Volume 24 Issue 4 Article 4 December 2022 Corruption and Non-Performing Loans Corruption and Non-Performing Loans Ardit Gjeçi University of Ljubljana, School of Economics and Business, PhD student, Ljubljana, Slovenia Matej Marinč University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia, matej.marinc@ef.uni-lj.si Follow this and additional works at: https://www.ebrjournal.net/home Part of the Finance Commons, and the Growth and Development Commons Recommended Citation Recommended Citation Gjeçi, A., & Marinč , M. (2022). Corruption and Non-Performing Loans. Economic and Business Review, 24(4), 240-259. https://doi.org/10.15458/2335-4216.1312 This Original Article is brought to you for free and open access by Economic and Business Review. It has been accepted for inclusion in Economic and Business Review by an authorized editor of Economic and Business Review. ORIGINAL ARTICLE Corruption and Non-Performing Loans Ardit Gjeçi a , Matej Marinc b, * a University of Ljubljana, School of Economics and Business, PhD Student, Ljubljana, Slovenia b University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia Abstract This article empirically evaluates the impact of corruption on the level of non-performing loans (NPLs) using the international bank-level data, spanning over the period 2000e2016 and across 140 countries. We find a positive and statistically significant relationship between corruption and NPLs. We also analyze the channels through which cor- ruptionaffectsNPLs.WefindthattherelationshipbetweencorruptionandNPLsbecomesmorepronouncedduringand aftertheglobalfinancialcrisisandismorepronouncedforsmallerbanks.TheassociationbetweencorruptionandNPLs is stronger in countries characterized by a high level of collectivism. The link between corruption and NPLs is higher where the legal environment is weak and where economies are market-based. Keywords: Bank lending, Non-performing loans, Financial crisis, Corruption perception index, Economic development JEL classification: G20, D73, O40 Introduction T he 2008e2010 global financial crisis pressured theregulators,policymakers,andacademicsin a meticulous search for its drivers. In an extensive review of the evidence, Thakor (2018) concludes that insolvency risk was the main cause for the crisis. Bank balance-sheets deteriorated due to financial market's related losses but also due to increasing levels of non-performing loans (NPLs). IntheU.S.,theaverageproportionofNPLspertotal gross loans increased from 1.4 percent in 2007 to 4.96 percent in 2009. In the EU, the level of NPLs increased from 2.4 percent in 2007 to 6.6 percent of total assets in 2016. 1 At the same time, several financial scandals caught public attention and called for further scrutiny of unsound practices in banking. In a 2015 survey, 47 percent of 1200 financial services professionals in the U.K. and the U.S. claimed that it is necessary to engage in an illegal or unethical activity at least one time to succeed and to gain an edge in the market. 2 Excessive risk taking and failed management and board control functions were cited as causes of failure of the U.K. bank HBOS that led to its acquisition by Lloyds and government bailout (FCA & PRA, 2015). The regulator investigated the wider allegation of corruption following a criminal inves- tigation in which two former HBOS bankers and four business associates were found guilty of cor- ruption,moneylaundering,andfraud.Inparticular, theleaddirectorofHBOS'simpairedassetsdivision was taking advantage of small businesses in threatening to terminate loans unless a bribe was being paid to the restructuring consultancy com- pany led by his accomplices. This compelled HBOS to write off £266 m in loans. 3 Other examples point to the anecdotal evidence between corruption and badloans,includingthearrestbytheIndianCentral Bureau of Investigation (CBI) in 2019 of eight senior executives from different financial institutions for Received 16 September 2020; accepted 14 December 2020. Available online 1 December 2022 * Corresponding author. E-mail address: matej.marinc@ef.uni-lj.si (M. Marinc). 1 These data are available at: World Bank, IMF Financial Soundness Indicators, and ECB. 2 University of Notre Dame's Mendoza College of Business and Labaton Sucharow LLP. Available at: https://www.corporatecomplianceinsights.com/ historic-survey-of-financial-services-professionals-reveals-widespread-disregard-for-ethics/. 3 Jane Croft, HBOS bankers bribed with sex parties guilty of corruption, Financial Times, 30 January 2017. The article is available at: https://www.ft.com/ content/4149dcd6-e70b-11e6-967b-c88452263daf. https://doi.org/10.15458/2335-4216.1312 2335-4216/© 2022 School of Economics and Business University of Ljubljana. This is an open access article under the CC-BY-NC-ND license (http://creativecommons. org/licenses/by-nc-nd/4.0/). taking bribes for a loan. 4 An older example is the bankruptcy of Hanbo Steel Industry Co. in South Koreawhereitsfounderwasarrestedforcorruption offences, including bribing bankers and high- ranking politicians to keep granting loans to the second largest steelmaker in South Korea. 5 The above examples indicate that the level of corruption can be intrinsically related to the prolif- eration of NPLs in banking. If corruption is wide- spread, corrupted bank officers might grant loans even to companies that do not fulfill loan re- quirements, with a subsequent decrease in loan portfolioquality.Inacross-countrysetting,thelevel ofcorruptioninaparticularlycountrycouldbeseen as a potential factor contributing to the surge of NPLs (Park, 2012). To address this issue, we combine the bank financial data across 140 countries and 7773 banks with a corruption index from Transparency Inter- national (2017) and several macroeconomic and institutional variables. We focus on the period 2000e2016, which comprises of the pre-financial crisis, financial crisis period, and post-financial crisis period. As a preliminary investigation, we plot the coun- tries'meanvaluesofthecorruptionindexandmean values of NPLs for the period 2000 to 2016 in Fig. 1. Fig. 1 indicates that there exists a positive relation- ship between the corruption index and the level of non-performing loans. Weinvestigatewhetherhighercorruptionleadsto the lower loan quality measured by the level of NPLs. We find robust support that higher corrup- tion is associated with higher levels of NPLs. This finding is statistically significant across several econometricspecificationsandrobustnesstests.The result is also economically significant. In particular, anincreaseincorruptionforonestandarddeviation wouldleadtoanexpectedincreaseinNPLsfor0.111 standard deviations. Ourarticleiscloselyrelatedtothecontributionby GoelandHasan(2011),whichshowsthatcorruption is positively related to the level of NPLs. Whereas Goel and Hasan (2011) use cross-country data, we employ bank-level data in order to determine the channels through which corruption affects the level of NPLs. ALB AND AIA ATG ARM ABW AUS AUT AZE BHS BHR BGD BRB BLR BEL BOL BWA BRA BGR KHM CAN CYM CHL CHN COL CRI HRV CUB CYP CZE DNK DMA DOM ECU SLV EST ETH FIN FRA GEO DEU GHA GRC GTM GUY HTI HND HKG HUN ISL IND IDN IRQ IRL ISR ITA JAM JPN JOR KAZ KEN KOR KWT KGZ LVA LBN LTU LUX MAC MKD MWI MYS MLT MUS MEX MDA MNG MAR NAM NLD NZL NIC NGA NOR OMN PAK PAN PER PHL POL PRT PRI QAT ROU RUS RWA SAU SRB SVK SVN ZAF ESP LKA SWE CHE SYR TWN TJK THA TGO TTO TUR UGA UKR ARE GBR USA URY UZB VUT VEN YEM ZMB ZWE 0 5 10 15 20 25 30 Average Non-performing Loans 0 1 2 3 4 Average Corruption Index RMSE = 4.93916 AvgNPL= = 2.86 + 1.7076*AvgCI R 2 = 11.7% Fig.1.PlotoftheaveragecorruptionindexandNPLs.Note:Weusestandardregressionequationasanestimationmethod.Thehorizontalaxisonthe graphrepresentstheaveragecorruptionindex(AvgCI)andtheverticalaxis(AvgNPL)representstheaverageNPLswheretheaverageiscomputedfor different countries. Source: World Bank and Fitch Connect. 4 Sangita Mehta, Bribe-for-loan scam: Time for more transparency, The Economic Times, updated 09 July 2019. The article is available at: https:// economictimes.indiatimes.com/industry/banking/finance/banking/bribe-for-loan-scam-time-for-more-transparency/articleshow/7062623.cms. 5 Reuters,HanboSteelFounderGiven15YearsinKoreanScandal, TheNewYorkTimes,2June1997.Thearticleisavailableat:https://www.nytimes.com/ 1997/06/02/business/hanbo-steel-founder-given-15-years-in-korean-scandal.html ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 241 When analyzing the channels through which the corruption affects the level of NPLs, we find some evidence that the effect of corruption on NPLs is bigger for smaller banks than for larger banks. This isconsistentwiththeexplanationthatsmallerbanks are less hierarchical and transfer bigger decision- making powers to bank officers. An additional discretion then leads to a stronger link between the corruption index and NPLs. We also find some evidence that the impact of corruption on NPLs is less pronounced for banks that are high profitable. A potential reason for this result is that more profitable banks have more prudentlendingpolicyandmoreeffectivelyallocate funds and, consequently, can overcome high levels of corruption. Our evidence shows that the impact of corruption on the level of NPLs is stronger for high-collectivist countries. This indicates that the impact of corrup- tion on the level of NPLs is stronger for countries wherepeople tend tohave lessinterdependentself- construal. Our findings confirm the importance of the quality of legal and institutional environment. We find that the well-functioning legal environ- ment, evident in the strong rule of law in a given country, weakens the link between corruption and NPLs. We investigate the effect of corruption on NPLs, before the crisis, during the crisis, and after the crisis. We show that the effect of corruption on the NPLs becomes more pronounced during the crisis period and in the post-crisis period. This is aligned with the view that during the financial crisis when the survival is at stake the impact of corruption on bank lending is more pronounced, causing a dete- rioration of bank asset quality. Hanousek et al. (2019) show that corruption negatively affects firm efficiency but the negative effectislesspronouncedforforeigncontrolledfirms and for firms run by female CEOs. Wellalage et al. (2018) find that corruption leads to increased credit constraints for small and medium enterprises in South Asia. They find that the effect of corruption varies across the male and female owners high- lighting gender differences in the relationship be- tween corruption and lending. Van Vu et al. (2018) find that various forms of corruption differently affect firm performance. Our findings complement their research by pointing to the mitigating role of firm size and the institutional environment com- plementing the literature that discusses how ethical behavior is implemented in organizations (Batten et al., 2017, 2018). Overall,ourfindingsprovidesomerelevantpolicy implications for the role of corruption as a potential factor that affects loan quality in the banking sys- tem.Theresultsofourpaperindicatethatmeasures taken against corruption have implications for financial stability. The paper is organized as follows. Section 1 pro- vides the theoretical framework and hypothesis development. The data and research method are described in Section 2. Section 3 presents and dis- cussestheresultsoftheempiricalanalysis.Section4 provides robustness checks. Section 5concludes the article. 