Abstract
The article examines the corruption–growth relationship in a non-linear framework using panel fixed effects (FE) and system generalized methods of moments (SGMM) model for over 110 countries for the period 1984–2009. The results reveal that the least corrupt countries enjoy higher growth rates, whereas highly corrupt countries experience low growth. Furthermore, corruption has a positive and significant effect on economic growth up to a certain level and thereafter it reduces growth. The results are robust under various methodology and an alternative measure of corruption.
Keywords
Introduction
Is corruption growth-enhancing or always sand-in-a-wheel? The recent high growth performances of many emerging economies challenge the existing literature which has established that the impact of corruption is detrimental to all societies, especially in the context of developing eco-nomies (Bardhan, 1997; Brunetti, Kisunko, & Weder, 1998; Klitgaard, 1988; Knack & Keefer, 1995; Mauro 1995; Rose-Ackerman, 1978). Whether corruption affects economic growth negatively or positively does not seem to be uniformed. Moreover, with a few exceptions, the studies have mainly focused on detecting the linear relationship between corruption and growth. But it remains unclear whether an increase in corruption consistently decreases/increases corruption across countries. While the linear negative relationship between corruption and economic growth has been noted in the literature, the degree of the level of corruption impact on growth is not uniform and straight forward. The overall long-term trend of the entire process may resemble the downward slope portrayed by a linear function, but the non-linear function can discriminate the experiences of less/medium corrupt countries from that of the high corrupt countries. This propels challenges to the claims of a negative/positive linear relationship between economic growth and corruption. Hence, this article evaluates the nexus between growth and corruption in a non-linear framework using fixed effect (FE) and system generalized methods of moments (SGMM) over 110 countries for the period 1984–2009.
There has been a debate in the literature about the impact of corruption, that is, ‘sand’ versus ‘grease’ the wheels of growth, over the last few decades. Theories regarding the impact of corruption on efficiency and growth have sometimes been mutually conflicting. The ‘sand-the-wheels’ view is that corruption has adverse effects on investment and economic growth (Myrdal, 1968; Shleifer & Vishny, 1993). A payment of a bribe to get an investment licence, for example, clearly reduces the incentive to invest (Bardhan, 1997, p. 1327). Another negative growth effect follows from the allocation of talent. When rent-seeking sectors offer most able people higher returns than the productive sectors offer them income, growth can be much lower than possible (Murphy, Shleifer, & Vishny, 1991). In contrast, other studies (led by Huntington, 1968 and Leff, 1964;) claim that bribery and corruption can have positive effects on efficiency. The efficiency-enhancing strand views corruption as increasing bureaucratic efficiency because corruption ‘greases the wheels’. The argument is that in the context of pervasive and cumbersome regulations in the developing countries, corruption may actually improve efficiency and help growth by reducing red tape.
Some of these growth effects have been statistically substantiated from cross-country data by Mauro (1995) on the basis of the corruption ranking data assembled from the Business International correspondents in about 70 countries for the period 1980–1983 and his seminal work finds that there is a negative and significant association between corruption and the investment rate, and the magnitude of the effect is considerable. The other subsequent studies commonly find that corruption is detrimental to growth (Gupta, Mello, & Sharan, 2001; Kaufmann & Wei, 1999; Mauro, 1997, 1998; Mo, 2001; Pelligrini & Gerlagh, 2004; Tanzi, 1998). All these studies mainly find a linear and negative correlation between the level of corruption and the average rate of per capita income growth. These studies use cross-sectional models with average rate of economic growth as the dependent variable and set of economic control variables such as initial income, investment, primary and secondary enrolment ratio, population growth, government expenditure and political stability.
Interestingly, Svensson (2005, p. 38) argues that although most of the theoretical literature as well as case study and micro evidence appear to suggest that corruption severely hamper development, however, measuring corruption in a cross-country setting does not affect growth. He questioned the validity of Mauro’s (1995) findings and concludes that an unanswered puzzle remains in the macro context. 1 Recent study by Rock and Bonnett (2004) supports Svensson’s claim. This study checks the robustness of the negative effect of corruption on growth and investment using four different corruption measures and finds that corruption slows growth and reduces investment in most developing countries but corruption significantly promotes economic growth in newly industrialized economies of East Asia. This finding indicates a possibility of positive effects of corruption.
