AFRICAN GOVERNANCE AND DEVELOPMENT
INSTITUTE
A G D I Working Paper
WP/12/031
On the effect of foreign aid on corruption
Simplice A. Asongu
African Governance and Development Institute,
P.O. Box 18 SOA/ 1365 Yaoundé, Cameroon.
E-mail: asongusimplice@yahoo.com
1
© 2012 African Governance and Development Institute
WP/12/031
AGDI Working Paper
Research Department
On the effect of foreign aid on corruption
Simplice A. Asongu1
February 2012
Abstract
The Okada & Samreth(2012, EL) finding that aid deters corruption could have an
important influence on policy and academic debates. This paper partially negates their
criticism of the mainstream approach to the aid-development nexus. Using updated
data(1996-2010) from 52 African countries we provide robust evidence of a positive aidcorruption nexus. Development assistance fuels(mitigates) corruption(the control of
corruption) in the African continent. As a policy implication, the Okada & Samreth(2012, EL)
finding for developing countries may not be relevant for Africa.
JEL Classification: B20; F35; F50; O10; O55
Keywords: Foreign Aid; Political Economy; Development; Africa
1
Simplice A. Asongu is Lead economist in the Research Department of the AGDI (asongus@afridev.org).
2
1. Introduction
The purpose of this comment is to stress some policy and methodological issues
resulting from Okada & Samreth(2012). The methodological basis of the paper is the
following: “previous research has primarily been based on Ordinary Least Squares(OLS),
instrumental variables and panel estimation. These approaches have disadvantages, as they
only estimate the parameters of interest at the mean evaluation by a conditional distribution
of the dependent variable(Billger & Goel,2009)”(p.240). To confirm this assertion we peruse
Billger & Goel(2009) and find the following: “ many previous studies of the determinants of
corruption employ OLS estimation, therefore reporting parameters estimates at the
conditional mean of corruption. While mean effects are certainly important, we expand upon
such findings using quantile regression. In addition, on underlying assumption for OLS
regression is that the error term and the dependent variables are normally distributed…..OLS
estimation can yield unreliable estimates, but quantile regression does not require a normally
distributed error term”(pp.300-301). Three facts result from this cross-examination:
- Billger & Goel(2009) do not invalidate panel instrumental variable estimation techniques;
-if the classical conditions for the validity of OLS are satisfied, that is, if the error term is
independently and identically distributed, conditional on the independent variables, then
quantile regression is redundant: all the conditional quantiles of the dependent variable will
march in lockstep with the conditional mean;
-while the Okada & Samreth(2012) criticism is valid with respect to OLS, it is short of
substance when extended to some instrumental and dynamic panel estimation techniques.
In this comment we assess the effect of foreign aid on corruption using two panel
estimation techniques in the context of Africa. The choice of Africa is based on the substantial
reliance of the continent on the ‘Big-Push’ development(poverty-reduction) policy. The rest
3
of the paper is organized as follows. Section 2 presents data and outlines the methodology.
Section 3 covers the empirical analysis. Section 4 concludes.
2.Data and Methodology
2.1 Data
We investigate a panel of 52 African countries with data from African Development
Indicators (ADI) of the World Bank (WB) ranging from 1996 to 2010. Okada &
Samreth(2012) have used data(1995 to 2009) from 120 developing countries. The outcome
variables are
the ‘control of corruption’ and the ‘corruption perception’ indexes. The
explaining variable is Net Official Development Assistance(NODA). For robustness checks
we use total NODA, NODA from Multilateral donors and NODA from the Development
Assistance Committee(DAC) countries. In the estimations we control for openness(trade),
autocracy and democracy. The choice of control variables is contingent on the degrees of
freedom necessary for overidentifying restrictions tests at second-stage regressions(more than
two control variables will result in exact or under-identification; meaning instruments are
either equal to or less than the number of endogenous explaining variables respectively). The
aid and trade variables are in percentage of GDP. Instrumental variables include: legalorigins, income-levels and religious-dominations. These instruments have been substantially
documented in the economic development literature (La Porta et al., 1997; Beck et al., 2003).