1 Theoretical framework and hypothesis development The relationship between the legal environment and finance has gained prominence following the seminal article by La Porta et al. (1997) which ana- lyzes how the quality of the legal and institutional framework affects financial systems. Several articles find that corruption is negatively related to the economic growth (see Mo, 2001; Lizal & Kocenda, 2001; Mauro, 1998) and that economic growth and firmfailuresarenegativelyrelated(Ali&Daly,2010; Ghosh, 2015). Corruption can contribute to a reduction in growth, investments, and firm produc- tivity in developed and developing countries (Mauro, 1995; Meon & Weill, 2010). Qi and Ongena (2019) show that firms engaged in bribery practices lose access to bank credit which impedes firm growth(seealsoBecketal.,2013;Ghosh,2015;Klein, 2013; Mohaddes et al., 2017; Vithessonthi, 2016). Fan et al. (2009) and Treisman (2000) show that highly developed countries have lower levels of corruption comparedtodevelopingcountries.Corruptionleads to weaker and less efficient banking systems (Chen et al., 2015; Goel & Hasan, 2011; Park, 2012). In most countries that are greatly affected by corruption, even uncreditworthy firms could get a bank loan (for example, by bribing credit loans of- ficers; Fungacova et al., 2015), 6 which can subse- quently lead to excessive corporate leverage (J~ oeveer,2013;Weill,2011a)andloandelinquencies. Using cross-country data, Goel and Hasan (2011) find that countries with higher corruption are associated with a higher level of NPLs. Park (2012) finds that corruption aggravates the problems of bad loans in the banking sector. Similarly, findings 6 Chen et al. (2013) find that bribery determines the access of private firms to bank credit in China (see also Weill (2011b)). 242 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 in Bougatef (2015) confirmthepositivelinkbetween corruptionandNPLsintheIslamicbankingsystem. Chen et al. (2015) analyze the impact of corruption on bank risk-taking behavior in 35 emerging econ- omies during the period 2000e2012. They show that an increase in corruption is associated with more pronounced risk-taking behavior of banks. Whereas the studies above provide an argument forthepositiverelationshipbetweencorruptionand NPLs,itmightbepossibletoconstrueanalternative explanation. If good borrowers do not receive the credit needed from banks or financial institutions due to the market frictions, they might resort to corruptive practices. To avoid adverse selection is- sues, good firms may bribe to signal their quality to lenders, enabling good borrowers 7 to receive the credit needed and improving theefficiency ofcredit allocation. In this alternative explanation, corrup- tion would reduce the level of NPLs. This leads to the following hypothesis: H1. Corruption affects the level of NPLs. Various determinants affect the quality of bank lending, as reflected in the level of NPLs. The in- ternal determinants consist of bank size, manage- ment quality and efficiency, and bank capitalization (Dimitrios et al., 2016a, b; Louzis et al., 2012; Espi- noza & Prasad, 2010; Hasan & Wall, 2004; Salas & Saurina, 2002; Berger & DeYoung, 1997). The external determinants consist of bank regulations andsupervision(Barthetal.,2004;Godlewski,2004), lawofenforcement(Barthetal.,2004;Boudrigaetal., 2009), and national culture (Dheera-aumpon, 2019; ElGhouletal.,2016;Zhengetal.,2013).Ouraimisto analyze how the internal and external determinants affect the link between corruption and NPLs. Theliteraturerelatedtotheorganizationalstructure of banks emphasizes that large banks with more centralized and hierarchical structures can perform betterwheninformationcanbehardenedandeasily transmitted across the hierarchical lines. 8 Whereas small banks have an advantage in soft information processing where discretion of a loan officer is of paramount importance (see Stein, 2002), 9 the prob- lem is that discretion might be abused in environ- ments with high levels of corruption. Skrastins and Vig (2015) show that more hierarchical organiza- tions operated better in environments with high levels of corruption, showing that hierarchy can restrain rent-seeking activities. This leads to the following hypothesis: H2. The link between corruption and bank size is less pronounced for larger banks. Therelatedissuereferstotheroleofbankcapitalfor the relationship between corruption and NPLs. One ofthemainrolesofbankcapitalistoprovideproper incentives to banks to internalize their risk taking strategy (see Admati & Hellwig, 2013). Thinking alongtheselineswouldpositthatweaklycapitalized banks have little incentive to engage in safe lending practices and would be prone to corruption. How- ever, an alternative view is also possible. Weakly capitalized banks are exposed to intensive scrutiny from the bank supervisor. Anticipating heavy su- pervision, weakly capitalized banks could restrain from corruption. H3. The link between corruption and NPLs is influenced by bank capital. The effect of corruption on NPLs is likely to be less pronounced for countries with a healthy legal environment and efficient regulatory systems (Danisman & Demirel, 2018). These countries are less likely to be adversely affected when the rules are better and efficiently enforced. La Porta et al. (1998) and Levine (1998) show that countries with weaker legal structure to protect borrowers have a smaller number of performing banks in their econ- omy. Moreover, Barth et al. (2009) find that owner- ship of banks and firms, legal environment, and firm competition significantly reduce lending cor- ruption. We analyze how the link between corrup- tion and NPLs interacts with the quality of the legal and regulatory framework. This leads to the following hypothesis: H4. The link between corruption and NPLs is weaker in countries with a strong legal environment. Corruption alters the effectiveness of the banking system. Beck et al. (2006b) show that the most effi- cient strategy to reduce corruption in bank lending is to expose banks to private monitoring and disci- pline by the disclosure of accurate information. Accordingtothisview,bankswouldhavetobehave 7 Even though thefirmfulfillsthe required conditionsto obtain a loan, itmight occur that aloan officer might require a bribe as an incentive toprocess theclient'sloanfile,intheconditionthathisbasesalaryisreducedduetovolatilebanks'operating income.Consequently,thisbribery wouldincreasethe cost of the loan and the burden is borne solely by the borrower (see Beck et al., 2006a). 8 Larger banks have better-defined procedures and possess sophisticated capabilities for loan assessment (Hu et al., 2004; Lis et al., 2000). 9 CanalesandNanda(2012)showthatbankmanagersinadecentralizedorganizationalstructurearebetteratbanklendingtosmallfirmsandfirmswith a plethora of soft information. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 243 more cautiously and avoid bad practices in lending. In addition, countries with high-powered supervi- soryagenciestendtohavelowerlevelsofcorruption (Beck et al., 2006b; Demirguc-Kunt et al., 2004). Akins et al. (2017) find that timely loan loss recog- nition by banks reduces the level of corruption in lending. Awartani et al. (2016) show that better institution quality leads to longer maturity of corporatedebt.Wehypothesizethatcorruptionalso makes the regulatory framework less effective. H5. Corruption reduces the effectiveness of bank regu- lation in lowering the level of NPLs. The relationship between corruption and NPLs mightbedrivenbythenationalculturaldimensions (i.e. individualism/collectivism, uncertainty avoid- ance, masculinity/femininity, and power distance), as construed by Hofstede (2001) and Hofstede (1983). In the extant literature, Zheng et al. (2013) analyze the association between national culture and corruption in bank lending. They argue that compared to other cultural dimensions, the role of national culture and collectivism in particular is an importantfactoraffectingthecorruptionofthebank officials. They confirm a positive and significant association between the level of collectivism and corruption of bank officials. El Ghoul et al. (2016) document that corruption is associated with collec- tivistic culture affecting bank lending, and Haider et al. (2018) report that corruption impacts the relation between financial constraints and firm performance. Triandis (2001) highlights individu- alism/collectivism as the most significant driver of cultural differences across countries. H6. The association between corruption and NPLs is stronger in countries characterized by a high level of collectivism. 2 Data and methodology 2.1 Data description The annual bank-levelfinancial data are retrieved from the Fitch Connect database. We consider consolidated data. 10 Our final unbalanced panel data sample includes 7773 banks from 140 countries over the period 2000e2016. We include in the database commercial, savings, and co-operative banks. Our bank and macroeconomic variables are 'winsorized' at a 1% interval to mitigate the impact ofextremevaluesandtoexcludepotentialoutliers. 11 We combine bank-level data, macroeconomic in- dicators,andinstitutionaldatafromvarioussources. Macroeconomic data are obtained from World Bank. Two indices of corruption are obtained from the Transparency International Corruption Percep- tion Index and the World Bank. Table A1 in Appendix A reports the definitions and the data sources of the variables used. 2.2 Empirical model To analyze the effect of corruption on NPLs, we estimate the following regression model: NPL i;c;t ¼aþbNPL i;c;t 1 þgCI c;t 1 þdBank i;c;t 1 þwMacro c;t 1 þkInstitutional c;t 1 þmYear t þe i;c;t ð1Þ where the dependent variable is non-performing loans ratio NPL i,c,t measured by the loans and ad- vances that are more than 90 days overdue divided bytotalloansforbank i,locatedincountry c,atyear t. We use NPL i,c,t as a proxy for banks' loan quality (Chaibi & Ftiti, 2015; Ghosh, 2015; Tarchouna et al., 2017; Vithessonthi, 2016). Ourmainexplanatoryvariableiscorruptionindex CI c,t-1 , which indicates the level of corruption in country c, in year t-1. We employ two alternative corruption indexes, the corruption perception index (CPI c;t ), obtained from the Transparency Interna- tional Corruption Perception Index and the control of corruption of the World Bank. First, the score of the CPI ranges on a scale from 0 to 10. The higher the corruption in the country is, the lower the CPI score the country gets. For interpretation reasons, we define a new index Control of corruption, CI c,t , where CI c,t ¼ 10-CPI c,t . Higher values of CI indicate higherlevelsofcorruption.Forexample,thehighest value of CI is 8.43 points for Bangladesh and the lowest is 0.61 points for Denmark. Second, we employ control of corruption (hereinafter CC) from the World Bank e Worldwide Governance Indica- tor. CC is widely used in related literature (Kauf- mannetal.,2010).Theindexrangesfromthelowest valueof 2.5tothehighestvalueofþ2.5.Thelower value represents the highest level of corruption and 10 Forastandardizedresearchoncorruption,theinclusionofsubsidiariesprovidesabettermeasureofthefirm'spropensitytocorruption(Pantzalisetal., 2008; Zeume, 2017). We have also performed the analysis on the unconsolidated data, obtaining qualitatively similar results. 