George O. Aragon Arizona State University. Vikram Nanda University of Texas at Dallas. Haibei Zhao Lehigh University. December 06, 2016.

Transcription

1 DO CORRUPTION PERCEPTIONS MATTER FOR INVESTORS? EVIDENCE FROM HEDGE FUNDS George O. Aragon Arizona State University Vikram Nanda University of Texas at Dallas Haibei Zhao Lehigh University December 06, 2016 Abstract We argue that in countries perceived as corrupt, hedge funds will be more willing to distort reported returns since they are less exposed to legal and reputational risks. We hypothesize and show that when these funds report poor performance, investors respond with substantially greater outflows, possibly because they infer the actual performance as being far worse and difficult to camouflage. Consistent with our arguments, we find country-level corruption perceptions are positively related to suspicious patterns in reported returns, restrictions on investor withdrawals, and managerial investment of personal capital. JEL Classification: G15, G23 Keywords: Corruption, hedge funds, flow-performance, governance Aragon is with the Finance Department, W. P. Carey School of Business, Arizona State University, Tempe, AZ , Nanda is with the Finance Department, Naveen Jindal School of Management, University of Texas at Dallas, Richardson, Texas 75080, Zhao is from the Perella Department of Finance, College of Business and Economics, Lehigh University, Bethlehem, PA 18015, We thank Vikas Agarwal, Andrew Karolyi, Bing Liang, Pedro Matos, and seminar participants at Texas Tech University, DePaul University, the first Drexel, Lehigh, Temple and University of Delaware Research Symposium, and the 2016 FMA Asia for valuable comments. This work was conducted in part while Haibei Zhao was at Georgia State University. The Securities and Exchange Commission, as a matter of policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed herein are those of the author and do not necessarily reflect the views of the Commission or of the author s colleagues upon the staff of the Commission. 1

2 1. Introduction The literature on law and finance highlights the critical role of the legal system and investor rights in fostering the development of a country s financial markets. After all, participation in financial markets can occur only if the legal institutions assure investors of a reasonable chance to profit from their investments. In this paper, we examine the functioning of an important investment vehicle hedge funds across countries that differ substantially in the quality of their institutions, as reflected in country-level measures of corruption perceptions. Historically, hedge funds are lightly regulated investment vehicles with minimal registration and disclosure requirements. Hence, it is plausible that investor concerns about hedge funds, in particular, will be amplified when funds are managed in countries with high corruption perceptions. For instance, extant studies argue that suspicious patterns in reported returns by US hedge funds reflect return manipulation and a heightened risk of fraud (Brown et al., 2008; Bollen and Pool, 2008, 2009, 2012; Cici, Kempf, and Puetz, 2013). This risk is likely to be exacerbated when the hedge funds are managed in environments where investor protection is limited and corruption is salient. In our study we develop and test three sets of hypotheses. Our initial hypotheses concern the behavior of investors, particularly their response to fund performance. We argue that investors will expect managers to be more willing to distort information in corrupt environments, where funds may be less concerned about legal jeopardy. In these environments, a hedge fund reporting good performance might be viewed skeptically; while a negative reported performance might be regarded as an attempt to hide a far worse performance that fund managers find difficult to camouflage. Therefore, we expect investors in corrupt 2

3 environments to react asymmetrically to reported performance, with disproportionately greater outflows in response to poor performance news. 1 Second, prior studies argue that hedge fund managers may have an incentive to distort their performance figures, and that such behavior is indicated from suspicious patterns in reported returns. When corruption perceptions are high, the reputational and legal consequences of returns management may be lower to the extent that this behavior is the norm and/or managers face less legal risk. Furthermore, if managers anticipate investors heightened sensitivity to poor reported performance then they may be even more reluctant to report poor outcomes, leading to a self-fulfilling bad equilibrium (Stein, 1989). 2 Therefore, we examine whether the evidence of returns management is stronger among funds managed in countries with high corruption perceptions. 3 Finally, we hypothesize about how corruption perceptions could shape the structure of funds. We argue that if the costs imposed by investor flow-performance response are high, then managers may take steps to curb this behavior by, for example, imposing restrictions on investor withdrawals and/or making a large investment of personal capital in the fund. In the presence of high corruption perceptions, this could lead to fund size being smaller and investors, in effect, requiring a return premium (arguably a corruption premium ) to invest in hedge funds. For our empirical analysis, we use the Corruption Perception Index (CPI) from Transparency International as our main measure of corruption. The Corruption Perception 1 Related studies from the accounting literature also predict an asymmetric earnings response coefficient. For example, Kothari, Shu, and Wysocki (2009) predict a much stronger stock price response to bad earnings disclosures if managers immediately disclose good news but accumulate and withhold bad news up to a threshold. 2 The bad equilibrium we envisage is one in which investors correctly expect funds to distort information and in which managers incentives to distort are made stronger by the investor response they expect. 3 Leuz, Nanda, and Wysocki (2003) find that earnings management is more common among firms located in countries with weak investor protection, which they attribute to managers having greater private benefits of control in such environments. 3

