Does easy availability of cash affect corruption Evidence from [617773]

Does easy availability of cash affect corruption? Evidence from
a panel of countries
Sunny Kumar Singh *, Kaushik Bhattacharya
Business Environment Area, Indian Institute of Management, Lucknow, India
A R T I C L E I N F O
Article history:
Received 1 September 2015
Received in revised form 5 June 2016
Accepted 10 June 2016
Available online 6 March 2017
JEL classi fications:
D73E51
Keywords:
Control of corruption index
ICRG corruption index
Currency in circulation
Large denomination banknotes
Static panel data model
Dynamic panel data model
Panel Granger causalityA B S T R A C T
Using annual panel data of 54 countries for the period 2005-14, we examine whether
currency in circulation, both aggregate and in large denominations, affects the level of
corruption in a country. Standard panel data models suggest that the ratios of (i) aggregate
currency in circulation to M1 and, (ii) large denomination banknotes to M1 are both
statistically significant determinants of corruption. Tests for reverse causality within a
panel Granger framework reveal a uni-directional causality of corruption with the first
variable, but a bi-directional one with the second. These findings suggest that a limitation
in the supply of high-denomination banknotes, inter alia, could be a tool to fight corruption,
and bring to the fore the important role of payment systems, extending an earlier study by
Goel and Mehrotra (2012). The results also highlight that, along with the government, the
central bank of an economy can also play an important role in the fight against corruption.
© 2017 Elsevier B.V. All rights reserved.
1. Introduction
There are many studies in the economics literature on corruption and its cross-country determinants (Abramo, 2008;
Aidt, 2003; Bardhan,1997; Elbahnasawy and Revier, 2012; Svensson, 2005; Treisman, 2000, 2007 ), based on which the cross-
country determinants of corruption could be categorized into various economic, socio-cultural and political factors. In most
of these studies, the economic factors taken into consideration are, among others, real GDP per capita, investment, inflation,
government size, openness, population growth, educational attainment. Among the sociocultural and political factors are
measures of ethnicity, government type, freedom of press, judicial efficiency, religion and others. Interestingly, most of these
determinants, especially the socio-cultural and political ones, have a limited short-run impact but tend to influence
corruption significantly in the long run.
It may be noted that the existing literature on corruption has mostly focused on the role of the government while largely
ignoring the role of the central bank and the payment system in a country. Financial transactions are at the heart of
corruption. The examination of this angle is important in the fight against corruption because, in contrast to the
governments ’ role of ushering in institutional changes, a few changes in policies by central banks may bring quick results.
While rigorous academic studies in this area are limited, many media reports argue that transactions in cash are used to
* Corresponding author.
E-mail addresses: sunnysingh.econ@gmail.com (S.K. Singh), kbhattacharya@iiml.ac.in (K. Bhattacharya).
http://dx.doi.org/10.1016/j.ecosys.2016.06.002
0939-3625/© 2017 Elsevier B.V. All rights reserved.Economic Systems 41 (2017) 236 –247
Contents lists available at ScienceDirect
Economic Systems
journal homepa ge: www.elsev ier.com/locate /ecosys

avoid taxes, generate black money and facilitate petty corruption.1Some of these reports also recommend that governments
should avoid printing or withdraw from circulation high-denomination banknotes and simultaneously promote larger
transactions via electronic payment system only.2In the recent period, there have been some instances where economies
withdrew high-denomination banknotes to deter corruption and black markets.3
In a cross-sectional study, Goel and Mehrotra (2012) have attempted to relate corruption to measures concerning the
payment system in a country. They find that an increased use of paper-based transactions and cheques adds to corruption, while
card-based transactions reduce the prevalence of corruption. However, the scope of their study is limited, covering only 12
advanced economies over the period 2004 –08. In another study, Goel et al. (2013) , using a sample of Croatian data, study the
effects of bureaucratic monopoly on the timing and nature of bribe payments and cash bribes in particular. Their findings
suggest that a monopolist bureaucrat is more likely to demand bribes in cash. Adhikari and Bhatia (2010) probe whether the
government of India ’s shift from cash payment of wages under the Mahatma Gandhi National Rural Employment Guarantee
Scheme (MGNREGA) to settlement through bank accounts prevents defrauding workers, thereby reducing leakage of public
money. Based on a survey in a limited part of the Uttar Pradesh and Jharkhand states in India, they find that the direct transfer
of wages into workers ’ bank accounts is a substantial protection against embezzlement, provided that banking norms are
adhered to and that workers are able to manage their own accounts. In a similar study, using micro-level data from the Indian
state of Andhra Pradesh, Muralidharan et al. (2014) evaluate the impact of a biometrically authenticated payments
infrastructure on public employment and pension programs. Their results suggest that this new technology delivered a faster,
more predictable, and less corrupt payment process without adversely affecting program access.
Regarding the theoretical grounding of this issue, researchers attempting to investigate the causes of corruption generally
borrow from the broader literature on crime and punishment that considers lawbreakers (bribe payers and bribe payees in
our case) as economic agents weighing the relative costs and benefits of their actions (Becker, 1968; Goel and Mehrotra,
2012 ). However, there has been a scarcity of literature investigating the direct relation between currency in circulation and
the prevalence of corruption. Nevertheless, there are many separate studies related to currency demand and the shadow
economy showing that a large shadow economy results in increased cash demand (Drehmann et al., 2002; Schneider and
Enste, 2000 ) and shadow economy and corruption , where corruption and the shadow economy are found to ‘complement ’
each other in low-income countries (Dreher and Schneider, 2009 ). Dreher and Siemers (2009) , in related research on
corruption and the financial system, try to identify a link between corruption and capital account restrictions. They find that
more corrupt countries are more likely to impose capital controls because they are less able to collect tax revenue. This is
because, in the presence of capital controls, individuals seeking to make international transactions may offer bribes to avoid
restrictions, adding to corruption.
Based on the above discussion, in this paper we examine whether the increased use of cash and large banknotes affects
corruption. It is well known that illegal transactions thrive on anonymity. Common sense suggests that the overwhelming
majority of such transactions will avoid the banking channel and any payment involved is expected to be carried out through
cash only. The role of cash in economic transactions, relative to other assets that leave traces, could therefore be one of its
important determinants. The role of large denominations in illegal transactions has repeatedly been highlighted in the
literature on money laundering (Rogoff, 2002; Rogoff et al.,1998; Drehmann et al., 2002 ). It is well known that the availability
of large banknotes reduces the transaction costs of corruption. This brings to the fore the important role a country ’s central
bank could play in the fight against corruption by reducing the availability of large denomination banknotes.
