The variables are the number of countries in the database that were a colony of the named country. As a consequence of the merger between the database A and database B, a lot of countries that were a colony are removed from the database.
A greater problem is in the dummy for common colony of Portugal. Table 2 shows the number of countries that has been a colony of the United Kingdom, France, Spain or Portugal in database A and database C. The loss of observations as a consequence of the merging of database A and database B caused fewer countries in each dummy category. Database A contains 35 development countries which were a colony of Portugal, while in database C there is only one country left that was a colony of Portugal, which is Brazil. As a consequence of this there are multiple variables used as instrument who are zero for every country except Brazil. Thus the effect of these variables cannot be estimated. Therefor the following
12
variables are excluded from the regression: dummy for common colony of Portugal, Dummy for common colony of Portugal * measure for openness and log of difference in population * dummy for common colony of Portugal * measure for openness. The 10 remaining instruments will from here an on be referred to as instrument set A13.
The budget surplus, ratio trade of the GDP and the black market premium are not included in the dataset of the study of Rajan & Subramanian (2008). This data are added to their dataset and comes from the world development indicators measured by the World Bank14. However this data are not available for every country. Data are especially missing in the period 1960-1970. Therefore the period 1960-2000 will not be estimated with the new policy index. For the other three periods are respectively 14, 7 and 4 countries less, compared to the original database B.
13
List of variables in instrument set A in appendix IV
13
3. Model
The regression from the study of Rajan & Subramanian (2008) will be augmented on three different ways. Firstly, the IV-regression from the study of Rajan & Subramanian (2008) will be redone using instrument set A instead of fitted aid as instrument
𝑦𝑖 = 𝛽1+ 𝛽2𝑝𝑖+ 𝛽3𝑑𝑎𝑖 + 𝛽4𝑝𝑖𝑑𝑎𝑖 + 𝑑′𝑤𝑖+ 𝑢𝑖. (3.1)
The dependent variable 𝑦𝑖 is the real growth in GDP per capita. In this model 𝑝𝑖 represents the policy index, while the percentage development aid is given by 𝑑𝑎𝑖.
Additionally we include the interaction term between these two variables. Finally a vector 𝑤𝑖
of control variables is included15. The instruments also need to be interacted with the exogenous part of the interaction term, as mentioned before. Thus there are another ten instrument used.
Secondly, since the use of the openness measure designed by Sachs & Werner (1995) is questionable, we will also construct a policy index. This policy index will be constructed using the method of Easterly (2003) as discussed before. Model (3.1) will be estimated with an IV-estimator with the new policy index based on the method of the study of Easterly (2003). The instruments that will be used for this regression are fitted aid and fitted aid * policy index.
Third, model (3.1) will be estimated with an OLS-estimator with either the policy index of Sachs & Werner (1995) and the newly created policy index based on the model from the study of Easterly (2003). As a consequence of using an OLS-estimator instead of an IV-estimator, we expect a reduction of the standard error of the interaction term. This reduction may be enough for the policy-aid interaction term to become significant.
The OLS-estimator of the coefficient of the interaction term needs to comply with two conditions in order to be consistent and asymptotically normally distributed, according to the study of Bun & Harrison (2014). Firstly, the coefficient of the interaction term obtained with an OLS-estimator is only consistent if
𝐸(𝑑𝑎𝑖 ∗ 𝑝𝑖𝑖) ∗ 𝐸(𝑑𝑎𝑖 ∗ 𝑝𝑖𝑖2) − 𝐸(𝑝𝑖2) ∗ 𝐸(𝑑𝑎𝑖2∗ 𝑝𝑖) = 0.
Which implies that: 𝐸(𝑑𝑎𝑖∗ 𝑝𝑖𝑖2) = 𝐸(𝑑𝑎𝑖2∗ 𝑝𝑖) = 0. Since 𝐸(𝑑𝑎𝑖∗ 𝑝𝑖𝑖) is the covariance between the policy index and development aid and 𝐸(𝑝𝑖2) is the variance of the policy index. Secondly, the variance of the policy index plus the variance of development aid needs to be normally distributed in order for the coefficient of the interaction term to be asymptotically normally distributed.
14
To test these conditions, we will the same Hausman test as the study of Bun & Harrison (2014). With this Hausman test the coefficient of the interaction term that is obtained using an IV-estimator will be compared with the same coefficient that is obtained using an OLS-estimator based on the variance. The IV-residuals are used to estimate the variance of the OLS-estimator16. The Hausman test is as follows17:
𝐻𝐼𝑉,𝑂𝐿𝑆 = 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝐼𝑉 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆
𝑉̂𝐻
, Finally, the Hausman test will also be used to test the assumption that development aid is endogenous. Instead of testing one coefficient of the estimators with the Hausman test, now the entire vector of coefficients will be tested. This Hausman statistic becomes:
𝐻𝐼𝑉,𝑂𝐿𝑆 = (𝛽̂𝑝𝑖𝑑𝑎𝑖,𝐼𝑉 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆 )′ 𝑉̂𝐻−1(𝛽̂
𝑝𝑖𝑑𝑎𝑖,𝐼𝑉 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆 ) ,
16
Bun & Harrison (2014)
15
4. Results and analyses
In this section we report the results from the empirical analysis. The first paragraph analyzes the results exploiting the instruments separately, thus not in one index. The second section analyzes the results of the OLS-estimator. In addition does it verify whether a OLS-estimator can be used to estimate the coefficient of the interaction term. The third section exploits both the IV-estimation and OLS-estimation in combination with the new policy index.
