Fraud Recovery and Country Governance: Evidence from Operational Losses of. U.S. Bank Holding Companies

Transcription

1 Fraud Recovery and Country Governance: Evidence from Operational Losses of U.S. Bank Holding Companies Filippo Curti Atanas Mihov This Draft: August 15, 2015 Abstract Using regulatory data from large financial institutions on their operational losses related to fraud in foreign markets, we find large differences in recovery rates across countries. In particular, we find that losses in countries with poor governance have lower recovery rates than losses in countries with good governance. Such results are driven by dimensions such as control of corruption, rule of law, regulatory quality and government effectiveness as opposed to other dimensions such as political stability, and voice and accountability. In addition, we document important interaction effects between country governance and bank risk management quality suggesting strong governance effects on fraud recovery for institutions with poor risk management. This paper presents unique evidence that ties in international market characteristics with operational risk beyond what has been documented in prior literature. Keywords: Fraud; Recovery; Operational risk; Risk management; Internationalization; Banking; Country governance; Filippo Curti and Atanas Mihov are Financial Economists at the Federal Reserve Bank of Richmond and can be reached at and respectively. We thank Azamat Abdymomunov, José Fillat, Scott Frame, Jeff Gerlach, Mike Gibson, Joseph Nichols and Richard Rosen for helpful comments. We are especially indebted to Nada Mora for valuable discussions and suggestions. We also thank Leandro Sanz for excellent research assistance. All remaining errors are our own. The views expressed in this paper do not necessarily reflect the position of the Federal Reserve Bank of Richmond or the Federal Reserve System.

2 1 Introduction Multinational companies, and especially financial firms, have various forms of exposure to the foreign markets they operate in. For instance, a recent article in the Wall Street Journal discusses a number of fraud cases in China that have affected Western banks operating there. 1 Such occurrences are linked, at least partially, to governance dimensions such as the country legal environment: Foreign lenders in China have been stung by a string of suspected fraud cases and problem loans in the country as Beijing investigates company executives and seizes assets in a crackdown on corruption. In this study we consider a specific form of operational risk, internal and external fraud, and ask several questions. What country-specific factors are associated with fraud at corporations? Do country governance and legal environment matter? Are good internal risk management practices a substitute for country governance? How much can be recovered? To address such questions, we use supervisory operational loss data reported by U.S. bank holding companies (BHCs) related to internal and external fraud cases in international markets. Such losses are reported by large banking organizations to the Federal Reserve System for stress testing purposes in accordance with the Dodd-Frank Wall Street Reform and Consumer Protection Act. As discussed in De Fontnouvelle et al. (2006), public sources of operational loss data, including fraud cases, are biased towards larger losses, and may not report all significant cases. In contrast, we have the advantage of more comprehensive and complete loss data at the individual firm level. Using such data from 2002 to 2012, we indeed document important relationships between fraud event outcomes and the governance and legal environment of countries. Our main findings can be summarized as follows. We document a significant association between fraud recovery rates and country governance. Specifically, we show that recovery 1 See Wall Street Journal: Troubles in China Rattle Western Banks (E. Curran, October 27, 2014). 1

3 rates are higher in countries with good governance relative to countries with bad governance. In terms of economic magnitude, a one standard deviation increase in our country governance index is associated with 0.9 percentage points increase in recovery rates. Such an increase is economically significant given an unconditional average recovery rate of 3.3 percentage points in our sample. To precisely identify the effects of country governance in our setting, we need an exogenous shock to governance, at the country level, independent of other national factors. Such an event is highly unlikely. Moreover, if one could identify such an exogenous change in country governance, it is improbable that the results would generalize to the majority of countries where banks incur fraud-related losses. For these reasons, we choose to take an approach as general as possible, and to control for alternative explanations using instrumental variables. Specifically, to mitigate endogeneity concerns, we instrument for country governance using countries legal origins. Since legal traditions are typically introduced into different countries through colonization and conquest, legal origin can be considered largely exogenous (La Porta et al. (1997) and Porta et al. (1998)). In addition, while legal origin should be correlated with the quality of country governance, its effect on fraud recovery rates should not be through spurious uncontrolled for effects. Using this approach, we find similar results. We then investigate the governance dimensions driving this relationship. In particular, we decompose our country governance measure in six components: control of corruption, regulatory quality, rule of law, government effectiveness, political stability, and voice and accountability. Arguably, such governance dimensions capture disparate aspects of governance and thus might have differential influence on fraud recovery. While the capacity of the government to effectively formulate and implement sound policies should matter more, political stability or the process of selection and replacement of governments are likely less important. Consistent with such a hypothesis, we show that the positive association between 2