2.2 Methodology
2.2.1 Endogeneity
While development assistance affects the quality of institutions in the recipient
countries, some foreign-aid is also contingent on the quality of institutions in the beneficiary
countries. We are thus faced with an important issue of endogeneity owing to reversecausality and omitted variables. To address this concern we shall assess the presence of
4
endogeneity with the Hausman-test and selection of estimation technique will depend on the
outcome of the test.
2.2.2 Estimation techniques
a) HAC Two-Stage Least Squares(TSLS) Instrumental Variables(IV)
The TSLS is preceded by the Hausman test for endogeneity. The null hypothesis of
this test is the stance that OLS estimates are efficient and consistent; therefore a rejection of
this null hypothesis points to the presence of endogeneity and hence an estimation approach
that incorporates it. Before estimation we verify that the instruments are exogenous to the
endogenous components of explaining variables(aid channels) conditional on other
covariates(control variables). Borrowing from Beck et al.(2003) with use the TSLS-IV with
Heteroscedasticity and Autocorrelation Consistent(HAC) standard errors. The validity of the
instruments is assessed by the Sargan Overidentifying Restrictions(OIR) test. The null
hypothesis of this test is the position that the instruments are not correlated with the error term
in the equation of interest(do not suffer from endogeneity).
b) System Generalized Methods of Moments(Dynamic Panel)
Blundell & Bond(1998) proposed another approach to the issue of endogeneity with
an application of the Generalized Method of Moments(GMM) that exploits all the
orthogonality conditions between the dependent lagged variables and the error term. We
prefer the second-step GMM since it corrects the residuals for heteroscedasticity. In the firststep the residuals are homoscedastic. The hypothesis of no auto-correlation in residuals is
crucial as past differenced variables are to be used as instruments for the dependent variables.
This concern is addressed with the second-order autocorrelation test: AR(2). Also the
estimation depends on the assumption that the lagged values of the outcome variable and
5
other explaining variables are valid instruments in the estimation. The validity of the
instruments is investigated by the Sargan over-identifying restrictions test(OIR).
2.2.3 Robustness checks
To ensure robustness of the analysis, the following checks will be carried out: (1)
usage of alternative NODA indicators ; (2) employment of two distinct interchangeable sets of
moment conditions that encompass every category of the instruments; (3) usage of alternative
corruption indicators; (4) account for the concern of endogeneity; (5) estimation with robust
Heteroscedasticity and Autocorrelation Consistent(HAC) standard errors;(6) application of
restricted and unrestricted regressions.
3.Empirical results
3.1 Instrumental panel(TSLS)
Table 1 below presents results in HAC standard errors for restricted(panel A) and
unrestricted(panel B) TSLS-IV regressions. Rejection of the null hypothesis of the Hausman
test in all regressions confirms the presence of endogeneity and hence the choice of the IV
estimation approach. Failure to reject the hull hypothesis of the Sargan-OIR test lends credit
to the validity of the instruments. Clearly, it could be noticed that foreign aid significantly
diminishes the control of corruption and the CPI. Reduction in the CPI indicates increase in
corruption(see Transparency International CPI computation). These results are robust to the
alternative set of instrumental variables.
3.2 Dynamic Panel(System GMM)
Table 2 presents dynamic panel system GMM estimation results for restricted(panel
A) and unrestricted(panel B) regressions. Failure to reject the null hypotheses of the AR(2)
and Sargan-OIR tests for the most part confirms the absence of autocorrelation in the
6
residuals and validity of the instruments respectively. The results broadly confirm those in
Table 1.