4.1 New instruments
16
Table 3. - Summary statistics of database C
Notes: Summary statistics reflect information for four different periods of relations between on average 22 donor countries and 70 developing countries. All variables used in this Table are reported with additional information in appendix I.
(1) (2) (3) (4) (5) (6) (7) (8) Variable Mean Std. Dev. Min Max Mean Std. Dev. Min Max
1960 – 2000 (1628 Obs) 1970 – 2000 (1716 Obs) Real annual average per capita GDP growth 1,6 1,8 -3,4 6,8 1,2 2,0 -4,9 6,4 Aid to GDP 5,9 6,4 0,1 26,4 5,5 6,2 0,1 27,6 Fitted aid to GDP 5,1 4,5 -7,0 15,7 4,7 4,4 -7,7 20,4 Initial level of per capita (PPP) GDP 7,4 0,7 6,0 9,0 7,7 0,8 5,8 9,3 Openness (Sachs & Werner) 0,3 0,3 0,0 1,0 0,3 0,3 0,0 1,0 World development indicator 48,6 9,7 32,4 71,7 52,9 9,7 34,4 71,2 Average frost days * fraction of land in tropics -0,6 0,7 -1,0 1,5 -0,5 0,8 -1,0 1,5 Institutional quality 0,5 0,1 0,2 0,9 0,5 0,1 0,2 0,9 Average inflation 14,3 32,4 -0,8 173,2 18,9 34,3 -0,8 198,7 Average M2/GDP 19,7 12,6 2,6 73,0 23,5 12,3 6,3 73,0 Average BB/GDP -3,6 4,4 -20,0 5,8 -3,8 4,8 -23,1 5,8 Average number of revolutions 0,2 0,2 0,0 0,8 0,2 0,2 0,0 1,1 Ethnic fraction 0,5 0,3 0,0 0,9 0,5 0,3 0,0 0,9 Dummy for African countries below the Sahara 0,4 0,5 0,0 1,0 0,4 0,5 0,0 1,0 Dummy for countries in the East, 0,1 0,3 0,0 1,0 0,1 0,3 0,0 1,0 Dummy for colony of donor land 0,0 0,2 0,0 1,0 0,0 0,2 0,0 1,0 Mean country still colony of donor country 0,0 0,0 0,0 0,5 0,0 0,0 0,0 0,3 Dummy for common language as donor country 0,2 0,4 0,0 1,0 0,2 0,4 0,0 1,0 Dummy for common colony of United Kingdom 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of France 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of Spain 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of Portugal 0,0 0,0 0,0 1,0 0,0 0,0 0,0 1,0 log of difference in population 0,8 2,1 -7,7 6,2 1,2 2,0 -4,9 6,4
1980 – 2000 (1650 Obs) 1990 – 2000 (1540 Obs) Real annual average per capita GDP growth 1,0 2,1 -2,9 6,3 1,1 2,7 -10,2 7,4 Aid to GDP 4,9 5,6 0,0 21,3 4,6 5,9 0,0 26,9 Fitted aid to GDP 4,2 3,8 -6,5 13,9 4,2 3,5 -5,9 13,2 Initial level of per capita (PPP) GDP 7,9 0,8 6,1 9,3 8,0 0,8 6,3 9,8 Openness (Sachs & Werner) 0,1 0,3 0,0 1,0 0,4 0,5 0,0 1,0 World development indicator 57,8 9,6 35,4 74,6 61,6 9,8 35,2 76,5 Average frost days * fraction of land in tropics -0,5 0,8 -1,0 1,5 -0,4 0,8 -1,0 1,5 Institutional quality 0,5 0,1 0,2 0,9 0,5 0,1 0,2 0,9 Average inflation 31,9 57,7 -0,8 352,0 195,9 834,3 1,1 6425,0 Average M2/GDP 31,0 15,6 8,1 78,4 34,9 21,2 3,9 117,1 Average BB/GDP -4,7 6,0 -39,1 3,4 -1,8 3,7 -10,5 12,7 Average number of revolutions 0,2 0,3 0,0 1,3 0,3 0,4 0,0 1,6 Ethnic fraction 0,4 0,3 0,0 0,9 0,4 0,3 0,0 0,9 Dummy for African countries below the Sahara 0,4 0,5 0,0 1,0 0,3 0,5 0,0 1,0 Dummy for countries in the East, 0,1 0,3 0,0 1,0 0,1 0,3 0,0 1,0 Dummy for colony of donor land 0,0 0,2 0,0 1,0 0,0 0,2 0,0 1,0 Mean country still colony of donor country 0,0 0,0 0,0 0,0 0,0 0,0 0,0 0,0 Dummy for common language as donor country 0,2 0,4 0,0 1,0 0,2 0,4 0,0 1,0 Dummy for common colony of United Kingdom 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of France 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of Spain 0,0 0,1 0,0 1,0 0,0 0,1 0,0 1,0 Dummy for common colony of Portugal 0,0 0,0 0,0 1,0 0,0 0,0 0,0 1,0 log of difference in population 0,5 2,1 -7,9 5,9 0,2 2,1 -8,0 5,9
17
The inflation database C, in the period 1990-2000 has got an outlier with 6425% inflation, which is caused by the republic of Congo. But even without the republic of Congo has the average inflation increased.
