Corruption and growth in “developmental states”

MASTER’S THESIS IE&B EWM068A25

Corruption and growth in “developmental states”

Submitted to Prof. Dirk Bezemer Prof. Padma Rao Sahib

Faculty of Economics University of Groningen

Submitted by Hilde Anna de Vries

International Economics and Business Student University of Groningen

s1323776 hildeanna@gmail.com

August 31, 2006

Corruption and growth in “developmental states”

Abstract

This paper investigates the effect of corruption on economic growth in developmental

states. A developmental state is characterised by high public sector involvement in economic

growth and was distinguished in the sample by creating an index. Contrary to our

expectations, the results indicate that corruption has a negative impact on economic growth in

developmental state countries, while it has no effect in non-developmental states. These

results held once an additional interaction term model was set up. Overall this confirms that

the effect of corruption on economic growth depends on the institutional context and cannot

be considered a bad under all circumstances.

CONTENTS

1 Introduction 5

2 Literature Review 6

3 Problem Statement 11

4 Variables and Methodology 13

4.1 DS Variables 13

4.1.1 Description and Sources 13

4.1.2 Correlation DS Variables 15

4.2 Methodology - DS Indices 15

4.2.1 Unweighted and Weighted Indices 16

4.2.2 Threshold Index 17

4.2.3 Relative Ranking Index 18

4.3 Similar Research – Méndez and Sépulveda 18

4.4 Growth Model Variables 19

4.4.1 Description and Sources 19

4.4.2 Endogeneity 21

4.4.3 Correlation and Collinearity Growth model Variables 22

4.5 General Regression Analyses 23

4.6 Interaction Term Analysis 24

4.6.1 Interaction Term 24

4.6.2 Collinearity Interaction Term Variables 24

5 Results 25

5.1 DS Indices 25

5.2 Results General Regression Analyses 26 5.3 Comparison – Méndez and Sépulveda 29 5.4 Results Interaction Term Regression Analyses 31

6 Conclusions and Limitations 31

6.1 Conclusions 31

6.2 Limitations 33

References 35

Appendices 38

List of Appendices

Appendix 1 Ranking of Developmental State characteristics Appendix 2 Data Sources of all variables

Appendix 3 Data DS variables

Appendix 4 Data Growth model variables Appendix 5 Correlation CPI and PSEI

Appendix 6 Scatter plots with regression GDPgrowth/CPI/

PSEI

Appendix 7 Descriptive statistics table DS variables Appendix 8 Cross-correlation table DS variables

Appendix 9 Descriptive statistics growth model variables Appendix 10 Cross-correlation table growth model variables Appendix 11 Collinearity growth model variables – Results of

auxiliary regressions

Appendix 12 DS Unweighted Ordinal Ranking scores Appendix 13 DS Weighted Ordinal Ranking scores Appendix 14 DS Threshold scores

Appendix 15 DS Relative Ranking Scores

Appendix 16 Overview of all scores with indication of DS Appendix 17 Complete sample regressions

Appendix 18 Unweighted Ordinal Ranking regressions Appendix 19 Weighed Ordinal Ranking regressions Appendix 20 Threshold regressions

Appendix 21 Relative Ranking regressions Appendix 22 Overview of regression results

Appendix 23 Collinearity Interaction Term model variables -

Results of auxiliary regressions

Appendix 24 Threshold regressions – Interaction Term model Appendix 25 Relative Ranking regressions – Interaction Term

Model

1 INTRODUCTION

“Corruption lowers economic growth, biases the tax system to favour the rich and well-connected, reduces the effectiveness of targeting of social programs, biases government policies towards favouring inequality in asset ownership, lowers social spending, reduces access to education by the poor, and increases the risk of investment by the poor.”

-Gupta, Davoodi, and Alonso-Terme (1998)

This quote indicates the severity of the effect of corruption on an economy. Even though it has gone by unnoticed for a long period of time, during the past few decades it has become a hot item (Wei, 2001). It is also an issue that affects practically all societies, whether developed or developing, whether in a nation with a strong or a weak state. The extent to which there is a free playing field for this sub-optimal way of allocating means and resources differs across societies, typically the poorest countries also have the highest corruption perception indexes (Salierno, 2005). While focussing on the public sector, a commonly used definition is the sale of government property for personal gain. The negative impact this has on overall economic development results from the shift of the focus on areas that are important for economic growth, like for instance education and health, to areas that are less important, together with less than optimal domestic and foreign direct investment (Wei, 2001). Because of this negative impact on economic development, a number of transnational institutions have initiated anti-corruption strategies and long term international goals in order to combat this outcome (World Bank).

As a research area, corruption has been extensively dealt with, looking at for instance the effect it has on the macro-economic level (Soon, 2006), including overall employment, compensation levels, GDP, foreign direct investment, and productivity, to the effect of anti- corruption policies. Understandably, what the main underlying assumption is in the existing literature, is that in general corruption always has a negative influence on economic growth, which has been pointed out by for instance Mauro (1995) and Shleifer and Vishny (1993).

However, a handful of academics, Leff (1964) and Huntington (1968) and more recently, Méndez and Sepúlveda (2006) have pointed out that some corruption might be desirable.

The purpose of this research paper is to explore whether there is some truth in the

latter assumptions of a positive effect of corruption. What we presuppose is that the effect of

corruption depends on the institutional context, more specifically that in countries with high

government involvement, corruption has a neutral or positive influence on economic growth.

Since this environment is characterised by high regulations, opportunities for rent seeking arise, while in the opposite situation the government leaves the economic motion to the forces of the market, and opportunities arise for a pool of individual bureaucrats to earn additional rents, which is completely uncontrolled and has devastating effects (Bardhan 1997). In the former case corruption is probably merely a by-product of a good thing, while in the latter it can be considered a “bad” no matter what.

In order to define government involvement in this analysis, the concept of

“developmental states” was brought in, which was first introduced during the 1980’s. In a few words a developmental state is a type of state that involves a proactive public sector with a number of characteristics like for instance a high export rate, fostering of industries and low tax rates (Sindzingre, 2004). The idea is that a developmental state is a package for which the whole is larger than the sum of its parts, a more thorough explanation will be provided later on. Based on this concept this research will attempt to find out whether corruption has a negligible effect on economic growth in societies that can be considered developmental states.

