• No results found

The effect of corruption on economic growth in least developed countries

N/A
N/A
Protected

Academic year: 2021

Share "The effect of corruption on economic growth in least developed countries"

Copied!
34
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

University of Amsterdam

Faculty of Economics and Business

Bachelor thesis for BSc Economics and Business Economics Track: Economics

The effect of corruption on economic growth in least

developed countries

Author: Martin Peˇceˇn´ak

Supervisor: Dr. P´eter F¨oldv´ari

(2)

This document is written by Martin Peˇceˇn´ak who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

Using panel data on the least developed countries from a 2005-2018 period, this study estimates a dynamic panel estimator to evaluate the effect of corruption on economic growth, and FE/IV estimator to explore its different transmission chan-nels. Whilst the impact of corruption on economic growth was estimated in various empirical studies since the data on corruption became widely accessible, they sel-dom focused explicitly on the least developed, most corrupt countries. I find that the effect of corruption on economic growth in LDCs is substantially higher than previously reported - a 10-unit increase in CPI score results in a 2.5% increase in GDP growth. The effect is mainly transmitted through the investment channel, which accounts for 31% of the total effect. The findings suggest that the marginal impact of corruption is more severe in the most corrupt countries.

(4)

Contents

1 Introduction 1

2 Theoretical Framework 3

2.1 What determines economic growth? . . . 3

2.2 What is Corruption? . . . 4

2.3 Greasing the Wheels . . . 5

2.4 Sanding the Wheels . . . 6

2.5 Filling the Gap . . . 7

3 Methodology and Estimation Method 7 3.1 Standard Model . . . 7 3.2 Indirect Effect . . . 8 3.3 Estimation Method . . . 9 3.3.1 Growth equation . . . 9 3.3.2 Transmission channels . . . 11 4 Data 11 4.1 Data Description . . . 12 4.1.1 Independent Variable . . . 12

4.1.2 Corruption Perception Index . . . 12

4.1.3 Transmission Channels . . . 13

4.1.4 Control Variables . . . 16

4.2 Data Exploration . . . 16

5 Empirical Results 18 5.1 Growth Equation Regression . . . 18

5.2 Transmission Channels . . . 20 5.2.1 Investment Channel . . . 21 5.2.2 Human Capital . . . 22 5.2.3 Trade Openness . . . 22 5.2.4 Government Expenditure . . . 23 5.2.5 Political Stability . . . 24

6 Total Effect of Corruption 24

(5)

1

Introduction

The general economic effect of corruption has been assessed numerous times in academic literature. Mauro (1995) used newly available data to discover its detrimental impact in decreasing the effectiveness of the government or undermining its legitimacy, influencing the size of private investments; and thus, growth. Building on his findings, Mo (2001) observed the direct and indirect effects of corruption through its transmission channels. He claimed that corruption adversely affects human capital and contributes to increased political instability, therefore indicating a possible corruption trap. Whilst well educated citizens impede the growth of widespread corruption in governmental institutions, subse-quently increasing their efficiency, the lack thereof, does the opposite (Svensson, 2005).

An alternative view was initially represented by Leff (1964), who claimed corruption could serve as a potential cure to bureaucratic inefficiency as it allows for albeit expensive, but faster achievement of goals. Additionally, it functions as an extra incentive to increase the productivity of officials. More recently, M´eon and Weill (2010) presented empirical findings on the positive effects of corruption in poorly governed countries, nonetheless warning about potential long-term detrimental effects of an uncontrolled spread of cor-ruption.

Governments of the least developed countries (LDCs) are often inefficient and sub-ject to untimely cabinet changes or political violence, with statistics claiming that the worst performing countries regarding corruption perception were indeed LDCs (Trans-parency International, 2020). To illustrate, in Sudan, 67% citizens were exposed to cor-ruption in 2011. Furthermore, from 1998 to 2002, over four billion USD - 26% of Angola’s GDP - got lost on their way from the Angolan treasury. These are countries with average schooling years of 3.7 and 5.1, respectively.

(6)

Corruption may be especially damaging to LDCs through an exacerbation of already present societal issues, ultimately furthering the development gap. However, mainly due to a lack of data, they are seldom a focal point of academic studies. Whilst Gyimah-Brempong (2002) empirically supported the adverse impact of corruption, he did not provide a detailed account of how corruption transmits in developing countries. Therefore, in this paper, I aim to discuss the extent of corrup-tion’s impact on economic growth and its possible transmission channels in the least developed countries.

Whilst Mauro (1995) claimed that corruption harms investments in physical capital Papyrakis and Gerlagh (2004), using the framework of Mo (2001), added four other transmission channels of harmful impact of corruption. I will add a government expenditure channel, as Barro (1991) reported a negative effect of government consumption on growth. Furthermore, corruption due to its inherent illegality often needs to be conducted by the public sector, that has power over legal control mechanisms. Moreover, governments are responsible for investments of substantial propor-tions, which in connection with weak controlling authorities, attracts corruption. Hence, the effect of corruption on government expenditure may play a significant role in corruption transmission. Therefore, in total I will analyze five possible mechanisms of corruption transmission, including investment in physical capital, trade openness, political stability, human capital, and government expenditure. Based on previous research, I expect the relationship to be negative for all transmission channels but for the latter, for which the effect remains ambiguous.

To conduct an empirical growth analysis, I will use a dynamic panel model which allows me to estimate possible growth regressions without undesired effects provoked by the presence of dynamics and endogenous regressors. Moreover, as I lack suitable instrumental variables, a dynamic panel model allows the predetermined variables to act as such. Later, as the models used to estimate the effect of corruption on transmission channels are not dynamic, I will use FE/IV model to account for possible endogeneity of corruption.

I focus on LDCs because of their vulnerability to corruption stemming from their weak in-stitutions, as a large number of countries experience political instability, a lack of political rights, low educational attainment, or under-investments in crucial areas. Currently, Africa is subject to economic restructuring and a large influx of foreign aid. Thus, their inefficient implementation may waste resources, deepen an already large development gap, and further fuel the public discourse on the negative effects of development aid efforts. Thus, if corruption proves to enhance such eco-nomically debilitating problems, correctly tailored policies may prove to be more valuable than in developed economies.

The following section introduces working definitions and a literature review. Section three presents the model used for empirical analysis. Section four introduces data and presents summary statistics and correlation matrix. Section five discusses empirical results and the transmission mechanisms of corruption, while section six presents an overview of a total effect of corruption. Finally, the last section concludes the paper.

(7)

2

Theoretical Framework

2.1

What determines economic growth?

The Solow (1956) model, which introduced basic determinants of economic growth, is often a base of empirical studies on economic growth. The model is displayed in equation (1).

Y (t) = K(t)α(A(t)L(t))α−1 (1)

Solow (1956) argued, that income of a country (Y) is determined by a share of physical capital (K), share of labour (L), and the level of technology (A), while the latter two were considered exogenous. He introduced the idea of conditional convergence by asserting that countries sharing the same characteristics are on the same growth path; and hence, achieve the same steady state. While on the path to the same steady state, the poorer countries should grow faster, indicating that the level of initial income has a negative effect on temporary economic growth. Nonetheless, as Solow (1956) assumed that there is no long-term economic growth other than that induced by exogenous technological progress, he did not solve the issue of determinants of long-term economic growth.

Mankiw, Romer, and Weil (1992) argued that the model was suffering from omitted variable bias; thus, produced coefficients for growth determinants that were too large. They stressed that human capital is an important determinant of economic growth, as by its inclusion, their model explained approximately 80% of economic growth. While they agree, that differences in savings, population growth and education explain the differences in GDP per capita, they challenge the exogeneity of variables in Solow model and provide some potential determinants of the variables, such as differences in policies or political stability.