4 country governance and recovery rates is driven by governance dimensions such as control of corruption, regulatory quality, rule of law and government efficiency. In contrast, dimensions such as government stability, and voice and accountability are economically and statistically insignificant. Finally, we also investigate the link between banks internal risk management practices and fraud recovery in an international context, and examine potential interaction effects between risk management quality and country governance. Leveraging a risk management rating internally developed and maintained by the Federal Reserve System, we indeed find significant relationships. Specifically, we show that country governance has stronger positive effects on recoveries for banks with poor practices. Conversely, banks with poor risk management face incrementally lower recoveries in countries with bad governance. In terms of economic magnitude, a one standard deviation increase in country governance increases recoveries for institutions with poor management significantly - by 5.0 percentage points. To check the robustness of our results, we re-run our analyses using alternative variable formulations and different estimation methodologies, where we find evidence in support of our main finding that country governance is associated with higher recovery in international fraud-related losses. In addition, we also design and implement a falsification test to enhance the credibility of our results. Specifically, we re-estimate the relationship between country governance and operational loss recovery rates for types of losses whose recoveries we a priori believe should not be related to country governance. Consistent with our hypothesis, we do not find a significant relationship, suggesting that our main empirical results positively linking country governance to fraud recovery are not driven by latent operational loss factors unrelated to country governance. Our paper contributes to several research streams. First, our paper extends the large and growing literature linking country governance and investor protection to economic activity and outcomes. For example, La Porta et al. (2006) examine the effect of securities laws 3

5 on stock market development in 49 countries, finding strong evidence that laws mandating disclosure and facilitating private enforcement through liability rules benefit stock markets. Similarly, La Porta et al. (1997) show that countries with poorer investor protections, measured by the character of legal rules and the quality of law enforcement, have smaller and narrower capital markets. Antras et al. (2009) provide evidence that links multinational firms capital flows to country investor protection environment. 2 La Porta et al. (2002b) show evidence of higher valuation of firms in countries with better protection of minority shareholders. Djankov et al. (2007) examine the cross-country determinants of private credit, showing that both creditor protection through the legal system and informationsharing institutions are associated with more private credit. Nenova (2003) shows the value of control-block votes varies widely across countries with the legal environment, law enforcement, investor protection, takeover regulations, and power-concentrating corporate charter provisions explaining the majority of variation in the value of control-block votes. Davydenko and Franks (2008) show that bank debt recovery rates across countries are sharply different, reflecting different levels of creditor protection. In contrast, our study provides unique evidence documenting a significant positive relationship between the quality of country governance and loss outcomes from internal and external corporate fraud. Second, our study also contributes to a large literature in accounting, banking and finance focusing on fraud. Choi (2007), Griffin et al. (2001) and Thompson and Sale (2003) provide evidence on the frequency and the cost imposed by fraud. Burns and Kedia (2006) and Efendi et al. (2007) examine the characteristics of firms involved in fraud. Palmrose and 2 Related literature including King and Levine (1993), Levine and Zervos (1998), Rajan and Zingales (1998), Wurgler (2000), and Acemoglu et al. (2009) among others, shows that financial market conditions have bearing on corporate investment behavior, economic growth, and industry composition. More broadly, our research also relates to the literature on firm exposure to foreign country risk (e.g., Hughes et al. (1975), Agmon and Lessard (1977), Amihud and Lev (1981), Michel and Shaked (1986), Bartov et al. (1996), Reeb et al. (1998), Burgman (1996), Cuervo-Cazurra et al. (2007), Reeb et al. (1998), Kwok and Reeb (2000), Fillat et al. (2015), Fillat and Garetto (2015), Berger et al. (2004), Amihud et al. (2002), Chari and Gupta (2008), Berger et al. (2001), Li and Guisinger (1992), Shapiro (1985), Brewer and Rivoli (1990), Buch and DeLong (2004)). 4

6 Scholz (2004) document the impact of fraudulent financial reporting on firm value. Dyck et al. (2010) identify the most effective mechanisms for detecting corporate fraud. Karpoff et al. (2008a) and Karpoff et al. (2008b) focus on the costs borne by firms and managers upon fraud detection. La Porta et al. (2003) provide evidence that related loans are more likely to default and have lower recovery rates when they do. Our study complements this strand of literature with evidence on fraud recovery by banking institutions in an international context. In addition, we show unique evidence that strong legal environment and country governance mitigate fraud recovery experience especially for institutions with poor risk management practices. Our work is also related to the literature on the determinants of operational risk of financial institutions. Chernobai et al. (2012) show that most operational losses can be traced to a breakdown of internal controls. The findings in the paper highlight the correlation between operational risk and credit risk, and focus on the role of corporate governance and managerial incentives in mitigating operational risk. Allen and Bali (2007) provide evidence of cyclical components in operational risk. Moreover, the study suggests that approximately 18% of financial institutions returns compensate for operational risk, with this number growing to 39% for depositary institutions. Cope et al. (2012) investigate the relationship between operational loss severity and various regulatory, legal, geographical, and economic indicators. Hess (2011) and Cope and Carrivick (2013) analyze the impact of the financial crisis on operational risk and losses in the financial services industry. In contrast, we take a different direction and show that the outcome from a specific type of operational losses, fraud, is associated with the governance of the foreign markets where fraud occurs. Such findings suggest that that the legal underpinnings of markets could be an important factor to consider when it comes to firm operational risk. Finally, our findings have supervisory policy implications regarding operations and risk taking in foreign countries. We provide results suggesting that foreign country governance 5