In order to test this, a “developmental state index” will be composed in order to distinguish the respective countries in the sample. After that an OLS regression analysis will be set up with economic growth as the dependent variable.

The guideline for this research will be an analysis conducted by Méndez and Sepúlveda (2006) who looked into the effect of corruption on economic growth, while making a distinction based on political freeness. Their main findings already indicated that a positive optimal level of corruption is beneficial for growth. However, more empirical analysis with respect to his finding is necessary to possibly put an end to the existing assumption that corruption is always a detrimental phenomenon.

The following section will discuss the existing literature on this topic, which is followed by the problem statement. Subsequently the variables and methodology will be explained. Section 5 deals with the Results, and the final section elaborates on the conclusions and limitations.

2 LITERATURE REVIEW

As pointed out in the introduction, an extensive amount of research is available on corruption. The main emphasis has been on the detrimental effects on economic development.

Therefore, this literature review will first highlight some of the main influential articles in this

area, which will be followed by an overview of the literature that contradicts the negative existing view. Furthermore, literature concerning the “developmental state” concept will be covered as well.

In 1995, Mauro published a very influential paper in the corruption research area. He conducted a statistical analysis using a subjective corruption index, a composite grade based on nine indicators of bureaucratic efficiency in order to find out what effect corruption has on economic growth. What he found was that corruption had a significant strong negative correlation with both investment and growth. Also, he claimed that his results did not support Huntington’s view, one of the first proponents of corruption, who alleged that corruption would become beneficial in case of slow bureaucracy. A more elaborate overview of his work will be provided later on. Mauro’s findings have been used as a benchmark for many scholars when discussing the effects of corruption on economic growth.

Murphy, Shleifer and Vishny (1993) were the first to impose a general corruption framework. This model gave more insight into the effects of corruption on inequality and economic growth. By adapting this framework Li, Xu and Zou (2000) also found evidence that corruption has a negative effect on growth, however, the results were less pronounced, and relatively insignificant. This undermined Mauro’s strong claims. In addition, they pointed out that corruption indices were unable to explain cross-country differences in growth rates.

Shleifer and Vishny (1993) showed that corruption, even if well-organized, works as an extortionary tax. Due to the secrecy that has to be kept, attention of government officials shifts to unnecessary projects or discourages innovation, which has a harmful effect on economic development (also Wei, 2001). Therefore they stood by Mauro, confirming that no positive level of corruption can be justified for economic growth. What they did point out was that in countries with weak governments, the negative effects are more pronounced.

Rock and Bonnett (2004) once more conducted a cross-country regression analysis.

This way of comparison led to the following results; Small, rather than large countries seem to be affected the most by corruption. Both growth rates and investment rates were affected more as the country decreased in terms of size. One very striking finding was that this rule applied to all developing countries except for Asia. In this case corruption seemed to increase the growth rates overall. Interesting to mention here, is that Asian countries are in general characterized by high public sector involvement.

These results, where however contradicted by Wei (2001). In one of his articles,

written for the World Bank, he argues that corruption always has a negative effect on

economic development, even in Asian countries. He points out that after controlling for the

size of the market, the lower wages and the high growth rate of the economy, which has attracted scores of foreign investors, corruption still negatively affects foreign direct investment in these countries. Therefore, he claims that the Asian paradox does not really exist.

Even though most academics agree that any level of corruption has harmful effects on

economic growth, there seems to be a handful of scholars that suggest the opposite, however, under certain circumstances. It might be said that Huntington was the first academic

that brought up the possible advantageous externalities of corruption. In his paper, which was published in 1968 a number of often quoted statements can be discerned, like;

"In terms of economic growth the only thing worse than a society with a rigid, over centralized, dishonest bureaucracy is one with a rigid, over centralized, honest bureaucracy."

-Huntington, 1968.

What he obviously meant is that a certain level of corruption in an economy is always better than no corruption at all. The underlying assumption is that bribery can be used to speed up bureaucratic activities, which enables individuals applying for a certain service to be more efficient, of which the benefits are especially strong in countries with intricate public sectors.

In addition, it can also be used as an incentive to make people work harder, which ultimately has the same efficiency-effect.

Instead of merely suggesting a certain optimum level of corruption, a recently published article, by Méndez and Sepúlveda (2006) tried to empirically confirm the possible positive consequences of corruption. Their assumption was that in countries that fall into the

“politically free”-category, a certain level of corruption will stimulate economic growth, while in countries that have a less liberal political regime, corruption will still have a negative effect on economic development. Their expectations were confirmed; for low levels of corruption the effect is beneficial, while for high levels it is detrimental for economic growth. In addition they found evidence of a non-linear relationship between corruption and growth, which undermines the main assumption of a negative linear correlation between corruption and economic growth. This paper will be used as a guideline for the regression analysis in this investigation, therefore more details about their approach will be included in the methodology chapter.

Further proponents of a positive optimal level of corruption include Leff (1964) and

Lui and Ehrlich (1999). Even though Leff his ideas are relatively dated, he did point out that

governments, specifically in case of developing countries, might not have ideological goals - in the sense of economic growth- at all. Very often, these bureaucrats are “traditional elite”, which are indifferent or even hostile towards innovation or growth. Therefore, he claims, the only way for entrepreneurs to obtain permits or licences is to “persuade” officials. Bribery therefore often facilitates a higher overall level of investment and innovation, which in turn positively affects economic development. Lui and Ehrlich again tried to model the relationship between the government, corruption and growth by altering the political regime.

They distinguished two types of investment, either in human or political capital, where only the former positively affects economic growth. What they concluded was that a lot of government intervention has a negative impact on economic growth in very poor countries. In addition, in this model, a large government also has a negative influence on the GDP level, but this does not seem to affect the economic growth rate. Corruption seemed to affect both the GDP level and the economic growth rate in a similar way. Therefore, even though they refute the notion that a strong state is good for very poor developing countries, they do point out that corruption might not have a negative influence on the economic growth rate at all.