Table 3 also shows again why the three variables that have been removed from instrument set A18. Dummy for common colony of Portugal has got a mean and variation of zero. This is because Brazil is the only country in the database C that was a colony of Portugal.
The results from the study of Rajan and Subramanian (2008) and the results of replacing the instrument fitted aid with instrument set A are given in Table 4. Row 4 and 5 of this Table show which instruments have been used for the regression.
The effect of a policy-aid interaction term is still not significant, when using instrument set A. In fact the effect and the standard error of the policy-aid interaction term is with instrument set A for every period 0,00. In addition is R-squared -, instead of a number. This might be a consequence of the lack of variance in the variables used, since all variables that were originally from database B have got the same value for every 22 observation. Finally instrument set A fails a Sargan test. Thus all together instrument set A does not seem like a good instrumental choice in this model. The rest of this study I will use fitted aid as instrument, because it is more clear what the effect of an OLS-estimator is when not all the coefficients and standard errors are 0,00.
The coefficients have been tested with a Hausman test, the results are in row 6. The value in column 5 is the results of comparing the coefficient of column 1 with column 5, etc. The lowest p-value is 0,22 and is given by column 7. Thus the IV-estimator with instrument set A is not significantly different from the estimator with fitted aid as instrument.
.
18
These are dummy for common colony of Portugal, Dummy for common colony of Portugal * measure for openness and log of difference in population * dummy for common colony of Portugal * measure for openness
18
Table 4. – Results new instruments
(1) (2) (3) (4) (5) (6) (7) (8) 1960-2000 1970-2000 1980-2000 1990-2000 1960-2000 1970-2000 1980-2000 1990-2000 (1) Aid/GDP * policy -0,36 -0,14 1,75 0,61 0,00 0,00 0,00 0,00 (2) (standard error) (0,34) (0,31) (1,44) (0,61) (0,00) (0,00) (0,00) (0,00) (3) R2 0,58 059 0,32 0,24 - - - - (4) Fitted aid Yes Yes Yes Yes No No No No (5) Instrument set A No No No No Yes Yes Yes Yes (6) Hausman statistic
1,07 0,21 1,48 0.99
Notes: Regression coefficient are reported in row 1, along with standard errors in parentheses in row 2. Results are from database C. Each coefficient reported in this Table is from a different regression. Column 1-4 are the results from the study of Rajan & Subramanian (2008). Column 5-8 reflect IV-regressions using instrument set A as instruments.
4.2 Ordinary least squares
For the OLS-regression only database B is used, since the variables from database A were only used as instruments. This means that row 1 of Table 1 contains the number of observations for every period. Table 5 contains the results of the regressions.
Table 5. – Ordinary least squares results
(1) (2) (3) (4) (5) (6) (7) (8) Variable 1960-2000 1970-2000 1980-2000 1990-2000 1960-2000 1970-2000 1980-2000 1990-2000 (1) Aid/GDP * policy -0,36 -0,14 1,75 0,61 0,04 -0,02 -0,60 0,06 (2) (standard error) (0,34) (0,31) (1,44) (0,61) (0,10) (0,11) (0,31) (0,12) (3) R2 0,58 059 0,32 0,24 0,8 0,7 0,7 0,6 (4) IV-regression Yes Yes Yes Yes No No No No (5) OLS-regression No No No No Yes Yes Yes Yes (6) Hausman statistic
1,49 0,18 2,82 0,82
Notes: Regression coefficient are reported in row 1, along with standard errors in parentheses in row 2. Results are from database B. Each coefficient reported in this Table is from a different regression. Column 1-4 are the results from the study of Rajan & Subramanian (2008). Column 5-8 reflect IV-regressions using instrument set A as instruments.
The standard errors have declined as expected, especially in column 7 and 8. However the coefficient of the interaction term has also changed. There is a sign change in column 5,6 and 7 compared to column 1, 2 and 3. But for every OLS-regression applies that the coefficient has moved closer to zero. Thus with this new method there are still no significant results.
19
The coefficient of the interaction term of the OLS-estimator is compared with the IV-estimator using a Hausman test. Thus for example column 1 is compared with column 5. The result of this test is given in row 8. The highest Hausman statistic is reported in column 8 and has a p-value of 0,09. Thus H0 is not rejected and the coefficients estimated with the
OLS-estimator are valid compared to IV-OLS-estimator.