Table 1: Two-Stage Least Squares Instrumental Variable regressions
Panel A: Restricted regressions(HAC standard errors)
NODAgdp
Control of Corruption
-----
NODAMDgdp
-0.035***
(0.000)
---
NODADACgdp
---
-0.082***
(0.000)
---
Democracy
0.101*
(0.086)
-0.032
(0.773)
-0.005
(0.223)
234.028***
(0.000)
0.024
(0.875)
0.106
16.099***
488
0.119*
(0.078)
-0.000
(0.999)
-0.007
(0.169)
255.223***
(0.000)
0.109
(0.741)
0.098
14.177***
488
Autocracy
Trade
Hausman
Sargan-OIR
Adjusted R²
Fisher
Observations
---0.062***
(0.000)
0.087
(0.116)
-0.058
(0.575)
-0.004
(0.322)
233.669***
(0.000)
0.000
(0.996)
0.094
17.011***
488
Corruption Perception Index(CPI)
-----0.032*
(0.060)
-----0.074*
(0.068)
-----0.058*
(0.057)
0.261
0.275
0.248
(0.105)
(0.104)
(0.110)
0.171
0.200
0.145
(0.516)
(0.459)
(0.577)
0.027*
0.025*
0.028**
(0.050)
(0.075)
(0.035)
501.364***
495.951***
504.967***
(0.000)
(0.000)
(0.000)
2.122
2.290
1.982
(0.145)
(0.130)
(0.159)
0.180
0.177
0.178
148.337***
158.260***
138.526***
368
368
368
Panel B: Unrestricted Regressions(HAC standard errors)
Control of Corruption
Constant
Corruption Perception Index
-0.649***
(0.000)
---
-0.621***
(0.000)
---
NODAMDgdp
-0.631***
(0.000)
-0.023**
(0.014)
---
---
NODADACgdp
---
-0.053**
(0.017)
---
-0.041**
---0.125***
(0.013)
(0.000)
0.107**
0.104**
0.255***
0.259***
0.252***
(0.017)
(0.018)
(0.000)
(0.000)
(0.000)
50.302***
49.910***
115.635***
118.12***
118.09***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
0.695
0.214
2.383
2.086
3.825
(0.706)
(0.898)
(0.303)
(0.352)
(0.147)
0.172
0.167
0.241
0.235
0.225
6.315***
6.400***
21.499***
20.853***
21.255***
514
514
388
388
388
Constant; English ; Christianity; Middle Income; Lower Middle Income
Constant; French; Islam; Lower Income; Upper Middle Income
NODAgdp
Democracy
0.105**
(0.017)
Hausman
49.346***
(0.000)
Sargan-OIR
0.039
(0.980)
Adjusted R²
0.177
Fisher
6.416***
Observations
514
First-Set of Instruments
Second-Set of Instruments
2.782***
(0.000)
-0.068***
(0.000)
---
2.727***
(0.000)
---
2.813***
(0.000)
---
-0.150***
(0.000)
---
---
*;**;***: significance levels of 10%, 5% and 1% respectively. OIR: Overidentifying Restrictions test. NODAgdp: NODA on GDP.
NODAMD: NODA from Multilateral Donors on GDP. NODADACgdp: NODA from DAC countries on GDP. OIR: Overidentifying
Restrictions test. P-values in brackets.
7
Table 2: Dynamic System GMM regressions
Panel A: Restricted regressions
Initial
Control of Corruption
0.789***
0.790***
(0.000)
(0.000)
-----
NODAMDgdp
0.785***
(0.000)
-0.005***
(0.004)
---
NODADACgdp
---
-0.010**
(0.042)
---
Democracy
0.001
(0.576)
1.324
(0.185)
47.079
(0.347)
547.996***
(0.000)
334
0.001
(0.694)
1.272
(0.203)
46.156
(0.383)
420.894***
(0.000)
334
NODAgdp
AR(2)
Sargan-OIR
Wald
Observations
---0.007***
(0.001)
0.0005
(0.873)
1.366
(0.171)
45.410
(0.413)
648.423***
(0.000)
334
Corruption Perception Index
0.873***
0.870***
0.874***
(0.000)
(0.000)
(0.000)
0.005
----(0.108)
----0.015*
(0.081)
----0.008
(0.130)
0.045*
0.045**
0.046*
(0.055)
(0.022)
(0.055)
1.812*
1.821*
1.799*
(0.069)
(0.068)
(0.072)
44.902
44.891
44.769
(0.966)
(0.966)
(0.967)
6836.4***
6437.15***
6876.4***
(0.000)
(0.000)
(0.000)
335
335
335
Panel B: Unrestricted regressions
Control of Corruption
Initial
NODAMDgdp
0.681***
(0.000)
-0.250***
(0.000)
-0.001***
(0.005)
---
NODADACgdp
---
-0.003
(0.133)
---
Democracy
0.021***
(0.000)
1.353
(0.175)
46.024
(0.388)
175.78***
(0.000)
334
0.023***
(0.000)
1.332
(0.182)
45.935
(0.392)
137.485***
(0.000)
334
Constant
NODAgdp
AR(2)
Sargan-OIR
Wald
Observations
0.668***
(0.000)
-0.267***
(0.000)
---
0.689***
(0.000)
-0.248***
(0.000)
-----0.002***
(0.005)
0.020***
(0.000)
1.347
(0.177)
45.431
(0.412)
172.401***
(0.000)
334
Corruption Perception Index
0.776***
(0.000)
0.597***
(0.001)
-0.003
(0.148)
---
0.776***
(0.000)
0.594***
(0.003)
---
0.780***
(0.000)
0.582***
(0.000)
-----
---
-0.008
(0.153)
---
0.026**
(0.016)
1.943*
(0.051)
44.569
(0.969)
377.631***
(0.000)
335
0.026**
(0.027)
1.933*
(0.053)
44.553
(0.969)
376.473***
(0.000)
335
-0.005
(0.144)
0.024**
(0.019)
1.949*
(0.051)
44.759
(0.967)
385.711***
(0.000)
335
*;**;***: significance levels of 10%, 5% and 1% respectively. OIR: Overidentifying Restrictions test. NODAgdp: NODA on GDP.