In order to distinguish countries with high government involvement, one could think about for instance the number of import policies a government implements, where high involvement would imply a large amount of policies and vice versa. However, this might not completely cover the concept. A better suggestion would be the notion of a “developmental state”, a term that has been explained by a number of scholars. According to Doner, Ritchie and Slater (2005), Castley (1996), Amsden (1998), Thompson (1996), Weiss and Hobson (1995), Guillen (2005), Kriekhaus (2002) and Sindzingre (2005) a developmental state is an

“organizational complex in which expert and coherent bureaucratic agencies collaborate with

organized private sectors to spur national economic transformation.” The state still plays a

large role, and leads rather than follows the market (Yoshimatsu, 2003). They suggest that

these well-organized and rare bureaucratic arrangements are the result of specific settings,

which until now have only been evident in a number of Asian countries namely South-Korea,

Taiwan and Singapore. As a result of the collaboration between the public and the private

sectors these countries tend to have high export levels together with high levels of innovation,

or in other words, high economic growth, which suggests that it is a package for which the

outcome is larger than the sum of its parts. Sindzingre (2004) discussed the characteristics of

these states more specifically. What she mentions is that developmental states can be

distinguished by their low tax rates, fostering of industries in order to promote rapid

industrialization, protectionist policies, high exports, high saving rates, limited foreign

shareholding, incentives for the banking sector and firm financing, training in technology, high public expenditure on health and education, low income inequalities (Levi Faur (1998), Weiss and Hobson, (2005)) and long term relationships between political power and the private sector and also between banks and the public and private owned firms. Furthermore, they are deeply embedded in society (Sindzingre, 2004). In addition, though most scholars put an emphasis on the export rates, the fostering of industries, protection and the tax rate, strong nationalism (Levi Faur, 1998) is mentioned, just as a large public sector and little aid flows seem to characterise developmental states as well (Sindzingre, 2004). The only downside is the small number of countries that fall into this category. In addition, critics have pointed out that the collaborative ties were infested with corruption, which may have caused the recent Asian crisis (Yoshimatsu, 2003). For an overview of the characteristics and their relative importance, please take a look at Appendix 1.

Another important notion pointed out by Kang (2002) is that one should be careful when labelling certain countries as developmental states, since they are not merely the result from a list of universal characteristics. He points out that each example should be analyzed in its own political-economical context. For instance South Korea is characterized as a developmental state, while the public goods that have been produced are not the result of goodwill on behalf of the public apparatus, but merely the result from deals between small interest groups. However, the fact that this country experienced extremely high growth rates before the Asian crisis struck, has been enabled by these investments, but Kang points out that these arrangements were still far from efficient.

After discussing the existing literature, the main impression is that there are still a number of question marks when it comes to the overall effect of corruption. Both sides, proponents and opponents of the overall negative effects, have been represented extensively, however a clear view is not really the case. Of course it is naive to think that there is only one

“golden rule” when talking about corruption, since the effect always has to be interpreted in its own specific context. However, this research will attempt to add some additional insight to this ongoing debate, while keeping a door open to the possible positive effects of corruption.

In addition, developmental states seem to have a large number of characteristics, however, not

all of them are represented equally in the literature, therefore only the most prominent features

will be focused on from now on.

3 PROBLEM STATEMENT

As already pointed out, the mainstream idea about corruption is that it is by definition bad for economic growth. However, a number of scholars have suggested that there might be some exceptions to this rule. Méndez and Sepúlveda (2006) discussed, building on Leff and Huntington’s assumptions, that there might be an optimal level of corruption, larger than zero.

For instance in case of a strong public apparatus, which means that the government is highly involved in the economic activities of its country, it might simply be a by-product of the system. Since in this environment, it is very difficult to for instance start up a business due to the high regulations, opportunities arise for rent-seeking. For developing countries, a strong state is however the best remedy for slow economic growth and development. Therefore the assumption is that it is merely a side-effect, or the “grease for the squeaking wheals of a rigid administration” (Huntington, 1968) which might even stimulate growth, and is not really bad for the economy itself. It is also to a certain extent predictable, which does certainly not apply to the way corruption is perceived in a weak state where the government is not highly involved in the economy.

In a weak state, opportunities for corruption will also arise, however the effects in this case are appalling. According to Bardhan (1997), the fact that corruption in countries with a weak public apparatus has a more devastating effect, results from the vast pool of independent profit maximizing players. Due to the lack of ability on the state’s behalf to stop this ongoing process, since more and more opportunities to obtain rents are created in this case, this results in a very inefficient public sector. The Russian public sector illustrates both types of states, since during the Communist regime the public sector was highly regulated and monitored, in which case the collection, and the actual amounts of bribes were fairly consistent. However, during the more recent post-Communist times, an unpredictable individually profit- maximizing state is more applicable.

As pointed out in the introduction the assumption is that corruption has a different

effect on economic growth under different types of involvement of the public sector. The

distinction will be based on the concept of a “developmental state”. As already pointed out in

the literature review, according to Sindzingre (2002), Guillén (2005), Yoshimatsu (2003),

Krieckhaus (2002), Levi Faur (1998) and several other academics, a developmental state is a

type of state that involves a proactive public sector, that stimulates for instance high exports,

low imports, high protection of domestic industries, limited foreign shareholding, high

savings, training in technology, high public expenditure on health and education and low tax

rates. Furthermore the public sector is known to pick certain industries and fosters them with policies, and is deeply embedded in society (Weiss and Hobson, 1995) which is obvious when looking at the collaboration and long term relationships between the private and public

sectors. In addition, Levi-Faur (1998) points out that developing states are characterised by nationalism and have low income disparity.

By definition this type of state is typically an Asian phenomenon, and both Sindzingre and Yoshimatsu only point out Singapore, Taiwan and South Korea as perfect examples.

However, a number of these characteristics can also be found in other developing countries, which also becomes evident from looking at relevant papers, which provides opportunities to create a “developmental state index” which will function as a dummy variable to make a distinction in the sample. For instance Botswana and Mauritius were pointed out as possible developmental states by Sindzingre (2005), just as Israel was put forward by Levi-Faur (1998).

Putting it all together, this would lead to the possibility that in countries with high government involvement, in this context, countries which portray more “developmental state” characteristics, do not experience a very negative impact of corruption.

After defining the problem, the resulting hypothesis is;

H1: In an economy with many developmental state characteristics, corruption does not have a negative effect on growth.

The resulting research questions are;

Does the effect of corruption on economic growth depend on the existence of a developmental state in a certain developing country?

And if the answer is yes, the following question is ;

For developmental states, do the results of the regression model support a positive optimal level of corruption?