7 is a relevant dimension for U.S. bank holding companies operational losses, which should be considered when analyzing banks foreign exposure. Specifically, increased attention to foreign exposures in countries with poor governance practices might be warranted. The results are particularly resounding given that major fraud cases have recently rattled several high-profile banking institutions in the United States, with fraud losses sometimes booked through foreign branches. For example, as recently as 2012, JPMorgan Chase lost billions of dollars amid a rogue trading scandal in its London office. 3 The remainder of the paper is organized as follows. Section 2 describes our sample, provides variable definitions, and presents descriptive statistics. Section 3 lays out our main empirical results. Section 4 checks for robustness. Finally, Section 5 concludes. 2 Fraud Sample and Variable Definitions 2.1 Sample Construction In this paper, we use a supervisory data set of operational losses provided by U.S. bank holding companies in accordance with the Dodd-Frank Wall Street Reform and Consumer Protection Act. The reported data complies with FR Y-14Q reporting requirements and covers BHCs with total consolidated assets of $50 billion or more. 4 For every individual operational loss, BHCs are required to report data points such as gross amounts, recovery amounts, loss event type, and a description of the event that generated the loss. The Basel Committee on Banking Supervision defines 7 types of operational losses: Internal fraud (IF), External Fraud (EF), Employment Practices and Workplace Safety (EPWS), Clients, Products and Business Practices (CPBP), Damage to Physical Assets (DPA), Disruption of Business or System Failures (BDSF), and Execution, Delivery and Process Man- 3 See Wall Street Journal: Making Waves Against Whale (K. Burne, April 10, 2012). 4 A detailed description of the FR Y-14Q reporting forms and instructions can be found at: http: // 6

8 agement (EDPM). Since the focal point of our study is to uncover the relationship between country governance and the outcome of fraud, the analysis concentrates on the first two Basel categories - internal fraud and external fraud. In the last section of our study, as a robustness check, we test whether the relationship holds within the 5 remaining event type categories. The distinctive characteristic that differentiates internal from an external fraud events is whether banks employees are involved in the fraudulent action. Events where banks employees are directly involved are considered internal fraud, while events where banks employees are not directly involved are considered external fraud. In our sample, internal fraud type events are mostly comprised of fraud instances in the form of fraudulent deposits or loan issuances. Cases of embezzlement by personal bankers and rogue traders are also common. External fraud type events are mostly comprised by fraudulent payment order, wire transfers, checks, and credit card transactions. In order to detect which country originates an operational loss we employ a regular expression algorithm that reads the individual loss description field and matches it with a list of country names. We then manually check every entry to verify that our algorithm properly paired the operational loss with a particular country. The final data set includes 1,270 fraud-related operational losses from 13 unique BHCs and from 115 countries Characteristics of Fraud Events First, we investigate the time series variation in fraud occurrence. Table 1, Panel A reports the number of operational loss events by year of discovery. One can notice a significant variation in the number of fraud events across years. For instance, there are 187 fraud instances 5 Of these, internal fraud accounts for 66 loss events, while external fraud accounts for 1,204. Due to the small number of internal fraud instances in our sample, we do not differentiate between internal and external fraud in this study. 7

9 in 2012, while there are as few as 21, 23 and 30 events in 2002, 2003 and 2004, respectively. 6 Table 1, Panel B shows a breakdown of fraud loss events by country. The countries with the highest number of fraud occurrences are ones that are either geographically close or economically linked to U.S. (e.g., Canada, Mexico and the United Kingdom), or alternatively, tax haven countries such as Bermuda, Maldives and Seychelles (Gravelle (2015)). 7 [Insert Table 1 about here] Table 2, Panel A reports descriptive statistics for our data set on international fraud. For comparison purposes, Panel B of the same table reports descriptive statistics for a random sample of 10,000 fraud events (incurred by the same 13 BHCs) that include both domestic and international instances. The mean loss amount for a fraud loss occurring outside the U.S. is $428,000, while the mean in the random sample is only $10,000. Such differences potentially reflect banks disparate reporting thresholds for domestic and international losses. The mean recovery rate of fraud losses occurring outside the United States is 3.3%. 8 In contrast, the mean recovery rate for our random sample is 2.3%. Overall, such results indicate that loss recovery related to fraud events is fairly low. [Insert Table 2 about here] 6 As a result of the implementation of Basel II requirements, banks data collection and reporting standards have substantially improved since 2005 and the effect is reflected in the time series frequency of our sample. Prior to 2005 the average number of events in a year is around 25, while after 2005 it is closer to 150. While we do not expect the improvement in data collection and reporting practices of U.S. BHCs to have direct confounding effects on the relationship between operational loss recovery rate and foreign country governance, we verify that excluding data before 2005 from our analysis produces directionally and statistically similar results. 7 Two countries, the Maldives and Canada, stand out in terms of number of fraud occurrences. Specifically, 158 and 106 instances of fraud occurred in the Maldives and Canada, respectively. Both of these counts are significant relative to the total number of observations in our sample, We verify our main results are robust to the exclusion of both the Maldives and Canada from our sample. 8 Fraud loss recovery rate, RR, is formally defined as the total amount recovered in an event of fraud, expressed as a percentage of the gross amount lost. 8