Comparing the results of the OLS-estimator with the results of column 5-8 in Table 4 shows that the results from instrument set A are very close to the OLS-regression results. Thus instrument set A is perhaps not so bad after all.
When checking the assumption that development aid is endogenous, we found no evidence for endogeneity in our model. Using a Hausman test for every period we found that the lowest P-value is 0,93 and is from the period 1970-2000. Thus the assumption that development aid is endogenous is incorrect. However this might be a consequence of the use of fitted aid as an instrument. As mentioned before is the use of fitted aid questionable.
4.3 New policy index
Table 6 contains the summary statistics of a modified database B, since it includes three new variables to construct the policy index and some countries are excluded due to lack of data.
Comparing Table 3 with Table 6 does not show big differences. The countries that are no longer part of the database seem to be random, when one might expect that the poorer countries drop out, since keeping your administration in order costs money.
Both the black market premium and the external debt ratio doubles from 1970-2000 compared to 1980-2000. But the external debt growth decreases, while the black market premium keep growing exponentially. However in the period 1990-2000 Nicaragua has got a black market premium of 10990. In contrast with the other two new variables, the ratio trade of the GDP decreases between 1980-2000 and 1990-2000.
20
Table 6. – Summary statistics of the modified database.
Notes: Summary statistics from modified database B are slightly different from Table 2. Some countries are left out as a consequence of missing values.
Next the weights of the different variables have been calculated for every period with model (2.1), the results of this regression are given in Table 7. The ratio trade of the GDP is never significant. In column 1 only one regressor is significant on a 95% level.. this is the ratio M2 of the GDP. The sign of M2/GDP is positive as expected.
Column 2 contains the results for 1980-2000. Ratio M2 of the GDP is again significantly larger than zero. In addition is the external debt is now significantly smaller than
(1) (2) (3) (4) (5) (6) (7) (8) Variable Mean Std. Dev. Min Max Mean Std. Dev. Min Max
1970 – 2000 (64 Obs) 1980 – 2000 (68 Obs)
Real annual average per capita GDP growth 1,25 2,02 -4,90 6,36 1,00 2,07 -2,93 6,27 Aid to GDP 5,52 6,21 0,07 27,63 4,93 5,61 0,05 21,26 Fitted aid to GDP 4,67 4,44 -7,66 20,38 4,24 3,86 -6,45 13,92 Initial level of per capita (PPP) GDP 7,65 0,77 5,80 9,26 7,93 0,80 6,09 9,35 World development indicator 52,88 9,77 34,36 71,21 57,79 9,65 35,40 74,60 Frost days * fraction of land in tropics -0,55 0,77 -1,04 1,53 -0,48 0,81 -1,04 1,53 Institutional quality 0,53 0,13 0,23 0,86 0,54 0,12 0,24 0,86 Inflation 18,86 34,54 -0,84 198,72 31,94 58,06 -0,84 351,97 M2/GDP 23,48 12,39 6,32 72,98 30,98 15,74 8,13 78,36 BB/GDP -3,83 4,84 -23,15 5,84 -4,75 6,08 -39,09 3,38 Black market premium 26,90 47,78 -21,19 348,89 43,36 111,09 -5,35 818,26 External debt/GDP 30,57 18,48 0,00 84,00 70,67 59,69 4,40 385,90 Trade/GDP 50,06 33,34 6,33 244,33 66,22 47,77 10,59 373,48 Number of revolutions 0,23 0,22 0,00 1,10 0,25 0,27 0,00 1,29 Ethnic fraction 0,47 0,29 0,00 0,90 0,45 0,30 0,00 0,90 Dummy for African countries below Sahara 0,41 0,50 0,00 1,00 0,37 0,49 0,00 1,00 Dummy for countries in the East 0,12 0,32 0,00 1,00 0,12 0,33 0,00 1,00
1990 – 2000 (66 Obs)
Real annual average per capita GDP growth 1,11 2,75 -10,15 7,41 Aid to GDP 4,58 5,98 0,03 26,95 Fitted aid to GDP 4,24 3,51 -5,92 13,17 Initial level of per capita (PPP) GDP 8,03 0,85 6,35 9,79 World development indicator 61,64 9,90 35,20 76,54 Frost days * fraction of land in tropics -0,43 0,83 -1,04 1,53 Institutional quality 0,55 0,12 0,23 0,86 Inflation 195,94 840,05 1,11 6424,99 M2/GDP 34,88 21,30 3,86 117,14 BB/GDP -1,82 3,73 -10,53 12,71 Black market premium 238,54 1319,88 0,00 10990,70 External debt/GDP 81,01 87,18 13,50 658,30 Trade/GDP 62,38 48,25 15,16 372,62 Number of revolutions 0,28 0,40 0,00 1,60 Ethnic fraction 0,42 0,30 0,00 0,90 Dummy for African countries below Sahara 0,34 0,48 0,00 1,00 Dummy for countries in the East 0,13 0,34 0,00 1,00
21
zero. As expected has the external debt of a country got a negative effect on the growth of a country.