In order to solve this problem, first of all the countries within the sample have to be

ranked according to their developmental state characteristics, for which purpose an index will

be composed. There are a number of ways to do this, which will be outlined in the Variables

and Methodology section. Secondly, based on this index, the sample should be split up in a

developmental state group and a non-developmental states group. Subsequently a number of

regression analyses with the GDP growth rate as a dependent variable have to be set up in

which both a single corruption term as well as a squared corruption term will be added in

order to find out what the relationship, and the shape of the relationship is. The expectation is

then that the parameters in front of the corruption variable in both groups are different in size

and/or sign. More specifically, positive or zero in case of the developmental state group and negative in case of the non-developmental state group.

The regressions are initially based on a standard growth model, which has been used by Méndez and Sépulveda (2006) as well, in order to enable a comparison of the results. The overall model entails that growth depends on a number of explanatory variables through a non-linear equation. As a dependent variable, the average GDP growth rate for each country over 15 years will be calculated, and the independent variables will include the population growth rate, initial GDP per capita of 1960, secondary school enrolment, capital formation, or investment, government expenditure and political stability. The final three variables were added later on, since they are generally not included in economic growth model research, however, according to both Mauro (1995) and Méndez and Sépulveda (2006) these factors affect the speed with which an economy converges to it’s steady state and thereby the growth rate of an economy at large. The sample includes ten or eleven countries for several country regions, with the actual number of countries for each region between brackets; Asia (10), Eastern Europe (11), Africa (11), Latin America (10) and the Middle East(10), which amounts to a total of 52 countries. For more details concerning the sample please take a look at Appendix 3.

4 VARIABLES AND METHODOLOGY

This chapter will discuss the variables that were used to compose the developmental state indices just as the regression model variables. Furthermore, the methodology will be explained in detail. From this point on developmental states will be indicated by DS in some instances as well.

4.1 Developmental State Variables 4.1.1 Description and sources

As already pointed out, according to Sindzingre (2004) Yoshimatsu (2003) and Levi-

Faur (1998) a developmental state has a large number of characteristics. For instance there is

a lot of collaboration and long term relationships between the private and public sectors, the

export levels are high, the import levels are low, the tax rates are low as well, domestic

industries are protected, there is a high saving rate, high nationalism, embeddedness in

society, limited foreign shareholding, incentives for the banking sector and firm financing,

training in technology, high public expenditure on health and education, low income

inequalities and long term relationships between the banks and public and private owned firms (Sindzingre, 2004). Not all characteristics are represented equally in the literature, for instance the main emphasis seems to be on exports, low income disparity and protection of industries, which comes forward from Levi-Faur (1998) and Yoshimatsu (2003). For an easier overview, please take a look at Appendix 1 for a ranking of relative importance, based on the presence of these characteristics in the literature. This ranking will also provide an arbitrary basis for picking the most important variables for the developmental state index. It has to be mentioned that the articles were arbitrarily selected from the Economic Literature and Business Source Premier, therefore it is probably a good sample of all the literature available on this topic.

In addition, not all of these characteristics are easy to measure across the large sample of developing countries which is a familiar problem. One very important aspect is the robustness of the data, therefore time series data should be retrieved concerning as many of these factors as possible during the period of 1990 until 2004, after which the average will be calculated.

Unfortunately some of the data were missing, however in almost every case an average could be calculated, even though quite often for less than 15 observations. After evaluating each factor, which is first of all based on the representation in the literature, appropriate weight can be attributed to each one, which will enable the composition of an index. Of course this is not the best way of attributing weight, but it does provide a certain guideline.

The most important factors for which the data could be found across the whole sample

were -sorted according to relevance-; the export rate, as the average value of all exports of

goods and services for the year 1990 up to 2004 as a percentage of GDP, fostering of

industries, as the amount of subsidies as a percentage of total expenditure as an average for

the year 1990 up to 2004, the tax rate as in the amount of tax revenue as a percentage of GDP

from 1990 until 2004, the amount of protection, measured as the customs and import tax

revenues as a percentage of total tax income which is also an average for the period 1990-

2004, followed by inequality using the Gini coefficient which has a range from 0 to 100,

where a high value means a high inequality. Then the saving rate, as the gross domestic

savings as a percentage of GDP average for 1990 until 2004, the import rate average for the

same time period as the value of all imports of goods and services as a percentage of GDP, the

share of manufacturing in GDP average over 1990 to 2004, the average public expenditure on

education as a percentage of total expenditure and finally the foreign shareholding measured

as the percentage of FDI to GDP for the same time period of 1990 until 2004. Unfortunately it was practically impossible to find data on the embeddedness of the developmental state, while the training in technology, just as high public expenditure on health, nationalism, incentives for the banking sector, aid flows and the size of the public sector where only mentioned once in the literature, which is why they were left out of the analysis. Therefore only the 10 variables that were indicated will be used from now on as the main variables of a developmental state. For an additional overview please take a look at Appendix 1,3 and 7.

4.1.2 Correlation DS variables

The extent to which correlation exists between factors is in general very important, since this might lead to wrong conclusions, and low informational value. Nevertheless, this does not apply to the composition of an index , since the variables that are taken into account for the index do not enter a regression analysis individually. Consequently, a correlation would not influence the least squares estimator and significance of any parameters later on.

Still for illustrative purposes a correlation matrix is provided in Appendix 8. There seems to be a relatively high correlation between imports and exports of 0.839907. Furthermore, a moderate correlation was found between fostering of industries and both protection and manufacturing; of -0.467347 and 0.420882 respectively. Protection also had a moderately strong negative correlation with manufacturing; -0.560618. Imports and exports had a correlation of 0.839907which is quite high. This can be interpreted the following way;

countries that have high exports also seem to have high imports, which undermines the assumption that developmental states have high exports together with low imports, since in this sample it would be hard to find a country that has both characteristics. However, by attaching a higher weight to exports because of its relative importance in the weighted approach, this might be overcome. One could also look at the trade deficit or surplus to subdue this problem, however since the relative importance of exports is significantly higher than imports in one of the approaches, this will not be incorporated.