10 2.3 Country Governance and Control Variables To examine whether the quality of country governance has any influence on fraud recovery rates, we follow Alcala and Ciccone (2004) and Karolyi and Taboada (2015), and use the Worldwide Governance Indicators (WGI) calculated by Kaufmann et al. (2009) under the auspices of the World Bank. 9,10 Such data, available for 215 countries over the period [ ], comprises six aggregate governance indicators based on hundreds of specific individual variables gauging various aspects of governance, drawn from 35 data sources and constructed by 33 different organizations worldwide. These data reflect the governance perceptions of public sector, private sector and NGO experts, including thousands of citizen and company respondents across the world. Kaufmann et al. (2009) broadly define governance as the traditions and institutions by which authority in a country is exercised. This includes the process by which governments are selected, monitored and replaced; the capacity of the government to effectively formulate and implement sound policies; and the respect of citizens and the state for the institutions that govern economic and social interactions among them. This definition is molded into six governance indicators: control of corruption, rule of law, regulatory quality, government effectiveness, political stability, and voice and accountability. Control of corruption (CC Index) captures the extent to which public power is exercised for private gain. Both petty and grand forms of corruption are considered, including political and governmental capture by private interests. Rule of law (RL Index) measures the quality of contract enforcement, property rights, the police (violent and organized crime), and the courts (fairness and expediency of judicial process). Regulatory quality (RQ Index) measures the ability of the 9 Other recent studies using WGI data include Beck et al. (2014), Caprio et al. (2011), Berden et al. (2014), Badinger (2008), and Kurtz and Schrank (2007) among others. 10 In unreported results, we check the robustness of our findings to using alternative measures of country governance. We follow Easton and Walker (1997), La Porta et al. (2002a) and Lothian (2006), and leverage data developed and maintained by the Fraser Institute. Specifically, we use legal structure, quality of property rights and burden of business regulation indices to verify that our main results hold using alternative country governance indicators. 9

11 government to formulate and implement sound policies and regulations that promote private sector development. Government effectiveness (GE Index) captures the quality of public and civil services, the degree of their independence from political influences as well as the quality of policy formulation and implementation. The effectiveness of public institutions and the excessiveness of red tape receive particularly strong consideration. Political stability (PS Index) measures the likelihood of political instability and politically motivated violence. Specific considerations include disorderly and violent transfers of power, armed conflict, social unrest and the threat of terrorism. Lastly, voice and accountability (VA Index) captures the extent to which a country s citizens are able to participate in government election, freedom of expression and association, and freedom of the press. 11 As originally constructed, each of these indicators is bounded between -2.5 and 2.5 with higher values indicative of better governance (Kaufmann et al. (2009)). To accommodate our empirical strategies and for presentation purposes, we rescale all indices by adding 2.5 and then dividing by 10. Using these six transformed indicators we then calculate our main aggregate governance measure, defined as the equal-weighted average of all six governance indicators. We call the variable CG Index. Table 2 presents the summary statistics. All governance indices have positive sample means, centered around 0.28, with standard deviations close to The means, the standard deviations and the spreads between the 1 st and the 99 th percentiles suggest adequate sample variation in the governance of the countries, where losses are incurred. To give a sense of the relative strength of governance across countries, we compare the top 15 countries by number of fraud occurrences. Specifically, Figure 1 presents bar charts of country governance, as measured by CG Index, for each of the 15 countries, with Figure 1A presenting values for 2002 and Figure 1B for Additional details, including methodology, data sources and measurement error analysis among others, can be found in Kaufman et al. (2004) and Kaufmann et al. (2009). 10

12 [Insert Figure 1 about here] From the countries represented, those with relatively stronger governance include Canada, the United Kingdom and France. In comparison, the countries with relatively poorer governance include Nigeria, Russia and China. Importantly, by comparing Figure 1A and 1B, one can also notice that country governance (CG Index) is relatively stable during the period without large swings in individual country governance indicators. In addition to country governance measures, our multivariate regression analysis also uses a number of variables to control for confounding effects. These include country level (financial development and economic growth), bank level (size of subsidiary operations in foreign countries; risk management quality), and fraud event level (loss gross amount) variables. Definitions of all the variables can be found in Appendix A of our study. 3 Main Empirical Results 3.1 Fraud Recovery and Country Governance We start our examination of the relationship between fraud recovery and country governance quality with simple correlation analysis. Table 3 presents correlations coefficients between the major variables in our analysis. [Insert Table 3 about here] There are several notable observations. There is a strong positive correlation between economic development and country governance, consistent with the interpretation that more developed countries have better governance quality. Economic growth, on the other hand, is negatively correlated with country governance and economic development, suggesting that 11