In column 3 is the ratio M2 of the GDP and the external debt still significant. But in addition are the black market premium and the inflation significant. The black market premium has got a positive effect on the growth of a country, which is not as expected. The black market premium should have a negative effect on the growth of a country according to the economic literature. Inflation has as expected a negative sign.
Most coefficients have got a small effect, except for the ratio m2 of the GDP. This is familiar to the criticism on the openness indicator given by the study of Rodriquez & Rodrik (2001). It raises the question if the variables that are used to construct the policy index effect the growth of a country.
Table 7. – Constructing a policy index
(1) (2) (3) Variable 1970-2000 1980-2000 1990-2000 Trade / GDP -0,01 -0,01 0,00 (standard error) (0,01) (0,01) (0,01) Black Market premium 0,00 0,00 0,00 (standard error) (0,01) (0,00) (0,00) External debt -0,01 -0,01 -0,02 (standard error) (0,02) (0,00) (0,01) Inflation 0,01 0,00 -0,00 (standard error) (0,01) (0,00) (0,00) M2 / GDP 0,10 0,06 0,03 (standard error) (0,03) (0,02) (0,01) Constant -0,24 0,43 1,59 (standard error) (0,72) (0,65) (0,69) R2 0,19 0,24 0,50 Observations 64 68 66
Notes: Regression coefficients with standard errors in parentheses are reported in column 1-3. The estimations from every column has been obtained using a different regression.
The policy index can now be calculated with equation (2.2). The summary statistics of the new policy index are given in Table 8. As expected does the policy index drop in 1980-2000 as a consequence of the external debt. However in 1990-1980-2000 the policy index increases to almost its old level. The difference between countries increases as time passes. In 1960-2000 the worst performing country on the policy index had -0,65, while in 1990-1960-2000 the
22
worst country has got an index value of -8,86. However there are still countries with an index number around one. Thus the new policy index has definitely improved on this compared to the measure of openness that was previously used.
Table 8. – Summary statistics of the new policy index
(1) (2) (3) (4) (5) Period Obs Mean Std. Dev. Min Max 1970-2000 64 1,03 0,86 -0,65 3,74 1980-2000 68 0,91 1,01 -1,76 3,29 1990-2000 66 1,00 1,94 -8,86 4,45 Notes: Summary statistics for the new policy index for the periods 1970-2000, 1980-2000 and 1990-2000.
The results of the IV-regression and the OLS regressions are given in Table 9. The policy-aid interaction term is not significant on a 95% level in both the IV-regression as the OLS-regression. The sign of the interaction term is in the IV-regressions negative. The standard errors of the interaction term estimated with the OLS-estimator have been reduced drastically. However the coefficient of the interaction term has also moved closer to zero. Thus the effect of the interaction term is still not significant.
R2 is close to zero for column 1 and 2. In contrast to column 3 where R2 has got a value of 0,51. These IV-regressions are done with the instrument fitted aid. The OLS-regressions have got higher values for R2.
Again the coefficient of the interaction term obtained using an OLS-estimator is tested using an Hausman test. Thus for example column 1 is compared with column 4. The highest Hausman statistic is reported in column 4 and has a p-value of 0,11. Thus H0 is not rejected
and the coefficients estimated with the OLS-estimator are valid compared to IV-estimator. However the p-value for column 5 and 6 are respectively 0,83 and 0,95. The policy index seems to score better for the more recent periods.
The results of Table 9 are very similar to the results of Table 5. The difference between the IV-estimator and the OLS-estimator is almost the same. But the measure for openness was less significant according to the Hausman statistic for the periods 1980-2000 and 1990-2000. This may be a consequence of previous mentioned problems on the openness indicator. The new policy index however is barely significant on a 90% level for 1970-2000, but for the 1980-2000 and 1990-2000 it is very significant. Thus the new policy index seems to be more correct overall than the measure for openness.
23
Table 9. –Regression results with the new policy index
(1) (2) (3) (4) (5) (6) Variable 1970-2000 1980-2000 1990-2000 1970-2000 1980-2000 1990-2000 Aid/GDP * policy -0,29 -0,23 -0,11 -0,02 0,01 -0,02 (standard error) (0,18) (0,21) (0,07) (0,05) (0,04) (0,03) R2 0,00 0,02 0,51 0,65 0,61 0,64 IV-regression Yes Yes Yes No No No OLS-regression No No No Yes Yes Yes Hausman statistic 2,50 0,05 0,00 Notes: Regression coefficient are reported in row 1, along with standard errors in parentheses in row 2. Results are from a modified database B. Each coefficient reported in this Table is from a different regression. Column 1-3 are the results from an IV-estimation using fitted aid as instrument. Column 4-6 are the results from an OLS-estimation
24
5. Conclusion
Based on this study we cannot conclude that the policy of a country affects the effectiveness of development aid. The effect of a policy-aid interaction term have been analysed using two different policy indexes and with two different estimators in four different periods. In not a single estimation was the coefficient of the policy-aid interaction term significant.