4.2 Methodology - DS Indices

The hypothesis will be tested using four approaches in order to be able to take different points of view on which countries would fall into the “developmental state”

category. In addition, only using one approach might lead to biased results. First of all, an

unweighted index will be calculated, which will rank the countries within the sample based on

the sum of their score on all variables. This score on each variable depends on whether they

are in the lowest, second, third of fourth quartile of the group on that variable, and accordingly they received 1, 2,3 or 4 points. Secondly a weighted index will be calculated, which is similar to the previous approach, however more weight will be applied to the score on more important characteristics in this case. Then a third approach will be used, where a threshold for each characteristic will be designed, and countries that are above or on the threshold for 5 or more characteristics will be labelled as developmental states. The final approach is based on assigning relative scores to each variable, which ranks every observation within the sample after which the sample is divided in two groups based on the sum of all these scores. For all the analyses the data was sorted based on the applicable index or score, using Eviews software, after which two regression analyses were set up, one for the highest scoring half, and one for the lowest scoring half. For all regressions an Ordinary Least Squares Method was used, with the assumption of a non linear relationship between corruption and GDP growth.

Each approach will get a subsection for a more detailed explanation, starting with the unweighted ordinal index. In all situations, the variables import rate, the Gini coefficient and FDI as a percentage of GDP, where low imports, low inequality and little FDI are characteristics of a developmental state, the ranking or score will be reversed, in order to indicate that a low value means a high score on the developmental state scale.

4.2.1 Unweighted and weighted indices

In order to arrive at a certain index, each variable should be given a certain amount of weight, however there is no general rule for this application. First of all an unweighted index will be composed. Then a weighted index will be composed, which attributes the largest weight to exports, since it is the most important characteristic, according to the literature, and the smallest weight to the percentage of FDI. As an arbitrary guideline the number of academics that mentioned this variable in the literature will be added up, and each variable will receive its share in the total as weight. Each variable’s weight can also be found in Appendix 1; Exports received a 7/36 weight, fostering a 1/6 weight, the tax rate, protection and the Gini coefficient received a 1/9 weight each, gross domestic savings was attributed a weight of 1/12, and the import rate, the share of manufacturing, public expenditure on education and FDI all received a 1/18 share, naturally, the total adds up to 1.

The index was calculated using an ordinal scale, which was done by creating 4

categories, and then assigning the highest quartile to category four, and the lowest quartile to

the first category. The higher the category the better, and therefore in case of dimensions for

which a low value characterises a developmental state, the lower the value the higher the category to which that country was assigned on each dimension. One of the reasons for using this approach is that the scores do not depend on the highest values, but only on four categories of small medium and high scores, which makes it rather objective. For the outcome of this assignment, please look at Appendix 12. In this case both an unweighted and a weighted ordinal index was composed, the scores on the weighted ordinal index can be found in Appendix 13, in addition, for a better overview, the overall scores can both be found in Appendix 16 as well. The unweighted index was simply the total of the scores on all dimensions. In the regression analyses, the data were sorted as explained based on their score, and two regressions were set up for either the highest scoring 50% or the lowest scoring 50%

of the sample. In formulae the unweighted index can be seen as;

Unweighted index = Σ (Ordinal score on each variable) Expression 1

Where the ordinal score is either 1,2,3 or 4, and depends on the respective quartile.

The weighted index was composed by attributing the weight -which can be found in Appendix 1- to each ordinal value on each dimension. Again the total was added up, which lead to the weighted ordinal score, on basis of which the data were again sorted and split up before entering the regression analyses. The overall scores for each country can be found in Appendix 13.

In formulae the calculation of the weighted index can be expressed as;

Weighted index = (7/36 * OS Exports) + (1/6 * OS Fostering) + (1/9 * (OS Tax Rate + OS Protection + OS Inequality)) + (1/12 * OS Savings) + (1/18 * (OS Import Rate + OS Manufacturing Share +OS Expenditure on Education

+ OS FDI)) Expression 2

Where OS means the ordinal score on each dimension in the unweighted situation.

4.2.2 Threshold index

In the threshold situation, the mean value in the sample for each dimension was used

as a threshold, and the countries that were either on or above this threshold were given one

point, while the countries below this threshold were given no points on that dimension. Those

countries that received a point on at least five dimensions were labelled as developmental

states in this case, added to the interaction term in the regression model. The idea in this situation is that there are no “more important” characteristics of a developmental state; each dimension plays an equally large role. The points were added up, and this was considered the threshold score, according to which the data were again sorted and two regression models were set. For the overall scores within the sample of this approach please take a look at Appendix 14 and 16. The mean values for each dimension respectively were; Export rate:

35.95% of GDP; Fostering: 37.78% of total expenditures; Tax revenue: 16.29% of GDP;

Protection: 14.33% of tax revenue; Inequality: a Gini coefficient of 42.18 ; Saving rate:

19.42% of GDP; Import rate: 39.55% of GDP; Manufacturing: 19.94% of GDP; School enrolment: 70.79% of the population; Public expenditure on education: 4.02% of GDP and FDI: 3.14% of GDP. For an easier overview of these mean values please take a look at Appendix 7.

4.2.3 Relative ranking index

The third approach assigned each country a grade on each dimension based on their score within the sample. In order to do that the actual value minus the minimum value is divided by the maximum value minus the minimum value, in other words, the total range of this dimension. A consequence of this is that the maximum score for every variable is 1, and the minimum score will be 0. Each country receives a score on each dimension, after which all scores were added up, and added to the interaction term in the regression model find out what effect corruption has in this sample.

In formulae this can be expressed as;

k

– k

k

– k

= Index for observation i of dimension j Expression 3

where i indicates one country of the sample, and j indicates the variable (tax level etc.) The underlying reason for this approach is that this calculation includes all observations, and ranks them within the whole sample. Overall scores from this approach can be found in Appendix 15 and 16.

4.3 Similar Research - Méndez and Sépulveda

When setting up a model to find out what effect corruption has had on economic

growth, it is highly important that the results can be compared to similar research. Therefore

the growth model that was implemented by Méndez and Sépulveda (2006) will be used in this

case too. They implemented a standard, often used growth model, and added some frequently used additional variables. In total this meant that initial GDP, in this case the GDP per capita for each country in 1960, was included, just as the population growth rate, and school enrolment. Furthermore they added investment, political stability and a measure of government expenditure. In order to study the shape of the relationship between corruption and economic growth they added a squared corruption term. Many authors have pointed out in the past that when more and more variables are added to a regression model, the parameter for the investigated variable often becomes insignificant. Therefore, in their investigation, as well as in this one the variables will be added stepwise, which means that first of all 4 variables will be included, then 5 and finally all 7. As pointed out in the literature review, their findings were that in politically free countries a positive optimal level of corruption is growth maximizing, in addition they found that for low levels of corruption it is beneficial, while for high levels it is detrimental for economic growth.