13 more developed countries with good governance have lower rates of economic growth. As expected, banks have bigger operations in developed countries with strong governance. Bigger losses (higher gross amounts) are incurred in developed countries, where bank holding companies have more robust representation. More importantly, the results also indicate strong positive association between country governance and fraud loss recovery. The correlation coefficients of CG Index with RR is 7.8%, statistically significant at the 5% level. Such evidence suggests, at least at a first glance, that bank holding companies have higher fraud recovery rates in countries with higher governance quality. We next investigate this hypothesis in a more rigorous regression setting. We utilize the following ordinary least squares (OLS) specifications: E(RR it CG Index it, Ctrl it ] = β i β p + β 1 CG Index it + β 2 Ctrl it (1) where i indexes the bank holding company and t indexes the year of fraud. RR is the fraud loss recovery rate, defined as the total amount recovered in an event of fraud, expressed as a percentage of the gross amount lost. 12 CG Index is a governance index at the countryyear level. Ctrl is a vector of control variables. Specifically, we include the following: gross amount of the loss (Ln(Loss)), country economic development (EcoDev), country economic growth (EcoGrowth), and country presence (ForSubSize). β i β p are interactions of bank holding company and time period fixed effects, where two time periods are defined - pre and post The dramatic change in macro and regulatory environments before and after 2008 affected banks policies, procedures and operating environment unequally across institutions. Allowing for different BHC level fixed effects during the two periods controls for such effects One could argue that the skewness of RR (as indicated by summary statistics) might affect the statistical inferences in our analysis. In robustness tests, we show that the results hold when using specifications with a natural log transformation of RR to mitigate skewness concerns. 13 We verify the robustness of our main results to alternative fixed effect schemes such as, for example, using BHC and year level fixed effects separately. 12

14 The error terms are clustered at the β i β p level, and are heteroscedasticity-consistent. Table 4, Panel A presents the results. [Insert Table 4 about here] Our country governance measure, CG Index, has a positive sign and is statistically significant at the 5% level. Such a positive association implies that banks have higher fraud recovery rates in foreign countries with good governance. In terms of economic significance, a one standard deviation increase in CG Index is associated with 0.9 percentage points increase in RR. Given an average recovery rate of 3.3 percentage points, this translates into a 29% increase. In dollar terms, a one standard deviation in country governance translates into $4,000 incremental recovery. Overall, this evidence suggests that country governance is a statistically and economically important factor for bank fraud recovery. Specifically, banks have higher fraud recovery rates in countries with better governance. Most of the control variable coefficient signs are intuitive. Fraud loss size, Ln(Loss), is positively related to recovery rates, suggesting that banks recover more on larger losses. Such an association potentially reflects procedural safeguards against material fraud events at banks as well as increased incentives to pursue the resolution of large fraud cases. In addition, banks have higher loss recovery rates in countries where they have larger presence as indicated by the positive coefficient of ForSubSize. Increased political and business clout in the foreign markets where banks have presence, better knowledge of local norms and laws, as well as more local resources and infrastructure to be used to recover fraud losses might account for such a relationship. There is also some limited evidence that country economic development might be a factor for fraud loss recovery. Specification (3) shows positive correlation between EcoDev and RR, statistically significant at the 5%. However, once we control for other factors, this relationship turns insignificant at conventional levels as shown 13

15 in Specification (6). Admittedly, a potential problem with estimating Equation (1), is that the relationship between fraud recovery rates and country governance could be contaminated by endogeneity, and thus can be spurious. For instance, governance could proxy for omitted country level factors. We estimate a two-stage least squares (2SLS) system to address such concerns. La Porta et al. (1997) and Porta et al. (1998) provide empirical evidence that legal rules, and more generally governance practices, vary systematically among legal traditions and origins. Arguably, since legal traditions are typically introduced into different states through colonization and conquest, they can be considered largely exogenous. Based on such assumptions, we use legal origins as an instrument for country governance quality in a two-stage procedure, where the second stage explains fraud recovery rates in foreign markets. Table 4, Panel B reports the results. The second stage estimation in Column (2) shows that the association between fraud recovery and country governance is robust to accounting for endogeneity concerns. Econometrically, a major requirement for an instrument to be valid is that it should be correlated with the potentially endogeneous explanatory variable. Column (1) shows evidence of this. The instrumental variables have highly significant coefficients, which is consistent with the arguments in La Porta et al. (1997) and Porta et al. (1998) that legal origins indeed play a role in country governance profiles. The adjusted R 2 s are high and the F-statistics are well above the threshold of 10 prescribed by Stock et al. (2002). Such evidence supports the claim that our instrumental variable estimations do not suffer from weak instrumental variable problems. 3.2 Country Governance Dimensions In general, the concept of country governance relates to the conventions, traditions and institutions through which authority is applied and exercised in a given country. Our main measure of country governance, CG Index, weights equally 6 governance dimensions: control 14