Barro (1996b) furthered the discourse on determinants of economic growth and offers additional variables, such as trade improvements, rule of law or schooling years that are able to explain the differences between countries’ incomes.

Finally, Islam (1995) opted for the use of panel data and estimated the models using both dynamic and static panel data estimators. Such an approach allowed accounting for country differences, which were in previous cross-section studies omitted.

While I base my research on Solow (1956), I follow the methodology of Islam (1995), and I add a limited list of variables proposed by Barro (1996b), of which corruption is the focal point of my analysis.

(8)

2.2

What is Corruption?

The ambiguity of the term often poses issues when modeling the impacts of corruption. It may take various forms and differ in its extent or origin; however, it is usually a case of abuse of entrusted power for private gain (Jain, 2001). Even though corruption is not specifically present in public sector, the survival of illegal activities within the private sector that do not directly concern public officials often depend on the cooperation of public institutions (Svensson, 2005).

Jain (2001) recognizes three types of corruption that differ on the type of bribe-seeker and the source of their power. The first type of corruption, marked by its presence within the highest dimensions of society, is characterized by highly influential government officials acting in opposition to general civic interests. Rulers divert the country’s funds to areas that can yield the highest revenue to them personally (Della Porta & Vannucci, 1997). To illustrate, Equatorial Guinea invests around 80% of its annual budget into infrastructure projects, many being a lavish depiction of irresponsible resource management. In com-parison, their education sector, reflecting decreasing investments, releases children after nine years of schooling, three years earlier than other comparable economies. Moreover, statistics show that they die ten years earlier than their counterparts in countries with similar levels of development (IMF, 2019). Nevertheless, a lack of explicit bribes impedes on the measurement and detection of this type of corruption. In addition, government leaders have, especially in non-democracies, influence over anti-corruption institutions. This ultimately furthers the issue and allows agents to escape legal consequences.

Second, petty or bureaucratic corruption is present in any regulated procedure of product and service acquisition (Jain, 2001). It is a luxurious advantage that allows for the faster execution of transactions, a necessary fee to survive in overly bureaucratic public institutions, or a method to obtain services officially unobtainable. Bureaucrats victimize citizens by requiring additional rents for services that they are entitled to. Al-ternatively, citizen induced bribes may diminish legal consequences for illegal activities. Hence, bureaucratic corruption can be seen as a two-way mechanism revealing institu-tional deficiencies and possibly over-regulated procedures.

Third, legislative corruption concerns bribing the public to enact desired parties into official institutions. The briber is subsequently able to profit from abusing the power of the favored party in executive positions. Therefore, vote-bribing may fall under this category (Jain, 2001).

Individuals participate in corrupt activities solely when the benefits outgrow the risks. Hence, the presence of corruption is argued to depend on the size of the bribe

(9)

relative to the subject’s usual income, in addition to a presumed efficiency of controlling institutions. Furthermore, Ades and Di Tella (1999) proved that countries with low levels of competition are more corrupt, be it due to the presence of poor competition regulation or limited entries of foreign investors. With lower competition, private entities enjoy more substantial revenues; thus, to stay in a non-competitive environment, they are more willing to bribe controlling bureaucrats. Treisman (2000) confirmed five other theoretical determinants of corruption, including democracy or institutional efficiency. Democracy has no significant instantaneous effect on corruption; nevertheless, long-term exposure to democracy decreases corruption. Press freedom, a lower persecution rate for non-cooperative citizens, or increasingly efficient institutions impose greater risks on bribe-taking and make the consequences of corruption more severe.

2.3

Greasing the Wheels

Leff (1964) was one of the first to hypothesise about corruption’s beneficial ’grease the wheels’ effect, and a year later, Leys (1965) wondered what the problem about corruption is. He referred to potential positive impact of corruption on inefficient institutions by providing the officials an additional incentive. Lui (1985) illustrated this mechanism on lower queue times, specifically on the example of administrative services. According to him, officials are eager to maximize their income from bribes by minimizing the time spent on individual subjects in the queue, ultimately accelerating the administrative process. Furthermore, he claimed that the possibility of bribe-taking increases the total income of bureaucrats, which may attract workers of higher quality that because of low base wages would not be interested in similar line of work. Nye (1967) expanded on the inefficient institution theory by arguing that corruption may alleviate losses resulting from inefficient discriminatory regulations. Bribe-seeking bureaucrats are more inclined to disobey official guidelines which consequently lessens the negative effect of disadvantageous regulations. Beck and Maher (1986) explored parallels between bribery and bidding. For instance in a business environment, where a license is necessary for firms to function in a market, companies compete with each other to acquire said license. Similarly in the bribing market, firms willing to forgo the highest costs of corruption, thus the most efficient ones, obtain the license. Therefore, corruption can be seen as a sorting mechanism through which only the most efficient firms with the highest quality of investment prospects earn the right to exist in a market.

Ehrlich and Lui (1999) concluded that corruption in autocracies, where leadership is intertwined with administration, may incentivize rulers to minimize losses resulting from

(10)

corruption. This is in contrast to democracies, where governance is substantially more decentralized, bureaucrats are not as concerned about overall productivity and thus more prone to corruption.

Finally, M´eon and Weill (2010) presented empirical evidence for the ’grease the wheels’ hypothesis regarding efficiency, supporting the arguments above. Nonetheless, they warned about the adverse effects of letting corruption run its course. Furthermore, arguments of Ehrlich and Lui (1999) solely illustrate how corruption may prevent its own negative effects, inadvertently indicating the possible drawbacks of graft. Hence, in the next section I will clarify the disadvantages of corruption by providing counter-arguments to above-specified claims.

2.4

Sanding the Wheels

Using the queuing model of Lui (1985), Kurer (1993) contrastingly argues that bureau-crats taking bribes have no incentive to accelerate the administrative process. If the original queuing model holds, the increased efficiency and hastened procedures decrease waiting costs. Nonetheless, in order to provoke bribes, bureaucrats aim to preserve high waiting costs and distort the process. Even if corruption would increase the efficiency of bribe-seeking clerks, it would solely diminish the negative effects of artificially inflated waiting times. Therefore, the efficiency of administration tends to decrease, or at most stay the same, rather than increase.

Rose-Ackerman (1997) challenged the efficiency of bribing auctions for licences or government contracts. She argued the auctions do not necessarily prioritize the most efficient firms, but rather firms with compromised ethical values, or those connected to the government. A firm with scruples, however efficient it is, may refuse to participate in corrupted auctions. For instance, winning firms may be less concerned about the quality of their products or services, and hence are willing to offer higher bribes at the expense of consumers. The quality of public investments suffers as well, as corruption tends to divert investments into less efficient sectors where opportunities for bribes are higher.

Nur-Tegin and Czap (2012) concluded that autocracies are usually more corrupt than democracies, disproving the arguments of Ehrlich and Lui (1999). Assiotis and Sylwester (2014) further the anti-corruption narrative by empirically demonstrating the substan-tially higher negative effects of corruption in autocracies rather than in democracies. Hence, whilst corruption may serve as a self-destructive mechanism in non-democracies, its impacts are still disruptive.

(11)

on the topic, the effect of corruption seems to be dependent on different research methods. Moreover, only a few focused specifically on the LDCs and none presented a detailed analysis of corruption transmission.

2.5

Filling the Gap

The increased availability of corruption measures have contributed to a surge of empirical analyses regarding the impact of corruption on economic growth. Mo (2001) created a framework to analyze both the direct and indirect effects of graft, whilst emphasizing the transmission of the latter through political stability and investment channels. Research regarding the combined effects of transmission channels managed to explain over 80 per-cent of the total impact of corruption. Pellegrini and Gerlagh (2004) followed the same framework, adding trade as an additional transmission channel and found approximately the same results. While both analyses were conducted using the 2SLS regression tech-nique, accounting for the endogeneity of the growth model, none of the previously cited studies consider the dynamic nature of the economic growth equation.