Empirically, we test two hypotheses. First, we test whether the ratio of currency in circulation to narrow money (M1) is a
statistically significant explanatory variable of corruption across countries. Second, we test whether the total value of high-
denomination banknotes relative to M1 is another significant cross-country determinant of corruption. We also examine the
possibility of reverse causality in both these cases, i.e. whether the prevalence of corruption might influence the use of cash.
The paper proceeds as follows: Section 2 describes the data, Section 3 discusses the empirical methodology, and Section 4
presents the results. Finally, Section 5 concludes the paper.
2. The data
Our sample consists of 54 countries and covers the period 2005 –2014. We were not able to include the time period prior
to 2005 in our study due to the unavailability of denomination-wise data of cash in circulation for most of the countries,
1For example, a media report on political corruption in India finds some evidence that political parties disburse cash to voters prior to elections, for which
a huge amount of cash is held and transported from one place to another (http://indiatoday.intoday.in/story/it-is-raining-cash-in-andhra-pradesh-bypolls/
1/199369.html ). Similarly, there are reports on the popularity of 500 euro banknotes among criminals and how they facilitate global crime (http://www.
dailymail.co.uk/home/moslive/article-1246519/How-500-euro- financing-global-crime-wave-cocaine-traf ficking-black-market-tax-evasion.html ).
2http://www. firstpost.com/business/economy/we-should-abolish-rs-500-and-rs-1000-notes-completely-354908.html and http://digitalmoney.shift-
thought.co.uk/digital-money-in-india-a-path-to-better-governance/
3For instance, the 500 euro note has been withdrawn from circulation in the United Kingdom from May 2010 following concerns that it is the
denomination of choice for criminals, tax evaders and terrorists due to its cost-effective movement and storage (http://www.telegraph.co.uk/news/uknews/
crime/7714809/500-euro-notes-withdrawn-over-organised-crime-fears.html ). Similarly, with effect from October 2014, the Monetary Authority of
Singapore has stopped issuing 10000 Sg notes to help deter money laundering, with critics complaining that the note is the bill of choice for bribe-payers in
neighboring Indonesia (http://www. firstpost.com/world/singapore-to-stop-issuing-sg-1 0000-notes-to-deter-money-laundering-1612645.html ).S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 237

especially the developing ones. Our choice of countries is also constrained by data availability (e.g., countries from the euro
area are excluded due to the unavailability of individual country-speci fic data related to cash in circulation). However, the
sample is a fair mixture of high income (26), upper middle income (17) and lower middle income (11) countries.4The list of
sample countries along with their income groups is provided in Table A1 in Appendix A.
Corruption is a variable that is not measured directly. However, there are a number of indices that measure the perceived,
rather than the actual, level of corruption in a country. This paper uses two alternative measures of corruption, (1) the
Control of Corruption Index (CC) published by the World Bank ’s Worldwide Governance Indicators and (2) the International
Country Risk Guide ’s corruption index (ICRG) .5In comparison to Transparency International ’s Corruption Perception Index
(CPI), these indices are more suitable when it comes to cross-country comparisons and comparisons over time (Kaufmann
et al., 2011; Treisman, 2007 ).6According to Kaufmann et al. (2011) , the main objective of the CC is “to capture perceptions of
the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as
‘capture ’ of the state by elites and private interests ”. The CC has a range from /C02.5 (representing the highest corruption) to 2.5
(representing no corruption). On the other hand, the ICRG has a range of 0–6, with a higher score indicating less corruption.
The ICRG is captured from statements like ‘high government officials are likely to demand special payments ’ and ‘illegal
payments are generally expected throughout lower levels of government ’ in the form of ‘bribes connected with import and export
licences, exchange rate controls, tax assessment, police protections, or loans ’ (Seldadyo and de Haan, 2006; Tanzi and Davoodi,
1997 ).
We take the ratio of aggregate currency in circulation to M1 (CIC) as well as large denomination banknotes to M1 (LCIC) as
a measure of cash. Here M1 is the sum of total currency and demand deposits. We define large denomination banknotes as
the ratio of the sum of the two largest denomination banknotes to aggregate currency in circulation. The reason for taking
the two largest denomination banknotes is that some of the largest or second largest denomination banknotes came into
existence during 2005-14, while others were withdrawn from circulation. In this study, the data related to currency in
circulation and M1 were taken from the International Monetary Fund (IMF). However, it is difficult to get data on
denomination-wise banknotes from a single source like the IMF, so we compiled the data from the annual reports of the
respective countries ’ central banks. Table A2 reports the values of the two largest denomination banknotes in terms of US
dollars (USD) for each sample country. It is revealed that the largest denomination banknotes, in almost one third of the
countries in our sample, have a value of more than or equal to USD 100.
Regarding the controls to be included in our model, there is no broadly accepted theory of determinants of corruption to
guide the selection of those variables in the model. The control variables used in this study are income, government size,
openness, inflation, internet and freedom of press.7
We use GDP per capita (PCY) as a measure of income. Data for GDP per capita (in logarithmic form) is adjusted for
purchasing power parity and comes from the World Bank ’s World Development Indicators (WDI) database. Government
size (GOV) is measured as the ratio of general government final expenditure to GDP. The share of trade as a percentage of
GDP is used as a measure of openness (OPENNESS ). Inflation (INFLATION ) is measured in terms of the annual percentage
change in the consumer price index. Internet (INTERNET) is measured as the number of internet users per 100 people. The
data on government size, openness, inflation and internet are also taken from the World Bank ’s WDI database.
Freedom House publishes a data index for freedom of the press (PRESS ). The freedom of the press index ranks countries on
a scale from 0 to 100, where a lower value indicates a free press and vice versa. Tables A3 and A4 in the Appendix A provide
the summary statistics and the correlation matrix for all the variables respectively.
3. Empirical methodology
3.1. Static panel data model
Due to the lack of a strong theoretical framework for corruption, there is a lack of consensus on the proper regression
model for the analysis of corruption (Seldadyo and de Haan, 2006 ). Apart from the pooled ordinary least squares model
(pooled OLS), the two most widely used techniques for panel data analysis are the fixed effects model and the random effects
model. The fixed effects model assumes that the unobservable country-speci fic effects are fixed parameters to be estimated
along with the coefficients of the model, while the random effects model assumes the unobservable country-speci fic effects
4The classi fication is based on the World Bank Income Classi fication 2015, which is based on gross national income (GNI) per capita. High income (HI)
countries are those with a GNI of more than $12,736 in 2014, upper middle income (UMI) with a GNI of between $4125 to $12,756, and lower middle (LMI)
with a GNI between $1045 to $4125 in 2014. Those with a GNI of $1045 or lower in 2014 are low income (LI) countries.