4.4 Growth Model Variables

In this section the focus will be on the growth model variables, with respect to their definitions and sources. Furthermore, endogeneity and the correlation and collinearity between the variables will be discussed.

4.4.1 Description and sources

Since we are trying to find out what effect corruption has on economic growth in developmental state countries, the dependent variable in each regression model is the GDP growth rate in each country. Instead of using one year data, the average value across a time period of 15 years (1990-2004) was calculated in order to control for strong fluctuations.

These data were obtained from the World Development Indicators query of the Economic Library.

Furthermore, data was necessary that covers how corrupt the public sector of each

country is. As pointed out in the introduction, by definition corruption concerns the public

sector only (Transparency International), however there are a number of indices that attempt

to measure this concept. For instance the PSEI, or the public sector ethics index, which is the

percentage of firms in that specific country that give satisfactory ratings (answers 5, 6 or 7) to

questions relating to the honesty of politicians, government favouritism in procurement,

diversion of public funds, trust in postal office and average bribe frequencies for permits,

utilities and taxes, which is published by the World bank. A commonly used index is the CPI

which is a subjective measure based on several sources of how corrupt a certain country is

and annually published by Transparency International. The World bank publishes a number of other indices concerning corruption as well, however in many cases these involved the business level as well rather than only the public sector, and since this investigation only focuses on corruption at the second level, these measures were not added to the dataset.

In general corruption indices are not expected to greatly differ from one year to another, which is why only one year indices were collected. In most cases they have only been composed once for each country anyway. For both indices, higher values are an indication of less corruption, therefore when adding them to the growth model, the indices were reversed in order to see the effect of higher corruption on economic growth. This means that the CPI index will be recalculated according to the following formula:

Reversed CPI = 10 – CPI Expression 4

Since the CPI index has a range from 0 up to 10, for the PSEI index this means that the reversed PSEI is recalculated as follows;

Reversed PSEI = 100 – PSEI Expression 5

Since the PSEI index has a range from 0 up to 100.

Even though Méndez and Sépulveda (2006) and most other academics in the corruption field used the raw CPI data, these newly calculated indices will make the interpretation of the results easier, since a higher index in this case means an increase in corruption. As can be seen in Appendix 5, there is a strong correlation between both indices, which was expected since they both measure the same concept. In addition, in Appendix 6, two scatter plots with regression were set up in order to find out what the relationship between the reversed corruption terms and the GDP growth rate is. For both indices an higher corruption is associated with a lower economic growth.

In order to find out what the shape of the relationship between corruption and economic growth was, in other words if the effect was either linear or non-linear, Méndez and Sépulveda added a squared corruption term as well. Initially this investigation will also include this variable, to stay comparable.

The standard growth model variables that were also used by Méndez and Sépulveda

were the population growth rate, school enrolment, initial GDP per capita, capital formation,

government expenditure and political stability.

Initial GDP was in this research captured by also taking the GDP per capita in 1960, measured in 1994 dollars, which were found in the Penn World Tables online. For the population growth rates, and total public sector expenditures as a percentage of GDP an average was calculated again for the time period 2000 up to 2004. When considering the overall school enrollment, secondary school enrollment was chosen as an indicator rather than tertiary school enrollment, since data on the latter were not available for the whole dataset. In addition, there is still a lot of variance in this secondary enrollment ratio across the sample, which probably still makes it a good explanatory factor.

Investment can be captured by looking at gross capital formation. According to the World Bank’s definition, gross capital formation consists of outlays on additions to the fixed assets of the economy, net changes in the level of inventories, and net acquisitions of valuables, where fixed assets include land improvements (fences, ditches, drains, and so on);

plant, machinery, and equipment purchases; and the construction of roads, railways, and the like, including schools, offices, hospitals, private residential dwellings, and commercial and industrial buildings. Inventories are stocks of goods held by firms to meet temporary or unexpected fluctuations in production or sales, and “work in progress.”

Furthermore, a measure of political stability was added as well, just as measure of government consumption. The political stability index was based on a mean value of zero, ranging from -2,5 until 2,5, where a higher value means more stability, which was found on the World Bank website as well. Government consumption as a percentage of GDP covered the government consumption variable in this case, once again the average value across a time period of 15 years was calculated from the WDI query.

Data on all growth model variables can be found in Appendix 4 and descriptive statistics for both the developmental state and the growth model data can be found in appendices 7 and 9 respectively. An overview of the data sources can be found in Appendix 2.

4.4.2 Endogeneity

Méndez and Sépulveda pointed out that using average values for all variables for the whole sample outweighed the need for a control variable, since average values across time periods control for endogeneity which is due to certain political shocks or higher growth rates.

In addition, they argue that an often used control variable -ethno linguistic fractionalization- is in fact highly correlated with economic growth (Easterly and Levine, 1997), which makes it an unreliable instrument.

4.4.3 Correlation and collinearity growth model variables

In addition to the correlation between the developmental state variables it is also highly important to check the correlation and collinearity between the explanatory variables in the growth model; since this would influence the least squares estimators and the standard errors of the regression model (Hill, Griffiths and Judge, 2001). What they point out is that if there exist nearly exact linear relationships between some of the variables, as a result some of the variances, co-variances and standard errors of the least squares estimators may be very large, which in turn leads to insignificant parameters, and sensitivity to additional variables.

As can be seen in Appendix 10 there is no strong -as in larger than 0.80- correlation between the variables. Moderate negative correlation was found between political stability and the reversed CPI and between political stability and the reversed PSEI; -0.677654 and - 0.581723 respectively, higher political stability goes together with lower corruption and vice versa. Furthermore the population growth rate and secondary school enrolment were also moderately negatively correlated; -0.597294. In addition, the reversed CPI and the reversed PSEI as expected have a relatively strong correlation of 0.870738 , which was already uncovered in Appendix 5, however still a separate regression model will be composed for each different measure in order to stay objective.