4 Index is widely used in the literature, including cross-country studies of mutual funds. In addition, CPI is strongly correlated with alternative measures of corruption perceptions, such as the indicator by the International Country Risk Guide and the Control of Corruption (Kaufmann, Kraay and Mastruzzi, 2003; Fisman and Gatti, 2002; and Svensson, 2005). 4 Our hedge fund data consist of a large sample of funds obtained by merging four datasets from Lipper TASS (Tremont Advisory Shareholder Services), HFR (Hedge Fund Research), Morningstar and Eureka from January 1994 to December Our empirical analysis shows that hedge funds in countries with high CPI face greater investor redemptions in response to poor performance. The effect of corruption perceptions on the flow-performance relation is economically significant. In our baseline result, a one standard deviation change in CPI leads to around 22% change in the flow-performance sensitivity. Importantly, the greater flow-performance sensitivity associated with higher CPI is only evident for poor performance. This evidence is consistent with investors imputing an even worse performance, resulting in greater outflows. Such voting with their feet will, in effect, serve as a means to monitor and discipline fund managers when corruption perceptions are high. 5 We also find that CPI is positively related to the propensity of funds to fail due to poor performance. This finding provides additional evidence that investors are more responsive to negative performance signals in these countries. We conduct a number of tests to better identify the effect of corruption perception. Previous literature suggest that investors can exhibit more run-like behaviors when the asset investments are more illiquid due to payoff complementarities (Chen, Goldstein and Jiang, 2010; Goldstein, Jiang and Ng, 2015). If the asset investments in more corrupt countries are 4 We include results using alternative measures of corruption perceptions in the Appendix. 5 The flow-performance results parallel the recent theory and empirical evidence on governance through exit in the corporate finance literature (see Admati and Pfleiderer, 2009; Edmans and Manso, 2011; Edmans, Fang and Zur, 2013; and Bharath, Jayraman and Nagar, 2013). 4

5 more illiquid, the run-like behaviors can drive our finding on the concave flow-performance relation. We construct a sample of twin hedge funds that are sold in different countries but share the same pool of asset investments. Within each fund company, we compute the correlations of returns for all possible fund pairs, and label them twin funds if the two funds have similar performance (i.e. return correlation over 99%) but with different assets under management and located in two different countries. Since the twin funds have almost exactly the same returns and pool of assets, this test can help isolate the effect of corruption and operational risk, and alleviate a lot of concerns regarding omitted variables such as fund s asset liquidity and risk exposures. We find that our flow-performance results are robust using our twin funds sample. We also show that our inference using the full sample of hedge funds is unchanged when we control for fund lockup, restriction period and autocorrelation of fund returns as measures of fund liquidity. We further show that the flow-performance results are robust to (i) using ethno-linguistic fractionalization as an instrument for country level corruption (Mauro, 1995); (ii) cross-country differences in hedge fund regulations using the measures from Cumming and Dai (2009); (iii) alternative measures of corruption such as the World Bank s Control of Corruption and the number of corruption convictions per 1 million population on the state level in the U.S. (Butler, Fauver and Mortal, 2009; Dass, Nanda and Xiao, 2016); and (iv) controls for the funds region of asset investment. Next, we study whether corruption perceptions influence the behavior of fund managers. We first construct an index of return quality based on the historical returns of each fund, including spikes in December returns (Agarwal, Daniel, and Naik, 2011), an unusual number of zero and positive returns (Bollen and Pool, 2012), conditional autocorrelation of returns (Bollen and Pool, 2008), and discontinuities in the return distribution (Bollen and Pool, 2009). 5