16 of corruption, rule of law, regulatory quality, government effectiveness, political stability and absence of violence, and voice and accountability (Kaufmann et al. (2009)). These dimensions capture substantially disparate aspects of country governance and, thus, might have a differential effect on the link between governance and fraud recovery. In this section, we examine each component separately. We hypothesize that the dimensions driving the documented positive relationship between governance and fraud recovery should be through the capacity of the government to effectively formulate and implement sound policies (i.e., law enforcement, control of corruption, regulatory quality, and the efficiency of a country s administrative body). In contrast, we expect that dimensions related to political stability or the process of selection and replacement of governments (i.e., voice and accountability) should matter less. We start with decomposing CG Index into its six different component indices: CC Index (control of corruption), RL Index (rule of law), RQ (regulatory quality), GE Index (government effectiveness), PS Index (political stability and absence of violence), and VA Index (voice and accountability). We then proceed by examining the relationship between fraud recovery and the six individual components. Table 5, Panel A shows strong correlation betweens RR and each of CC Index, RL Index, RQ Index and GE Index. The correlation coefficients are positive and highly statistically significant. In contrast, the correlation between RR and each of PS Index and VA Index, albeit positive, is statistically insignificant at conventional levels. [Insert Table 5 about here] We next show similar results in a simple portfolio analysis setting. In particular, we examine fraud recovery rate portfolios formed on the six governance component indices. We sort fraud events into below/above median groups based on CC Index, GE Index, RQ Index, 15

17 RL Index, PS Index and VA Index ranking. We then calculate averages of our fraud recovery variables for every group and test for mean difference between the groups. Table 5, Panel B present results. Notably, the groups with above median governance have higher average recovery rates than those in the below median groups for CC Index, RL Index, RQ and GE Index. In contrast, average recovery rates in the above/below median groups are indistinguishable from each other at conventional statistical levels for PS Index and VA Index. Correlation and univariate sort analyses thus suggest that only certain dimensions of country governance quality matter for fraud loss recovery. We next engage into multivariate regression analysis to further examine the relationship between different country governance dimensions and fraud recovery. To do so, we estimate regressions similar to Equation (1), but we replace CG Index with each of its components: CC Index, GE Index, RQ Index, RL Index and VA Index. Table 6 presents results. [Insert Table 6 about here] Consistent with Table 5, the results in Table 6 support our hypothesis that different dimensions of country governance have unequal effect on the link between country governance and fraud recovery. Specifically, while law enforcement, control of corruption, regulatory quality, and the efficiency of a country s administrative body are positively related to recovery rates, political stability and democratic principles of rule are not. Economically, a one standard deviation increase in CC Index, RL Index, RQ and GE Index is associated with 0.9, 1.4, 1.7 and 0.7 percentage points increase in RR, respectively. Regression coefficients are statistically significant at the 5% level or higher. 16

18 3.3 Country Governance and Bank Risk Management Chernobai et al. (2012) examine the role of corporate governance in mitigating operational risk. Their findings suggest that firms with weak corporate governance and internal controls are associated with a higher incidence of operational risk events. Consistent with such findings, regulators and supervisors have recognized and accentuated risk management as essential for risk mitigation. For example, the second pillar of the Basel II Capital Accord mandates the regular reporting of operational risk exposure and loss experience to senior management and the board of directors, and regular review of risk management systems and processes by internal and external auditors. Furthermore, international finance literature suggests that country and firm level corporate governance (and internal control mechanisms) might have important interaction effects. Evidence exists that country and company governance processes are substitutes of each other. Bris and Cabolis (2008) document, in mergers and acquisitions context, that superior shareholder protection and better accounting standards in an acquirer s country are related to higher merger premium in cross-border mergers relative to matching domestic acquisitions. Similarly, Dahya et al. (2008) find a positive relation between corporate value and the fraction of independent directors especially in countries with weak legal protections for shareholders. Such observations motivate us to examine the association between internal bank risk management and fraud recovery, and especially explore potential interaction effects of bank risk management practices on the link between fraud recovery rates and country governance. Specifically, in this section we examine whether good country governance offsets effects of poor bank risk management practices on recovery experience. To test this hypothesis, we use a risk management rating system developed and internally maintained by the Federal Reserve Bank. 14 The rating system is at the bank holding 14 A detailed description of the Bank Holding Company Rating System can be found at: federalreserve.gov/boarddocs/srletters/2004/sr0418.htm. 17