5The number of sample countries is reduced to 51 when we use the ICRG corruption index as a corruption measure, as the index is not available for Bosnia
and Herzegovina, Korea and Kyrgyzstan.
6According to the report of the Transparency International Corruption Perception Index (2012), the CPI score is calculated based on an updated
methodology from 2012 onward. Under the previously used methodology, CPI scores are not comparable over time (http://www.transparency.org/ files/
content/pressrelease/2012_CPIUpdatedMethodology_EMBARGO_EN.pdf ).
7Initially, we also took the index for democracy as a control. However, we had to drop it because of a high correlation of around 0.93 with freedom of
press. The reason for dropping democracy rather than freedom of press is because of its inconsistent effect on corruption (Lambsdorff, 2005).238 S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247

to be a random disturbance. Despite being used widely, the fixed effects model and the random effects model both have their
advantages and limitations (Baltagi, 2008 ). To select between the two models, we rely on the Hausman (1978) speci fication
test to test the null hypothesis of no correlation between the regressors and the individual country-speci fic random effects.
Hence, to analyze the impact of cash in circulation, both aggregate as well as high-denomination banknotes, on the level of
corruption, the functional form of the panel data model is as follows:
Corruption it= a + ui+ bCICit+ gX'it+ eit;i = number of countries, t = time period (1)
Corruption it= a + ui+ bLCIC it+ gX'it+ eit;i = number of countries, t = time period (2)
Here Corruption itis the control of corruption index (CC) and the ICRG corruption index (ICRG) , CICitis the ratio of aggregate
currency in circulation to narrow money, LCIC itis the ratio of large denomination banknotes to narrow money, X 'itis a vector
of control variables, uiindicates an unobservable time-invariant country-speci fic effect that is not included in the model and
eitis the error term. Further, ui/C24 IIDð0; s2
u) and ei/C24 IID 0; s2e/C0 /C1.
As we already mentioned above regarding the inclusion of control variables in corruption studies, there is no broadly
accepted theory to guide the selection of those variables in the model. However, a variable that has been found to be a robust
and consistent determinant of corruption is GDP per capita (Serra, 2006 ). It is found that with an increase in per capita GDP,
corruption in a country tends to decrease (Bardhan, 1997 ).
Government size contributes to corruption by increasing bureaucracy and red tape and can also lower corruption when a
larger government is associated with greater checks and balances (Elbahnasawy and Revier, 2012; Rose-Ackerman, 1999 ).
Treisman (2000, 2007) argues that openness to trade is also an important determinant of corruption. He finds that greater
openness to trade increases market competition and discourages rent-seeking behavior of corrupt officials by reducing the
monopoly power of domestic producers. Inflation, measured by the consumer price index, is also a robust predictor of
corruption. It is found that countries with higher inflation have greater corruption (Treisman, 2007 ). Recent studies
(Andersen, 2009; Goel et al., 2012; Lio et al., 2011 ) have tried to show whether internet use has any impact on the level of
corruption in a country. The findings suggest that an increase in internet use has the capacity to reduce corruption; however,
its full potential is yet to be realized. Furthermore, press freedom is also found to be a significant determinant of corruption. It
is found that better press freedom enhances transparency and elevates the risk of corrupt acts (Chowdhury, 2004; Freille
et al., 2007; Serra, 2006; Treisman, 2007 ).
To choose between the fixed effects model and the random effects model, the Hausman test suggests using the fixed
effects model in the case of CC as the dependent variable and the random effects model in case the dependent variable is ICRG
in Eqs. (1) and (2).
3.2. Dynamic panel data model
We have the additional concern that corruption in a country may be highly persistent. Most of the studies related to cross-
country determinants of corruption have used lagged dependent variable models to address serial correlation in corruption
levels (Chowdhury, 2004; Dreher and Siemers, 2009; Elbahnasawy, 2014; Lio et al., 2011 ). Moreover, one of the limitations of
the fixed effects model (the random effects model) is that it assumes exogeneity of all explanatory variables with the fixed
(random) country effects. However, the disturbances contain unobservable, time-invariant country effects that may be
correlated with explanatory variables. The dynamic panel data method allows for such endogeneity by employing the
instrumental variables technique (Baltagi, 2008 ). Following this, the functional form of dynamic panel data is written as:
Corruption it= a + ui+ bCICit+ gX'it+ uCorruption i(t-1)+ eit; i = number of countries, t = time period (3)
Corruption it= a + ui+ bCICit+ gX'it+ uCorruption i(t-1)+ eit; i = number of countries, t = time period (4)
where uiis assumed to be random and independent of eitand ui/C24 IIDð0; s2
u) and ei/C24 IID 0; s2e/C0 /C1.
To estimate Eqs. (3) and (4), Arellano and Bond (1991) have suggested a generalized method of moments (GMM)
procedure where the orthogonality conditions that exist between the lagged dependent variable and the disturbances eitare
utilized to obtain additional instruments. The GMM estimator uses the lagged values of the endogenous explanatory
variables as instruments to address the endogeneity problem. Using Arellano and Bond ’s (1991) and Blundell and Bond ’s
(1998) GMM framework, we have applied the two-step system GMM8with robust standard errors proposed by Windmeijer
(2005) to estimate Eqs. (3) and (4). As compared to one-step system-GMM, two-step system GMM is asymptotically more
efficient.
8For estimating system GMM, we use the xtabond2 package in STATA developed by Roodman (2006) .S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 239

4. Empirical results
4.1. Preliminary results
To get a general idea of the relationship between corruption and cash in circulation, we plot the control of corruption
index and measures of cash in circulation, both aggregate and in large denominations, in Fig. 1. The figure shows the negative
relation between the control of corruption index and cash in circulation by using their averages over the period 2005-14. It
also reveals that most of the high and upper middle income countries fall under the upper left part of the graph, indicating
that these countries are characterized by low levels of corruption and cash in circulation simultaneously. In contrast, most of
the lower middle income countries fall under the lower right part of the graph, indicating that they are experiencing a higher
level of corruption and cash in circulation simultaneously.
Columns (1) to (3) in Table 1 also provide the exact relation between corruption and cash in circulation through pooled
OLS in the presence of all the controls. The coefficient of CIC and LCIC is around 0.61. This means that corruption in a country
tends to increase almost in a similar way, in terms of absolute values, with an increase in CIC and LCIC. Adjusted R-squared
Fig. 1. Scatterplot along with the best linear fit of the relation between corruption and cash in circulation.