6 We also consider delays in the reporting of fund returns to commercial databases as a measure of operational risk within the fund (Aragon and Nanda, 2015). Our evidence shows that corruption perceptions are associated with both a greater incidence of suspicious return patterns and longer delays in the disclosure of fund returns. These results also suggest that the stronger flow-performance relation among funds that are managed in high CPI countries could exacerbate the agency problem between managers and fund investors that is, facing greater flow induced incentives, managers might attempt to delay outflows by managing reported returns. We then test hypotheses about what investor flow responsiveness and managerial incentives to distort returns imply for fund structure and contracting, and fund performance. Investor outflows can induce the fire-sale of assets (Edelen, 1999; Coval and Stafford, 2007) and present a drag on fund performance. We would expect fund managers to anticipate a greater flow response to poor performance when corruption perceptions are high, and adjust the fund s contractual features ex-ante to lower the impact of potential outflows. Our results suggest that fund attributes are chosen to discourage outflows. We find that higher levels of CPI are associated with greater restrictions on redemptions, such as longer lockup periods and lower redemption frequencies. These results are consistent with Hombert and Thesmar (2014) where they show that hedge fund managers can reduce the sensitivity of flows to poor performance by imposing redemption restrictions. We also find a positive relation between CPI and managerial investment of personal capital in their funds which, by virtue of providing a signal for the quality of fund assets (Leland and Pyle, 1977), helps to lower the flow-performance sensitivity. Finally, we examine fund performance and find some evidence that investors require higher returns from investing in funds managed in countries with high corruption perceptions. In particular, in these environments, funds operate with a smaller amount of assets under management and the returns to investors are correspondingly higher. 6

7 Our study joins a growing literature on the cross-country studies of fund management. Khorana, Servaes and Tufano (2009) study how country characteristics explain the different in mutual fund fees across countries. Lin, Massa and Zhang (2014) demonstrate that mutual funds in countries with poor governance rely more on semi-public information. Aragon, Liang, and Park (2013) study fund operations for onshore and offshore U.S. hedge funds due to differences in regulation and taxation. In addition, our paper is related to the flow-performance literature and has implications on fund capital formation and investor capital allocation. Previous studies show that the flow-performance relation in mutual funds is convex (Ippolito, 1992). 6 We show that the flow-performance relation in hedge funds is more concave in countries perceived to be more corrupt, suggesting that country characteristics such as corruption have an important impact on investor s capital allocation decisions. Our paper also contributes to the growing literature on the influence of institutional quality and corruption on financial markets. The study of corruption and its potential impact on economic activity has garnered substantial attention in the economics and finance literatures (see, e.g., Rose-Ackerman, 1975; Bardhan, 1997; and Svensson, 2005). Corruption perceptions are usually regarded as a drag on the economy because it tends to distort economic decisions and leads to a misallocation of resources (Shleifer and Vishny, 1993). When the legal system is too weak to define property rights and enforce contracts, we expect to see stronger market forces to ensure contract performance (Klein and Leffler, 1981). Our paper investigates how corruption perceptions influence investors response to fund performance, fund structure, and the incidence of suspicious patterns in the returns reported by fund managers. 2. Related Literature and Hypotheses 6 The non-linear flow-performance relation can be due to search cost (Sirri and Tufano, 1998; Huang, Wei, Yan, 2007), change in fund manager or strategy (Lynch and Musto, 2003), fund brokers (Bergstresser, Chalmers, and Tufano, 2009), or asset liquidity (Chen, Goldstein, and Jiang, 2010). 7