19 company level and provides an evaluation of risk management factors that are common to all BHCs (with a strong focus on uniformity across institutions). Specifically, the ratings constitute an assessment of the ability of the BHCs boards of directors and senior management, as appropriate for their respective positions, to identify, measure, monitor, and control risk. These ratings range on a numeric scale from 1 to 5. A rating of 1 indicates the strongest risk management practices and least degree of supervisory concern. In contrast, a rating of 5 indicates the lowest rating and highest degree of supervisory concern. Based on such a scale, we categorize banks with a rating greater or equal to 3 as institutions with bad risk management practices. We then employ a regression framework similar to Equation (1), where we include interaction terms between country governance and our bank risk management quality indicator variable (Bad RM ) along with country governance and the risk management indicator that enter the regressions separately. Table 7 presents the results. [Insert Table 7 about here] In Panel A, specifications (1) and (2) indicate that although Bad RM has the expected negative sign, the coefficients are not statistically significant at conventional levels. 15 Importantly, however, the interaction effect between country governance and bank risk management is positive and significant (at the 10% level). In terms of economic magnitude, a one standard deviation increase in country governance increases recoveries for institutions with poor management by 5.0 percentage points - a significant increase. The results thus indicate that banks with poor risk management recover more from losses incurred in good governance countries, an observation consistent with the interpretation that strong country institutions are particularly important and beneficial for companies with weaker internal mechanisms (at 15 Changing the definition of Bad RM to categorize banks with a rating equal to 5 as institutions with bad risk management produces statistically significant coefficients for Bad RM. However, such results should be interpreted with caution since the subsample of banks rated 5 is substantially thin. 18

20 least with regards to fraud recovery experience). Notably, in both Specifications (1) and (2), the coefficient of country governance, CG Index, remains positive and statistically significant at conventional levels. Even though the relationship between fraud recovery and governance is ampliefied for institutions with poor risk management, the effects extend beyond just banks with weak controls. In Panel B, we also run estimations interacting the subcategories of country governance with risk management. All the interaction terms with the exception of Bad RM*VA Index are positive and statistically significant at least at the 5% level. While the majority of country governance components are particularly relevant for financial firms with weaker internal controls, voice and accountability seems to be the exception. The coefficient estimates of CC Index, RL Index, RQ Index and GE Index remain positive and statistically significant, consistent with results from Table 6. Interestingly, while country political risk is not associated with fraud recovery rates on average, these results suggest it might be relevant for institutions with weak risk management practices. 4 Robustness and Falsification Tests In this section, we provide several robustness checks of our results. First, we examine whether our main empirical findings are robust to accounting for alternative estimation methodologies and different variable formulations. Second, we design and implement a falsification test to enhance the credibility of our results. 4.1 Robustness Checks Our fraud recovery data are bounded between 0 and 1, with a large concentration of observations at the boundary of 0 and to a lesser extent at the boundary of 1. An OLS estimation of Equation (1) may have similar limitations to using a linear probability specification to 19

21 model binary response variables. Specifically, it cannot guarantee the predicted values from the model will map to the bounded [0,1] interval as noted in Wooldridge (2002). In addition, OLS estimation of Equation (1) would ignore important nonlinearities that might exist in our fraud recovery data. A common econometric method used to address similar situations is to utilize a transformation G(.) of the data that maps the [0-1] recovery rates onto the whole real line [ ω, +ω] (McCullagh and Nelder (1989)). While several functional forms for G(.) are available, one of the most commonly used is the cumulative normal distribution function, which we employ in our analysis. Similar to Papke and Wooldridge (1996), we use quasi-maximum likelihood estimation (QMLE) to estimate Equation (1), where the non-linear estimation procedure maximizes the Bernoulli log-likelihood function given by: l i (β) = y i ln(g i ) + (1 y i )ln(1 Gi) (2) QMLE produces consistent and asymptotically normally distributed estimates as noted by Gourieroux et al. (1984). 16 Table 8, Panel A presents the results from our estimations. [Insert Table 8 about here] The results are directionally similar to those in Tables 4 and 6, and statistically significant at conventional levels; strong country governance is associated with higher fraud recovery rates. As an alternative to the quasi-maximum likelihood estimation methodology of Papke and Wooldridge (1996), we also re-estimate Equation (1) applying a natural log transformation to fraud recovery rates. The intention is to address concerns that skewness of our recovery 16 For an application of the Papke and Wooldridge (1996) QMLE method in banking-related topics, see, for example, Dermine and Neto de Carvalho (2006) and Khieu et al. (2012). 20

22 rate variable might affect the statistical inferences in our analysis. Table 8, Panel B presents the results. The results are again consistent with Tables 4 and 6, indicating a positive association between governance and fraud recovery. 4.2 Falsification Test In this section, we implement a falsification test, which examines whether country governance is a factor for recovery rates of categories of operational losses in foreign markets, which we a priori believe should not be related to the strength of country governance. The intuition behind such a test is to check whether the documented correlations between country governance and fraud recovery in our study are a spurious manifestation of some latent unobservable factor that induces an association between governance and operational loss outcomes in general. To do this, we gather data of operational losses in foreign countries from five event types that are not related to fraud: (1) employment practices and workplace safety (losses arising from acts inconsistent with employment, health or safety laws or agreements, from payment of personal injury claims, or from diversity and discrimination events) (2) clients, products, and business practices (losses arising from an unintentional or negligent failure to meet a professional obligation to specific clients (including fiduciary and suitability requirements), or from the nature or design of a product) (3) damage to physical assets (losses arising from loss or damage to physical assets from natural disaster or other events) (4) business disruption and system failures (losses arising from disruption of business or system failures) (5) execution, delivery, and process management (losses from failed transaction processing or process management, from relations with trade counter parties and vendors). We then run estimations similar to Equation (1) in order to determine whether country governance 21