NODAMD: NODA from Multilateral Donors on GDP. NODADACgdp: NODA from DAC countries on GDP. OIR: Overidentifying
Restrictions test. AR(2): Second order auto correlation test. Wald: statistics for joint significance of estimated coefficients. Initial: lagged
endogenous variable. P-values in brackets.
4. Conclusion
The Okada & Samreth(2012, EL) finding that aid deters corruption could have an
important influence on policy and academic debates. This paper partially negates their
criticism of the mainstream approach to the aid-development nexus. Using updated
8
data(1996-2010) from 52 African countries we provide robust evidence of a positive aidcorruption nexus. Development assistance fuels(mitigates) corruption(the control of
corruption) in the African continent. As a policy implication, the Okada & Samreth(2012, EL)
finding for developing countries may not be relevant for Africa.
References
Blundell, R.W., & Bond, S.R., (1998), “Initial Conditions and Moment Restrictions in
Dynamic Panel Data Models”, Journal of Econometrics, 87, pp.115-143.
Beck, T., Demirgüç-Kunt, A., & Levine, R.,(2003), “Law and finance: why does legal origin
matter?”, Journal of Comparative Economics, 31, pp. 653-675.
Billger, S.M., & Goel, R.K., (2009), “Do existing corruption levels matter in controlling
corruption?cross-country quantile regression estimates”, Journal of Development
Economics, 90, pp. 299–305.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R.W., (1997). “Legal
Determinants of External Finance”, Journal of Finance, 52, pp. 1131-1150.
Okada, K., & Samreth, S.,(2012), “The effect of foreign aid on corruption: A quantile
regression approach”, Economic Letters, 11, pp.240-243.
9
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
1. Introduction
The purpose of this comment is to stress some policy and methodological issues
resulting from Okada & Samreth (2012). The methodological basis of the paper is the
following: “previous research has primarily been based on Ordinary Least Squares (OLS),
instrumental variables and panel estimation. These approaches have disadvantages, as they
only estimate the parameters of interest at the mean evaluation by a conditional distribution
of the dependent variable (Billger & Goel, 2009)”(p.240). To confirm this assertion we
peruse Billger & Goel (2009) and find the following: “ many previous studies of the
determinants of corruption employ OLS estimation, therefore reporting parameters estimates
at the conditional mean of corruption. While mean effects are certainly important, we expand
upon such findings using quantile regression. In addition, an underlying assumption for OLS
regression is that the error term and the dependent variables are normally distributed…..OLS
estimation can yield unreliable estimates, but quantile regression does not require a normally
distributed error term”(pp.300-301). Three facts result from this cross-examination:
- Billger & Goel (2009) do not invalidate panel instrumental variable estimation techniques;
-if the classical conditions for the validity of OLS are satisfied, that is, if the error term is
independently and identically distributed, conditional on the independent variables, then
quantile regression is redundant: all the conditional quantiles of the dependent variable will
march in lockstep with the conditional mean;
-while the Okada & Samreth (2012) criticism is valid with respect to OLS, it is short of
substance when extended to some instrumental and dynamic panel estimation techniques.
In this comment we assess the effect of foreign aid on corruption using two panel
estimation techniques in the context of Africa. The choice of Africa is based on the substantial
reliance of the continent on the ‘Big-Push’ development (poverty-reduction) policy. The rest
of the paper is organized as follows. Section 2 presents data and outlines the methodology.