As LDCs are often institutionally troubled, bribing or using one’s network of con-nections may be one of the few efficient means to survive in such a dysfunctional admin-istration. While it is unreasonable to consider Africa - the origin of 70% of LDCs - as ’genetically’ corrupt, corruption follows the logic of gift-giving or negotiation embedded in the culture of a substantial part of the continent (Olivier de Sardan, 1999). The sudden rejection of such cultural norms could ignite new, or fuel already existing cases of political instability.

Whilst Gyimah-Brempong (2002) rooted the adverse impact of corruption on eco-nomic growth in African countries, accounting for both endogeneity and dynamics of the economic growth equation, he did not provide an overview of transmission channels. Thus, the mechanisms of corruption in LDCs remain ambiguous and under-researched.

3

Methodology and Estimation Method

3.1

Standard Model

I estimate the effect of corruption on economic growth using an extended conditional convergence growth equation proposed by Solow (1957), which was further empirically proven and augmented by Mankiw et al. (1992), and which Islam (1995) adapted to panel data, the data format this analysis will be based on. Moreover, some other variables

(12)

proposed by Barro (1996b) are going to appear in the model. The model to be estimated is following:

yi,t= β0+ β1Yi,t−1+ β2Ci,t+ β3Xi,t+ β4Zi,t+ µi,t+ i,t (2)

The dependent variable (yi,t) denotes annual economic growth, which is represented

as a logarithmic change in income per capita for observed periods. The initial GDP level in the preceding period (Yi,t−1) is included as the first independent variable. If

conditional convergence holds, I expect the effect of initial level of real income on the economic growth to be negative. Corruption is a focal point of my analysis, and occurs as a second independent variable in the regression. Vector X includes explanatory variables often seen in traditional economic growth analyses: investments (Inv), human capital (HC), trade openness (Open), governmental expenditure (GE) and political stability (P S). Vector Z contains control variables that are not a focus of my research, but may have a significant effect on the dependent variable: population growth (P op), contribution of natural resources rent to economic growth (N R) and political rights (P R). Error term is denoted by , and the unobserved country and time fixed effect is represented by µ. Subscripts i (i=1, ..., N ) and t (t=1, ..., T ) represent country and time, respectively. Introduction to the above mentioned variables is in section 4.

3.2

Indirect Effect

Corruption may influence economic growth through direct and indirect channels. In my analysis, indirect channels consist of investment, human capital, trade openness, political stability, and governmental expenditure channels. As those channels may have an effect on the economic growth, the impact of corruption may be transmitted through them. The general equation for transmission channels may be found below:

Chi,t= β0+ β1Yi,t−1+ β2Ci,t+ β3Zi,t+ µi,t+ i,t (3)

Dependent variable (Ch) denotes an individual possible transmission channel. First independent variable is an initial real GDP per capita (Yi,t−1) in preceding period, which

is followed by the focal variable Corruption (C) and a vector Z of control variables which is identical to the vector Z in (2). Error term is denoted by () and the unobserved country and time fixed effect are represented by (µ). Subscripts i (i=1, ..., N ) and t (t=1, ..., T ) represent country and time, respectively.

(13)

3.3

Estimation Method

3.3.1 Growth equation

Empirical literature on the effect of corruption on economic growth was predominantly built upon the usage of a cross-sectional regression. Following the methodology of Mauro (1995), Mo (2001) observed these effects throughout the years 1970 to 1996 using 2SLS technique, instrumenting corruption with ethnolinguistic fractionalization. Cross-section regressions produce biased estimates if included independent variables are endogenous, which often proves to be the case in growth regressions. Whilst 2SLS accounts for en-dogeneity problems, it is biased when omitted constant variables are not included in the regression equation or if the regression equation includes dynamics (Bond, Hoeffler, & Temple, 2001).

Even though panel data models avoid the issue of omitted constant variables by including fixed country and time effects, they also suffer from bias due to endogeneity and dynamics, by which, according to Caselli, Esquivel, and Lefort (1996), growth equations are intrinsically characterized. Moreover, it eliminates long run variations, which may be useful when estimating dynamic growth models. To avoid the previously mentioned issues, I decided to estimate the equation using a dynamic panel model that performs well in the presence of dynamics and is consistent when variables are endogenous.

Arellano and Bond (1991) proposed a first-difference GMM estimator that addresses the previously mentioned issues. First, to remove the omitted variable problem arising from country and time-specific effects, it first differences the equation.

∆yi,t = β0+ β1∆Yi,t−1+ β2∆Ci,t+ β3∆Xi,t+ β4∆Zi,t+ ∆i,t (4)

Second, while assuming no serial correlation of error terms in equation (4), the lagged values of independent variables may be used as instruments for their previous values. The levels of endogenous variables lagged two or more periods serve as valid instruments for the first-difference equation due to their correlation with the values of previous endogenous variables, rather than with the error term. Moreover, as predetermined variables are not correlated with current error term values, they may be instrumented by their values lagged one or more periods. In contrast, exogenous variables uncorrelated with the error term are instrumented with their current values.

Nevertheless, Bond et al. (2001) have shown that the lagged values of variables are weak instruments for their subsequent values under the conditions of persistent times series and relatively small amount of periodical observations hence, making Difference GMM a potentially poor estimator for finite samples. Ergo, I will use a System GMM

(14)

estimator developed by Arellano and Bover (1995) that solves the issue of finite samples experienced by Difference GMM.

System GMM operates with two equations. Similarly to Difference GMM, it first creates a difference equation showcased in (4). Afterwards, it estimates the level equation (5) and thus avoids the elimination of cross-country and time variances, ultimately solving another drawback of Difference GMM.

yi,t = β0+ β1Yi,t−1+ β2Ci,t+ β3Xi,t+ β4Zi,t+ µi,t+ i,t (5)

Nevertheless, to allow the use of lagged two or more periods first-differences as instruments in level equation, System GMM requires an additional assumption which would specify the stationarity of the independent variable. Whilst this is often not the case with GDP growth, the inclusion of time dummies functions as a transformation of variables into deviations from time means, hence making it consistent over time and allowing the use of lagged first-difference series as instruments for the level equation (Bond et al., 2001). The validity of the instruments will be assessed using tests proposed by Arellano and Bover (1995). The first test should reject the auto-correlation of the first order, whilst the second test should not reject the auto-correlation of the second order. If this is the case, the GMM estimator is consistent. Moreover, the Hansen J-test will be reported to assess the validity of instruments. If it fails to reject the null hypothesis, the over-identifying excluded instruments are not correlated with the error term. Moreover, if the assumption that the identifying instruments are exogenous holds as well, the instruments are correctly excluded from the regression equation. Nonetheless, Roodman (2009) argues that high p-values of the Hansen J-test may indicate that the results of the test are biased, advocating for the p-values in 0.1 - 0.25 region. The large p-values of the test may be caused by a large number of instruments compared to the set of groups in the sample, and can result in unsuccessful reduction in endogeneity, producing biased coefficients. As I have relatively large time span, I will limit the amount of instruments and collapse their sets.

The System GMM proposes two different estimators: one and two-step. Even though the latter estimator should be more efficient, Bond et al. (2001) argue that the marginal efficiency bonus is small. They issue warnings regarding a possible slow convergence to asymptotic distribution, which may cause a downward bias in finite samples. Nevertheless, by using Windmeijer’s finite sample correction of standard errors, the two-step estimator will no longer produce downward biased results (Windmeijer, 2005). Furthermore, as first-differentiation tends to magnify gaps in datasets, I will use a forward orthogonal

(15)

deviation method, which preserves observations when gaps are present.