23 is significantly associated with recovery rates of non-fraud related losses. 17 Table 9 presents the results. [Insert Table 9 about here] Consistent with our hypothesis, country governance is not associated with operational loss recovery rates for non-fraud related categories. In both Specifications (1) and (2), the coefficient of CG Index is economically and statistically close to zero. Such results are consistent with an interpretation that our main empirical results positively linking country governance to fraud recovery are not likely to be driven by common latent operational loss factors (unrelated to country governance). 5 Conclusion This study makes an important contribution to the strand of literature portraying a country s legal and governance underpinnings as important factors for its business climate, and economic and financial development. We focus on a particular type of firms, large financial institutions, for which a regulatory framework, the Dodd-Frank Wall Street Reform and Consumer Protection Act, provides us with data to carry out tests of our hypothesis. Using a sample of 1,270 loss events related to fraud in international markets from 13 large financial institutions over the period, we present several pieces of evidence that relate the quality of country governance to fraud loss recovery. First, we show that country governance, on average, is positively associated with fraud recovery. We then show that different dimensions of governance are not equally important 17 One difference with Equation (1) is that in addition to bank holding company period fixed effects, we also include loss category fixed effects. However, loss category fixed effects do not play an important role in the analysis and results are similar if we do not include them. 22

24 for loss outcomes. Control of corruption, rule of law, regulatory quality and government effectiveness all have robust positive effects on fraud recovery. In contrast, political stability, and voice and accountability do not. This is consistent with the interpretation that while fraud recovery is positively related to the capacity of the government to effectively formulate and implement sound policies, political stability and democratic principles of government rule matter less. Finally, we also document how internal bank risk management practices and country governance interact with respect to international fraud recovery. Specifically, we find an amplification of the positive association between country governance and fraud recovery particularly for institutions with weak internal controls. We conclude that country governance characteristics can have a significant impact on loss outcomes related to fraud. In addition to being directly relevant to corporate risk management practices, our findings also have policy and supervisory implications. In particular, operational risk analysts should take into account country governance quality as a factor when analyzing banks risk related to foreign exposure and operations. Increased scrutiny of foreign exposures in countries with poor governance might be warranted given the increased risks associated with operations in such locales. 23

25 References Acemoglu, D., Johnson, S., and Mitton, T. (2009). Determinants of vertical integration: Financial development and contracting costs. The Journal of Finance, 64(3): Agmon, T. and Lessard, D. R. (1977). Investor recognition of corporate international diversification. The Journal of Finance, 32(4):pp Alcala, F. and Ciccone, A. (2004). Trade and productivity. The Quarterly Journal of Economics, 119(2): Allen, L. and Bali, T. G. (2007). Cyclicality in catastrophic and operational risk measurements. Journal of Banking & Finance, 31(4): Amihud, Y., DeLong, G. L., and Saunders, A. (2002). The effects of cross-border bank mergers on bank risk and value. Journal of International Money and Finance, 21(6): Amihud, Y. and Lev, B. (1981). Risk reduction as a managerial motive for conglomerate mergers. The Bell Journal of Economics, 12(2):pp Antras, P., Desai, M. A., and Foley, C. F. (2009). Multinational firms, FDI flows, and imperfect capital markets. The Quarterly Journal of Economics, 124(3): Badinger, H. (2008). Trade policy and productivity. European Economic Review, 52(5): Bartov, E., Bodnar, G. M., and Kaul, A. (1996). Exchange rate variability and the riskiness of U.S. multinational firms: Evidence from the breakdown of the Bretton Woods system. Journal of Financial Economics, 42(1): Beck, T., Lin, C., and Ma, Y. (2014). Why do firms evade taxes? The role of information sharing and financial sector outreach. The Journal of Finance, 69(2): Berden, K., Bergstrand, J. H., and van Etten, E. (2014). Governance and globalisation. The World Economy, 37(3): Berger, A. N., Buch, C. M., DeLong, G., and DeYoung, R. (2004). Exporting financial institutions management via foreign direct investment mergers and acquisitions. Journal of International Money and Finance, 23(3): Banking, Development and Structural Change. Berger, A. N., Klapper, L. F., and Udell, G. F. (2001). The ability of banks to lend to informationally opaque small businesses. Journal of Banking & Finance, 25(12): Brewer, T. L. and Rivoli, P. (1990). Politics and perceived country creditworthiness in international banking. Journal of Money, Credit and Banking, 22(3):pp Bris, A. and Cabolis, C. (2008). The value of investor protection: Firm evidence from cross-border mergers. Review of Financial Studies, 21(2): Buch, C. M. and DeLong, G. (2004). Cross-border bank mergers: What lures the rare animal? Journal of Banking & Finance, 28(9): Burgman, T. A. (1996). An empirical examination of multinational corporate capital structure. Journal of International Business Studies, 27(3):pp