Section 3 covers the empirical analysis. Section 4 concludes.
2. Data and Methodology
2.1 Data
We investigate a panel of 52 African countries with data from African Development
Indicators (ADI) of the World Bank (WB) ranging from 1996 to 2010. Okada & Samreth
(2012) have used data (1995 to 2009) from 120 developing countries. The outcome variables
are the ‘control of corruption’ and the ‘corruption perception’ indexes. The explaining
variable is Net Official Development Assistance (NODA). For robustness checks, we use total
NODA, NODA from Multilateral donors and NODA from the Development Assistance
Committee (DAC) countries. In the estimations, we control for openness (trade), autocracy
and democracy. The choice of control variables is contingent on the degrees of freedom
necessary for overidentifying restrictions tests at second-stage regressions (more than two
control variables will result in exact or under-identification; meaning instruments are either
equal-to or less-than the number of endogenous explaining variables respectively). The aid
and trade variables are in percentage of GDP. Instrumental variables include: legal-origins,
income-levels and religious-dominations. These instruments have been substantially
documented in the economic development literature (La Porta et al., 1997; Beck et al., 2003).
2175
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
2.2 Methodology
2.2.1 Endogeneity
While development assistance affects the quality of institutions in the recipient
countries, some foreign-aid is also contingent on the quality of institutions in the beneficiary
countries. We are thus faced with an important issue of endogeneity owing to reversecausality and omitted variables. To address this concern we shall assess the presence of
endogeneity with the Hausman-test and selection of estimation technique will depend on the
outcome of the test.
2.2.2 Estimation techniques
HAC Two-Stage Least Squares (TSLS) Instrumental Variables (IV)
The TSLS is preceded by the Hausman test for endogeneity. The null hypothesis of
this test is the stance that OLS estimates are efficient and consistent; therefore a rejection of
this null hypothesis points to the presence of endogeneity and hence, an estimation approach
that incorporates it. Before estimation, we verify that the instruments are exogenous to the
endogenous components of explaining variables (aid channels) conditional on other covariates
(control variables). Borrowing from Beck et al. (2003), with use the TSLS-IV with
Heteroscedasticity and Autocorrelation Consistent (HAC) standard errors. The validity of the
instruments is assessed by the Sargan Overidentifying Restrictions (OIR) test. The null
hypothesis of this test is the position that the instruments are not correlated with the error term
in the equation of interest (do not suffer from endogeneity).
System Generalized Methods of Moments (Dynamic Panel)
Blundell & Bond (1998) proposed another approach to the issue of endogeneity with
an application of the Generalized Method of Moments (GMM) that exploits all the
orthogonality conditions between the lagged dependent variables and the error term. We
prefer the second-step GMM since it corrects the residuals for heteroscedasticity. In the firststep, the residuals are homoscedastic. The hypothesis of no auto-correlation in residuals is
crucial as past differenced variables are to be used as instruments for the dependent variables.
This concern is addressed with the second-order autocorrelation test: AR(2). Also the
estimation depends on the assumption that the lagged values of the outcome variable and
other explaining variables are valid instruments in the estimation. The validity of the
instruments is investigated by the Sargan over-identifying restrictions test (OIR).
2.2.3 Robustness checks
To ensure robustness of the analysis, the following checks will be carried out: (1)
usage of alternative NODA indicators ; (2) employment of two distinct interchangeable sets of
moment conditions that encompass every category of the instruments; (3) usage of alternative
corruption indicators; (4) account for the concern of endogeneity; (5) estimation with robust
Heteroscedasticity and Autocorrelation Consistent (HAC) standard errors; (6) application of
restricted and unrestricted regressions.
2176
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
3.Empirical results
3.1 Instrumental Panel (TSLS)
Table 1 below presents results in HAC standard errors for restricted (Panel A) and
unrestricted (Panel B) TSLS-IV regressions. Rejection of the null hypothesis of the Hausman
test in all regressions confirms the presence of endogeneity and hence the choice of the IV
estimation approach. Failure to reject the hull hypothesis of the Sargan-OIR test lends credit
to the validity of the instruments. Clearly, it could be noticed that foreign-aid significantly
diminishes the ‘control of corruption’ and the Corruption Perception Index (CPI). Reduction
in the CPI indicates an increase in corruption (see Transparency International CPI
computation). These results are robust to the alternative set of instrumental variables.