Finally, to achieve transparency due to high sensitiveness of GMM to model spec-ifications, I will report the specifications of the models in the footnotes of the result tables.

3.3.2 Transmission channels

As the model specifying equation for transmission channels is no longer dynamic, I will substitute GMM for more suitable static panel model. While, Fixed Effects Model (FEM) or Pooled OLS (POLS) are both appropriate candidates, the time invariant unobserved variables render POLS inconsistent. Thus, to control for such country and time fixed ef-fects, I will use FEM, which functions similarly to Difference GMM, illustrated in equation 4. Nonetheless, it is possible that some transmission channels, such as human capital, may reversely influence corruption hence, introducing reverse causality. To account for such issue, I will estimate Fixed Effects IV regression extended to panel data using lagged val-ues of corruption as instruments for its simultaneous valval-ues. For the correct use of lagged endogenous variables as instruments, they need to not be able to explain changes in the dependent variable in equation of interest, while they should be a determinant for the instrumented variable. The validity of instrument exogeneity will be evaluated by Hansen test of overidentifying restrictions defined in previous section. Regarding the second as-sumption, Morris and Klesner (2010) argued, that corruption is a cyclical mechanism, which by deteriorating the trust of public creates a breeding ground for itself. Hence, previous values of corruption should be a determinant of the extent of current corrup-tion. In addition, Kleibergen-Paap statistics will be reported as an evaluation for under identification. Not rejecting the null hypothesis of both of the tests above is essential in confirming the validity of the instruments.

4

Data

The dataset consists of panel data of the 38 least developed countries according to the UN, recorded over a 2005-2018 time period. Whilst the UN reports 47 LDCs, countries such as Djibouti, Eritrea, Kiribati, Sao Tome and Principe, Solomon Islands, Somalia, South Su-dan, Tuvalu, and Yemen were not included due to a significant lack of data. All variables except for proxies for Human Capital, Democracy, and Corruption were taken from the World Bank. Human Capital is proxied by data representing the average schooling years, taken from the UNESCO Institute for Statistics; Democracy is represented by data on political rights obtained from Freedom House; and Corruption is analyzed through data from the well-known Corruption Perception Index created by Transparency International.

(16)

Table 1: Variable names and descriptions

Variable Description Data Source

y GDP per capita growth World Bank

Y Initial GDP per capita at the beginning of individual periods World Bank

C Corruption Perception Index TI

Inv Gross capital formation-to-GDP ratio World Bank

HC Average schooling years UIS

Open Trade exposure of the country to GDP ratio World Bank

GE Domestic government expenditure to GDP ratio World Bank

P S Political Stability Index World Bank

P op Annual growth in population World Bank

P R Political rights as a proxy for country regime Freedom House

N R Natural resources rents to GDP ratio World Bank

4.1

Data Description

4.1.1 Independent Variable

The independent variable GDP per capita growth is a logarithmic difference between Real

GDP per capita, which is a sum of value of products and services in a country’s currency,

in current and previous period.

4.1.2 Corruption Perception Index

Corruption is notoriously difficult to quantify due to its subjective and secretive na-ture, stemming from its inherent illegality. Corruption Perception Index is an index that aggregates perceptions of businesses and experts on the corruption levels of individual countries. Transparency International published the first report in 1995, however, a ma-jority of LDCs were included in the index only since 2005, which explains the time frame of this analysis. Fundamentally, the index is constructed from several sources and surveys which seek to achieve objective measurements. It evaluates countries on a scale from 0 to 100, with 100 being the least corrupt.

While perception indices are often criticized for being subjective, there is no measure that is able to report corruption levels completely objectively. Moreover, Ahmad and Aziz (2001) analyzed different ways of measuring corruption, ultimately finding high correlations between various corruption measurements and consistently similar regression results, hence increasing the credibility of such indices.

(17)

4.1.3 Transmission Channels

Gross capital formation is reported based on SNA 1993, which measures the fixed capital

formations plus changes in inventories and acquisitions minus disposals of valuables, as a share of GDP. Weak institutions resulting in or from corruption may prohibit larger investments, thus I expect that corruption will have a negative relationship with Gross capital formation.

Figure 2: Relationship between Investments and CPI

Human capital is proxied by the average schooling years for citizens older than 25

years. In his paper, W¨oßmann (2003) concluded that the most suitable human capital proxies include data regarding the quality of education and the amount of returns to schooling. Whilst data on average schooling years does not directly reflect the returns to schooling, nor does it contain information about the quality of education, it is a rather significant measure that provides the largest amount of data for LDCs. Due to the limitations mentioned above, I expect the coefficient of the variable to be positive, however it may be small and insignificant.

Corruption shifts investments from productive activities to activities that offer the highest amount of rents (Shleifer & Vishny, 1993). As investments in schooling do not produce high immediate returns, the flow of funds is diverted from them. Moreover, cor-ruption prohibits efficient collection of taxes hence, reduces the available country’s capital that could have been allocated to schooling. Nonetheless, while the effect of corruption on human capital is expected to be detrimental, I find no unconditional relationship between the two variables.

(18)

Figure 3: Relationship between Human Capital and CPI

Openness to trade is measured as the sum of exports and imports as a share of GDP.

Even though it does not belong to basic growth regressions, Sachs, Warner, ˚Aslund, and Fischer (1995) argued that countries who underwent trade liberalization and market integration ultimately grow faster. Additionally, Dollar and Kraay (2003) eliminated the endogeneity of growth equation and concluded that trade itself plays a large role in increasing GDP. While trade enhances economic growth, corruption may demotivate the cross-border movement of goods by imposing costs on trade through rent-seeking tariffs and regulations, consequently impacting economic growth.

Figure 4: Correlation between Trade Openness and CPI

Government expenditure is reported as government consumption as a share of GDP.

Landau (1983) argued that government consumption has a negative relationship with economic growth. Whilst Wu, Tang, and Lin (2010) rejects the presence of a negative relationship between government expenditure and economic growth overall, he concludes

(19)

that an inverse relationship is prevalent in low income countries due to weak institu-tions. Even though disorganized public institutions decrease governmental expenditure, they also create space for corrupt and large-scale rent-seeking investments, rendering the impact of corruption on government expenditure unclear.

Figure 5: Relationship between Government Expenditure and CPI

Political stability is an index with a range of -2.5 to 2.5, with 2.5 being the most

politically stable entity. It measures the perceptions of political instability and politically-motivated violence. Alesina, ¨Ozler, Roubini, and Swagel (1996) concluded that political stability has a significantly detrimental effect for economic growth. Furthermore, Uddin, Ali, and Masih (2017) using a dynamic panel model, analyzed the impact of political instability on economic growth and argued that political instability has a significant negative effect on economic growth. Prevalent corruption may fuel political instability and therefore restrict economic growth.

(20)

4.1.4 Control Variables

Population growth rate is the annual change in population, figuring in the growth

regres-sion as a proxy for labour growth rate. The impact of population growth rate on economic growth is negative in the short-run, however, it may have positive effect in the long-run.

Political rights is an index that assesses people’s rights based on the Universal

Decla-ration of Human Rights, used as a proxy for the type of country’s regime. The measure is quantified based on scores of political pluralism and participation, the electoral process, and the functioning of the government. It can be seen on a scale from 1 to 7, with 1 attributed to countries with the highest levels of political freedom.

There is still no academic consensus on the impact of political rights on economic growth. Nevertheless there has been some evidence, albeit frail, that a shift to a more democratic regime boosts economic growth (Barro, 1996a).