In addition we found that the validity of the instrument fitted aid is questionable. We found that three variables had to be removed as a consequence of a lack of data variety. In addition, when using the variables that were used to construct fitted aid as instrument, separate and not as an index, we found insignificant effects. Finally did this instrument set not pass the Sargan test.
When estimating the coefficient of the policy-aid interaction term with an OLS-regression, there were still no significant results. The OLS-regression is consistent and
asymptotically normally distributed for the coefficient of the policy-aid interaction term, even though we assume that development aid is endogenous.
The assumption that development aid is endogenous has been tested with a Hausman test. According to this test we have no reason to assume endogeneity in our model. This might be a consequence of the use of fitted aid as instrument. In the growth literature it is commonly accepted that in contrast to this test development aid is endogenous.
The new policy index based on the model of the study of Easterly (2003) does not seem to significantly change the effect of the interaction term. The coefficient is even closer to zero compared to the results with the openness indicator. The policy index scored better in the periods 1980-2000 and 1990-2000 and the openness indicator scored better in the period 1960-2000. The most important improvement of the new policy indicator is that it does not diverge over time, like the measure for openness did after 1980. The variables which were used to construct the policy index were not always significant. In fact the ratio of trade to GDP was never significant. A policy index based on different variables, but created with the same method seems like a good idea.
For future research I suggest using a different measure for growth. GDP may not be the best measurement for growth in development countries. Because for example it adds nothing to the GDP if a farmer eats his own harvest. Different measure for growth can be the life expectancy or the education rate. Finally, not all development aid is included in the
25
References
Boone, P. (1996). Politics and the effectiveness of foreign aid. European economic review, 40(2), 289-329.
Bun, M. J., & Harrison, T. D. (2014). OLS and IV estimation of regression models including endogenous interaction terms (No. 2014-3). LeBow College of Business, Drexel University.
Burnside, C., & Dollar, D. (2000). Aid, policies, and growth. American economic review, 847-868.
Buchinsky, M. (1995). Estimating the asymptotic covariance matrix for quantile regression models a Monte Carlo study. Journal of Econometrics, 68(2), 303-338.
Canarella, G., & Pollard, S. (2004). Parameter heterogeneity in the neoclassical growth model: a quantile regression approach. Journal of Economic Development, 29, 1-32. Dalgaard, C. J., Hansen, H., & Tarp, F. (2004). On the empirics of foreign aid and growth*.
The Economic Journal, 114(496), F191-F216.
Doucouliagos, H. & Paldam, m. (2008). Aid effectiveness on growth: A meta study. European journal of political economy, 24(1), 1-24.
Durlauf, S. N., & Johnson, P. A. (1995). Multiple regimes and cross‐country growth behaviour. Journal of Applied Econometrics, 10(4), 365-384.
Easterly, W. (2003). Can foreign aid buy growth?. Journal of Economic Perspectives, 23-48.
Hansen, H., & Tarp, F. (2001). Aid and growth regressions. Journal of development Economics, 64(2), 547-570.
Ram, R. (1986). Government size and economic growth: A new framework and some evidence from cross-section and time-series data. The American Economic Review, 191-203.
Rajan, R. G., & Subramanian, A. (2008). Aid and growth: What does the cross-country evidence really show?. The Review of economics and Statistics, 90(4), 643-665.
Ratha, D., Mohapatra, S., & Silwal, A. (2009). Migration and Remittance Trends 2009: A better-than-expected outcome so far, but significant risks ahead.
Rodriguez, F., & Rodrik, D. (2001). Trade policy and economic growth: a skeptic's guide to the cross-national evidence. In NBER Macroeconomics Annual 2000, Volume 15 (pp. 261-338). MIT PRess.
Sachs, J. D., Warner, A., Åslund, A., & Fischer, S. (1995). Economic reform and the process of global integration. Brookings papers on economic activity, 1-118.
Tavares, J., & Wacziarg, R. (2001). How democracy affects growth. European Economic Review, 45(8), 1341-1378.
Wacziarg, R., & Welch, K. H. (2003). Trade liberalization and growth: New evidence. The World Bank Economic Review, 22(2), 187-231.