Notes : + = high income, * = upper middle income, ~ = lower middle income.240 S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247

also tends to increase from 0.81 to 0.83 as we move from no measure of cash in circulation to LCIC as a measure of cash in
circulation. In columns (4)-(6), we run a similar regression with ICRG as the dependent variable for the robustness of our
results. The coefficient of CIC (0.78) and LCIC (0.59) is also highly significant; however, the absolute values are different due to
the different scale range of ICRG . Adjusted R-squared also tends to increase from 0.71 to 0.73 as we move from no measure of
cash in circulation to LCIC as a measure of cash in circulation. Apart from that, most of the controls in both equations are
significant with the expected signs of the coefficients. However, we may not rely on the pooled OLS results completely
because of the presence of endogeneity between some of the explanatory variables and possible reverse causality between
our variables of interest.
4.2. Static panel data model
Table 2 presents the results of the static panel data models. Columns (1)-(2) and columns (3)-(4) present the estimation
results with CC and ICRG as a measure of corruption, respectively. Based on the robust Hausman test statistics, columns (1)-
(2) are estimated using the fixed effects model, while columns (3)-(4) are estimated using the random effects model. The
heteroskedasticity-robust standard error is used to deal with heteroskedasticity. The signs of all the control variables confirm
the findings of the previous literature and are significant at least in one of the speci fications at least at the 10% level.
Using CC as the dependent variable in Table 2, GDP per capita is significant in each speci fication, indicating that poor
countries are more prone to corruption than rich ones. A one percent increase in GDP per capita tends to increase the control
of corruption index by around 0.26-0.36 unit in the presence of all the controls. Similarly, inflation also turns out to be
significant at least at the 10% level in one of the speci fications, which means that countries with high inflation also suffer
from higher corruption. One of the controls that are consistently highly significant across all speci fications is internet use.
This suggests that the spread of the internet has the potential to decrease the level of corruption in a country, because it
creates transparency by removing information asymmetry. Similarly, press freedom is also found to be consistently
significant at least at the 10% level. This means that an increase in press freedom tends to reduce the level of corruption in a
country. Government size and trade openness are weakly significant in at least one of the speci fications.
The sign of our main variables, i.e. cash in circulation, confirms our hypothesis to be true. CIC and LCIC are significant at
least at the 10% level. Interestingly, the impact of CIC is similar to that of LCIC probably because large banknotes cover a
substantial portion of aggregate currency in terms of value. However, there is a difference in the magnitude of the impact.
With CC as a dependent variable, one unit increase in CIC decreases the control of corruption index by 0.20 unit, i.e. it
increases the perception of corruption in a country and vice versa. In other words, frequent use of cash, rather than electronic
payment systems that can be utilized only if one maintains a deposit account in a bank, in day-to-day transactions seems to
increase the level of corruption in a country. Similarly, one unit increase in LCIC decreases the control of corruption index by
0.30 unit. In comparison to aggregate currency in circulation, the impact of large banknotes on the level of corruption seems
to be relatively high. However, with ICRG as a dependent variable, CIC is significant, whereas LCIC is insigni ficant.Table 1
Estimation results of pooled OLS.
Dependent Variable /C0 CC Dependent Variable /C0 ICRG
(1) (2) (3) (4) (5) (6)
CIC /C00.614*** /C00.775***
(/C07.34) (/C06.11)
LCIC /C00.609*** /C00.584***
(/C05.70) (/C03.88)
PCY 0.508*** 0.476*** 0.498*** 0.489*** 0.464*** 0.481***
(20.38) (18.97) (19.42) (12.33) (11.65) (11.58)
GOV /C00.019*** /C00.018*** /C00.017*** /C00.027*** /C00.028*** /C00.028***
(/C04.41) (/C04.17) (/C03.97) (/C03.80) (/C04.01) (/C03.80)
INFLATION /C00.006 /C00.005 /C00.004 /C00.013*** /C00.011** /C00.011**
(/C01.56) (/C01.30) (/C01.12) (/C02.65) (/C02.20) (/C02.14)
OPENNESS 0.002*** 0.002*** 0.002*** 0.000 0.000 0.000
(5.19) (5.60) (5.31) (0.74) (0.75) (0.50)
INTERNET 0.001 0.001 0.001 0.005*** 0.004** 0.005***
(1.04) (0.42) (0.89) (2.93) (2.22) (2.71)
PRESS /C00.015*** /C00.014*** /C00.014*** /C00.015*** /C00.014*** /C00.015***
(/C013.85) (/C013.15) (/C013.75) (/C011.51) (/C010.33) (/C010.32)
CONSTANT /C03.454*** /C03.012*** /C03.307*** /C00.651** /C00.173 /C00.461
(/C014.48) (/C012.66) (/C013.70) (/C02.01) (/C00.54) (/C01.32)
R-Squared 0.81 0.83 0.83 0.72 0.73 0.73
F Statistics 466.59 417.47 414.28 248.30 221.01 204.68
No. of Observations 508.00 504.00 488.00 478.00 476.00 460.00
Notes : *, ** and *** indicate statistical significance at the 10%, 5% and 1% level, respectively. Figures in parentheses are their respective t-statistics with robust
standard errors.S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 241

4.3. Dynamic panel data model
Tables 3 and 4 present the results of the dynamic panel data model by utilizing the two-step system GMM procedure with
CC and ICRG as a measure of corruption, respectively. In each case, columns (1) to (4) present the results with different
speci fications. Columns (1) and (2) provide the results of simple two-step system GMM, while columns (3) and (4) present
the results by collapsed instruments, which are used to limit the number of instruments generated in system GMM and avoid
bias in the results.9
Based on these alternative estimation options, our estimation results in terms of the direction of the coefficients remain
almost same. The effect of the past level of corruption is statistically significant at the 1% level with a positive sign in all
models. Therefore, corruption does seem to have inertia, and that part of present corruption contributes to its initial
conditions significantly. The presence of lagged levels of corruption in the explanatory variables reduces the magnitude of
CIC and LCIC including the controls significantly. Using CC as a dependent variable, one unit increase in CIC significantly
decreases the control of corruption index by 0.16-0.20 unit, while LCIC is insigni ficant in both cases. Using ICRG as a
dependent variable, one unit increase in CIC significantly decreases the control of corruption index by 0.31-0.22 unit, while
one unit increase in LCIC significantly decreases the ICRG corruption index by 0.22-0.14 unit. Hence, we may conclude that an
increase in cash in circulation increases the level of corruption in a country even after taking care of endogeneity.
Among the controls, the effects of GDP per capita, government size, inflation, and freedom of press are statistically
significant, while openness and internet are statistically insigni ficant. However, the direction of the impact is more or less
similar to the previous models.
4.4. Reverse causality
In some circumstances, it is possible that the prevalence of corruption in a country may pressurize the central bank to
supply the desired amount of cash. To empirically investigate the causal relationship between cash in circulation, CIC as well
as LCIC, and corruption, we employ the panel Granger causality test, which utilizes both cross-sectional and time series data
and is therefore more efficient than solely utilizing time series data (Dumitrescu and Hurlin, 2012 ). Following Lio et al.