Natural resources rents as a share of GDP are the sum of oil, natural gas, forest,

mineral, and coal rents divided by GDP. Whilst natural resources are possibly connected with higher economic growth over the short-term, its over-exploitation neglects other aspects necessary for economic growth and indirectly decreases it. Nevertheless, I expect a positive relationship between economic growth and natural resources.

4.2

Data Exploration

Using kernel weighted local polynomial smoothing, figure 7 captures nonlinearities in the relationship between corruption and economic growth and presents their unconditional marginal relationship. The shaded area depicts the range of CPI score from 12-32 range, into which falls 75% of the analyzed sample. The finding is supported by the graph de-picting corruption density. Considering the steep slope of the curve in the shaded area; and hence, the specialized nature of the sample, the figure hints, despite depicting un-conditional relationship, that the coefficient for corruption may be higher than previously reported in the literature.

(21)

Table 2 presents correlations of the variables used in regressions, combined with descriptive statistics of the individual variables. The unconditional relationship of trans-mission channels is moderate to strong, with exceptions for proxies for trade openness and human capital, which relationships with corruption is small. Finally, noteworthy is high mean and low standard deviation of CPI confirming that results are going to be oriented on marginal impacts of changes in CPI in the highly corrupted countries.

Table 2: Correlations and description statistics

y Y C Inv HC Open GE P S P op P R N R y 1 Y 0.04 1 (0.36) C −0.06 −0.22∗ 1 (0.19) (0.00) Inv 0.22∗ 0.32∗ 0.43∗ 1 (0.00) (0.00) (0.00) HC −0.06 0.24∗ 0.05 0.13∗ 1 (0.89) (0.00) (0.22) (0.00) Open 0.04 0.22∗ 0.16∗ 0.19∗ 0.18∗ 1 (0.44) (0.00) (0.00) (0.00) (0.00) GE −0.08 0.14∗ 0.400.170.01 0.341 (0.07) (0.00) (0.00) (0.00) (0.85) (0.00) P S 0.09 0.29∗ 0.59∗ 0.28∗ 0.26∗ 0.26∗ 0.30∗ 1 (0.05) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) P op −0.13∗−0.31−0.09−0.13−0.22−0.04−0.10−0.05 1 (0.00) (0.00) (0.04) (0.00) (0.00) (0.40) (0.03) (0.20) P R −0.08 −0.29∗ −0.34−0.10−0.22−0.01 −0.16−0.330.241 (0.09) (0.00) (0.00) (0.01) (0.00) (0.75) (0.00) (0.00) (0.00) N R −0.05 0.08 −0.14∗−0.15−0.06 0.360.17−0.140.330.141 (0.29) (0.06) (0.00) (0.00) (0.16) (0.00) (0.00) (0.00) (0.00) (0.00) Obs 560 560 535 558 560 523 488 560 560 560 518 M ean 2.56 24.99 27.67 24.18 3.90 71.16 14.08 −0.69 2.44 4.54 13.50 SD 5.04 18.00 9.34 11.28 1.14 35.25 6.55 0.91 0.85 1.57 12.04 M in −36.56 5.29 8 1.40 1.30 0.17 3.46 −3.00 −0.49 2.00 0.38 M ax 38.70 103.69 68 73.00 7.10 311.35 40.44 1.38 4.13 7.00 0.74 P-values are reported in parentheses.

(22)

5

Empirical Results

5.1

Growth Equation Regression

In this section I present estimates regarding the effect of corruption on GDP per capita growth, both with and without including independent variables, which should provide an overview of the total effect of corruption. Afterward, I will analyze the different transmission channels.

Model (1) in Table 3 is an initial regression that, except for corruption and initial income, additionally includes control variables. With the exclusion of all transmission channels, I aim to estimate the total effect of corruption on economic growth. The coefficient for the logarithm of the initial level of GDP per capita is -0.0585, and while not significant, its negative sign supports the theory of conditional convergence. Population growth has a negative, but an insignificant coefficient of -2.293, which further supports conditional convergence. Proxies for the political regime and the presence of natural resources are insignificant with a positive coefficient of 0.0120 and 0.0013, respectively. Finally, the coefficient of corruption is negative, equal to -0.025 and significant at a 5% significance level. Therefore, a ten unit increase in CPI score, a shift from the position of Togo to the position of Rwanda, results in a substantial increase in temporary economic growth by 2.55%. This is significantly higher than previously estimated by Mo (2001) or Pellegrini and Gerlagh (2004), which may be caused by the different composition of the samples discussed in the previous section. The model is evaluated by the second order serial correlation test that does not reject the null hypothesis and hence, supports the use of second lagged instruments. Nonetheless, its low p-value invites for caution while interpreting the results, indicating that the model might still suffer from endogeneity. Finally, Hansen test confirms a lack of correlation between the excluded instruments and error terms.

Model (2) omits corruption and introduces vector X from regression (2) with all the possible transmission channels. The coefficient for the logarithm of initial income per capita remains negative; however, it becomes significant. A one percent decrease in the initial GDP per capita increases the temporary economic growth of GDP per capita by 0.0582%. On the other hand, a one percent increase of investment to GDP ratio, which is positive and significant at 10% significance level, increases temporary economic growth by 0.038%. While the coefficient for human capital is not significant, it has an expected positive sign and based on its size, is important in determining economic growth as suggested by Mankiw et al. (1992). Trade openness is significant at a 5% significance level with a coefficient of 0.0006; hence, a one percentage point increase in trade results in an increase of temporary economic growth by 0.06%.

(23)

Table 3: Growth equation regression as in Eq.2

Dependent Variable: yi,t(log) (1) (2) (3)

Constant 0.0702 0.3780 0.3940 (0.74) (0.18) (0.19) Yi,t−1(log) −0.0585 −0.0582∗∗−0.0689∗∗ (-0.80) (-0.04) (-0.04) Ci,t −0.0025∗∗ −0.00115 (-0.03) (-0.18)

Invi,t(log) 0.0380∗ 0.040∗

(0.09) (0.06) HCi,t 0.0156 0.0199∗ (0.25) (0.07) Openi,t 0.0006∗∗ 0.0005∗∗ (0.03) (0.018) GEi,t −0.0027 −0.0026 (0.20) (-0.18) P Si,t 0.0238∗∗ 0.0225∗ (0.04) (0.06) P opi,t(log) −2.293 −2.171 −2.167 (-0.46) (-0.45) (-0.29) P Ri,t 0.0021 −0.0021 −0.0006 (0.83) (0.83) (-0.94) N Ri,t 0.0013 0.0015∗ 0.0017∗∗∗ (0.20) (0.07) (0.01)

Time dummies Yes Yes Yes

Number of observations 424 424 424

Number of groups 39 39 39

Number of instruments 17 23 25

Second order serial correlation test 0.130 0.190 0.192 Hansen test of over-identification 0.105 0.577 0.761 P-values are reported in the parentheses. The asterisks of *, ** and *** denote significance at 0.10, 0.05 and 0.01 level, respectively. Instruments: Endogenous: Twice lagged levels and differences of the dependent variable, corruption and other explanatory variables except for PR and NR, which are taken as predetermined and their levels are lagged only once. All instrument sets were collapsed. Standard instruments: Time dummies in level equation.