26
Appendix I
Variable name Description (if necessary)
Rgdpchg Real annual average per capita GDP growth Aid_gdp Percentage development aid of GDP
Faid_gdp Percentage fitted development aid of GDP Yc_penn Initial income from Pennworld
Sw1 Measure for openness
Le_wdi World development indicator
Geog6099 Average frost days * fraction of land in tropics Gadp6099 Institutional quality
Inf5_ES Average inflation for first 5 non-missing years M2_GDP5_WDI Average M2/GDP for first 5 non-missing years Revol Average number of revolutions
Ethfrac_pap Ethnic fraction, based on Soviet Atlas
Safrica Dummy for African countries below the Sahara
East Dummy for countries in the East, for example China and India Colony_uk Dummy for if a country was a colony of the United Kingdom Colony_fr Dummy for if a country was a colony of France
Colony_sp Dummy for if a country was a colony of Spain Multi_gdp Total Multilateral aid/GDP
Fmulti_gdp Fitted Multilateral aid/GDP Bilat_gdp Bilateral aid/GDP
Fbilat_gdp Fitted bilateral aid/GDP Social_gdp Social disbursement /GDP Fsoc_gdp Fitted social aid/GDP
Economic_gdp Economic disbursement /GDP Geco_gdp Fitted economic aid/GDP
Aid_l Long impact aid
27
Appendix II
Control variables used by Rajan et al. (2008): - initial income from Pennworld
- Measure for openness designed by Sachs and Warner (1995) - World development indicator
- Average frost days * fraction of land in tropics - Government Anti-Diversion Policy
- Inflation estimate - Average M2 - Average BB - Revolution
- Ethnic fraction based on soviet atlas - regional dummy for Sub-Saharan Africa - regional dummy for East Asia
28
Appendix III
𝐻0: 𝑝𝑙𝑖𝑚(𝛽̂𝑝𝑖𝑑𝑎𝑖,𝐼𝑉2 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆 ) = 0 𝐻𝑎: 𝑝𝑙𝑖𝑚(𝛽̂𝑝𝑖𝑑𝑎𝑖,𝐼𝑉2 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆 ) ≠ 0 The Hausman test statistic becomes:
𝐻𝐼𝑉2,𝑂𝐿𝑆 =
𝛽̂𝑝𝑖𝑑𝑎𝑖,𝐼𝑉2 − 𝛽̂𝑝𝑖𝑑𝑎𝑖,𝑂𝐿𝑆 𝑉̂𝐻,𝑝𝑖𝑑𝑎𝑖
With under homoscedasticity: 𝑉̂𝐻= 1𝑛(𝑦 − 𝑋𝛽̂𝐼𝑉2)′(𝑦 − 𝑋𝛽̂𝐼𝑉2) (𝑋′𝑃𝑧2𝑋)−1
− 1
𝑛(𝑦 − 𝑋𝛽̂𝑂𝐿𝑆)′(𝑦 − 𝑋𝛽̂𝑂𝐿𝑆) (𝑋𝑃𝑧2𝑋)−1(𝑋′𝑋)−1 This is asymptotically disturbed as 𝛸12.
In the case of heteroscedasticity the following applies according to Bun et al. (2014): 𝑉̂𝐻 = 𝑉̂22+ 𝑉̂00− 2𝑉̂20 With: 𝑉̂22 = 𝑛 (𝑋′𝑃𝑧2𝑋)−1𝑋′𝑍𝐼𝑉2(𝑍𝐼𝑉2′ 𝑍𝐼𝑉2)−1𝑆̂22(𝑍𝐼𝑉2′ 𝑍𝐼𝑉2)−1𝑍𝐼𝑉2′𝑋(𝑋′𝑃𝑧2𝑋)−1 𝑉̂00= 𝑛(𝑋′𝑋)−1𝑆̂00(𝑋′𝑋)−1 𝑉̂20 = 𝑛 (𝑋′𝑃𝑧2𝑋)−1𝑋′𝑍 𝐼𝑉2(𝑍𝐼𝑉2′ 𝑍𝐼𝑉2)−1𝑆̂20(𝑋′𝑋)−1 And with: 𝑆̂22 = 1 𝑛∑ 𝑢̂𝐼𝑉2,𝑖2 (1 − ℎ𝐼𝑉,1)2𝑧𝑖 (𝐼𝑉2)𝑧 𝑖 (𝐼𝑉2)′ 𝑛 𝑖=1 𝑆̂00 = 1 𝑛∑ 𝑢̂𝐼𝑉2,𝑖2 (1 − ℎ𝐼𝑉,1)2𝑋𝑖𝑋𝑖′ 𝑛 𝑖=1 𝑆̂20= 1 𝑛∑ 𝑢̂𝐼𝑉2,𝑖2 (1 − ℎ𝐼𝑉,1)2 𝑛 𝑖=1 𝑧𝑖(𝐼𝑉2)𝑋𝑖′
29
Appendix IV
Instruments used to construct fitted aid:
- Common language - Current colony - Common colony
- Dummy for if a country was a colony of the United Kingdom - Dummy for if a country was a colony of France
- Dummy for if a country was a colony of Spain - Dummy for if a country was a colony of Portugal
- Log of the difference in population between donor country and receiving country - Product between log of the difference in population between donor country and
receiving country and common colony
- Product between log of the difference in population between donor country and receiving country and dummy for if a country was a colony of the United Kingdom - Product between log of the difference in population between donor country and
receiving country and dummy for if a country was a colony of France
- Product between log of the difference in population between donor country and receiving country and dummy for if a country was a colony of Spain
- Product between log of the difference in population between donor country and receiving country and dummy for if a country was a colony of Portugal
Instrument set A:
- Common language - Current colony - Common colony
- Dummy for if a country was a colony of the United Kingdom - Dummy for if a country was a colony of France
- Dummy for if a country was a colony of Spain
- Log of the difference in population between donor country and receiving country - Product between log of the difference in population between donor country and
30
- Product between log of the difference in population between donor country and receiving country and dummy for if a country was a colony of the United Kingdom - Product between log of the difference in population between donor country and