(2011) , we apply the system GMM estimator (Arellano and Bond, 1991; Arellano and Bover, 1995; Blundell and Bond, 1998 ),Table 2
Estimation results of the static panel data model.
Dependent Variable- CC Dependent Variable- ICRG
Fixed effect Fixed effect Random effect Random effect
(1) (2) (3) (4)
CIC /C00.198*** /C00.365**
(/C02.64) (/C02.01)
LCIC /C00.299** /C00.305
(/C01.99) (/C00.79)
PCY 0.266* 0.362** 0.392*** 0.413***
(1.75) (2.20) (6.20) (6.15)
GOV 0.007 0.009* 0.026 0.023
(1.50) (1.91) (1.59) (1.39)
INFLATION /C00.003** /C00.002* /C00.005 /C00.005
(/C02.39) (/C01.75) (/C01.12) (/C01.12)
OPENNESS /C00.000 /C00.000 0.002* 0.001
(/C00.08) (/C00.16) (1.86) (1.34)
INTERNET 0.003** 0.003** 0.004* 0.005*
(2.45) (2.33) (1.67) (1.90)
PRESS /C00.002 /C00.002 /C00.008** /C00.007*
(/C01.14) (/C01.15) (/C02.28) (/C01.67)
CONSTANT /C01.853 /C02.735* /C01.199 /C01.515
(/C01.42) (/C01.91) (/C01.36) (/C01.61)
R-Squared (Within) 0.11 0.11 0.12 0.1
R-Squared (Overall) 0.75 0.75 0.69 0.68
F Statistics (p-value) 4.82 5.83 17.14 13.47
Robust Hausman Test (p-value) 0 0 0.07 0.10
No. of Countries 54 54 51 51
No. of Observations 504 488 476 460
Notes : *, ** and *** indicate statistical significance at the 10%, 5% and 1% level, respectively. Figures in parentheses are their respective t-statistics with robust
standard errors clustered by country. The robust Hausman test is based on Schaffer and Stillman (2011) , user written command in xtoverid , which is
recommended when robust standard errors are used in panel data.
9A large instrument collection, as in system or difference GMM without collapsed instruments, overfits endogenous variables even as it weakens the
Hansen test of the instruments ’ joint validity. For more details on the implementation of this technique, see Roodman (2009) .242 S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247

which employs an instrumental variables technique to estimate autoregressive equations. The null hypothesis is that cash in
circulation does not Granger cause corruption and vice versa. Using the model and moment selection criteria (MMSC) for
GMM estimation (Andrews and Lu, 2001 ), we calculate the MMSC Akaike information criteria (MMSC-AIC) and the MMSC
Bayesian information criteria (MMSC-BIC) to choose the optimal number of lags, and taking into consideration the limited
time period, we use one to two lags for testing panel Granger causality based on measures of corruption and cash in
circulation.10
Table 5 presents the panel Granger causality results. The Wald tests for CIC and corruption show that the direction of
causality is from CIC to corruption with an appropriate lag. This means that CIC in the past period leads to an increase in
corruption in the present period, not the other way around. Similarly, the Wald tests for LCIC and corruption show evidence
of bi-directional causality, which means that an increase in LCIC in the past increases current corruption. Also, past
corruption increases LCIC. We therefore conclude that easy availability of large banknotes facilitates corruption, and a
corrupt environment could also sustain their availability, implying that the institutional environment of printing decisions of
large banknotes could be an as yet unexamined determinant of corruption.
The presence of reverse causality may bias the estimation results based on pooled OLS and the static panel data model.
We have tried to solve this issue in the dynamic panel framework by applying system GMM. However, the problem of reverse
causality has to be addressed in a separate work studying the impact of corruption on cash in circulation.
5. Conclusion
In this paper, we examine whether cash in circulation affects the level of corruption in a country. The results suggest that
the ratios of (i) aggregate currency in circulation to M1 and, (ii) large denomination banknotes to M1 are both statistically
significant determinants of corruption across countries. Tests for reverse causality within a panel Granger framework reveal
a uni-directional causality of corruption with the first variable, but a bi-directional one with the second variable. From aTable 3
Estimation results of the dynamic panel data model.
Dependent Variable- CC
Two-step Sys-GMM Two-step Sys-GMM (CL)
(1) (2) (3) (4)
CC(-1) 0.743*** 0.792*** 0.667*** 0.728***
(6.13) (7.41) (4.54) (5.08)
CIC /C00.157* /C00.198*
(/C01.71) (/C02.03)
LCIC /C00.077 /C00.116
(/C00.86) (/C01.03)
PCY 0.117* 0.095* 0.166* 0.143
(1.84) (1.69) (1.96) (1.66)
GOV /C00.006* /C00.006* /C00.008** /C00.008*
(/C01.94) (/C01.71) (/C02.31) (/C01.89)
INFLATION /C00.003** /C00.003** /C00.003*** /C00.003**
(/C02.44) (/C02.57) (/C02.70) (/C02.25)
OPENNESS 0.000 0.000 0.000 0.000
(1.01) (0.52) (1.43) (0.99)
INTERNET 0.000 0.001 0.000 0.001
(0.36) (0.99) (0.07) (0.75)
PRESS /C00.003* /C00.003 /C00.004** /C00.003*
(/C01.91) (/C01.66) (/C02.34) (/C01.90)
CONSTANT /C00.709* /C00.604 /C01.030* /C00.959
(/C01.69) (/C01.54) (/C01.79) (/C01.52)
F Statistics 1054.23 890.87 1156.70 841.49
Hansen Test (p-value) 0.34 0.29 0.54 0.38
AR(2) (p-value) 0.15 0.14 0.18 0.15
No. of Instruments 28 28 18 18
No. of Countries 54 54 54 54
No. of Observations 454.00 442.00 454.00 442.00
Notes : *, ** and *** indicate statistical significance at the 10%, 5% and 1% level, respectively. Figures in parentheses are their respective t-statistics with
Windmeijer-corrected cluster-robust standard errors. CL denotes two-step system GMM estimation with collapsed instruments. The row for the Hansen
test reports the p-values for the null hypothesis of instrument validity. The values reported for AR(2) are the p-values for second order autocorrelated
disturbances in the first differences equations.