3.2 Dynamic Panel (System GMM)
Table 2 presents dynamic panel system GMM estimation results for restricted (Panel
A) and unrestricted (Panel B) regressions. Failure to reject the null hypotheses of the AR(2)
and Sargan-OIR tests for the most part confirms the absence of autocorrelation in the
residuals and validity of the instruments respectively. The results broadly confirm those in
Table 1.
2177
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
Table I: Two-Stage Least Squares Instrumental Variable regressions
Panel A: Restricted regressions (HAC standard errors)
NODAgdp
Control of Corruption
-----
NODAMDgdp
-0.035***
(0.000)
---
NODADACgdp
---
-0.082***
(0.000)
---
Democracy
0.101*
(0.086)
-0.032
(0.773)
-0.005
(0.223)
234.028***
(0.000)
0.024
(0.875)
0.106
16.099***
488
0.119*
(0.078)
-0.000
(0.999)
-0.007
(0.169)
255.223***
(0.000)
0.109
(0.741)
0.098
14.177***
488
Autocracy
Trade
Hausman
Sargan-OIR
Adjusted R²
Fisher
Observations
---0.062***
(0.000)
0.087
(0.116)
-0.058
(0.575)
-0.004
(0.322)
233.669***
(0.000)
0.000
(0.996)
0.094
17.011***
488
Corruption Perception Index (CPI)
-0.032*
----(0.060)
---0.074*
--(0.068)
-----0.058*
(0.057)
0.261
0.275
0.248
(0.105)
(0.104)
(0.110)
0.171
0.200
0.145
(0.516)
(0.459)
(0.577)
0.027*
0.025*
0.028**
(0.050)
(0.075)
(0.035)
501.364***
495.951***
504.967***
(0.000)
(0.000)
(0.000)
2.122
2.290
1.982
(0.145)
(0.130)
(0.159)
0.180
0.177
0.178
148.337***
158.260***
138.526***
368
368
368
Panel B: Unrestricted Regressions (HAC standard errors)
Control of Corruption
Constant
NODAMDgdp
-0.631***
(0.000)
-0.023**
(0.014)
---
NODADACgdp
---
NODAgdp
Democracy
0.105**
(0.017)
Hausman
49.346***
(0.000)
Sargan-OIR
0.039
(0.980)
Adjusted R²
0.177
Fisher
6.416***
Observations
514
First-Set of Instruments
Second-Set of Instruments
Corruption Perception Index
-0.649***
(0.000)
---
-0.621***
(0.000)
---
-0.053**
(0.017)
---
---
2.782***
(0.000)
-0.068***
(0.000)
---
2.727***
(0.000)
---
2.813***
(0.000)
---
-0.150***
(0.000)
---
---
-0.041**
---0.125***
(0.013)
(0.000)
0.107**
0.104**
0.255***
0.259***
0.252***
(0.017)
(0.018)
(0.000)
(0.000)
(0.000)
50.302***
49.910***
115.635***
118.12***
118.09***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
0.695
0.214
2.383
2.086
3.825
(0.706)
(0.898)
(0.303)
(0.352)
(0.147)
0.172
0.167
0.241
0.235
0.225
6.315***
6.400***
21.499***
20.853***
21.255***
514
514
388
388
388
Constant; English ; Christianity; Middle Income; Lower Middle Income
Constant; French; Islam; Lower Income; Upper Middle Income
*;**;***: significance levels of 10%, 5% and 1% respectively. OIR: Overidentifying Restrictions test. NODAgdp: NODA on GDP.
NODAMD: NODA from Multilateral Donors on GDP. NODADACgdp: NODA from DAC countries on GDP. OIR: Overidentifying
Restrictions test. P-values in brackets.