11 We removed negative values for non-performing loans, total assets, and loans to customer deposits. 244 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 vice versa. 12 We follow Park (2012) and Goel and Hasan (2011) and compute WBCI as WBCI c,t ¼(5- (CC c,t þ2.5)) where CC c,t is control of corruption at time t four country c. This adjustment is employed for interpretation reasons such that higher values of WBCI are associated with lower values of control of corruption and higher values of corruption. 13 The highestvalueofWBCIwas3.98pointsforLibyaand the lowest for Denmark with 0.12 points. The averagevaluesofWBCIpereachcountryareshown in Table A2 in Appendix A. We include the following bank-specific variables as control variables. Bank size i,c,t-1 is computed as the natural logarithm of total assets for bank i, located in country c, in year t-1. 14 LTCD i,c,t-1 repre- sents the ratio of total loans divided by total customer deposits for each bank i, located in coun- try c, in year t-1. 15 ROAA i,c,t-1 represents the ratio of the return on average assets for each bank i, located in country c, in year t-1. 16 The capitalization ratio, Capitalization i,c,t-1 ,istheratiooftotalequitydivided bytotalassetsforbanki,locatedincountryc,inyear t-1,asinMakri et al.(2014),Klein (2013),andLouzis et al. (2012), 17 where LLP i,c,t represents the ratio of the loan loss provisions divided by gross loans for each bank i, located in country c, in year t. 18 We also include several macroeconomic in- dicators as controls. GDP growth c,t-1 denotes the annual percentage growth rate of GDP in country c, at time t-1. 19 UNEMP c,t-1 indicates the unemploy- ment rate for a country c, at time t-1 20 , and GCF c,t-1 is computed as the growth capital formation as percentage of GDP in country c, at time t-1. RIR c,t-1 presents the real interest rate in a country c, at time t-1. 21 We also include several institutional and regula- tory variables as controls. RoL c,t-1 represents the rule of law in country c, at time t-1, and serves as a proxy for the quality of contract enforcement, propertyrights,thepolice,andthecourts,aswellas the likelihood of crime and violence (Kaufmann et al., 2010). Cap_Str c,t-1 denotes capital stringency fromBarthetal.(2013)incountryc,attimet-1.Year t is the yearly dummy variable and is included in the model to control for the time-specific effects (Petersen, 2009). e i;c;t denotes the white noise error term. Hereinafter, we suppress the subscript in the variables. To test our second, third, and sixth hypotheses about thechannelsthroughwhich corruption might affect the level of non-performing loans, we esti- mate equation (2): NPL i;c;t ¼aþbNPL i;c;t 1 þgCI c;t 1 þdBank i;c;t 1 þwMacro c;t 1 þfCI c;t 1 *BankSize i;c;t 1 þtCI c;t 1 *Capitalization i;c;t 1 þuCI c;t 1 *ROAA i;c;t 1 þrCI c;t 1 *CLT c;t 1 þmYear t þe i;c;t ð2Þ where the control variables are the same as in equation (1).CI *Bank Size denotes the interaction term between corruption and the natural logarithm oftotalassetsforbank i,locatedincountry c,inyear t-1. CI*Capitalization denotes the interaction term between corruption and the ratio of total equity divided by total assets for bank i, located in country c,inyeart-1.CI*ROAAdenotestheinteractionterm between corruption and return on average assets for each bank i, located in country c, in year t-1. CI*CLT denotes the interaction term between cor- ruption and the level of collectivism (CLT) in a country c. 12 For further detailed methodology on how the indicator is constructed, see Kaufmann et al. (2010). 13 Whereasthecorruptionindexdescribedabovepresentstheoverallcorruptioninacountryandisnotspecificallytighttothebankingsectorcorruption, it is widely accepted that the corruption in the banking sector is highly correlated with the overall corruption in the economy; see Park (2012), Chen et al. (2015) and Goel and Hasan (2011). 14 Intheextantliterature,theeffectofbanksizeonthelevelofNPLsisambiguous.Ontheonehand,RanjanandDhal(2003)andSalasandSaurina(2002) showthatlargerbankshavemorediversificationpossibilitieswhichleadstolowerNPLs.Ontheotherhand,ChaibiandFtiti(2015),Louzisetal.(2012),and Brei and Gadanecz (2012) argue that large banks may take excessive risks which can result in higher NPLs. 15 Thehigherratioofloanswithrespecttodepositsindicatesmoreaggressivelendingpracticesofabankresultingineasierloangrantingand,therefore,a higher likelihood of establishing NPLs (Dimitrios et al., 2016b). It indicates an increased risk appetite of banks with a potential for higher levels of non- performing loans. 16 Return of average assets is used as a proxy for management'sefficiency. We anticipate that the relationship between the return on average assets and non-performing loans is negative (see Dimitrios et al., 2016a; Vithessonthi, 2016). 17 WeanticipatethatahigherlevelofcapitalizationpressuresbankstoavoidriskylendingpracticesandsubsequentlynegativelyeffectsthelevelofNPLs. Makri et al. (2014), Salas and Saurina (2002), and Keeton and Morris (1987) show that the level of capital negatively affects the level of NPLs. 18 We anticipate that the relationship between loan loss provisions and NPLs is positive, since banks use higher levels of provisioning when they predict loan defaults (Boudriga et al., 2009; Chaibi & Ftiti, 2015). 19 Love and Turk Ariss (2014), Ghosh (2015), Ali and Daly (2010), and Treisman (2000) find that the GPD growth is negatively related to NPLs. 20 A higher unemployment rate affects the borrowers' incomes and, hence, affects the delinquency rates (Al-Marhubi, 2000; Dimitrios et al., 2016a; Klein, 2013; Saha & Ben Ali, 2017; Rinaldi & Sanchis-Arellano, 2006). 21 Chaibi and Ftiti (2015) and Castro (2013) find a positive influence of real interest rate on the NPLs. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 245 To test our fourth and fifth hypotheses, we esti- mate the following equation: NPL i;c;t ¼aþgCI c;t 1 þdBank i;c;t 1 þwMacro c;t 1 þ4RoL c;t 1 þuCI c;t 1 *RoL c;t 1 þtCap Str c;t 1 þrCI c;t 1 *Cap Str c;t 1 þmYear t þe i;c;t ð3Þ where CI*RoL represents the interaction term be- tween the corruption index and the rule of law, located in country c, in year t-1. Cap_Str, and the interactiontermbetweenthecorruptionindexanda measure of capital stringency, located in country c, in year t-1 and all other control variables are the same as in equation (1). We apply different panel data estimation ap- proaches. First, to account for the unobserved het- erogeneity across banks, we setup the fixed effects and random effects (hereinafter, FE and RE) panel regression models with robust standard errors (Carlson et al., 2013; Kosak et al., 2015; Micco & Panizza, 2006). 22 Second, to take into consideration the time presence of NPLs, we include NPL i,c,t-1 as the independent variable of NPL of bank i, located in country c, in year t-1, and use the system GMM estimation to achieve consistent and efficient esti- mates (Blundell & Bond, 1998). In Table 1, we present the descriptive statistics, including the mean, standard deviation, and mini- mum and maximum values. In the first part of the table, we present bank- specific variables. We observe a large disparity in the NPL rate between banks with minimum of 0 percent and a maximum of 29.3 percent, and the standard deviation of the NPL rate is 3.196 percent. We observe a high variation in the mean ratio of loans to customer deposits of 80.61 percent. Regarding the management efficiency, ROAA ranges from a minimum of 6.0 percent to a maximum of 8.4 with a mean value of 1.11 percent during the period 2000e2016, whereas the standard deviation equals 1.35 percent, indicating that on average banks in our data are profitable but some banks have financial difficulties. The mean value of capitalizationandNPLequals11.66percentand1.81 percent respectively. Loan loss provisions divided by total loans, LLP, has a mean of 0.49 percent. In the second part of Table 1, we present the re- sults for macroeconomic indicators: growth rate of GDP, unemployment ratio, gross capital formation, andrealinterestrates.Themeanannualpercentage growth rate of GDP is 2.04 percent. We note that some countries have a negative GDP growth rate with a minimum value of 5.62 percent. The mean unemployment rate is 6.35 percent, the mean gross capital formation is 22.1 percent, and the mean real interest rate is 2.91 percent. In the third panel, we present the corruption index measured by the World Bank, Transparency International, and In- ternational Country Risk Guide. In our sample, the meanofWBCIis1.11,themeanofCIis2.74,andthe mean of ICRG is 1.95. The last column of Table 1 presents the maximum of a corruption index per Table 1. Descriptive statistics. Variable Obs. Mean Std. Dev. Min Max Dependent variable NPL (%) 105,708 1.809 3.195 0.000 29.300 Bank specific variables Bank size (log) 109,178 5.422 1.905 2.229 14.319 Capitalization (%) 109,178 11.662 6.889 1.480 74.240 LTCD 108,156 80.606 46.020 5.900 864.97 ROAA (%) 108,429 1.107 1.349 6.000 8.370 LLPGL (%) 106,431 0.490 0.955 1.500 8.570 Crisis 109,178 0.182 0.386 0.000 1.000 Macroeconomic indicators GDP growth (%) 108,854 2.037 1.829 5.619 10.636 Unemployment (%) 107,958 6.346 1.953 2.500 19.900 GCF (%) 108,697 21.074 2.513 14.428 42.894 RIR (%) 106,804 2.907 2.324 9.633 29.120 Corruption perception index CI 108,653 2.736 0.862 0.000 9.600 WBCI 103,079 1.108 0.447 0.030 4.222 ICRG 108,522 1.953 0.569 0.000 5.500 Institutional controls CLT 106,462 11.202 10.819 9.000 94.000 Overall capital stringency 19,147 4.980 1.649 1.000 7.000 RoL 103,084 0.994 0.419 0.400 4.678 Economies classification 105,392 0.022 0.148 0.000 1.000 Note: The sample covers the period from 2000 to 2016. The bank variables are NPL - non-performing loans; Bank size - natural logarithm of total assets; Capitalization - total equity as a pro- portion of total assets; LTCD - total loans as a proportion of total customer deposits; ROAA - return on average assets; LLP - loan loss provision divided by gross loans. The macroeconomic in- dicators are: GFC - global financial crisis; GDP growth ratio; Unemployment ratio; GCF - gross capital formation as a per- centage of GDP; RIR - real interest rate; CI - corruption percep- tion index from Transparency International; WBCI - corruption perception index from World Bank; CI-ICRG - International Country Risk Guide; CLT - an index of collectivism by using Hofstede data; Cap_Str - capital stringency; RoL - rule of law; Economies classification - a dummy variable equal to 1 if the financial system is a bank-based financial system or 0 if the financialsystemisamarket-basedfinancialsystem.Source:Fitch Connect, World Bank - World Development Indicators, and Transparency International. 