On the other hand, Méon and Sekkat (2005) find a significant negative impact of corruption on growth when governance is of poor quality. This study adds an interaction term between corruption and the quality of government to the standard model to examine the impact of corruption. The problem with this approach is that corruption and the quality of government is highly correlated and they should not be treated as individual distortions. Similar study by Méon and Weill (2008) has challenged these results. These authors analyse the interaction between aggregate efficiency, corruption and different dimensions of governance for a panel of 54 countries and show that corruption may be positively associated with efficiency where institutions are ineffective. This study finds evidence in favour of the grease-the-wheels hypothesis. This argument may lead to a non-linear relationship between growth and corruption.
Moreover, some authors challenge the so-called notion of a linear corruption–growth relationship and argue the possibility of a positive output-maximizing level of corruption. For example, Acemoglu and Verdier (1998) and Klitgaard (1988) show that in a second-best world the optimal level of corruption might be greater than zero because monitoring corruption is costly. Mendez and Sepulveda (2006) study the effects of corruption on long-run growth-incorporating measures of political freedom as a key determinant of the relationship. They find evidence of a non-monotonic relationship between corruption and growth after controlling for several other economic variables. They show that corruption has a beneficial effect on long-run growth at a low level of corruption but it is destructive at high levels, indicating that the growth-maximizing level of corruption is significantly greater than zero. However, this effect remains robust only in a sub-sample of countries that have achieved a high degree of political freedom. A similar study by Aidt, Dutta and Sena (2008) identify two governance regimes, and in the regime with high quality of institutions, corruption is found to have a significant negative impact on growth. However, in a low institutional quality framework corruption may not effect growth.
Hence, the existing literature remains inconclusive about corruption–growth relationship and therefore the non-linear relationship between growth and corruption in a multi-country framework should not be ruled out. In fact, Swaleheen (2011) investigates the effect of corruption on growth for a panel of countries for the period 1984–2007. Using FE and GMM techniques, the author finds that the effect of corruption is non-linear. Interestingly, the results show that corruption decreases growth at its low level and increases growth when corruption is high. In other words, corruption is growth enhancing even when incidence of corruption is high, which is inconsistent with the theory developed by Shleifer and Vishny (1993). These results might be due to the incorrect model specification because Swaleheen (2011) study considers (a) logarithm of real GDP (RGDP) per capita as a dependent variable instead of RGDP per capita growth, (b) both primary and secondary enrolment data (primary and secondary enrolments are highly correlated) are considered as independent variables in the same equation and (c) political stability is incorporated as one of the independent variables along with corruption but it is argued that political stability and corruption are highly correlated.
This article addresses some of the caveat of the past literature and examines the nexus between corruption and growth in a non-linear framework. Our purpose is to offer a systematic analysis of corruption–growth relationship which follows the theoretical consistency. In order to evaluate the relationship, we extend the standard econometric specification by including a quadratic term for corruption that allows a test for a positive growth-maximizing level of corruption. Following theoretical framework of Shleifer and Vishny (1993), it is expected that at a low level, corruption is growth enhancing, whereas it reduces growth when it is at a high level. 2 Our analysis incorporates more relevant economic and institutional independent variables in the model and deleted some of the variables with possible multicollinearity as discussed earlier. The principal part of our analysis draws on data reported by International Country Risk Guide (ICRG) and we supplement this with an additional analysis of a second dataset on corruption perception index (CPI) measured by Transparency International (TI) since 1995. The advanced panel estimation techniques have been utilized after controlling for FE and endogeneity biases by employing GMM, the advanced, robust and well-recognized technique in the literature. 3
Data, Models and Methodology
The empirical analysis to examine the non-linear effects of corruption on growth uses the ICRG indicator of corruption as the key explanatory variable. 4 The ICRG index is a measure of corruption within the political system that threatens foreign investment by distorting the economic and financial environment, reducing the efficiency of government and business by enabling people to assume positions of power through patronage rather than ability. This index has been used widely in the economics literature. The ICRG corruption index is measured in a scale ranging from 0 (most corrupt) to 6 (least corrupt). For ease of explanation, we rescale the index from 1 to 10, with the highest value indicating a maximum incidence of corruption. Moreover, TI’s CPI is used for robustness check. Like ICRG index, we rescale CPI from 1 to 10 and a high score indicates that corruption is perceived to be high.
Following the standard neoclassical growth model in the literature, the base model is structured as follows:
where RGDPPCYgr is growth rate of real per capita gross domestic product, and CORRP is incidence of corruption measured by ICRG and TI. The detailed descriptions of all variables are given in Table 1.