Since pairwise correlation is only a simple way to look at sample correlation, a

collinearity check has to be performed as well. The benefit of this procedure is that

involvement of more than two variables can be detected as well. In order to identify to what

extent collinearity between the variables existed, a number of auxiliary regressions were set

up. This implies that a number of least squares regressions were arranged, with one of the

explanatory variables as the dependent variable, and all the remaining variables as the

independent variables. The indication of collinearity is the R² for the model, if this value

happens to be larger than 0.80, there is evidence that collinearity in fact exists, and that a large

part of the variance of the dependent variable in this analysis, is explained by the variation in

the other explanatory variables of the model. With respect to Appendix 11, this means that if a

regression analysis would be carried out with both a single and a squared corruption term, for

all approaches, the parameter in front of the single corruption term could not be estimated,

because an almost exact linear relationship would exist between these terms. If this is the

case, the implication is that there standard errors and variances are very high, which leads to

unstable coefficients, and unreliable information. Since the single corruption term is of the

highest importance, and only a squared corruption term would not make sense, since then a

single term would again be necessary as well, the squared term had to be dropped. For an overview of the results of this analysis please take a look at Appendix 11.

4.5 General Regression Analyses

Before setting up the Ordinary Least Squares regression analyses both Korea and Chad were eliminated from the sample due to too many missing values. Therefore the total sample contained 50 countries.

Méndez and Sépulveda (2006) pointed out the importance of adding all variables in a three step fashion. The most important variables were added initially, which were the population growth rate, school enrolment and initial GDP per capita in 1960. In the second step capital formation or investment was added, and finally the government expenditure measure and political stability were included as well. What underlies this approach is the fact that in their own just as in similar large studies, for instance by Mauro (1995) and Li et al (2000) the corruption parameter becomes insignificant when more and more independent variables are added. The restricted model then provides some insight into the workings of corruption anyway.

Unfortunately the collinearity check of Appendix 11 pointed out some strong collinearity between the reversed CPI/PSEI and the other variables, which was only eliminated once the squared reversed CPI/PSEI terms were dropped from the equation.

Therefore, unlike the analyses of Méndez and Sépulveda this analysis will be carried out without the squared corruption term, in order to reach some unbiased and significant results.

The OLS estimation is in general very sensitive to outliers. Therefore first of all the economic growth rate was analysed for outliers, defined by values more than 3 standard deviations from the mean. This implied that all values should be between -4.602 and 11.092, which was the case. However, it has to be mentioned in the data collecting phase it became evident that Bosnia had an extremely large growth rate during the 1990’s, which might lead to a biased estimation. Thus, the two years with the highest values were eliminated while calculating the average growth rate during the 1990-2004 period in order to obtain a more

‘customary’ growth rate.

Furthermore the corruption indices should be analysed for outliers as well, which

implied for the CPI that the scores should lie between 1.695 and 11.505, and for the PSEI

respectively between 18.905 and 117.895. Unfortunately, Singapore, the least corrupt country,

has a score outside the range for both indices. Still, this value will be included in the analysis,

since it is first of all a country that is pointed out in the literature as one of the absolute

developmental states, and secondly because it is a measured score on a predefined range of a dimension that does not strongly fluctuate over time, like for instance economic growth. This indicates that it is not an error or a short term fluctuation.

4.6 Interaction Term Analysis

This section can be considered a bonus, since it involves an additional analysis, using an interaction term instead of splitting up the sample.

4.6.1 Interaction term

Even though the main emphasis in the methodology part has been on the Méndez and Sépulveda approach, it has to be pointed out that according to Brambor, Clark and Golder (2005) every hypothesis that is restricted to a certain condition should include an interaction term. This additional variable (X*Y) fits the purpose of the investigation quite well since it captures the composed effect of variable X on the dependent variable, given the value of variable Y. In this situation, instead of splitting up the sample an interaction term of the developmental state score and reversed corruption would be created which would capture the effect of corruption on economic growth as the developmental state score increases. Another strong point of this approach is the fact that the complete sample can be included rather than splitting it up, which leads to more robust results. This analysis will be included as a bonus in order to see to what extent the results of the main analysis hold when employing this methodology as well. A common mistake that is made by scholars is not adding the constitutive terms as separate variables, in other words, the terms that are used in the interaction term itself (Brambor et al., 2005). The main justification that is given is usually that the factor is not related to the dependent variable, however, leaving it out, as Brambor et al. argue leads to biased results.

4.6.2 Collinearity interaction term variables

Again, the variables that are included have to be analyzed for collinearity for the same purpose as pointed out in the previous section. Therefore a number of auxiliary regressions were set up with initially only the reversed CPI and the reversed PSEI as dependent variable, and the others as independent variables. The reasoning behind this was that at least the single corruption term has to be included in the variables, following Brambor et al.’s logic of including the constitutive terms. The results of this analysis can be found in Appendix 23.

Unfortunately the R-squared was almost 1 in all instances. After dropping the squared

corruption term, the R-squared was still higher than 0.90 for all regressions. Once the DS score for each approach was dropped as well, the R-squared for the Threshold approach became sufficiently low (<0.80) for both the reversedCPI and the reversedPSEI regressions to be able to set up a full model regression analysis. In case of the Relative Ranking, the political stability variable also had to be dropped before a full regression analysis became an option.

Dropping more variables would have been an option for the R-squared of the other approaches to become sufficiently low, but this would be at the cost of comparability to the main methodology.

As a result, only the Threshold approach and the Relative ranking approach provided the opportunity to set up a reliable regression analysis with an interaction term. Unfortunately, a consequence of the high collinearity was that the DS score had to be dropped as a variable, which violates Brambor et al.’s checklist of setting up an unbiased interaction term regression.

Perhaps we have found the reason why so many scholars did not include the interaction term variables as single variables either. Nevertheless, the full regression results can be found in Appendix 24 and 25 for the Threshold approach and the Relative ranking respectively, and will also be discussed in the following section.

5 RESULTS

This chapter will be divided into three subsections, in which first of all the developmental state scores will be discussed, secondly the regression results will be summarized, and subsequently, a discussion of the similarities and differences between these results and the findings of Méndez and Sépulveda will follow.