10The MMSC-AIC criterion selects the parameters and the instruments that minimize the following formula: MMSC /C0 BIC ¼ Ji/C0 log Nð Ț li/C0 Ki ð Ț, where Ji
refers to the Sargan test statistics used to test the validy of over-identifying restrictions evaluated under model i, Kiis the number of
parameters to be estimated, liis the number of moment conditions under model i and N is the sample size. Replacing log Nð Ț with 2 in the
above formula will give MMSC /C0 AIC. The number of instruments is based on the collapsed instruments.S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 243

policy perspective, we suggest that governments should evolve laws to prohibit cash transactions beyond a threshold level.
Lastly, central banks should also try to reduce high-denomination banknotes significantly, as the evidence suggests that
these are rarely used in legal transactions.
We stress that reducing the supply of large denominations could be a novel, relatively costless and practically
implementable option to limit corruption. In this context, it may be noted that earlier studies almost invariably highlighted
the important role of the government in curbing corruption. However, the implementation of government policies in this
area typically involves huge amounts of transaction costs. The results of Goel and Mehrotra (2012) extended the scope of
relatively costless anti-corruption policies by highlighting the role of the payment system. Our results extend the scope of
these policies further by bringing to the fore the important role of cash and hence the role of the central bank in dealing with
corruption in an economy.Table 4
Estimation results of the dynamic panel data model.
Dependent Variable- ICRG
Two-step Sys-GMM Two-step Sys-GMM (CL)
(1) (2) (3) (4)
ICRG(-1) 0.796*** 0.822*** 0.822*** 0.862***
(11.01) (12.41) (8.40) (10.78)
CIC /C00.307*** /C00.224**
(/C03.40) (/C02.10)
LCIC /C00.218** /C00.142*
(/C02.07) (/C01.78)
PCY 0.101** 0.093** 0.091* 0.070
(2.59) (2.39) (1.79) (1.63)
GOV /C00.006 /C00.004 /C00.006 /C00.004
(/C01.38) (/C01.14) (/C01.34) (/C01.10)
INFLATION /C00.005** /C00.005* /C00.005* /C00.005*
(/C01.99) (/C01.71) (/C01.93) (/C01.84)
OPENNESS 0.000 0.000 0.000 0.000
(0.21) (0.38) (0.68) (0.76)
INTERNET /C00.000 /C00.000 /C00.001 /C00.001
(/C00.33) (/C00.15) (/C00.96) (/C01.02)
PRESS /C00.003* /C00.003 /C00.003* /C00.003*
(/C01.91) (/C01.66) (/C01.86) (/C01.88)
CONSTANT 0.048 /C00.042 0.041 0.040
(0.24) (/C00.17) (0.28) (0.25)
F Statistics 385.10 428.41 846.48 1013.35
Hansen Test (p-value) 0.29 0.24 0.33 0.29
AR(2) (p-value) 0.54 0.58 0.52 0.56
No. of Instruments 43 43 16 16
No. of Countries 51 51 51 51
No. of Observations 427.00 415.00 427.00 415.00
Notes : *, ** and *** indicate statistical significance at the 10%, 5% and 1% level, respectively. Figures in parentheses are their respective t-statistics with
Windmeijer-corrected cluster-robust standard errors. CL denotes two-step system GMM estimation with collapsed instruments. The row for the Hansen
test reports the p-values for the null hypothesis of instrument validity. The values reported for AR(2) are the p-values for second order autocorrelated
disturbances in the first differences equations.
Table 5
Panel Granger causality test.
Direction of Causality Lags Joint Significance Sum of Coefficients MMSC-AIC MMSC-BIC No. of Observations No. of Countries
CIC ! CC [1/2] 4.54* /C02.13* /C04.36 /C016.56 432 54
CC !CIC [2/2] 1.39 /C01.09 /C03.98 /C020.20 427 54
LCIC ! CC [1/1] 6.16** /C02.48** /C03.14 /C011.45 472 54
CC ! LCIC [2/2] 8.12** /C01.47 /C07.55 /C039.65 409 54
CIC ! ICRG [1/1] 6.09** /C02.47** /C03.00 /C011.25 458 51
ICRG !CIC [1/1] 1.61 /C01.27 /C03.89 /C012.13 455 51
LCIC ! ICRG [1/2] 3.78* /C01.94* /C02.79 /C014.70 392 51
ICRG ! LCIC [2/1] 3.98** /C01.98** /C00.81 /C012.67 386 51
Notes :***,**and*denote statistical significance at the 1%, 5% and 10% level, respectively. We have tested lags [1/1], [1/2], [2/1] and [2/2] and presented the
best model based on the minimum value of MMSC-AIC and MMSC-BIC. The test statistics for joint significance follow the Chi-square distribution. The test
statistics for the test of the sum of coefficients follow the z distribution if the lag length is one, and the Chi-square distribution otherwise.244 S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247

A limitation of our study is that corrupt transactions may not always involve the domestic currency, but could also involve
foreign currencies, commodities like gold, or changes in bank deposits maintained in another country. This observation
limits the scope of changes in payment practices as a policy tool in fighting corruption unless such changes are carried out on
a global scale along with other regulatory measures. Despite these limitations, we believe that the unavailability of large
banknotes will increase the transaction costs of corruption substantially. In this context, an extension of the present study
would be to examine the detailed institutional mechanism of printing high-denomination banknotes in a country and to test
whether variables like central bank independence are influential in this context and can affect corruption in a country.
Appendix A.
Table A1
List of sample countries.
Australia (HI) Ghana (LMI) Malaysia (UMI) South Africa (UMI)
Azerbaijan, Republic of (UMI) Hong Kong SAR (HI) Mexico (UMI) Sudan (LMI)
Bahamas (HI) Hungary (HI) Namibia (UMI) Sweden (HI)
Bahrain (HI) India (LMI) New Zealand (HI) Switzerland (HI)
Bosnia and Herzegovina (UMI) Iraq (UMI) Nigeria (LMI) Thailand (UMI)
Botswana (UMI) Israel (HI) Norway (HI) Tunisia (UMI)
Brazil (UMI) Jamaica (UMI) Oman (HI) Turkey (UMI)
Bulgaria (UMI) Japan (HI) Pakistan (LMI) UAE (HI)
Canada (HI) Kenya (LMI) Poland (HI) UK (HI)
Chile (HI) Korea (HI) Romania (UMI) USA (HI)
China (UMI) Kuwait (HI) Russia (HI) Yemen ((LMI)
Congo Republic (LMI) Kyrgyzstan (LMI) Saudi Arabia (HI) Zambia (LMI)
Czech Republic (HI) Latvia (HI) Serbia (UMI)
Egypt (LMI) Lithuania (HI) Singapore (HI)
Notes: HI = High Income, UMI = Upper Middle Income, LMI = Lower Middle Income.
Table A2
Values of large banknotes (in local currency and US dollar).