(24)

The government expenditure coefficient is insignificant; nonetheless, negative. The sign can be possibly explained by unproductive resource management and weak capabili-ties of tax collection (Barro, 1996b). The proxy for political stability is significant at 5%, and an increase of 1 unit on the Political Stability scale ultimately leads to a substantial increase of temporary economic growth by 2.38%. While such a large effect seems im-plausible, the scale has a small 5-unit range, hence improvements in political stability are reflected with minimal change on the index. Population growth supports the conditional convergence theory with a negative but insignificant coefficient of -2.293. Political rights, a proxy for political regime is insignificant; nonetheless, positive. Finally, the presence of natural resources is significant at a 10% significant level. A ten percentage point increase in natural resources rents raises the temporary GDP per capita growth by approximately 1.6%. The model is evaluated by the second order serial correlation test that does not reject the null hypothesis, therefore supporting the use of second lagged instruments. The outcome of the Hansen test exceeds the p-values recommended by Roodman (2009), the number of instruments is not worryingly high; and thus, the failure to reject the null hypothesis, confirming the exogeneity of excluded instruments, can be considered valid.

Model (3) includes all the variables from equation (2). The coefficient for corruption decreased and a 10 point increase in the CPI scale results in a decrease in temporary GDP per capita growth by 1.5%. This supports the hypothesis that corruption itself does not affect economic growth in a solely direct manner, but is rather being transmitted through transmission mechanisms. Nonetheless, in contrast to results proposed by Mo (2001) or Pellegrini and Gerlagh (2004), the direct effect, whilst insignificant, is still relatively high as it represents 46.18% of the total effect. Hence, corruption may be transmitted through additional, different mechanisms in the analyzed sample. The indirect effect of corruption is discussed in the next section.

5.2

Transmission Channels

I this section, I analyze the indirect effects of regressions that are not visible in Table 3. I discuss how and to what extent corruption influences the transmission channels represented by vector X in equation (2), whilst referencing to Table 4 which displays the results of regressions for all the transmission channels as proposed by equation (3). Due to the use of a different regression method for transmission channel estimation, the indirect effects proposed in Table 4 do not account for the whole difference between total and direct effect.

(25)

Table 4: Transmission channels regressions

(4) (5) (6) (7) (8)

Dependent Variable Inv (log) HC Open GE PS

Yi,t−1(log) −0.324 0.591∗∗−4.737 −5.375∗ 0.505∗∗ (0.054) (0.001) (0.533) (0.019) (0.002) Ci,t 0.019∗∗∗ 0.020∗∗ 0.283 0.067 0.006 (0.000) (0.025) (0.337) (0.286) (0.343) P opi,t(log) 4.768 1.799 17.042 −69.689 3.075 (0.344) (0.783) (0.958) (0.235) (0.745) P Ri,t −0.063∗∗ 0.014 −0.095 0.036 −0.010 (0.002) (0.464) (0.901) (0.822) (0.696) N Ri,t −0.001 −0.001 1.085∗∗∗−0.003 −0.007∗∗ (0.650) (0.568) (0.000) (0.925) (0.021)

Time dummies Yes Yes Yes Yes Yes

Number of observations 379 347 329 382 424

Hansen test of over-identification 0.300 0.515 0.714 0.769 0.325 Under-identification test - F value 43.217 12.775 36.781 43.707 54.538 P-values are reported in the parentheses. The asterisks of *, ** and *** denote significance at 0.10, 0.05 and 0.01 level, respectively. Excluded instruments: Once and twice lagged values of endogenous variable - Corruption - except for Model (5), where corruption is instrumented by its second and third lags due to persevering effect of corruption on education.

5.2.1 Investment Channel

Corruption may be detrimental to gross capital formation through its tendency to impose additional operational costs on firms (Shleifer & Vishny, 1993). The costs do not only consist of the obvious rents firms have to pay the bribe-seeking regulatory officials, but also of less explicit costs of keeping corruption activities secret. Moreover, well estab-lished firms with powerful connections in the official administration of a country may try to illegally reduce competition and crowd out other sources of investments. However, building on the previous argument, companies seeking illegal enrichment may forego the additional costs of bribes and secrecy and increase their investments, especially with the cooperation of the public sector, with high illicit gains in mind Rose-Ackerman (1997).

In model (4), the corruption coefficient is negative and significant at a 1% significance level. A ten unit increase on a CPI scale results in a substantial 19%, or on average 4.75 percentage point increase in investment to GDP ratio, which in turn decreases temporary

(26)

economic growth by approximately 0.8%. While this result is higher than previously reported by Mo (2001) or Pellegrini and Gerlagh (2004), it is relatively similar to what Campos, Lien, and Pradhan (1999), who focused on developing countries, estimated. They discussed the dependency of the effect of corruption on its type and argued that corruption is associated with high uncertainty, therefore it can be said that disorganized corruption has the most substantial effect on investments. Hence, the results indicate that the corruption in the analyzed sample specified on the least developed countries may be highly disorganized and thus discouraging for investors.

The Hansen test of over-identification does not reject the null hypothesis, hence supporting the exogeneity of instruments. Under identification test does reject the null hypotheses, therefore confirm the instrument relevance.

5.2.2 Human Capital

As corruption diverts funds and attention from the productive activities to the rent-seeking activities and shrinks the capacity of governments to collect sufficient amount of funds for efficient distribution, the effect of corruption on human capital should be negative Shleifer and Vishny (1993). Moreover, the education sector is not perceived as attractive, as investments in education are not large-scale projects with substantial space for the diversion of funds and illicit enrichment.

In regression (5), the coefficient of corruption is significant and positive, as expected. A ten unit increase in CPI corruption results in 0.2 years or 73 additional days of schooling, or, considering the sample mean of 3.9 schooling years, an increase of 5%, which in turn decreases temporary GDP per capita growth by 0.4%. The result is consistent with results by Pellegrini and Gerlagh (2004), who estimated that a 25 increase in CPI increases schooling by half a year.

While Hansen tests does not reject the null hypothesis, the F-values of under identi-fication test are relatively low, inviting for a certain degree of caution while interpreting the results.

5.2.3 Trade Openness

Corruption may function as trade protectionism mechanism induced by border officials imposing rent-seeking tariffs on goods moving through the border, or by artificially inflat-ing waitinflat-ing times; and hence, creatinflat-ing higher barriers for trade (Bandyopadhyay & Roy, 2007). In contrast, if such delays and regulations are determined exogenously, bribing the officials may hasten the process by bypassing such inefficiencies. Nevertheless,

(27)

Kauf-mann and Wei (1999) proved that if trade tariffs are determined endogenously by border officials, bribes and harassment from officials are positively correlated, thus imposing increasingly large costs on subjects of exchange.

In model (6), corruption has no significant impact on trade openness, nevertheless its sign indicates a negative effect on trade exposure. A ten unit decrease in corruption would, based on this result, increase the trade to GDP ratio by 2.8 percentage points, which would increase the temporary economic growth by 0.1%.

The Hansen test rejected the null hypothesis, meaning that there is no correlation between the excluded instruments and the dependent variable, whilst the under identifi-cation test confirmed the relevance of instruments.

5.2.4 Government Expenditure

Corruption has been proven to increase public expenditure while decreasing its efficiency (Haque, Kneller, et al., 2008; Tanzi & Davoodi, 1998). Furthermore, corruption boosts public expenditure in the government operational sector, such as salaries and bonuses of government officials. Building on the education sector investment unattractiveness argu-ment, Mauro (1998) claimed that corruption diverts funds from productive sectors such as education to sectors that offer immediate returns, albeit often to chosen individuals that should act in accordance with public interest, however do not. Nevertheless, as corruption enlarges the shadow economy, governments suffer from smaller tax revenues which can either result in higher indebtedness or lower public expenditure.