8 Our hypothesis on the flow response to hedge fund performance is based on the notion that, when investors perceive the management environment as being corrupt, they will expect managers to distort information. The reason is that the typical compensation contract of a hedge fund manager consists of a fixed percentage (1-2%) of net assets and a percentage (15-20%) of any positive profits. Therefore managers have an incentive to report good performance to fund investors. Moreover, if the fund is managed in a country that is perceived as being corrupt, managers may be more willing to embellish their performance if they face little risk of legal consequences. Ipso facto, the reporting of positive performance would tend to be discounted by fund investors, while negative reported performance could be perceived as an indication that the actual performance, possibly accumulated over some time, was so poor that it could no longer be concealed thereby, triggering a much greater outflow of funds. In addition, we would also expect more investor outflows when performance is poor to the extent that poor performance reflected a greater risk of expropriation and managerial misbehavior. These economic arguments lead to the following prediction: Prediction 1: Investor flows are more sensitive to poor performance in countries with high corruption perceptions. Our next hypothesis is related to the distortion of performance information provided by hedge fund managers. When corruption perceptions are high, managers may feel less concerned about the legal consequences of distorting their performance figures, to the extent that these environments reflect inefficient and weak judicial systems (Djankov et al., 2003; Svensson, 2005). 7 Within a country, this behavior may be even more pronounced in the hedge fund industry, since hedge funds tend to be subject to less regulation as compared to other institutions such as mutual funds and banks. Prior literature shows that suspicious patterns in 7 This is in line with Becker s (1968) economic theory of crime: people commit crime because the expected gains overweight the costs of getting caught and punished. 8

9 reported returns are indicative of a manager s attempts to distort performance information. For example, Bollen and Pool (2012) find that hedge funds charged with legal or regulatory violations trigger suspicious reporting flags at a higher rate compared to other funds. Cici, Kempf, and Puetz (2013) find evidence that poorly performing hedge funds overvalue their equity positions. Taken together, we expect a greater incidence of suspicious patterns in reported returns among funds that are managed in countries with greater corrupt perceptions. In addition to the legal system, the pay-for-performance relation can also affect a manager s incentive to misreport. The literature has documented that stocks and options may have unintended consequences in corporations and lead to earnings management (Bergstresser and Philippon, 2006), misreporting and misstatement (Burns and Kedia, 2006; Efendi, Srivastava and Swanson, 2007), and accounting fraud (Erickson, Hanlon and Maydew, 2006). In the case of hedge funds, Agarwal, Daniel, and Naik (2011) find that managers with stronger incentive from pay-for-performance sensitivity engage more in returns management. Bollen and Pool (2008) show that the incentive contract motivates managers to underreport fund losses. In our setting, the higher flow-performance sensitivity in corrupt regions creates strong indirect incentives for fund managers through its effect on future fees (Lim, Sensoy, and Weisbach, 2014). Taken together, the weaker legal regime and stronger flow-induced incentives are expected to engender more returns management. Managers can also impact the timing of such performance disclosure. Aragon and Nanda (2015) find evidence of strategic timing of performance disclosure by hedge funds. Managers delay the disclosure of bad performance and wait for a future recovery to limit investor outflows. Such delay of bad news release to investors has also been documented in the accounting literature (Kothari, Shu and Wysocki, 2009). We argue here that such strategic timing can be more prevalent in funds that are managed in countries with high corruption perceptions. Our second prediction follows from the above discussions: 9

10 Prediction 2: Funds that are managed in countries with high corruption perceptions are associated with: 1) a greater incidence of suspicious patterns in reported returns; and 2) longer delays in reporting returns to commercial databases. Our third hypothesis is related to the contracts that are written between managers and fund investors. Edelen (1999) notes that U.S. mutual funds provide generous liquidity terms to their investors, but that investor flows lead to non-discretionary trading costs that negatively impact fund performance. In contrast, hedge fund managers often impose share restrictions that limit investor redemptions while allowing the manager to efficiently manage illiquid assets (Aragon, 2007). In the presence of heightened corruption perceptions among fund investors, managers can anticipate a greater sensitivity of investor flows to poor performance. Therefore, to reduce the potential costs of flow-motivated trading, managers may choose to impose more restrictions on investor redemptions at fund inception. 8 The share restrictions can be costly to the managers though, since investors may demand a higher premium due to these funding liquidity restrictions (Aragon, 2007). In addition, by investing their personal capital in the fund, managers can convey a positive signal to investors about quality of the fund (Leland and Pyle, 1977) and credibly commit to lower rates of diverting value from fund investors (Himmelberg, Hubbard, and Love, 2004). Taken together, in anticipation of a high flow response to poor performance, fund managers can impose redemption restrictions and invest personal capital ex-ante to lower the impact of potential outflows. Prediction 3: Hedge fund managers use more redemption restrictions and invest more personal capital in the fund when corruption perceptions are high. 8 Hombert and Thesmar (2014) show that hedge fund managers can use longer lockup and restriction periods to reduce the impact of outflows following poor performance. 10