5.1 Developmental State Indices

The individual scores in each approach, the unweighted and weighted ordinal ranking,

the threshold approach and the relative ranking can be found in Appendix 12 up to 15. In

addition, Appendix 16 provides an overview of all the scores and indices in order to enable an

easier comparison. By merely taking a glance at the developmental states in each approach –

which are indicated by the bold capital d-, it is obvious that this group rather changes in

composition from one approach to the other. This supports the idea that taking different points

of view with respect to what defines a developmental state, does in fact change the outcomes

as well, and it might lead to more informed conclusions. However, overall the emphasis

seems to be on both Asian and Eastern European countries, and countries in the Middle East

are well represented in the developmental state group very often as well. Not every approach made it possible to split the sample equally due to similar scores. For instance, in case of the Unweighted Ordinal Ranking approach, 30 countries fell into the developmental state category due to the large group that had at least a score of 25. Similarly, in Threshold approach, 33 countries had a score of 5 or higher, which all had to be assigned to the developmental state group. Furthermore, the composition of the DS group in the relative ranking approach was rather unrepresentative of the concept. After all, the literature defines developmental states as a typically Asian phenomenon, while this approach included the least Asian countries; only four, while in both the weighted and unweighted approach almost the whole country region was figured in. In addition, the threshold approach showed a similar partition, with only 5 Asian countries included. Therefore, after individually going over the results, the main emphasis will be on the unweighted and the weighted approach.

5.2 Results General Regression Analyses

Initially a complete sample regression model was set up for both corruption indices without making a distinction based on any developmental state index. The outcome can be found in Appendix 17. In practically all of the regression analyses corruption had no significant impact on economic growth, except in the four-variable PSEI model, where it significantly negatively affected economic growth. However, this effect became insignificant once more variables were added. Overall, the population growth rate and capital formation were the factors with a significant positive impact on economic growth, which even held once more variables were included in the regression model. Still, the sum of the squared residual of these regressions was quite large, while the adjusted R-squared was only ranging from 0.22 up to 0.40. Therefore the informational value is not very high.

Once the data were split in developmental state countries and non developmental state countries based on the unweighted ordinal ranking approach a relatively clear difference could be discerned between both groups (See Appendix 18). However, not in favour of the hypothesis, since in case of the developmental states the reversed corruption term was significantly negatively associated with economic growth for both corruption indices for both the 4-variable and the 5 variable-growth models. Despite the fact that both terms became insignificant once government expenditure and political stability were added it was quite a clear result. Moreover, the non-developmental state countries apparently were not affected by corruption at all. Still the sum of the squared residuals was relatively high in this approach.

The R-squared has improved when compared to the complete sample regression; it ranged

from 0.37 up to 0.44 for the DS countries and from 0.24 up to 0.39 for the non-DS countries.

The only variable that remained significant during the process of adding variables for the developmental state countries was the secondary school enrolment, even though with a negative parameter. In the non developmental state countries both the population growth rate and capital formation played a large positive role in economic growth.

Furthermore the sample was split up based on assigning more weight to some of the developmental state characteristics and again adding up the overall scores, in other words the weighted ordinal approach (See Appendix 19). In this case, a similar result came up. Again, for the developmental state countries, the corruption parameters were both negative and significant for the 4- and 5-variable models. School enrolment was again the only variable that played a large, however negative role for economic growth in these regressions. The adjusted R-squared was rather high; between 0.55 up to 0.60, which is relatively high, when compared to the previous regressions. For the non developmental state countries both corruption indicators did not seem to have an impact again, while instead of the population growth rate, initial GDP became significantly negative for both 4-variable models. Capital formation remained the significant influential factor throughout these regressions as well.

However, it can not be said that these non-DS regression models fitted the data very well since the adjusted R-squared only ranged from 0.14 up to 0.25.

The third approach was the threshold approach, were countries were assigned a point

on each dimension for which they had a value higher (or smaller if a small value indicated a

developmental state characteristic) than the mean value on that dimension. What came up

once the countries with a score equal to or larger than 5 were added to the regression model

was again that corruption had a negative impact on economic growth for DS countries and no

impact on non-DS countries (See Appendix 20), however only in two CPI regressions this

was significant and then even only at a 10% confidence interval. Furthermore, only the

population growth rate had a positive impact, for both groups except in case of the 7-variable

CPI model. In addition, school enrolment became significant and positive at this instant in the

non DS countries, for the 5- and 7-variable models. Initial GDP had a significant negative

impact in the 4- and 5- variable models for the PSEI index, and Capital formation stayed

significant in the non-DS group as well. When looking at the fit of the model it became clear

that the adjusted R-squared was not very high for the DS countries; only ranging from 0.24 up

to 0.31. Therefore the informational value for the DS regressions can not be considered very

high in this approach. Instead, the adjusted R-squared for the non-DS countries became rather

high; from 0.19 up to 0.62. However, as mentioned in the previous sub section, the group of

DS countries was not very representative of the concept, which is why these results will not be further taken into account.

The final approach was the relative ranking (See Appendix 21). A downside is that each country’s score in this case depends on both the highest and the lowest scoring countries, however, not one single approach is perfect. The sample in this case could be split up in two equal groups, which eliminates part of the problem with the relative aspect of this approach, since still the highest scoring countries were distinguished this way even though the differences might be very large. When looking at the regression results for the developmental state group it became evident that in this case all corruption parameters became insignificant, for both the CPI and the PSEI, and for both the DS- and non-DS countries. Yet the population growth rate parameter stayed significantly positive in the DS countries this time. For the non- DS countries only capital formation and government expenditure had a positive impact. The adjusted R-squared was rather high; around 0.54 for the DS countries, but in the 4-variable model of the non-DS countries this value even became negative. However, like in the previous approach, the results will not receive further attention due to the fact that the DS group is not representative of the DS concept.

In general the absolute values for the CPI parameter were higher than the PSEI parameter; however, the PSEI score for each country is based on a score from 0 to 100, while the CPI is only based on a score from 0 to 10. Consequently the parameter should be ten times smaller for the PSEI in order to compensate for the effect of the higher corruption score in this category.

Overall, based on the adjusted R-squared values, the weighted ordinal approach seemed to fit best; in addition, the relative ranking regressions also portrayed a strong fit.

However, when looking at the composition of the DS group, only the unweighted and the weighted approach appeared to be representative. Contrary to the hypothesis, corruption was in general -based on the unweighted and the weighted approach- negatively associated with economic growth in developmental state countries while it was not associated with economic growth in non developmental state countries. Unfortunately the parameters became insignificant once a 7-variable model was constructed. However, this is apparently a well known problem in growth model analyses.

Furthermore, different variables seem to play a significant role in economic growth in

both groups. While in the DS groups (in both the unweighted and the weighted approach) the

emphasis seems to be on corruption, and secondary school enrolment, the attention shifts to

the population growth rate and capital formation (in the unweighted approach) and initial