Description of Variables and Sources
The exogenous variables in the equation are secondary school enrolment (SEC), 5 population growth (POPgr), investment as a percentage of GDP (IRAT), government expenditure–GDP ratio (GRAT) and trade–GDP ratio (OPEN). eit is the error term, and subscripts i and t represent country and time, respectively. Coefficients β1 and β2 are the main focus of the analysis and measure the changes in RGDPPCYgr owing to a change in corruption, keeping all controls as constant. The expected signs of β1 and β2 should be positive and negative, respectively, to represent a bell-shaped parabolic relationship between corruption and growth. Therefore, before this threshold level of corruption the effect on growth is positive which becomes negative after the threshold level of corruption. Hence, an increase in corruption level increases growth and reaches a maximum level, that is, the threshold point, and thereafter with higher level of corruption it has a dampening (decreasing) effect on growth. To measure at what level of corruption the growth begins to decline, we estimate the threshold (turning) point of this parabolic relationship as follows:
where CORRP* is corruption at the threshold level.
Unbalanced panel data is used in this study. For the estimation procedure, both FE and SGMM methods are employed. In estimating growth equation, the existing literature uses OLS and 2SLS (Chervin & Sweder, 2010); however, there are advantages of GMM over 2SLS and OLS. According to Baum, Schaffer and Stillman (2003, p. 11), ‘if heteroskedasticity is not present, the GMM estimator is no worse asymptotically than the IV estimator’. The pooled OLS, however, fails to account for the potential endogeneity of the right-hand side variables. Specifically, it fails to account for potential unobserved country-specific variations. In general, the variables measured with an error term tend to display a bias towards zero, and OLS does not account for standard errors from the first stage estimator (Arellano, Schaffer, & Stillman, 2009). To overcome the issue, GMM estimation addresses potential endogeneity concerns between the set of cross-country regressors and other country-specific characteristics. Moreover, when the model consists of more moment conditions than model parameters and panel data set with a short time dimension and a larger country dimension, the use of GMM is more appropriate as it addresses potential endogeneity problems of the regressors and incorporates FE. Arellano and Bond (1991) pioneered the difference-GMM (DGMM) estimator, while the SGMM estimator is a product of the work done by Blundell and Bond (1998). Identification in both types of estimators is based on first-differencing and using lagged values of the endogenous variables as instruments. In the DGMM estimator, lagged levels are used to instrument the differenced right-hand side variables, while for the SGMM estimator, the estimated system is composed of a differenced equation instrumented with lagged levels, and additionally, a level equation, which is estimated using lagged differences as instruments (Bond, Hoeffler, & Temple, 2001; Rajan & Subramanian, 2008). Hence, the second-degree polynomial corruption–growth relationship is tested by employing panel FE as per Hausmann’s specification test and SGMM technique for the period 1984–2009. We test the instrument validity by using the Hansen’s J-test for over-identifying restrictions. The annual data is converted to five-year average data as five-year panel might provide better results because the impact of corruption on growth not to appear instantaneously or not even in one or two years (Acemoglu, Naidu, Restrepo, & Robinson, 2015).
Empirical Results
The polynomial fit of the scatter plots of RGDPPCYgr and CORRP in Figure 1 suggests a concave relationship between corruption and per capita growth of income. In other words, growth in per capita income increases when corruption level is low; however, after a threshold level, per capita income growth decreases as corruption increases. Hence, to estimate the impact of corruption on growth, a squared term of CORRP is incorporated. The results of the impact of corruption on growth of RGDP per capita using both FE are reported in Table 2 and the results obtained using GMM (Arellano–Bond Dynamic model; Arellano & Bond, 1991) for both ICRG and TI’s corruption indices are shown in Table 3. We start with a linear relationship in column (1). The FE results show that the coefficient for corruption is negative and insignificant indicating that there is insufficient evidence of a growth reducing effect of corruption, thus this raises the question of the validity of a linear relationship. 6 Therefore, a quadratic relationship is tested in columns (2)–(3). The results illustrate a positive linear and a negative squared term. The coefficients of corruption remain significant even after the inclusion of all control variables in column (3). This finding reveals the pattern one would expect: RGDPPCYgr increases when corruption is low; however, after a threshold level, RGDPPCYgr significantly reduces, which supports the view presented by Shleifer and Vishny (1993). The non-linear (bell-shaped parabolic) relationship is also consistent with Figure 1.