11 Finally, we examine the implications of corruption on fund performance. Prior studies find that firms from countries with stronger governance and investor protection have higher equity valuations (La Porta et al., 2002) and lower costs of capital (Hail and Leuz, 2006). Likewise, for hedge funds managed in countries with higher corruption perceptions, investors might demand higher returns as a compensation for the risk of expropriation, i.e., a corruption premium. 9 Prediction 4: Funds have better performance in countries with high corruption perceptions, an indication of a higher cost of capital. 3. Data 3.1. Corruption Perceptions Index (CPI) We use the Corruption Perception Index (CPI) from Transparency International as our main measure of corruption. The Corruption Perception Index is widely used in the literature (e.g. Djankov et al., 2002; Fisman and Miguel, 2007; Barth, et al., 2009; DeBacker, Heim and Tran, 2015), including cross-country studies of mutual funds (Lin, Massa and Zhang, 2014). Our focus on corruption perceptions has the advantage of obtaining a subjective measure of corruption for countries where other empirical data on actual corruption is difficult to obtain in absence of revelations through public investigations or prosecutions (Bardhan, 1997). The CPI combines multiple sources of survey data to capture the perceptions of corruption and investor protection from the perspective of business professionals and country experts. Survey questions that are relevant to our setting include 1) how likely it is for firms to make undocumented extra payments or bribes to obtain favorable judicial decisions (World Economic Forum Executive Opinion Survey); 2) whether there are suspiciously close ties 9 In general, fund performance is related to various factors such as flow-motivated trading (Coval and Stafford, 2007; Chen, Goldstein and Jiang, 2010), fund size ((Berk and Green, 2004; Fung et al., 2008), managerial incentives (Ackermann, McEnally and Ravenscraft, 1999; Liang, 1999; Agarwal, Daniel and Naik, 2009), and redemption restrictions (Aragon, 2007; Agarwal, Daniel and Naik, 2009). 11

12 between politics and business (Political Risk Services International Country Risk Guide); 3) whether there are adequate laws requiring financial disclosure and disallowing conflict of interest (Freedom House Nations in Transit); and 4) the extent to which the government officials use public office for private gains (World Justice Project Rule of Law Index; Transparency International Bribe Payers Survey). The original CPI measure ranges from zero to ten and a lower value suggests a higher level of corruption. Following the prior literature (e.g., DeBacker, Heim and Tran, 2015) we reverse the measure by subtracting the CPI from 11. The resulting transformed variable, coutcrpt, is increasing in the level of corruption perceptions. Table 1 reports the average coutcrpt (across years on our sample) for the matched sample of countries that are common to both the CPI and hedge fund databases. For example, Denmark is perceived as being the least corrupt among countries in our sample (coutcrpt = 1.90), while China is at the other extreme (coutcrpt = 7.00). We note that, in our analysis, we include year fixed effects in all regressions to account for any potential year-to-year changes in the methodology used by Transparency International to construct the CPI. All four hedge fund data sources report the country where the fund s management company is physically located and the domicile country for fund registration. For a given fund, the domicile country can be different from the physical location of fund operations. Empirically, the corruption measures based on domicile country and the physical location of fund operations are highly correlated (e.g., over 95% using fund-quarter observations). We use the corruption measure based on the management office location country of the hedge funds, and our inferences are unchanged using the domicile country. As noted earlier, CPI is highly correlated with alternative measures of country level corruption, such as the corruption indicator by the International Country Risk Guide and the Control of Corruption by Kaufmann, Kraay and Mastruzzi (2003) (also see Svensson, 2005). 12