Considering the previous empirical research, the results in model (7) are surprising. While corruption is statistically insignificant in determining government expenditure, it’s coefficient sign indicates that higher corruption decreases public investments. A ten unit increase in the CPI score would increase government expenditure by 0.067 percentage points and thus decrease temporary GDP per capita growth by 0.2%. This could be ex-plained through the channel of foreign direct investment (FDI) or foreign aid and economic growth. Highly corrupted countries are usually poorer than their counterparts, therefore, whilst FDIs are often subject to the costs of corruption, they are not high enough to offset the wage differences between the poor and rich economies (Egger & Winner, 2006). Moreover, aid is not dependent on the level of corruption, but it is dependent on the level of GDP (Alesina & Weder, 2002). Thus, some of the public expenditure may be covered either by FDI or foreign aid, clarifying the surprising sign of the corruption coefficient.

Both Hansen test and under identification test, used to evaluate the regression, con-firm its validity.

(28)

5.2.5 Political Stability

Alesina et al. (1996) argues that political instability creates uncertainty which may pro-voke both domestic or foreign economic subjects that would rather reconsider their fi-nancial decisions until uncertainty decreases in order to reduce savings and investments into any productive activity, ultimately decreasing economic growth. Corruption fuels political discontent between citizens who may demand changes regarding political regime or government officials. Moreover, Mo (2001) discusses the adverse impact of corruption on inequality, which induces civic unrest that might result in demands of political change and create uncertainty.

The coefficient of corruption in model (8) has the expected sign and is insignificant. A ten unit decrease in corruption increases a country’s score on the political stability index by 0.06, which in turn decreases temporary economic growth by 0.02%. The disclosed coefficient for corruption is slightly lower than results reported by Mo (2001).

The excluded instruments are jointly exogenous as confirmed by a Hansen test of over-identification, and relevant, as the null hypothesis of under identification are rejected.

6

Total Effect of Corruption

The total effect of corruption consists of the direct effect of corruption and indirect effects. Whilst the direct effect of corruption is visible in Regression (3) in Table 3, its indirect effects through transmission channels are computed by a summation of the products of coefficients of corruption on individual transmission channels, and the their coefficients on economic growth as seen in equation (6).

Eindirect= X n=5 (δChi δC · δy δChi ) (6)

The equation used to compute the total effect of corruption is outlined below.

Etotal=

δy

δC + Eindirect (7)

Table 5 offers an overview of the total effect of corruption on economic growth. Mo (2001) and Pellegrini and Gerlagh (2004) reported an approximate 80% contribution of indirect effects, whereas the indirect effects in this analysis are smaller and accounted for approximately 53.82% of the total effect; indicating a need for analyses of different transmission channels.

(29)

Table 5: Overview of the transmission channels

Table 3 Table 4 Contribution Relative Transmission Channels δChi δC δy δChi δChi δC · δy δChi Contribution Corruption direct 0.001 46.18% Investment 0.040 0.019 0.0008 31.32% Human Capital 0.0199 0.020 0.0004 15.98% Trade Openness 0.0005 0.283 0.0001 5.82% Government Expenditure −0.0026 0.067 −0.0002 −7.02% Political Stability 0.0225 0.006 0.0015 5.85% Error 0.00005 1.87% Total 100%

Except for the direct effect of corruption, the largest transmitter of corruption is investment. The channel constitutes over 31% of the total effect of corruption, which is in line with previous studies using similar framework (Mo, 2001; Pellegrini & Gerlagh, 2004). The second largest effect is associated with the human capital channel, which contributes almost 16% to the total effect of corruption, similar to results by Mo (2001). Trade openness and political stability have approximately the same contribution to the total effect, being at almost 6%. These transmission channels constitute a smaller percent-age of the total effect than previously reported. Finally, the transmission of corruption through government expenditure has the opposite sign, therefore, higher corruption means lower government expenditure, possibly due to reasons discussed in section 5.2.4 such as coverage of government expenditures by FDIs or corruption’s tendency to decrease tax revenue.

7

Conclusion

I aimed to observe the direct effect of corruption on economic growth and its transmis-sion channels. While the results show support for the ’sanding the wheels’ hypothesis, the total effect of corruption on economic growth surmounted those estimated by previous studies (Mauro, 1995; Mo, 2001; Pellegrini & Gerlagh, 2004). The divergence in outcomes continues with the total effect, of which only 53% is explained by transmission channels; hence, approximately 25 percentage points lower than estimated previously. The differ-ences may be caused by the specialized composition of the sample in this study, which,

(30)

due to its nature, may require further inquiries into the transmission mechanisms. Except for the direct effect of corruption, the total impact is mainly transmitted through the investment channel which accounted for 31% of the total effect and is ac-cordingly emphasised in preceding literature. The large importance of the investment channel may be attributed to the disorganized nature of corruption in LDCs as men-tioned previously (Campos et al., 1999). The other channels that conducted the negative effect of corruption are, in order of importance, human capital, political stability and trade. Higher levels of corruption, however, decrease government expenditure, and thus enhance economic growth. Nonetheless, such a phenomenon could be linked to a higher influx of FDIs that crowd out government expenditure; and hence, it would be premature to speak about positive impacts of corruption.

The LDCs suffer from underdeveloped education systems, rampant political instabil-ity, lack of private investments, insufficient trade infrastructure and unproductive use of government funds, almost all of which is worsened by corruption. While Mishra (2006) ar-gued that in a corrupt society, everyone is perceived corrupt; hence, non-compliance with law is a dominant strategy; and Herzfeld and Weiss (2003) empirically proved cyclical relationship between corruption and institutions, Seldadyo and De Haan (2011) argued otherwise and proved that corruption does change over time. Countries are, therefore, not doomed to stay corrupt. Thus, understanding of how corruption transmits and affects economic growth, may provide a guideline for policy makers to see the hidden ways of corruption and tailor policies accordingly.

Regarding the limitations of this research, the output of the dynamic panel models is a matter of slight worry. While the auto-correlation of second order is insignificant at 10% significance levels at all times, the p-values do not show a confident non-rejection of the null hypothesis. Therefore, this indicates that endogeneity may still be a slight concern and that the coefficients should be interpreted with caution. Moreover, the lack of data eliminated approximately a quarter of all LDCs from the sample; hence, the sample may not be entirely applicable to the whole population.

Finally, as the analyzed transmission channels failed to account for the large major-ity of the total effect, future research could focus on expanding the list of transmission channels and analyzing them in more detail. This could help understand how corruption influences certain aspects of specific transmission channels; for instance, how does it in-fluence investments or how does it prohibit progress in human capital. Furthermore, due to various natures and functions of corruptions inherent to different cultures, literature on the motivation of corruption and its cultural basis should be explored to understand and categorize the consequences of corruption elimination in different environments.

(31)

References

Ades, A., & Di Tella, R. (1999). Rents, competition, and corruption. American economic

review, 89 (4), 982–993.

Ahmad, N., & Aziz, D. (2001). Corruption perception indices: a comparative analysis [with comments]. The Pakistan Development Review, 813–830.

Alesina, A., ¨Ozler, S., Roubini, N., & Swagel, P. (1996). Political instability and economic growth. Journal of Economic growth, 1 (2), 189–211.

Alesina, A., & Weder, B. (2002). Do corrupt governments receive less foreign aid?

American Economic Review, 92 (4), 1126–1137.

Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte carlo evidence and an application to employment equations. The review of economic

studies, 58 (2), 277–297.

Arellano, M., & Bover, O. (1995). Another look at the instrumental variable estimation of error-components models. Journal of econometrics, 68 (1), 29–51.

Assiotis, A., & Sylwester, K. (2014). Do the effects of corruption upon growth differ between democracies and autocracies? Review of Development Economics, 18 (3), 581–594.

Bandyopadhyay, S., & Roy, S. (2007). Corruption and trade protection: evidence from panel data. Available at SSRN 988433 .

Barro, R. J. (1991). Economic growth in a cross section of countries. The quarterly

journal of economics, 106 (2), 407–443.