receiving country and dummy for if a country was a colony of France
- Product between log of the difference in population between donor country and receiving country and dummy for if a country was a colony of Spain
- Product between policy index and common language - Product between policy index and Current colony - Product between policy index and Common colony
- Product between policy index and Dummy for if a country was a colony of the United Kingdom
- Product between policy index and Dummy for if a country was a colony of France - Product between policy index and Dummy for if a country was a colony of Spain - Product between policy index and Log of the difference in population between donor
Product between policy index and country and receiving country
- Product between policy index, log of the difference in population between donor country and receiving country and common colony
- Product between policy index, log of the difference in population between donor country and receiving country and dummy for if a country was a colony of the United Kingdom
- Product between policy index, log of the difference in population between donor country and receiving country and dummy for if a country was a colony of France - Product between policy index, log of the difference in population between donor
31
Appendix V
Donor countries from database A
United Kingdom New Zealand Ireland France Denmark Netherlands Finland Germany Portugal Greece Austria Italy Switzerland United States Canada Sweden Norway Belgium Japan Luxembourg Spain Australia
32 Recipient countries from database A
Afghanistan Croatia Latvia Samoa
Albania Cuba Lebanon Sao Tome and Principe Algeria Cyprus Lesotho Saudi Arabia
Angola Czech Republic Liberia Senegal Anguilla Djibouti Libya Serbia Antigua Dominica Lithuania Seychelles Argentina Dominican Republic Macedonia, FYR Sierra Leone Armenia Ecuador Madagascar Singapore Aruba Egypt, Arab Rep. Malawi Slovakia Azerbaijan El Salvador Malaysia Slovenia Bahamas Equatorial Guinea Maldives Solomon Islands Bahrain Eritrea Mali Somalia
Bangladesh Estonia Malta South Africa Barbados Ethiopia -pre 1993 Marshall Islands,Rep Sri Lanka Belarus Falkland Islands Mauritania St. Helena
Belize Fiji Mauritius St. Kitts and Nevis Benin French Polynesia Mexico St. Lucia
Bermuda Gabon Micronesia, Fed.Sts.
St. Vincent and the Grenadines Bhutan Gambia, The Moldova Sudan Bolivia Georgia Mongolia Suriname Bosnia and Herzegovina Ghana Montserrat Swaziland
Botswana Gibraltar Morocco Syrian Arab Republic Brazil Grenada Mozambique Taiwan
Brunei Guatemala Myanmar Tajikistan Bulgaria Guinea Namibia Tanzania Burkina Faso Guinea-Bissau Nauru Thailand Burundi Guyana Nepal Timor Cambodia Haiti Netherlands Antilles Togo Cameroon Honduras New Caledonia Tonga
Cape Verde Hungary Nicaragua Trinidad and Tobago Cayman Islands India Niger Tunisia
Central African Republic Indonesia Nigeria Turkey Chad Indus Sterl Oceania Oman Turkmenistan Chile Iran Pakistan Tuvalu
China Iraq Palau US Pacific Islands China,P.R.:Hong Kong Israel Panama Uganda
China,P.R.:Macao Jamaica Papua New Guinea Ukraine
Christmas Island Jordan Paraguay United Arab Emirates Colombia Kazakhstan Peru Uruguay
Comoros Kenya Philippines Uzbekistan Congo, Dem. Rep. Kiribati Poland Vanuatu Congo, Rep. Korea, Rep. Qatar Venezuela, RB Cook Islands Kuwait Romania Vietnam Costa Rica Kyrgyz Republic Russia Yemen Cote d'Ivoire Lao PDR Rwanda Zambia
33 Recipient countries from database B
Algeria Mauritius Argentina Mexico Bangladesh Morocco Benin Namibia Bolivia Nicaragua Botswana Niger Brazil Nigeria Burkina Faso Pakistan Burundi Panama
Cameroon Papua New Guinea Chad Paraguay
Chile Peru
China Philippines Colombia Romania Congo, Dem. Rep. Rwanda Congo, Rep. Senegal Costa Rica Singapore Cote d'Ivoire South Africa Cyprus Sri Lanka
Dominican Republic Syrian Arab Republic Ecuador Thailand
Egypt, Arab Rep. Togo
El Salvador Trinidad & Tobago Ethiopia Turkey
Fiji Uganda
Gabon Uruguay Gambia, The Venezuela, RB
Ghana Zambia Guatemala Zimbabwe GuineaBissau Guyana Honduras India Indonesia Iran, Islamic Rep. Israel Jamaica Kenya Korea, Rep. Lesotho Madagascar Malawi Malaysia Mali Mauritania