22 When usingfixed and random effects panel regression methods, we drop the contemporaneous NPL variable from the regression equation in (1). We use the bank fixed effects model with robust standard errors, clustered at the country level. 246 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 each corruption measure. Higher values of these indexes indicate that various countries are suffering from severe corruption issues. The standard devia- tion of the corruption index measured bytheWorld Bank is near to the one documented in Chen et al. (2015) with the mean value of 0.704. In the fourth part of Table 1, we include institutional variables. The mean values of the index of the overall capital stringency, rule of law, regulatory quality, and the level of collectivism are 4.98, 0.99, 1.09, and 11.2 respectively. In Table 2, we present the correlations between our variables. In line with our empirical model, we find that NPL is positively and significantly corre- lated with corruption but negatively and signifi- cantly correlated with bank size and return on average assets (at the 1 percent significance level). The correlation between NPL and capitalization is positive and significant. Furthermore, variable NPL is negatively and significantly correlated with GDP growth and gross capital formation as a percentage of GDP but positively and statistically significantly correlated with unemployment rate. 3 Empirical results 3.1 Baseline results Webeginbyestimatingtheregressionin(1)using the fixed effects regression model (see Table 3). 23 The estimated regression coefficient between the level of corruption and NPL is positive and statisti- cally significant. This supports Hypothesis 1 and is consistent with the previous findings that corrup- tion negatively affects loan quality measured by the NPL (Bougatef, 2015; Goel & Hasan, 2011; Park, 2012). The result is also economically significant. That is, an increase in corruption for one standard deviationwouldleadtoanexpectedincreaseinNPL for 0.111 standard deviations (where 0.111 is computed as the estimated regression coefficient, associated with the level of corruption CI, 0.413, multiplied by the standard deviation of CI, 0.862, and divided by the standard deviation of NPL, 3.195). The regression coefficients of control variables have the anticipated signs. The bank size is signifi- cantly and positively related to NPL. This implies that larger banks take larger risks potentially due to the associated government guarantees due to their too-big-to-fail status, which leads to higher levels of NPLs. Our empirical evidence is consistent with the finding of Chaibi and Ftiti (2015) and Louzis et al. (2012). That the return on average assets is nega- tively associated with NPL is consistent with the explanation and findings of Dimitrios et al. (2016a), Dimitrios et al. (2016b), and Baselga-Pascual et al. (2015). Additionally, capitalization is negatively and significantly related to NPL. The GDP growth is negatively but not significantly related to NPL whereastheunemploymentrateispositivelyrelated to NPL. The GCF is negative but not significantly related toNPL.Wealsofindthatrealinterestrate is positively and statistically significantly related to NPL. We use the Hausman test (Hausman, 1978)t o choose between the FE or RE estimator. From the chi-squared test statistics, we conclude that the random effects model is rejected (with Table 2. Pearson's correlation matrix. NPL CI Bank Size LTCD ROAA Capitaliz. GDP growth Unemp. GCF RIR NPL 1 CI 0.333*** 1 Bank Size 0.221*** 0.415*** 1 LTCD 0.0672*** 0.130*** 0.230*** 1 ROAA 0.240*** 0.104*** 0.142*** 0.0513*** 1 Capitaliz. 0.00699* 0.106*** 0.0909*** 0.0608*** 0.0891*** 1 GDP growth 0.0417*** 0.263*** 0.112*** 0.0264*** 0.198*** 0.0390*** 1 Unemp. 0.325*** 0.245*** 0.129*** 0.0112*** 0.116*** 0.00766* 0.294*** 1 GCF 0.150*** 0.227*** 0.146*** 0.0819*** 0.165*** 0.0399*** 0.543*** 0.524*** 1 RIR 0.0171*** 0.112*** 0.00821** 0.0669*** 0.0568*** 0.0497*** 0.0530*** 0.283*** 0.319*** 1 Note:Thesamplecoverstheperiodfrom2000to2016.Thetablereportsthecorrelationmatrixbetweenthekeyvariableswhichareused in the model. NPL - non-performing loans; CI - corruption perception index from Transparency International; Bank size - natural logarithmoftotalassets;LTCD-totalloansasaproportionoftotalcustomerdeposits;ROAA-returnonaverageassets;Capitalization- total equity as a proportion of total assets; GDP growth ratio; Unemp - unemployment ratio; GCF - gross capital formation as a per- centageofGDP;RIR-realinterestrate.***,**,and*indicatesignificanceatthe1%,5%,and10%levels,respectively.Source:Ourown calculations. 23 To mitigate the reverse causality, we use one-year lag of each of the bank and macroeconomic variables. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 247 Prob > chi2(9) ¼ 0.0000 < 0.05). Hence, the FEs model should be employed rather than the RE model. In addition, we test for the time-fixed effects in the data and confirm that year dummies should be included in the model (the Breuche Pagan LM test shows the existence of the panel effect in the data for random effects). We estimate standard er- rors that are robust to heteroskedasticity. The fixed effects model is prone to the endoge- neity problem. The alternative explanation of our findingsmightbethatahighlevelofNPLscreatesa fertile ground for corruption to spread within the financial institution. For example, corrupt loan offi- cers can exclude the defaulted borrowers from the aberrationofpenalties(Boerner &Hainz,2004).The presence of endogeneity in the model can lead to inconsistent and biased estimators. To address the endogeneity problem, we use the dynamic panel- data setup to account for potential endogeneity of our dependent variable NPL i,c,t . We include the lagged variable NPL i;c;t 1 as an independent vari- able, as in regression in (1) (see Arellano & Bond, 1991; Baum, 2006; Chaibi & Ftiti, 2015; Tarchouna et al., 2017). Furthermore, Blundell and Bond (1998) suggesttorelyonthesystemgeneralizedmethodof moments (GMM) estimator for dynamic panel data. This method provides unbiased and efficient esti- mates. As endogenous instrument, we consider the lagged dependent variable NPL i;c;t 1 . All other var- iables are treated as exogenous instruments. Table 3. The relationship between corruption index and NPLs: the total sample with different estimation methods. Dependent variable: (1) (2) (3) (4) (5) (6) NPL Intercept 0.620 1.535 0.392 1.310 (-0.38) (-1.63) (-0.26) (-1.41) Corruption level CI 0.413*** 0.634*** 0.992** (3.21) (6.76) (2.52) WBCI 0.369*** 0.771*** 4.101*** (3.78) (2.88) (2.71) Bank-specific variables Lagged NPL 0.788*** 0.746*** (27.10) (15.03) Bank Size 0.186*** 0.197*** 0.105*** 0.185*** 0.213*** 0.0507 (6.28) (13.17) (2.59) (5.64) (11.13) (0.77) LTCD 0.00248 0.00270* 0.00339** 0.00259* 0.00293** 0.00240 (1.62) (1.89) (2.19) (1.76) (2.12) (1.23) ROAA 0.429*** 0.430*** 0.251*** 0.435*** 0.436*** 0.505** (-37.77) (-36.24) (-4.45) (-35.63) (-33.54) (-2.08) Capitalization 0.0352*** 0.0317*** 0.0381** 0.0351*** 0.0306*** 0.0274 (-21.60) (-9.90) (-2.03) (-22.47) (-9.76) (0.43) Macroeconomic indicators GDP growth 0.000390 0.000753 0.616 0.00714 0.0167* 1.121** (0.03) (0.06) (-1.32) (0.50) (1.74) (-2.19) Unemployment 0.219*** 0.223*** 1.101** 0.242*** 0.253*** 0.740** (5.26) (6.71) (2.40) (5.09) (7.11) (2.05) GCF 0.0535 0.0412 0.000751 0.0354 0.0193 0.282 (-1.07) (-1.17) (0.00) (-0.69) (-0.50) (0.87) RIR 0.0979*** 0.101*** 0.902*** 0.0852*** 0.0784*** 0.634*** (3.57) (4.28) (-3.85) (3.29) (3.52) (-3.22) Coefficient Estimates FE RE GMM FE RE GMM No. Obs. 102,905 102,905 102,420 97,314 97,314 96,850 Dummies Year Yes Yes Yes Yes Yes Yes R-squared within 0.2004 0.1998 0.1983 0.1973 Hansen J statistic (p-value) 21.08 (0.576) 22.13 (0.452) AB test AR(2) (p-value) 0.10 (0.923) 0.67 (0.501) Note:The sample coversthe periodfrom 2000 to 2016.The dependent variableis non-performingloans (NPL). The estimation methods are FE, RE and the twostep Arellano-Bond system GMM. CI - corruption perception index; WBCI - corruption perception index from WorldBank;Bank size- naturallogarithmof bankassets; LTCD- total loansas aproportionof totalcustomerdeposits;ROAA- return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomic indicators are: GDP growth ratio; Unemployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. The regressions include Year dummies. AR(2) reports the p-values for the null hypothesis that the errors in the first regression exhibit no second-order serial cor- relation. The independent variables are lagged one period. Robust-standard errors in parenthesis are clustered at the level of country. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. 248 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 Moreover, we use the set of lags from 2 to 5 to mitigate the over-identification problem of endog- enous instruments. Furthermore, to account for appropriate in- struments, the relevance condition states that the excluded instruments have to fulfill two conditions. They have to be correlated with the dependent variable (in our case NPL) and they have to be un- correlated with theresiduals(Baum et al.,2003). We also test for the over-identification restrictions by using Hansen's J statistic, as well as the presence of the first order autocorrelation AR(1) and the second order autocorrelation AR(2) of the residuals in the dynamic model. We confirm the validity of the in- struments chosen and no presence of the second order autocorrelation. This indicates that our GMM estimate is consistent. The regression coefficient of the first lag of NPLs is positive and significant (see column 3 of Table 3). The positive relationship in- dicates that NPLs are expected to increase when theymoveduptheyearbefore,asinChaibiandFtiti (2015). The results reported in column 4 and 6 of Table 3 indicate that the level of corruption is significantly and positively related to the NPLs even when we use an alternative corruption perception index from the World Bank. This confirms Hypothesis 1. 