Country D1 D2 D1* D2* Country D1 D2 D1* D2*
Australia AUD 100 AUD 50 74.63 37.31 Lithuania LTL 500 LTL 200 160.77 64.31
Azerbaijan, Republic of AZN 100 AZN 50 95.24 47.62 Malaysia MYR 500 MYR 100 131.58 26.32
Bahamas BSD 100 BSD 50 100.00 50.00 Mexico MXN 1000 MXN 500 63.53 31.77
Bahrain BHD 20 BHD 10 52.63 26.32 Namibia N200 CE N100 CE 16.10 8.05
Bosnia and Herzegovina KM 200 KM 100 113.64 56.82 New Zealand NZD 100 NZD 50 67.11 33.56
Botswana BWP 200 BWP 100 19.92 9.96 Nigeria NGN 1000 NGN 500 5.04 2.52
Brazil BRL 100 BRL 50 31.35 15.67 Norway NOK 1000 NOK 500 122.55 61.27
Bulgaria BGN 100 BGN 50 56.82 28.41 Oman OMR 200 OMR 100 526.32 263.16
Canada C1000 CE C100 CE 781.25 78.13 Pakistan PKR 5000 PKR 1000 49.14 9.83
Chile CLP 20000 CLP 10000 30.95 15.48 Poland PLN 200 PLN 100 53.33 26.67
China RMB 100 RMB 50 16.37 8.18 Romania RON 500 RON 200 124.07 49.63
Congo Republic FC 500, 2000 & 5000 FC 200 0.54 0.22 Russia RUB 5000 RUB 1000 88.32 17.66
Czech Republic CZK 5000 CZK 2000 203.92 81.57 Saudi Arabia SAR 500 SAR 200 133.33 53.33
Egypt EGP 200 EGP 100 25.54 12.77 Serbia RSD 5000 RSD 1000 46.16 9.23
Ghana GH 50 GH 20 14.49 5.80 Singapore SGD 10000 SGD 1000 7407.41 740.74
Hong Kong SAR HKD 1000 HKD 500 128.53 64.27 South Africa ZAR 200 ZAR 100 16.01 8.01
Hungary HUF 20000 HUF 10000 71.32 35.66 Sudan SDG 50 SDG 20 8.71 3.48
India INR 1000 INR 500 15.76 7.88 Sweden SEK 1000 SEK 500 118.91 59.45
Iraq IQD 25000 IQD 10000 20.72 8.29 Switzerland CHF 1000 CHF 500 1052.63 526.32
Israel ILS 200 ILS 100 52.91 26.46 Thailand B 1000 B 500 29.43 14.71
Jamaica JMD 5000 JMD 1000 42.85 8.57 Tunisia D 50 D 20 25.38 10.15
Japan JPY 10000 JPY 5000 81.63 40.82 Turkey TRY 200 TRY 100 74.91 37.45
Kenya KES 1000 KES 500 9.93 4.97 UAE AED 1000 AED 500 272.48 136.24
Korea KRW 50000 KRW 10000 44.25 8.85 UK GBP 50 GBP 20 78.13 31.25
Kuwait KWD 100 KWD 50 333.33 166.67 USA USD 100 USD 50 100.00 50.00
Kyrgyzstan KGS 5000 KGS 1000 79.66 15.93 Yemen YR 1000 YR 500 4.66 2.33
Latvia LVL 500 LVL 100 781.25 156.25 Zambia K 50000 K 20000 9.62 3.85
Notes : Here D1 = Largest denomination currency in local currency unit, D2 = Second largest denomination currency in local currency unit. D1* = Largest
denomination currency in USD, D2* = Second largest denomination currency in USD. Source : http://www.xe.com/currencyconverter/ and each country ’s
central bank website. Exchange rate per USD is taken as average of 2014.S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 245

References
Abramo, C.W., 2008. How much do perceptions of corruption really tell us? Economics: the open-access. Open Assess. E-J. 2, 3.
Adhikari, A., Bhatia, K., 2010. NREGA wage payments: can we bank on the banks? Econ. Polit. Wkly. 45 (1), 30–37.
Aidt, T.S., 2003. Economic analysis of corruption: a survey. Econ. J. 113 (491), F632 –F652. doi:http://dx.doi.org/10.1046/j.0013-0133.2003.00171.x .
Andersen, T.B., 2009. E-Government as an anti-corruption strategy. Inf. Econ. Policy 21 (3), 201 –210. doi:http://dx.doi.org/10.1016/j.infoecopol.2008.11.003 .
Andrews, D.W.K., Lu, B., 2001. Consistent model and moment selection procedures for GMM estimation with application to dynamic panel data models. J.
Econom. 101 (1), 123 –164. doi:http://dx.doi.org/10.1016/S0304-4076(00)00077-4 .
Arellano, M., Bond, S.,1991. Some tests of speci fication for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58
(2), 277 –297. doi:http://dx.doi.org/10.2307/2297968 .
Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-components models. J. Econom. 68 (1), 29–51. doi:http://dx.doi.
org/10.1016/0304-4076(94)01642-D .
Baltagi, B., 2008. Econometric Analysis of Panel Data, vol. 1. John Wiley & Sons .
Bardhan, P., 1997. Corruption and development: a review of issues. J. Econ. Lit. 35 (3), 1320 –1346 .
Becker, G.S., 1968. Crime and punishment: an economic approach. J. Polit. Econ. 76 (2), 169 –217.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econom. 87 (1), 115 –143. doi:http://dx.doi.org/
10.1016/S0304-4076 (98)00009-8 .
Chowdhury, S.K., 2004. The effect of democracy and press freedom on corruption: an empirical test. Econ. Lett. 85 (1), 93–101. doi:http://dx.doi.org/10.1016/
j.econlet.2004.03.024 .
Dreher, A., Schneider, F., 2009. Corruption and the shadow economy: an empirical analysis. Publ. Choice 144 (1–2), 215 –238. doi:http://dx.doi.org/10.1007/
s11127-009-9513-0 .
Dreher, A., Siemers, L.-H.R., 2009. The nexus between corruption and capital account restrictions. Publ. Choice 140 (1–2), 245 –265. doi:http://dx.doi.org/
10.1007/s11127-009-9423-1 .
Drehmann, M., Goodhart, C., Krueger, M., 2002. The challenges facing currency usage: will the traditional transaction medium be able to resist competition
from the new technologies? Econ. Policy 17 (34), 193 –228.
Dumitrescu, E.-I., Hurlin, C., 2012. Testing for Granger non-causality in heterogeneous panels. Econ. Modell. 29 (4), 1450 –1460. doi:http://dx.doi.org/
10.1016/j.econmod.2012.02.014 .