Control variables all follow the expected sign, that is, large government expenditure, a higher inflation rate and more unequal society reduce long-term growth, whereas a large SEC and more open society intensify RGDPPCYgr significantly. As expected, higher investment is beneficial to growth, although not significant. Population growth also shows the expected sign. The results for TI’s CPI are also consistent with ICRG index (columns (4)–(6)). SGMM methodology is also used to further establish the results.
Corruption and Growth: Dependent Variable Annual Growth Rate, 1984–2009 Using Fixed Effect Panel-data Estimation (Five-year Average)
Table 3 reports the estimated coefficients of equation (1) using SGMM estimation technique. 7 Columns (7)–(9) show the results of the estimations using ICRG data and columns (10)–(12) show the estimated results of the estimators where corruption is measured by TI’s CPI indexes. For the linear specification of corruption (column (7) and (10), Table 3), the estimated coefficient of corruption is negative and significant for ICRG data and positive but insignificant for CPI index data. For the full specification (in columns (9) and (12)) with linear and squared term, the linear coefficient of corruption is positive and significant and the squared-term coefficient is significantly negative implies that increase in corruption initially increases growth and after a certain threshold level of corruption (approximately 6–7 level of corruption), it reduces growth. Corruption is growth enhancing until around 6 (least corrupt) and growth demoting above 6 (most corrupt). In other words, corruption is growth augmenting and reducing at its lower and higher levels of corruption respectively. The non-linear results are consistent with both sand and oil theory of corruption. At its low level, corruption greases the wheel of growth; however, after a threshold level corruption acts as sand-in-the-wheels due to its pervasiveness. Moreover, our results validate the theoretical conjecture developed by Shleifer and Vishny (1993); however, it contradicts Swaleheen (2011) which finds that corruption decreases growth at its low level but increases growth when corruption is more pervasive; this result is against the theory.
Corruption and Growth: Dependent Variable Annual Growth Rate, 1984–2009 Using SGMM (Five-year Average)
Corruption and Growth: Dependent Variable—Annual Growth Rate, 1984–2009
Except population growth all other control variables are expected in sign and consistent with FE results. The negative population growth coefficients indicate that growth in population has negative impact on growth. The results also show that inflation reduces growth in all specification. Sargan tests for over-identifying restrictions indicate that the instrument choices for the SGMM models are valid. The model also passes the test for the absence of AR-2 process in the differenced error term.
Next, to examine the potentially different effects of corruption on growth for the least and most corrupt countries, the linear equation is estimated for two sub-samples (with corruption level between 1 and 6 for the least corrupt and above 6 for the most corrupt countries) using SGMM (Table 4). 8 The results for ICRG index show that corruption significantly increases growth for the least corrupt countries (columns (15)–(16)) and decreases growth (although not significantly) for the most corrupt countries (columns (13)–(14)). 9 Control variables demonstrate the correct signs and significant impacts. That is, secondary level of education and openness enhance per capita income growth but higher government expenditure reduces it. Population growth coefficient is negative and significant, indicating that a higher population growth is unfavourable for growth. Hansen’s J-tests for over-identifying restrictions indicate that the instrument choices for the SGMM models are valid. Interestingly, the magnitude of the coefficient of corruption decreases after incorporating IRAT in the estimation indicating that a high level of corruption erodes the beneficial effects of investment, thus growth rate decreases. Furthermore, IRAT coefficient is significant only for the least corrupt countries suggesting that the positive effect of investment on growth can be achieved only when corruption is low. Thus, this result is consistent with the theory. The results using CPI index are consistent and continue to support our quadratic growth–corruption model.
Conclusion
This article investigates the nexus between corruption and growth using panel data analysis in a non-linear framework. The results presented in this article illustrate that at a lower level, corruption increases growth but after a threshold level the effect is growth-inhibitory. In other words, corruption is beneficial for growth only when corruption is low. These findings remain robust to the use of various controls, estimations and alternative corruption indices. This non-linear pattern is consistent with both oil- and sand-in-the-wheel arguments that least corrupt countries enjoy higher growth rates but highly corrupt countries experience low growth. Some country-specific observations are also consistent with the findings. To reap the beneficial effect of corruption, anticorruption reform should be diverted to weeding out the pervasiveness of corruption. In policy perspective, government should take appropriate measures to reduce the pervasiveness of corruption to increase growth.