3.2 Channels through which corruption affects NPLs HavingidentifiedthemaindeterminantsofNPLs, we analyze the interaction terms between the cor- ruptionindexandbank-specificvariablesinorderto investigate the channels through which corruption impacts the level of NPLs. We introduce four interaction terms: corruption and bank size, cor- ruption and bank capitalization, corruption and ROAA, and corruption and collectivism. The results are presented in Table 4. As before, the level of corruption is positively and statistically significantly related to the level of NPLs. 24 The interaction term between the corruption index and bank size 25 is negatively and significantly (for GMM estimation but not for FE estimation) related to the level of NPLs. Wefind that a one-unit increase in CI*Bank size leads to reduction of NPLs by 0.0191 units. This indicates that the effect of corruption on the level of NPLs is less pronounced for larger banks than for smaller banks, providing some support for Hypothesis 2. The explanation may derive from more hierarchical structures of larger banks that preclude corruptive bank prac- tices, resulting in a weaker link between corruption and NPLs. The interaction term between corruption and capitalization has no significant effect on the level of NPLs. This indicates that the third hypoth- esis is not confirmed. The interaction term between corruption and ROAA is negatively and significantly (for GMM estimation but not for FE estimation) related to the level of NPLs. We show that a one-unit increase in CI*ROAA leads to a decrease of NPLs by 0.276 units. This implies that the effect of corruption on the level of NPLs is less pronounced for banks with high profitability. A potential explanation might be that as most profitable banks are less concerned with their revenue creation, they do not engage themselves toward risky lending, causing a cutback in the level of NPLs. We also analyze the association between corrup- tion and NPLs by using a specific dimension of cul- ture,namelycollectivismtocontrolforheterogeneity in national cultures across countries. 26 The interac- tion term between corruption and the level of collectivism (CLT) is positively and significantly (for FEestimationbutnotforGMMestimation)relatedto thelevelofNPLs.Wefindthataone-unitincreasein CI*CLTleads toanincreaseofNPLs by0.0433units. This indicates that the impact of corruption on the level of NPLs is stringer for countries where people tend to have less interdependent self-construal. This provides some support for our Hypothesis 6. 3.3 Corruption and NPLs during and after the global financial crisis We now evaluate how the global financial crisis affects the relationship between the corruption indexandNPLs.Weaddtheinteractiontermsofthe corruptionindexwithdummyvariablesthatequal1 during the three sub-periods: before the crisis (2000e2007), during the crisis (2008e2010), and the post-crisis (2011e2016), with 0 otherwise (see Allen et al., 2017). The estimation results in Table 5 show that the interaction term of corruption with before the crisis, CI*BEFORE GFC, is negatively and significantly associated with NPLs. This suggests 24 The p-values of Hansen's J statistics in Table 4 imply that the instruments satisfy the orthogonality conditions for all GMM regressions. 25 Skrastins and Vig (2015) approach to test the influence of bank size on the relation between corruption and NPLs. In their paper, the organizational hierarchy is measured by the managerial levels. We use the logarithm of total bank assets to test the impact. 26 Hofstede (2001) andHofstede (1983) constructed four cultural dimensions: individualism/collectivism (IDV), uncertainty avoidance (UAI), masculinity/ femininity (MAS), and power distance (PDI). ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 249 that before the global financial crisis, the effect of corruption on the NPLs is less pronounced. The interaction term of corruption index with the crisis dummy, CI*DURING GFC, is positively and significantly associatedwithNPLs.This impliesthat during the global financial crisis the effect of cor- ruption on the NPLs is more pronounced. The interaction term of the corruption index with the post-crisis dummy, CI*AFTER GFC, is positively andsignificantlyrelatedtotheNPLs,indicatingthat in the post-crisis period the positive effect of cor- ruption on NPLs is more pronounced. 3.4 Corruption and NPLs in bank-based and market-based financial systems We continue our investigation by exploring whether the impact of corruption on the NPLs is different in bank-based and market-based econo- mies (see Demirguc-Kunt & Levine, 1999). Table 6 showsthatcorruptionispositivelyrelatedtoNPLin thesubsample ofmarket-based economies, while in the subsample of bank-based economies the asso- ciation is statistically insignificant. Interestingly, the regression coefficient of the corruption index is Table 4. Channels through which corruption affects NPL. Dependent variable NPL (1) (2) (3) (4) (5) (6) (7) (8) Intercept 0.920 0.711 0.747 2.714*** (-0.63) (-0.41) (-0.43) (-4.81) Corruption level CI 0.517** 7.707*** 0.444** 1.098*** 0.441*** 1.964*** 0.295*** 1.270 (2.41) (3.16) (2.18) (2.89) (2.89) (3.79) (4.62) (0.79) Bank-specific variables Lagged NPL 0.752*** 0.792*** 0.774*** 0.752*** (25.32) (30.27) (26.69) (35.18) Bank Size 0.229** 2.008*** 0.185*** 0.0820** 0.183*** 0.146** 0.178*** 0.101** (2.30) (2.95) (6.37) (2.13) (6.34) (2.15) (7.40) (2.38) LTCD 0.00246 0.00353** 0.00250* 0.00422** 0.00247 0.00261 0.00169** 0.00235** (1.62) (2.31) (1.67) (2.45) (1.60) (1.54) (2.52) (2.39) ROAA 0.429*** 0.295*** 0.428*** 0.247*** 0.362*** 0.540* 0.424*** 0.249*** (-39.10) (-4.96) (-39.76) (-4.83) (-3.87) (1.76) (-22.34) (-5.98) Capitalization 0.0352*** 0.0215 0.0284 0.0570 0.0348*** 0.0640** 0.0343*** 0.0404** (-21.45) (-1.11) (-1.26) (0.36) (-25.40) (-2.12) (-9.35) (-2.29) Macroeconomic indicators GDP growth 0.000325 1.234** 0.000249 0.525 0.000294 0.564 0.00174 0.267 (-0.02) (-2.16) (0.02) (-1.09) (-0.02) (-0.74) (0.24) (-0.40) Unemployment 0.219*** 1.088** 0.220*** 1.076** 0.220*** 1.131** 0.236*** 0.810 (5.11) (2.39) (5.24) (2.38) (5.22) (2.02) (13.42) (1.27) GCF 0.0522 0.457 0.0532 0.171 0.0510 0.492 0.00684 0.0615 (-1.07) (1.24) (-1.06) (0.53) (-0.99) (-0.74) (-0.30) (0.15) RIR 0.0976*** 1.023*** 0.0979*** 0.712** 0.0968*** 1.253*** 0.109*** 0.610*** (3.52) (-3.61) (3.56) (-1.99) (3.60) (-2.76) (14.10) (-2.64) Interaction terms CI*Bank Size 0.0156 0.698*** (-0.38) (-2.77) CI*Capitalization 0.00255 0.0313 (-0.29) (-0.61) CI*ROAA 0.0231 0.276** (-0.63) (-2.41) CI*CLT 0.0433*** 0.00924 (6.06) (-0.50) Coefficient Estimates FE GMM FE GMM FE GMM FE GMM No. Obs. 102,905 102,420 102,905 102,420 102,905 102,420 102,013 101,610 Dummies Year Yes Yes Yes Yes Yes Yes Yes Yes Hansen J statistic (p-value) 16.38 (0.839) 30.17 (0.145) 21.10 (0.575) 48.51 (0.064) AB test AR(2) (p-value) 0.09 (0.928) 0.29 (0.768) 0.80 (0.425) 0.14 (0.887) Note:The sample coversthe periodfrom 2000 to 2016.The dependent variableis non-performingloans (NPL). The estimation methods are pooled FE and the twostep Arellano-Bond system GMM. CI - corruption perception index from Transparency International; Bank size - naturallogarithmof bank assets; LTCD- total loansas a proportion of total customer deposits;ROAA - return on average assets. Capitalization - total equity as a proportion of total assets. CLT - the level of collectivism. The macroeconomic indicators are: GDP growth ratio; Unemployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. The regressions include Year dummies.AR(2) reports the p-valuesfor the null hypothesisthat the errors in thefirst regressionexhibit no second-order serial correlation. The independent variables are lagged one period. Robust-standard errors in parenthesis are clustered at the level of country. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. 250 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 largerinthemarket-basedeconomies(0.434)thanin thebank-basedeconomies,indicatingthattheeffect ofthecorruptiononthelevelofNPLsispronounced in the market-based economies. 3.5 The legal environment and the effectiveness of the regulatory framework The relationship between the level of corruption and NPLs may be affected by cross-country differ- ences,especiallyinthelegalenvironment,andmight impacttheefficiencyoftheregulatoryframework.In Table 7, we add the interaction term between the corruptionindexandtheruleoflaw,CI*RoL. The interaction term between the corruption index and the rule of law is negatively and signifi- cantly related to the NPLs (see Table 7). This con- firms the view that in countries where the laws are better enforced, the impact of corruption on the NPLs becomes less pronounced. This finding is consistent with Hypothesis 4. We also evaluate the impact of corruption on the effectiveness of capital regulation by including the level of capital stringency, Cap_Str, and the inter- action term between the corruption index and a measure of capital stringency, CI*Cap_Str, in the regression model. Capital stringency has no Table 5. Corruption and the NPLs before, during, and after the global financial crisis. Dependent variable: (1) (2) (3) NPL Intercept 1.211 0.332 0.776 (0.89) (-0.23) (0.52) Corruption level CI 0.292*** 0.230** 0.299*** (3.19) (2.28) (2.98) Bank-specific variables Bank Size 0.123*** 0.173*** 0.145*** (3.46) (6.74) (3.85) LTCD 0.00248* 0.00232 0.00282** (1.73) (1.48) (2.00) ROAA 0.432*** 0.427*** 0.436*** (-31.88) (-36.24) (-31.37) Capitalization 0.0385*** 0.0365*** 0.0360*** (-21.22) (-21.04) (-19.35) Macroeconomic indicators GDP growth 0.0116 0.0656 0.130*** (0.89) (1.49) (-3.57) Unemployment 0.183*** 0.260*** 0.0903*** (5.50) (4.52) (4.47) GCF 0.0606 0.0609 0.0453 (-1.09) (-1.28) (-0.75) RIR 0.0672*** 0.0936*** 0.0734*** (5.77) (4.17) (5.00) CI*BEFORE GFC 0.344*** (-9.80) CI*DURING GFC 0.174** (2.53) CI*AFTER GFC 0.380*** (5.24) Coefficient Estimates FE FE FE No. Obs. 102,905 102,905 102,905 Dummies Year Yes Yes Yes Note:Thesamplecoverstheperiodfrom2000to2016.Thedepen- dent variable is non-performing loans (NPL). The estimation methodispooledFE.CI-corruptionperceptionindex;Banksize- naturallogarithmofbankassets;LTCD-totalloansasaproportion of total customer deposits; ROAA - return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomicindicatorsare:GDPgrowthratio;Unemployment ratio;GCF-grosscapitalformationasapercentageofGDP;RIR- real interest rate. The regressions include Year dummies. The in- dependentvariablesarelaggedoneperiod.Robust-standarderrors in parenthesis are clustered at the level of country. ***, **, and * indicatesignificanceatthe1%,5%,and10%levels,respectively. Table 6. Corruption and NPLs in bank-based and market-based economies. Dependent variable: (1) (2) NPL Intercept 12.16** 2.339* (2.09) (-1.83) Corruption level CI 0.652 0.434*** (-1.29) (5.00) Bank-specific variables Bank Size 0.120 0.203*** (0.60) (11.08) LTCD 0.00406 0.000917 (1.21) (1.64) ROAA 0.717** 0.415*** (-2.51) (-26.85) Capitalization 0.0708 0.0335*** (1.03) (-14.55) Macroeconomic indicators GDP growth 0.0979 0.00827 (-1.52) (0.71) Unemployment 0.200 0.273*** (-0.95) (12.00) GCF 0.140 0.000791 (-1.35) (0.02) RIR 0.0289 0.126*** (0.49) (11.92) Coefficient Estimates FE FE No. Obs. 788 100,901 Dummies Year Yes Yes Financial Systems Bank-based economy Market-based economy Note: The sample covers the period from 2000 to 2016. The dependent variable is non-performing loans (NPL). The estima- tion method is pooledFE.CI - corruption perceptionindex;Bank size - natural logarithm of bank assets; LTCD - total loans as a proportion of total customer deposits; ROAA - return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomic indicators are: GDP growth ratio; Unem- ployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. The regressions include Year dummies. The independent variables are lagged one period. Robust-standarderrorsinparenthesisareclusteredatthelevelof countries. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 251 significanteffectonthelevelofNPLs.Thisindicates that the fifth hypothesis is not confirmed. 3.6 High-corrupt vs. low-corrupt group of countries We now want to confirm that the relationship between the corruption index and NPL is positive for countries with high and low levels of corruption and countries with high and low levels of NPL. We use the average value as a reference unit to divide countries in the high-corrupt group (with the cor- ruption index above the mean value of 2.736 and NPL above the mean value of 1.809) and the low- corrupt group (with the corruption index below the mean value of 2.736 and NPL below the mean value of 1.809). Table 8 shows that corruption is positively related to NPLs for countries in the high-corrupt group and for countries in the low-corrupt group. Furthermore, we find a significant and positive ef- fect of corruption on the level of NPLs on both high and low NPL countries. This implies that country- levelcorruptionaffectspositivelythelevelofNPLin the bank. 4 Robustness checks 4.1 Alternative corruption index and loan quality indicator As a robustness check, we also use the value of corruption index based on the International Coun- try Risk Guide ICRG c,t . The score for the ICRG c,t ranges on a scale from 0 (least corrupt) to 6 (most corrupt). Table 9 provides the regression results using the alternative corruption indexes, corruption index from TI, CI, and ICRG. The estimated results remainwidelyunchangedandcorroborateourmain findingthatcorruptionispositivelyrelatedtoNPLs. Table 9 also provides the results of the alternative model in which we use loan loss provisions, LLP, as a loan quality indicator instead of NPL (following De Haan & Van Oordt, 2018; Tarchouna et al., 2017; Chaibi & Ftiti, 2015; Boudriga et al., 2009). The results in Table 9 confirm a positive and sig- nificant relationshipbetweenthelevel ofcorruption index and LLP. This implies that corruption aggra- vates loan quality which results in increased levels of LLPs. The effects of other variables on loan loss provisions remain similar to the basic model with NPLs as a dependent variable. 4.2 Subsamples of commercial banks We tested the robustness of our results based on the subsample of commercial banks (see Table 10). The results remain similar to the ones in the basic model. The level of corruption is positively and significantly related to the level of non-performing loan in both estimation methods (FE and GMM) to bank lending. This confirms the view that the cor- ruption is widespread across different bank specializations. Table 7. Corruption and NPLs and the quality of the legal and regu- latory environment. Dependent variable: (1) (2) NPL Intercept 0.111 6.124*** (0.16) (2.91) Corruption level CI 1.261*** 0.572 (13.96) (-0.74) Bank-specific variables Bank Size 0.153*** 0.241*** (5.85) (-4.55) LTCD 0.00245*** 0.00113 (2.98) (-1.47) ROAA 0.441*** 0.500*** (-21.81) (-34.93) Capitalization 0.0350*** 0.0647*** (-8.96) (-23.73) Macroeconomic indicators GDP growth 0.0292*** 0.257 (-2.64) (-1.29) Unemployment 0.202*** 0.144 (10.93) (0.94) GCF 0.0430* 0.153* (-1.67) (-1.72) RIR 0.00713 0.134 (0.42) (0.73) Legal system quality RoL 1.829*** (-7.40) CI*RoL 0.288*** (-6.81) Regulatory effectiveness Cap_Str 0.370 (1.20) CI*Cap_Str 0.0147 (0.17) Coefficient Estimates FE FE No. Obs. 91,657 18,187 Dummies Year Yes Yes Note: The sample covers the period from 2000 to 2016. The dependent variable is non-performing loans (NPL). The estima- tion method is pooledFE.CI - corruption perceptionindex;Bank size - natural logarithm of bank assets; LTCD - total loans as a proportion of total customer deposits; ROAA - return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomic indicators are: GDP growth ratio; Unem- ployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. The regressions include Year dummies. The independent variables are lagged one period. Robust-standarderrorsinparenthesisareclusteredatthelevelof country. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. 252 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 Table 8. Corruption and NPLs in low and high corrupt group of countries. Dependent variable: (1) (2) (3) (4) NPL Intercept 4.995*** 0.296 3.576* 0.368 (3.03) (0.93) (1.81) (-1.11) Corruption index CI 0.403** 0.598*** 0.554** 0.0928*** (2.23) (5.50) (2.23) (5.39) Bank-specific variables Bank Size 0.0522 0.214*** 0.0790 0.0571*** (0.44) (18.34) (0.90) (10.54) LTCD 0.00679*** 0.000632*** 0.00604*** 0.000243*** (3.62) (3.33) (3.37) (3.39) ROAA 0.614*** 0.342*** 0.527*** 0.00862*** (-14.11) (-63.99) (-13.74) (11.03) Capitalization 0.0763*** 0.0235*** 0.0355*** 0.00139*** (-6.33) (-20.99) (-3.37) (-6.01) Macroeconomic indicators GDP growth 0.126** 0.103*** 0.0425 0.00182 (-2.36) (-12.17) (-1.40) (-0.55) Unemployment 0.000532 0.274*** 0.168*** 0.0394*** (-0.01) (95.08) (3.37) (5.55) GCF 0.101** ( 2.31) 0.129*** ( 33.78) 0.113* ( 1.89) 0.00165 ( 0.17) RIR 0.0324 0.136*** 0.0699** 0.0255*** (1.36) (11.36) (2.42) (5.50) Coefficient Estimates FE FE FE FE No. Obs. 20,452 82,453 27,908 74,997 Group Low corrupt group High corrupt group Low NPL group High NPL group Note:Thesamplecoverstheperiodfrom2000to2016.Thedependentvariableisnon-performingloans(NPL).Theestimationmethodis pooled FE. CI - corruption perception index; Bank size - natural logarithm of bank assets; LTCD - total loans as a proportion of total customer deposits; ROAA - return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomic indicators are: GDP growth ratio; Unemployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. The regressions include Year dummies. The independent variables are lagged one period. Robust-standard errors in parenthesis are clustered at the level of countries. ***, **, * indicate significance at the 1%, 5%, and 10% level, respectively. Table 9. Alternative measures of corruption and loan quality. Dependent variable: (1) (2) (3) (4) NPL NPL LLP LLP Intercept 0.186 1.126*** (-0.13) (-2.83) Corruption level CI 0.104*** 0.637** (5.34) (2.27) ICRG 0.206*** 2.415*** (5.97) (3.46) Bank-specific variables Lagged NPL 0.798*** (24.94) Lagged LLP 0.182* (1.84) Bank Size 0.212*** 0.174*** 0.0600*** 0.0659*** (6.84) (4.08) (4.05) (3.83) LTCD 0.00258* 0.00209* 0.00132*** 0.00140*** (1.90) (1.81) (3.62) (3.16) ROAA 0.439*** 0.0171 0.113*** 0.0339 (-37.13) (0.37) (-9.77) (-0.99) Capitalization 0.0341*** 0.0238*** 0.0103*** 0.00236 (-22.49) (3.44) (17.53) (0.45) Macroeconomic indicators GDP growth 0.0208 0.955** 0.0941*** 0.640** (1.27) (-2.45) (-17.34) (-2.01) Unemployment 0.229*** 0.300 0.0324*** 0.00156 (5.44) (1.21) (2.87) (0.01) (continued on next page) ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 253 5 Conclusion To unveil the relationship between corruption and loan quality, we combine an unbalanced panel data of 109,178 bank level observations from 140 countries over the period from 2000 to 2016 with macroeconomic and regulatory indicators and the indexes of corruption. Our main finding is that corruption is positively and significantly related to the level of NPLs, indicating that cor- ruption leads to reduced loan quality in a banking system. We find that the relationship between corruption and NPLs is weaker for larger banks. Potential ex- planations reside in hierarchical practices of large banks and in the regulatory scrutiny. Larger banks may be less exposed to corruption due to their highly hierarchical structures. Hierarchical struc- tures give little discretion to loan officers, success- fully preventing corruptive behavior. Furthermore,wefindevidencethatcountrieswitha highlevelofcollectivismareassociatedwithahigher level of NPLs. Given that collectivist cultures are characterized by group and social cooperation, our findingssupportthehypothesisthatcountrieswitha high level of collectivism are associated with higher levelsofNPLs.Wealsoanalyzetheroleoflegalenvi- ronment on the impact of corruption on loan quality. We find that stronger rule of law makes the relationship between corruption and NPLs less pronounced. Ethical approval This article does not contain any studies with human participants or animals performed by any of the authors. Table 9. (continued) Dependent variable: (1) (2) (3) (4) NPL NPL LLP LLP GCF 0.0508 0.0122 0.0349*** 0.0280 (-1.02) (0.06) (3.67) (-0.17) RIR 0.0813*** 0.798*** 0.0498*** 0.145 (3.69) (-4.11) (4.13) (-1.59) Coefficient Estimates FE GMM FE GMM No. Obs. 102,810 102,328 103,012 96,234 HansenJstatistic(p-value) 27.94 (0.218) 39.37 (0.144) AB test AR(2) (p-value) 0.20 (0.840) 0.35 (0.724) Note:Thesamplecoverstheperiodfrom2000to2016.Thedependentvariablesarenon-performingloans(NPL)andloanlossprovision (LLP). The estimation methods are FE and twostep Arellano-Bond system GMM. CI - corruption perception index; ICRG - corruption indexfromtheInternationalCountryRiskGuide;Banksize-naturallogarithmofbankassets;LTCD-totalloansasaproportionoftotal customer deposits; ROAA - return on average assets; Capitalization - total equity as a proportion of total assets. The macroeconomic indicators are: GDP growth ratio; Unemployment ratio; GCF - gross capital formation as a percentage of GDP; RIR - real interest rate. TheregressionsincludeYeardummies.AR(2)reportsthep-valuesforthenullhypothesisthattheerrorsinthefirstregressionexhibitno second-orderserialcorrelation.Theindependentvariablesarelaggedoneperiod.Robust-standarderrorsinparenthesisareclusteredat the level of countries. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively. Table 10. Corruption and NPLs for the subsample of commercial banks. Dependent variable: (1) (2) NPL Intercept 0.683 0.408 (-0.42) (-0.27) Corruption level CI 0.468*** (2.99) WBCI 0.390*** (3.58) Bank-specific variables Bank Size 0.199*** 0.198*** (6.16) (5.50) LTCD 0.00375** 0.00384** (2.16) (2.30) ROAA 0.432*** 0.439*** (-34.02) (-31.92) Capitalization 0.0383*** 0.0384*** (-19.53) (-20.83) Macroeconomic indicators GDP growth 0.00552 0.00244 (-0.34) (0.15) Unemployment 0.207*** 0.233*** (5.05) (4.82) GCF 0.0582 0.0371 (-1.21) (-0.74) RIR 0.0993*** 0.0858*** (3.33) (3.05) Coefficient Estimates FE FE No. Obs. 88,386 83,620 Dummies Year Yes Yes R-squared within 0.1938 0.1919 Note: Thesamplecoverstheperiodfrom2000to2016.Thedepen- dent variable is non-performing loans (NPL). The estimation methodisFE.CI-corruptionperceptionindexfromTransparency International; WBCI - corruption perception index from World Bank-WorldDevelopmentIndicators;Banksize-naturallogarithm ofbankassets;LTCD-totalloansasaproportionoftotalcustomer deposits; ROAA - return on average assets. Capitalization - total equityasaproportionoftotalassets.Themacroeconomicindicators are: GDP growth ratio; Unemployment ratio; GCF - gross capital formationasapercentageofGDP;RIR-realinterestrate. There- gressions include Year dummies. Robust-standard errors in parenthesis are clustered at the level of country. ***, **,and* indicatesignificanceatthe1%,5%,and10%levels,respectively. 