Elbahnasawy, N.G., Revier, C.F., 2012. The determinants of corruption: cross-country-panel-data analysis. Dev. Econ. 50 (4), 311 –333. doi:http://dx.doi.org/
10.1111/j.1746-1049.2012.00177.x .
Elbahnasawy, N.G., 2014. E-government, internet adoption, and corruption: an empirical investigation. World Dev. 57, 114 –126. doi:http://dx.doi.org/
10.1016/j.worlddev.2013.12.005 .
Freille, S., Haque, M.E., Kneller, R., 2007. A contribution to the empirics of press freedom and corruption. Eur. J. Polit. Econ. 23 (4), 838 –862. doi:http://dx.doi.
org/10.1016/j.ejpoleco.2007.03.002 .
Goel, R.K., Mehrotra, A.N., 2012. Financial payment instruments and corruption. Appl. Financial Econ. 22 (11), 877 –886. doi:http://dx.doi.org/10.1080/
09603107.2011.628295 .
Goel, R.K., Nelson, M.A., Naretta, M.A., 2012. The internet as an indicator of corruption awareness. Eur. J. Polit. Econ. 28 (1), 64–75. doi:http://dx.doi.org/
10.1016/j.ejpoleco.2011.08.003 .
Goel, R.K., Budak, J., Rajh, E., 2013. Bureaucratic monopoly and the nature and timing of bribes: evidence from Croatian data. Compar. Econ. Stud. 55 (1), 43–
58. doi:http://dx.doi.org/10.1057/ces.2012.36 .
Hausman, J.A., 1978. Speci fication tests in econometrics. Econometrica 46 (6), 1251 –1271. doi:http://dx.doi.org/10.2307/1913827 .Table A3
Summary statistics of the variables included in the study.
Variable No. of Observations Mean Standard Deviation Minimum Maximum
Control of Corruption Index 540 0.27 1.06 /C01.58 2.46
ICRG Corruption Index 509 2.80 1.17 0.50 5.50
Currency in Circulation to M1 (Share) 535 0.35 0.22 0.01 1.02
Large banknotes to M1 (Share) 519 0.25 0.18 0.02 0.81
Log of GDP per Capita 538 8.84 1.37 6.17 11.14
Government Size 521 16.06 4.48 2.80 27.51
Inflation 529 5.54 5.56 /C010.07 53.23
Openness 529 96.48 70.76 19.12 455.28
Internet 537 43.87 27.85 0.90 96.30
Freedom of Press 540 44.92 22.97 10.00 87.00
Table A4
Correlation matrix for the variables under study.
CC ICRG CIC LCIC GDP GOV OPENNESS INFLATION INTERNET PRESS
CC 1
ICRG 0.931 1
CIC /C00.630 /C00.571 1
LCIC /C00.559 /C00.470 0.884 1
PCY 0.854 0.765 /C00.584 /C00.511 1
GOV 0.183 0.238 /C00.171 /C00.137 0.262 1
OPENNESS 0.278 0.200 /C00.071 /C00.112 0.240 /C00.235 1
INFLATION /C00.504 /C00.486 0.378 0.347 /C00.561 /C00.236 /C00.175 1
INTERNET 0.731 0.696 /C00.539 /C00.452 0.822 0.214 0.202 /C00.523 1
PRESS /C00.632 /C00.592 0.459 0.425 /C00.489 /C00.361 0.064 0.285 /C00.448 1
Note: All the correlation coefficients are significant at the 1% level.246 S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247

Kaufmann, D., Kraay, A., Mastruzzi, M., 2011. The worldwide governance indicators: methodology and analytical issues. Hague J. Rule Law 3 (02), 220 –246.
doi:http://dx.doi.org/10.1017/S1876404511200046 .
Lio, M.-C., Liu, M.-C., Ou, Y.-P., 2011. Can the internet reduce corruption? A cross-country study based on dynamic panel data models. Gov. Inf. Q. 28 (1), 47–
53. doi:http://dx.doi.org/10.1016/j.giq.2010.01.005 .
Muralidharan, K., Niehaus, P., Sukhtankar, S., 2014. Building State Capacity: Evidence from Biometric Smartcards in India (Working Paper No.19999)National
Bureau of Economic Research. . Retrieved from http://www.nber.org/papers/w19999 .
Rogoff, K., 2002. The surprising popularity of paper currency. Finance Dev. 39 (1), 56–57.
Rogoff, K., Giavazzi, F., Schneider, F., 1998. Blessing or curse? Foreign and underground demand for Euro notes. Econ. Policy 13 (26), 263 –303.
Roodman, D., 2006. How to Do Xtabond2: An Introduction to Difference and System GMM in Stata. Center for Global Development Working Paper, (103). .
Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=982943 .
Roodman, D., 2009. A Note on the theme of too many instruments. Oxford Bull. Econ. Stat. 71 (1), 135 –158. doi:http://dx.doi.org/10.1111/j.1468-
0084.2008.00542.x .
Rose-Ackerman, S., 1999. Corruption and Government: Causes, Consequences, and Reform. Cambridge University Press .
Schaffer, M.E., Stillman, S., 2011. XTOVERID: Stata Module to Calculate Tests of Overidentifying Restrictions After Xtreg, Xtivreg, Xtivreg2, XthtaylorStatistical
Software Components. . Retrieved from https://ideas.repec.org/c/boc/bocode/s456779.html .
Schneider, F., Enste, D.H., 2000. Shadow economies: size, causes, and consequences. J. Econ. Lit. 38 (1), 77–114.
Seldadyo, H., de Haan, J., 2006. The Determinants of Corruption: A Literature Survey and New Evidence, Working Paper. University of Groningen .
Serra, D., 2006. Empirical determinants of corruption: a sensitivity analysis. Publ. Choice 126 (1–2), 225 –256.
Svensson, J., 2005. Eight questions about corruption. J. Econ. Perspect. 19 (3), 19–42.
Tanzi, M.V., Davoodi, M.H.R., 1997. Corruption, Public Investment, and Growth. International Monetary Fund .
Treisman, D., 2000. The causes of corruption: a cross-national study. J. Publ. Econ. 76 (3), 399 –457. doi:http://dx.doi.org/10.1016/S0047-2727 (99)00092-4 .
Treisman, D., 2007. What have we learned about the causes of corruption from ten years of cross-national empirical research? Ann. Rev. Polit. Sci.10 (1), 211 –
244. doi:http://dx.doi.org/10.1146/annurev.polisci.10.081205.095418 .
Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient two-step GMM estimators. J. Econom. 126 (1), 25–51. doi:http://dx.doi.
org/10.1016/j.jeconom.2004.02.005 .S.K. Singh, K. Bhattacharya / Economic Systems 41 (2017) 236 –247 247