Barro, R. J. (1996a). Democracy and growth. Journal of economic growth, 1 (1), 1–27. Barro, R. J. (1996b). Determinants of economic growth: A cross-country empirical study

(Tech. Rep.). National Bureau of Economic Research.

Beck, P. J., & Maher, M. W. (1986). A comparison of bribery and bidding in thin markets. Economics letters, 20 (1), 1–5.

Bond, S. R., Hoeffler, A., & Temple, J. R. (2001). Gmm estimation of empirical growth models.

Campos, J. E., Lien, D., & Pradhan, S. (1999). The impact of corruption on investment: Predictability matters. World development, 27 (6), 1059–1067.

Caselli, F., Esquivel, G., & Lefort, F. (1996). Reopening the convergence debate: a new look at cross-country growth empirics. Journal of economic growth, 1 (3), 363–389. Della Porta, D., & Vannucci, A. (1997). The ‘perverse effects’ of political corruption.

Political studies, 45 (3), 516–538.

(32)

economics, 50 (1), 133–162.

Egger, P., & Winner, H. (2006). How corruption influences foreign direct investment: A panel data study. Economic Development and Cultural Change, 54 (2), 459–486. Ehrlich, I., & Lui, F. T. (1999). Bureaucratic corruption and endogenous economic

growth. Journal of Political Economy, 107 (S6), S270–S293.

Gyimah-Brempong, K. (2002). Corruption, economic growth, and income inequality in africa. Economics of governance, 3 (3), 183–209.

Haque, M. E., Kneller, R., et al. (2008). Public investment and growth: The role of corruption. Centre for Growth and Business Cycle Research discussion paper series,

98 .

Herzfeld, T., & Weiss, C. (2003). Corruption and legal (in) effectiveness: an empirical investigation. European Journal of Political Economy, 19 (3), 621–632.

IMF. (2019, Dec). Republic of equatorial guinea : Request for an extended arrangement

under the extended fund facility and second review under the staff-monitored program-press release, staff report, and statement by the executive director.

Retrieved from https://www.imf.org/en/Publications/CR/Issues/2019/12/ 20/Republic-of-Equatorial-Guinea-Request-for-an-Extended-Arrangement -Under-the-Extended-Fund-48904

Islam, N. (1995). Growth empirics: a panel data approach. The quarterly journal of

economics, 110 (4), 1127–1170.

Jain, A. K. (2001). Corruption: A review. Journal of economic surveys, 15 (1), 71–121. Kaufmann, D., & Wei, S.-J. (1999). Does” grease money” speed up the wheels of

com-merce? (Tech. Rep.). National bureau of economic research.

Kurer, O. (1993). Clientelism, corruption, and the allocation of resources. Public Choice,

77 (2), 259–273.

Landau, D. (1983). Government expenditure and economic growth: a cross-country study. Southern Economic Journal, 783–792.

Leff, N. H. (1964). Economic development through bureaucratic corruption. American

behavioral scientist, 8 (3), 8–14.

Leys, C. (1965). What is the problem about corruption? The Journal of Modern African

Studies, 3 (2), 215–230.

Lui, F. T. (1985). An equilibrium queuing model of bribery. Journal of political economy,

93 (4), 760–781.

Mankiw, N. G., Romer, D., & Weil, D. N. (1992). A contribution to the empirics of economic growth. The quarterly journal of economics, 107 (2), 407–437.

(33)

681–712.

Mauro, P. (1998). Corruption and the composition of government expenditure. Journal

of Public economics, 69 (2), 263–279.

eon, P.-G., & Weill, L. (2010). Is corruption an efficient grease? World development,

38 (3), 244–259.

Mishra, A. (2006). Persistence of corruption: some theoretical perspectives. World

Development, 34 (2), 349–358.

Mo, P. H. (2001). Corruption and economic growth. Journal of comparative economics,

29 (1), 66–79.

Morris, S. D., & Klesner, J. L. (2010). Corruption and trust: Theoretical considerations and evidence from mexico. Comparative Political Studies, 43 (10), 1258–1285. Nur-Tegin, K., & Czap, H. J. (2012). Corruption: Democracy, autocracy, and political

stability. Economic Analysis and Policy, 42 (1), 51–66.

Nye, J. S. (1967). Corruption and political development: A cost-benefit analysis.

Amer-ican political science review, 61 (2), 417–427.

Olivier de Sardan, J.-P. (1999). A moral economy of corruption in africa? Journal of

Modern African Studies, 37 (1), 25–52.

Papyrakis, E., & Gerlagh, R. (2004). The resource curse hypothesis and its transmission channels. Journal of Comparative Economics, 32 (1), 181–193.

Pellegrini, L., & Gerlagh, R. (2004). Corruption’s effect on growth and its transmission channels. Kyklos, 57 (3), 429–456.

Roodman, D. (2009). A note on the theme of too many instruments. Oxford Bulletin of

Economics and statistics, 71 (1), 135–158.

Rose-Ackerman, S. (1997). The political economy of corruption. Corruption and the

global economy, 31 , 60.

Sachs, J. D., Warner, A., ˚Aslund, A., & Fischer, S. (1995). Economic reform and the process of global integration. Brookings papers on economic activity, 1995 (1), 1– 118.

Seldadyo, H., & De Haan, J. (2011). Is corruption really persistent? Pacific Economic

Review, 16 (2), 192–206.

Shleifer, A., & Vishny, R. W. (1993). Corruption. The quarterly journal of economics,

108 (3), 599–617.

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

journal of economics, 70 (1), 65–94.

Solow, R. M. (1957). Technical change and the aggregate production function. The review

(34)

Svensson, J. (2005). Eight questions about corruption. Journal of economic perspectives,

19 (3), 19–42.

Tanzi, V., & Davoodi, H. (1998). Corruption, public investment, and growth. In The

welfare state, public investment, and growth (pp. 41–60). Springer.

Transparency International, C. P. (2020). Transparency international. URL: http://www.

transparency. org/news/feature/cpi 2013 now is the time for action.

Treisman, D. (2000). The causes of corruption: a cross-national study. Journal of public

economics, 76 (3), 399–457.

Uddin, M. A., Ali, M. H., & Masih, M. (2017). Political stability and growth: An application of dynamic gmm and quantile regression. Economic Modelling, 64 , 610–625.

Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step gmm estimators. Journal of econometrics, 126 (1), 25–51.

oßmann, L. (2003). Specifying human capital. Journal of economic surveys, 17 (3), 239–270.

Wu, S.-Y., Tang, J.-H., & Lin, E. S. (2010). The impact of government expenditure on economic growth: How sensitive to the level of development? Journal of Policy Modeling, 32 (6), 804–817.

Referenties

GERELATEERDE DOCUMENTEN

Still, the reduction to large item sizes can be extended to the non-uniform case, which might be of independent interest: As it turns out, in the non-uniform case we have to

A much different approach for probabilistic models is statistical model check- ing (SMC) [17,20,26]: instead of exploring—and storing in memory—the entire state space, or even a

The inclusion of the independent variable shows that there is a positive relation of .005 at a 1% significance level between the number of M&As and the host country

The effect of the Asian countries with a high level of judicial independence on the relation between the dependent and independent variable deteriorates the negative

insignificant. With the Hosmer and Lemeshow goodness of fit test stating the model was a poor fit, the same dependent and independent variables should be studied again under a

In contrast the results based on the OECD sample indicate that in case an economy with a TFP level that is higher than 96% of the US, an increase of the average years of

Concessionaire model for food and beverage operations in South African national parks | iii period of three weeks and consisted of four sections, namely a demographic section, a

Biomaterials Innovation Research Center, Division of Engineering in Medicine, Brigham and Women's Hospital, Harvard Medical School, Cambridge, MA 02139, USA.. Harvard-MIT Division