254 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 Conflict of interest Ardit Gjeçi declares that he has no conflict of in- terest. Matej Marinc declares that he has no conflict of interest. Acknowledgement The authors would like to thank Iftekhar Hasan, MarkoKosak,IgorMasten,VasjaRant,OrfeaDhuci, Adalbert Winkler as our discussant, and the par- ticipants at INFINITI 2018 in Poznan, Poland, and the 7th EBR Conference in Ljubljana for their valu- able comments and suggestions. Ardit Gjeçi grate- fully acknowledges the financial support from the European Commission under the Erasmus Mundus Project Green-Tech-WB: Smart and Green technol- ogies for innovative and sustainable societies in Western Balkans (551984-EM-1-2014-1-ES-ERA MUNDUS-EMA2). All views expressed in this publication are entirely those of the authors and should not be attributed to the Erasmus Mundus Project Green-Tech-WB. This work was supported by the Slovenian Research Agency under the research grant programme number P5-0161, enti- tled ‘The challenges of investors, firms, financial institutions and government in an uncertain Euro- pean economic environment’. References Admati, A., & Hellwig, M. (2013). The bankers' new clothes: What's Wrong with Banking and What to Do about it. Princeton Uni- versity Press. https://www.gsb.stanford.edu/faculty-research/ books/bankers-new-clothes-whats-wrong-banking-what-do- about-it Akins, B., Dou, Y., & Ng, J. (2017). Corruption in bank lending: The role of timely loan loss recognition. Journal of Accounting and Economics, 63(2e3), 454e478. Al-Marhubi, F. A. (2000). Corruption and inflation. Economics Letters, 66(2), 199e202. Ali, A., & Daly, K. (2010). Macroeconomic determinants of credit risk: Recent evidence from a cross country study. International Review of Financial Analysis, 19(3), 165e171. Allen, F., Jackowicz, K., Kowalewski, O., & Kozłowski, Ł. (2017). Bank lending, crises, and changing ownership structure in Central and Eastern European countries. Journal of Corporate Finance, 42, 494e515. Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies, 58(2), 277. Awartani, B., Belkhir, M., Boubaker, S., & Maghyereh, A. (2016). Corporate debt maturity in the MENA region: Does institu- tionalqualitymatter? International Review of Financial Analysis, 46, 309e325. Barth, J. R., Caprio, G., & Levine, R. (2004). Bank regulation and supervision: What works best? Journal of Financial Intermedia- tion, 13(2), 205e248. Barth, J. R., Caprio, G., & Levine, R. (2013). Bank regulation and supervision in 180 countries from 1999 to 2011. Journal of Financial Economic Policy, 5(2), 111e219. Barth, J. R., Lin, C., Lin, P., & Song, F. M. (2009). Corruption in bank lending to firms: Cross-country micro evidence on the beneficialroleofcompetitionandinformationsharing. Journal of Financial Economics, 91(3), 361e388. Baselga-Pascual, L., Trujillo-Ponce, A., & Cardone-Riportella, C. (2015).FactorsinfluencingbankriskinEurope:Evidencefrom thefinancialcrisis. The North American Journal of Economics and Finance, 34, 138e166. Batten, J. A., Loncarski, I., & Szilagyi, P. G. (2017). Financial market manipulation, whistleblowing and the common good: Evidence from the LIBOR scandal. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2964917 Batten,J.A.,Loncarski,I.,&Szilagyi,P.(2018).WhenKamaymet Hill:Organisationalethicsinpractice.JournalofBusinessEthics, 147(4), 779e792. Baum, C. F. (2006). An introduction to modern econometrics using stata. Stata Corp. Baum, C. F., Schaffer, M. E., & Stillman, S. (2003). Instrumental variables and GMM: Estimation and testing. STATA Journal, 3(1), 1e31. Beck, T., Demirguc-Kunt, A., Laeven, L., & Maksimovic, V. (2006a). The determinants of financing obstacles. Journal of International Money and Finance, 25(6), 932e952. Beck, T., Demirguc-Kunt, A., & Levine, R. (2006b). Bank super- vision and corruption in lending. Journal of Monetary Eco- nomics, 53, 2131e2163. Beck, R., Jakubik, P., & Piloiu, A. (2013). Non-performing loans: What matters in addition to the economic cycle? ECB Working Paper Series, 1515, 34. Berger, A. N., & DeYoung, R. (1997). Problem loans and cost ef- ficiency in commercial banks. Journal of Banking & Finance, 21(6), 849e870. Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econo- metrics, 87(1), 115e143. Boerner,K.,&Hainz,C.(2004).Thepoliticaleconomyofcorruption andtheroleoffinancialinstitutions.CESifoWorkingPaperNo. 1293https://www.cesifo.org/DocDL/cesifo1_wp1293.pdf Boudriga, A., Boulila, N., & Jellouli, S. (2009). Does bank super- vision impact nonperforming loans: Cross-country de- terminantsusingagregatedata. Munich Personal RePEc Archive (18068), 1e28. Bougatef,K.(2015).Theimpactofcorruptiononthesoundnessof Islamic banks. Borsa Istanbul Review, 15(4), 283e295. Brei,M.,& Gadanecz, B. (2012).Publicrecapitalisations andbank risk: Evidence from loan spreads and leverage. BIS Working paper. https://www.bis.org/publ/work383.pdf Canales, R., & Nanda, R. (2012). A darker side to decentralized banks: Market power and credit rationing in SME lending. Journal of Financial Economics, 105(2), 353e366. Carlson, M., Shan, H., & Warusawitharana, M. (2013). Capital ratios and bank lending: A matched bank approach. Journal of Financial Intermediation, 22(4), 663e687. Castro, V. (2013). Macroeconomic determinants of the credit risk in the banking system: The case of the GIPSI. Economic Modelling, 31(1), 672e683. Chaibi, H., & Ftiti, Z. (2015). Credit risk determinants: Evidence from a cross-country study. Research in International Business and Finance, 33,1e16. Chen, M., Jeon, B. N., Wang, R., & Wu, J. (2015). Corruption and bank risk-taking: Evidence from emerging economies. Emerging Markets Review, 24, 122e148. Chen, Y., Liu, M., & Su, J. (2013). Greasing the wheels of bank lending: Evidence from private firms in China. Journal of Banking& Finance, 37(7), 2533e2545. Danisman, G. O., & Demirel, P. (2018). Bank Risk-taking in Developed Countries: The influence of market power and bank regulations. Journal of International Financial Markets, In- stitutions and Money, 59, 202e217. https://doi.org/10.1016/j. intfin.2018.12.007 De Haan, L., & Van Oordt, M. R. C. (2018). Timing of banks' loan loss provisioning during the crisis. Journal of Banking & Finance, 87, 293e303. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 255 Demirguc-Kunt, A., Laeven, L., & Levine, R. (2004). Regulations, market structure, institutions, and the cost of financial inter- mediation.Journal ofMoney,Credit, andBanking,36(3),593e622. Demirguc-Kunt, A., & Levine, R. (1999). Bank-based and market- based finanical systems: Cross-country comparisons. Policy research Working Paper series 2143. The World Bank. Dheera-aumpon, S. (2019). Collectivism and connected lending. Research in International Business and Finance, 48(May 2018), 258e270. Dimitrios,A.,Helen,L.,&Mike,T.(2016a).Determinantsofnon- performing loans: Evidence from euro-area countries. Finance Research Letters, 18, 116e119. https://doi.org/10.1016/j.frl.2016. 04.008 Dimitrios, A., Helen, L., & Mike, T. (2016b). Non-performing loans in the euro area: Are core-periphery banking markets fragmented? Working Papers 219. Bank of Greece. El Ghoul, S., Guedhami, O., Kwok, C. C. Y., & Zheng, X. (2016). Collectivism and corruption in commercial loan production: Howtobreakthecurse?JournalofBusinessEthics,139(2),225e250. Espinoza,R.,&Prasad,A.(2010).NonperformingloansintheGCC banking system and their macroeconomic effects. IMF Working Papers,10(224),1.https://doi.org/10.5089/9781455208890.001 Fan,C.S.,Lin,C.,&Treisman,D.(2009).Politicaldecentralization and corruption: Evidence from around the world. Journal of Public Economics, 93(1e2), 14e34. FCA & PRA. (2015). The failure of HBOS plc (HBOS). https://www. bankofengland.co.uk/-/media/boe/files/prudential- regulation/publication/hbos-complete-report Fungacova,Z.,Kochanova,A.,&Weill,L.(2015).Doesmoneybuy credit? Firm-level evidence on bribery and bank debt. World Development, 68, 308e322. Ghosh, A. (2015). Banking-industry specific and regional eco- nomic determinants of non-performing loans: Evidence from US states. Journal of Financial Stability, 20,93e104. Godlewski, C. J. (2004). Bank capital and risk taking in emerging marketeconomies. Journal of BankingRegulation, 6(2),128e145. Goel,R.K.,&Hasan,I.(2011).Economy-widecorruptionandbad loans in banking: International evidence. Applied Financial Economics, 21(7), 455e461. Haider,Z. A.,Liu, M.,Wang,Y.,& Zhang, Y.(2018). Government ownership, financial constraint, corruption, and corporate performance: International evidence. Journal of International Financial Markets, Institutions and Money, 53,76e93. Hanousek, J., Shamshur, A., & Tresl, J. (2019). Firm efficiency, foreign ownership and CEO gender in corrupt environments. Journal of Corporate Finance, 59(15), 344e360. Hasan, I., & Wall, L. D. (2004). Determinants of the loan loss allowance: Some cross-country comparisons. The Financial Review, 39(1), 129e152. Hausman, J. A. (1978). Specification tests in econometrics. Econ- ometrica, 46(6), 1251e1271. Hofstede,G.(1983).Theculturalrelativityoforganizationalpractices andtheories.JournalofInternationalBusinessStudies,14(2),75e89. Hofstede, G. (2001). Culture's consequences: Comparing values, be- haviors, institutions, and organizations across nations. Sage. Hu, J.-L., Li, Y., & Chiu, Y.-H. (2004). Ownership and nonper- forming loans: Evidence from Taiwan's banks. The Developing Economies, 42(3), 405e420. J~ oeveer, K. (2013). Firm, country and macroeconomic de- terminants of capital structure: Evidence from transition economies. Journal of Comparative Economics, 41(1), 294e308. Kaufmann, D., Kraay, A., & Mastruzzi, M. (2010). The worldwide governance indicators: Methodology and analytical issues. Policy Research Working Papers. https://doi.org/10.1596/1813-9450- 5430 Keeton,W.,&Morris,C.(1987).Whydobanksloanlossesdiffer? Federal reserve bank of Kansas city. Economic Review,3e21. Klein, N. (2013). Non-Performing Loans in CESEE: Determinants and impact on macroeconomic performance. IMF Working Papers, 13(72), 1. Kosak,M.,Li,S.,Loncarski,I.,&Marinc,M.(2015).Qualityofbank capital and bank lending behavior during the global financial crisis.InternationalReviewof FinancialAnalysis,37,168e183. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. W. (1997). Legal determinants of external finance. The Journal of Finance, 52(3), 1131e1150. La Porta, R., Silanes, F. L. de, Shleifer, A., & Vishny, R. W. (1998). Lawandfinance.JournalofPoliticalEconomy,106(6),1113e1155. Levine, R. (1998). The legal environment, banks, and long-run economic growth. Journal of Money, Credit, and Banking, 30(3), 596e613. Lis, S. F. de, Pages, J. M., & Saurina, J. (2000). Credit growth, problem loans and credit risk provisioning in Spain.Working Papers BIS autumn central bank economists' meeting. Lizal, L., & Kocenda, E. (2001). State of corruption in transition: Case of the Czech republic. Emerging Markets Review, 2(2), 138e160. Louzis, D. P., Vouldis, A. T., & Metaxas, V. L. (2012). Macroeco- nomic and bank-specific determinants of non-performing loans in Greece: A comparative study of mortgage, business and consumer loan portfolios. Journal of Banking & Finance, 36(4), 1012e1027. Love, I., & Turk Ariss, R. (2014). Macro-financial linkages in Egypt:Apanelanalysisofeconomicshocksandloanportfolio quality.JournalofInternationalFinancialMarkets,Institutionsand Money, 28(1), 158e181. Makri, V., Tsagkanos, A., & Bellas, A. (2014). Determinants of non-performing loans: The case of Eurozone. Panoeconomicus, 61(2), 193e206. Mauro, P. (1995). Corruption and growth. Quarterly Journal of Economics, 110(3), 681e712. Mauro, P. (1998). Corruption and the compostion of government expenditure. Journal of Public Economics, 69(1), 263e279. Meon, P. G., & Weill, L. (2010). Is corruption an efficient grease? World Development, 38(3), 244e259. Micco, A., & Panizza, U. (2006). Bank ownership and lending behavior. Economics Letters, 93, 248e254. Mo, P. H. (2001). Corruption and economic growth. Journal of Comparative Economics, 29(1), 66e79. Mohaddes, K., Raissi, M., & Weber, A. (2017). Can Italy grow out of its NPL overhang? A panel threshold analysis. Economics Letters, 159, 185e189. Pantzalis, C., Park, J. C., & Sutton, N. (2008). Corruption and valuation of multinational corporations. Journal of Empirical Finance, 15(3), 387e417. Park, J. (2012). Corruption, soundness of the banking sector, and economic growth: A cross-country study. Journal of Interna- tional Money and Finance, 31(5), 907e929. Petersen, M. A. (2009). Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22(1), 435e480. https://doi.org/10.1093/rfs/hhn053 Qi, S., & Ongena, S. (2019). Will money talk? Firm bribery and credit access. Financial Management, 48(1), 117e157. Ranjan, R., & Dhal, S. C. (2003). Non-performing loans and terms of credit of public sector banks in India: An empirical assessment. Reserve Bank of India Occasional Papers, 24(3), 81e121. Rinaldi, L., & Sanchis-Arellano, A. (2006). Household debt sustain- ability. What explains household non-performing loans? An empirical analysis (Working paper series 570/January). Saha, S., & Ben Ali, M. S. (2017). Corruption and economic development: New evidence from the middle eastern and north African countries. Economic Analysis and Policy, 54, 83e95. Salas, V., & Saurina, J. (2002). Credit risk in two institutional re- gimes: Spanish commercial and savings banks. Journal of Financial Services Research, 22, 203e224. Skrastins, J., & Vig, V. (2015). How organizational hierarchy affects information production (IMFS Working Paper Series. No. 92). Stein, J. C. (2002). Information production and capital allocation. The Journal of Finance, 57(5), 1891e1921. Tarchouna,A.,Jarraya,B.,&Bouri,A.(2017).Howtoexplainnon- performing loans by many corporate governance variables simultaneously? A corporate governance index is built to US commercial banks. Research in International Business and Finance, 42(July), 645e657. 256 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 Thakor, A. V. (2018). Post-crisis regulatory reform in banking: Address insolvency risk, not illiquidity. Journal of Financial Stability, 37, 107e111. Treisman, D. (2000). The causes of corruption: A cross-national study. Journal of Public Economics, 76, 399e457. Triandis,H.C.(2001).Individualismcollectivismandpersonality. Journal of Personality, 69(6), 907e924. Van Vu, H., Tran, T. Q., Van Nguyen, T., & Lim, S. (2018). Cor- ruption, types of corruption and firm financial performance: New evidence from a transitional economy. Journal of Business Ethics, 148(4), 847e858. Vithessonthi, C. (2016). Deflation, bank credit growth, and non- performingloans:EvidencefromJapan. International Review of Financial Analysis, 45, 295e305. Weill,L.(2011a).Doescorruptionhamperbanklending?Macro and micro evidence. Empirical Economics, 41(1), 25e42. Weill, L. (2011b). How corruption affects bank lending in Russia. Economic Systems, 35(2), 230e243. Wellalage,N.H.,Locke,S.,&Samujh,H.(2018).Corruption,gender andcreditconstraints:EvidencefromSouthAsianSMEs.Journal ofBusiness Ethics,(2016),1e14. Zeume, S. (2017). Bribes and firm value. Review of Financial Studies, 30(5), 1457e1489. Zheng, X., Ghoul, S. El, Guedhami, O., & Kwok, C. C. Y. (2013). Collectivism and corruption in bank lending. Journal of Inter- national Business Studies, 44(4), 363e390. Appendix A Table A1. Definitions and data sources of variables. Variable Definition Source Bank-specific variables NPL Non-performing loans ratio (%). Measured as impaired loans divided by total loans. Fitch Connect Bank Size The natural logarithm of total assets of bank i at time t. Fitch Connect LTCD The ratio of loans to customer deposits Fitch Connect ROAA Return on average assets ratio Fitch Connect Capitalization The capitalization ratio represents the ratio of total equity to total assets in (%). Fitch Connect LLP Loan loss provisions divided by total loans in (%). Fitch Connect Macroeconomic indicators GDP Annual growth rate of GDP World Bank Unemployment The unemployment ratio (%) World Bank GCF Gross capital formation as percentage of GDP (%) World Bank RIR Real interest rates, measured as the difference between nominal interest rate and inflation rate. World Bank Economies classification Isadummyvariableequalto1ifthefinancialsystemisa bank-based financial system or 0 if the financial system is a market-based financial system. Demirguc-Kunt & Levine (1999) and own calculations Corruption indexes CI Corruption perception index ranges from 0 to 10. Transformed (10-CI c,t ). Higher values indicate more corruption. Transparency International WBCI World bank corruption perception index. Ranges from 2.5 to þ2.5. Transformed (5-(CI c,t þ2.5)). Higher value indicates more corruption. Kaufmann et al. (2010); World Bank ICRG Corruption index, higher value indicates higher corruption. International County Risk Guide; https://www.prsgroup.com/explore-our- products/international-country-risk-guide/ Institutional variables Capital stringency (Cap_Str) Indicates "whether the capital requirement reflects certain risk elements and deducts certain market value losses from capital before minimum capital adequacy is determined". Higher values demonstrate grater capital stringency. Barth, Caprio, & Levine (2013), Survey conducted in 2003, 2007 and 2011 Rule of law (RoL) The indicator captures perceptions "of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence". Kaufmann et al. (2010); World Bank Global Financial Crisis (GFC) Takes value 1 for the period 2008e2010 and 0 otherwise. Own calculations CLT We account for Hofstede's individualism index. For interpretation reasons, we define a new index named collectivism (CLT) as an index equal to 100 minus Hofstede's individualism (IDV). Higher values of CLT index indicate higher collectivism in the country. Hofstede (2001); Zheng et al. (2013); El Ghoul et al. (2016) ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 257 Table A2. Number of banks and average corruption index in sample countries from the period 2000e2016. Country Banks Avg. of WBCI Avg. of CI Country Banks Avg. of WBCI Avg. of CI Albania 4 3.20 7.05 Kuwait 14 2.34 5.51 Andorra 6 1.22 Kyrgyzstan 2 3.67 7.50 Anguilla 1 1.22 Latvia 1 2.28 6.00 Antigua and Barbuda 1 1.24 Lebanon 24 3.32 7.17 Argentina 1 2.87 7.00 Libya 1 3.98 8.23 Armenia 6 3.13 6.91 Liechtenstein 1 1.21 Aruba 2 1.31 Lithuania 2 2.16 5.55 Australia 18 0.58 1.63 Luxembourg 8 0.51 1.57 Austria 28 0.74 2.09 Macau 1 1.99 4.48 Azerbaijan 12 3.59 7.60 Macedonia 1 2.87 7.30 Bahamas 10 1.18 2.96 Madagascar 1 2.74 7.10 Bahrain 16 2.27 4.94 Malawi 3 3.11 6.86 Bangladesh 3 3.78 8.43 Malaysia 5 2.31 5.02 Barbados 3 1.04 2.75 Malta 1 1.59 3.90 Belarus 9 3.04 7.11 Mauritius 3 2.20 5.04 Belgium 19 1.01 2.75 Mexico 62 2.99 6.71 Benin 1 3.09 7.50 Moldova 2 3.27 7.33 Bolivia 1 2.95 7.13 Monaco 1 Botswana 2 1.59 3.90 Mongolia 1 3.00 6.20 Brazil 48 2.57 6.21 Morocco 2 2.76 6.57 Bulgaria 10 2.65 6.13 Myanmar 1 3.89 Burundi 1 3.47 7.65 Namibia 1 2.26 5.90 Cambodia 1 3.68 7.86 Netherlands 27 0.43 1.27 Canada 98 0.55 1.47 New Zealand 14 0.19 0.77 Cayman Islands 12 1.31 Nicaragua 2 3.27 7.22 Chile 26 1.08 2.95 Nigeria 5 3.64 7.81 China 35 2.97 6.42 Norway 10 0.39 1.30 Colombia 12 2.78 6.29 Oman 1 2.13 4.89 Costa Rica 11 1.88 5.06 Pakistan 1 3.41 7.55 Croatia 4 2.40 6.32 Panama 40 2.82 6.49 Cuba 2 2.25 5.80 Peru 2 2.77 6.52 Curacao 9 Philippines 9 3.07 7.09 Cyprus 10 1.45 4.03 Poland 13 2.04 5.21 Czech Republic 6 2.14 5.76 Portugal 17 1.44 3.68 Denmark 4 0.12 0.61 Puerto Rico 5 1.76 4.10 Dominica 2 1.92 4.63 Qatar 7 1.57 3.46 Dominican Republic 5 3.24 6.94 Romania 10 2.72 6.44 Ecuador 4 3.19 7.08 Russian Federation 101 3.42 7.49 El Salvador 8 2.89 6.22 Rwanda 2 2.31 5.33 Estonia 6 1.36 3.29 Saint Lucia 1 2.90 Ethiopia 3 3.09 7.28 San Marino 1 Finland 10 0.23 0.82 Saudi Arabia 11 2.55 5.80 France 81 1.13 3.09 Serbia 3 3.15 7.10 Georgia 13 2.44 5.99 Singapore 1 0.29 0.85 Germany 37 0.67 2.15 Slovakia 7 2.18 5.77 Ghana 6 2.61 6.00 Slovenia 8 1.57 4.13 Greece 6 2.18 5.69 South Africa 13 2.27 5.42 Guatemala 4 3.20 7.09 Spain 49 1.45 3.65 Guernsey 6 Sri Lanka 2 2.76 6.68 Guyana 1 3.12 7.04 Sweden 9 0.28 0.82 Haiti 3 3.80 8.22 Switzerland 19 0.41 1.25 Honduras 1 3.25 7.10 Syrian Arab Republic 7 3.75 7.77 Hong Kong 13 0.67 1.87 Taiwan 3 1.72 3.94 Hungary 18 2.00 4.97 Tajikistan 3 3.70 7.76 Iceland 14 0.36 0.99 Thailand 1 2.75 6.57 India 2 2.83 6.58 Togo 1 3.42 7.04 Indonesia 15 3.34 7.71 Trinidad and Tobago 6 2.63 6.34 Iran 1 3.22 7.10 Turkey 34 2.64 6.34 Iraq 3 3.84 8.33 Uganda 3 3.43 7.47 Ireland 18 0.93 2.55 Ukraine 26 3.43 7.66 Isle of Man 1 United Arab Emirates 20 1.48 3.59 (continued on next page) 258 ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 Table A2. (continued) Country Banks Avg. of WBCI Avg. of CI Country Banks Avg. of WBCI Avg. of CI Israel 3 1.57 3.56 United Kingdom 43 0.67 1.76 Italy 33 2.21 5.29 United States 6270 1.03 2.58 Jamaica 3 2.70 6.32 Uruguay 2 1.21 3.04 Japan 10 1.06 2.59 Uzbekistan 18 3.74 8.22 Jersey 2 Vanuatu 1 2.27 6.73 Jordan 9 2.32 5.09 Venezuela 4 3.65 7.86 Kazakhstan 16 3.48 7.40 Yemen 3 3.85 8.02 Kenya 5 3.47 7.57 Zambia 2 2.95 7.28 Korea (South) 3 3.86 5.25 Zimbabwe 1 3.89 7.93 Source:CI-corruptionperceptionindexfromTransparencyInternational;WBCI-corruptionperceptionindexfromWorldBank-World Development Indicators and our calculation. ECONOMIC AND BUSINESS REVIEW 2022;24:240e259 259