2178
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
Table II: Dynamic System GMM regressions
Panel A: Restricted regressions
Initial
Control of Corruption
0.789***
0.790***
(0.000)
(0.000)
-----
NODAMDgdp
0.785***
(0.000)
-0.005***
(0.004)
---
NODADACgdp
---
-0.010**
(0.042)
---
Democracy
0.001
(0.576)
1.324
(0.185)
47.079
(0.347)
547.996***
(0.000)
334
0.001
(0.694)
1.272
(0.203)
46.156
(0.383)
420.894***
(0.000)
334
NODAgdp
AR(2)
Sargan-OIR
Wald
Observations
---0.007***
(0.001)
0.0005
(0.873)
1.366
(0.171)
45.410
(0.413)
648.423***
(0.000)
334
Corruption Perception Index
0.873***
0.870***
0.874***
(0.000)
(0.000)
(0.000)
0.005
----(0.108)
--0.015*
--(0.081)
----0.008
(0.130)
0.045*
0.045**
0.046*
(0.055)
(0.022)
(0.055)
1.812*
1.821*
1.799*
(0.069)
(0.068)
(0.072)
44.902
44.891
44.769
(0.966)
(0.966)
(0.967)
6836.4***
6437.15***
6876.4***
(0.000)
(0.000)
(0.000)
335
335
335
Panel B: Unrestricted regressions
Control of Corruption
Initial
NODAMDgdp
0.681***
(0.000)
-0.250***
(0.000)
-0.001***
(0.005)
---
NODADACgdp
---
-0.003
(0.133)
---
Democracy
0.021***
(0.000)
1.353
(0.175)
46.024
(0.388)
175.78***
(0.000)
334
0.023***
(0.000)
1.332
(0.182)
45.935
(0.392)
137.485***
(0.000)
334
Constant
NODAgdp
AR(2)
Sargan-OIR
Wald
Observations
Corruption Perception Index
0.668***
(0.000)
-0.267***
(0.000)
---
0.689***
(0.000)
-0.248***
(0.000)
-----0.002***
(0.005)
0.020***
(0.000)
1.347
(0.177)
45.431
(0.412)
172.401***
(0.000)
334
0.776***
(0.000)
0.597***
(0.001)
-0.003
(0.148)
---
0.776***
(0.000)
0.594***
(0.003)
---
0.780***
(0.000)
0.582***
(0.000)
-----
---
-0.008
(0.153)
---
0.026**
(0.016)
1.943*
(0.051)
44.569
(0.969)
377.631***
(0.000)
335
0.026**
(0.027)
1.933*
(0.053)
44.553
(0.969)
376.473***
(0.000)
335
-0.005
(0.144)
0.024**
(0.019)
1.949*
(0.051)
44.759
(0.967)
385.711***
(0.000)
335
*;**;***: significance levels of 10%, 5% and 1% respectively. OIR: Overidentifying Restrictions test. NODAgdp: NODA on GDP.
NODAMD: NODA from Multilateral Donors on GDP. NODADACgdp: NODA from DAC countries on GDP. OIR: Overidentifying
Restrictions test. AR(2): Second order auto correlation test. Wald: statistics for joint significance of estimated coefficients. Initial: lagged
endogenous variable. P-values in brackets.
4. Conclusion
The Okada & Samreth (2012, EL) finding that aid deters corruption could have an
important influence on policy and academic debates. This paper partially negates their
criticism of the mainstream approach to the aid-development nexus. Using updated data
(1996-2010) from 52 African countries we provide robust evidence of a positive aidcorruption nexus. Development assistance fuels (mitigates) corruption (the control of
corruption) in the African continent. As a policy implication, the Okada & Samreth (2012,
EL) finding for developing countries may not be relevant for Africa.
2179
Economics Bulletin, 2012, Vol. 32 No. 3 pp. 2174-2180
References
Blundell, R. W and Bond, S. R. (1998) “Initial Conditions and Moment Restrictions in
Dynamic Panel Data Models” Journal of Econometrics, 87, 115-143.
Beck, T., Demirgüç-Kunt, A and Levine, R. (2003) “Law and finance: why does legal origin
matter?” Journal of Comparative Economics, 31, 653-675.
Billger, S. M and Goel, R. K. (2009) “Do existing corruption levels matter in controlling
corruption? cross-country quantile regression estimates” Journal of Development
Economics, 90, 299–305.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A. and Vishny, R. W. (1997) “Legal
Determinants of External Finance” Journal of Finance, 52, 1131-1150.
Okada, K. and Samreth, S. (2012), “The effect of foreign aid on corruption: A quantile
regression approach” Economic Letters, 11, 240-243.
2180