13 One benefit of using Control of Corruption is that it covers many offshore locations such as the Bermuda, Cayman Islands and the U.S. Virgin Islands, while the Corruption Perception Index from Transparency International does not. We show in the Appendix that our results are robust to the Control of Corruption measure and, hence, the inclusion of some countries that are excluded from our main sample Hedge funds In our analysis, we use a large sample of hedge funds by manually merging four datasets from Lipper TASS (Tremont Advisory Shareholder Services), HFR (Hedge Fund Research), Morningstar and Eureka from January 1994 to December Following the hedge fund literature we use data after 1994 to mitigate survivorship bias, since prior to 1994 we do not have information on defunct funds. We merge fund strategies using the classification scheme suggested by Agarwal, Daniel and Naik (2009) given that the four datasets use different nomenclature to classify fund strategies. The five broad strategies are Directional, Relative Value, Security Selection, Multi-process and Other Traders. Based on these fund strategies, we calculate the style-adjusted returns for each fund in each quarter. We use style-adjusted returns and raw returns (both net-of-fees) as our main performance measures. We consider two sets of variables that are related to the quality of reported returns. First, we construct two composite variables of return quality indexes index1 and index2 based on eleven indicators of suspicious patterns in reported returns. The first variable is the sum of the eleven indicator variable flags and the second is the first principal component of these flags. Suspicious return flags considered by Bollen and Pool (2008, 2009, 2012), Cumming and Dai (2010), and Agarwal, Daniel, and Naik (2011) include the autocorrelation of monthly returns, discontinuity of returns around zero, December return spikes, abnormal number of zero returns, abnormal number of negative returns, abnormal number of unique returns, abnormal length of a string of returns, abnormal number of recurring return blocks, abnormal distribution 13

14 of the second digit of returns, abnormal R-squared from factor models, and conditional autocorrelation of returns. Both index1 and index2 are calculated for the entire sample of hedge funds from 1994 to 2013 on the fund-year level. Table 2 shows that the median fund in our sample triggers two (out of a possible 11) suspicious return flags. Second, we construct a variable (replag) that captures the timeliness with which reports are reported to commercial databases, following the procedures in Aragon and Nanda (2015). Due to data availability, the reporting delay is based on TASS funds only from 2009 to 2013, while the frequency of observation is on the fund-month level. We adjust the return quality indexes and the reporting delay by their corresponding styles. Further details on the construction of index1, index2, and replag are included in Aragon and Nanda (2015). 4. Results 4.1 Flow-Performance Relation Baseline Results Table 3, Panel A reports the results on flow-performance relations in which quarterly net fund flows (flow1) are regressed on past fund performance and other characteristics. We calculate quarterly net flows for a fund each quarter per usual as the difference between the percentage growth rate in AUM minus the net return. We include fund age and its interaction with past performance as control variables. The reason is that managers of younger funds may have greater career concerns and worry about their perceived labor market value (Chevalier and Ellison, 1999). Alternatively, younger funds can be more nimble due to their relatively small size and their managers may have higher marginal utility of wealth (Aggarwal and Jorion, 2010). In addition, investors can observe past performance to learn about the fund manager s skill. Therefore, the age/performance interaction will control for the possibility that investors react less strongly to recent performance when the manager has a longer track record of returns. Finally, to control for the possibility that funds that are managed in U.S. experience greater net 14

15 flows, in specifications (2) and (4) we use an indicator variable that is equal to one if the fund s country of location is US. We also add fund contractual features to explain flows in these specifications. Panel A of Table 3 reports the results. Fund performance (perf) is measured using raw returns in Models (1) and (2) and style-adjusted returns in Models (3) and (4). We observe a positive and significant coefficient on the key variable of interest that is, the interaction between past performance and coutcrpt (crp_perf). In other words, we find a stronger flowperformance relation for countries that are perceived as more corrupt. The results are economically significant. For example, when we use raw return as our performance measure in the second specification, we estimate that a one standard deviation increase in corruption leads to a (=0.0875*1.35) increase in the flow performance sensitivity. In Panel B of Table 3 we allow for a nonlinear flow-performance relation to directly test our main prediction of a greater sensitivity of investor flows to poor performance when corruption perceptions are high. Specifically, we include the interaction variable perf_low and its interaction with the corruption measure. The variable perf_low is equal to the fund s performance measure (perf) times a dummy variable that equals one if the corresponding performance measure such as raw return and style adjusted return is below median, and zero otherwise. In Models (1) and (3) we include size, age, and age_perf as control variables, while in Models (2) and (4) we additionally include the remaining control variables from Panel A. We suppress all control variables in the table to conserve space. We estimate that the coefficient on perf_low is positive ( *1) among funds operating in the least corrupt countries (i.e., corruption score of unity). Therefore, even in the least corrupt countries, we find that the fund flows are more responsive to poor (vs. good) performance. This finding contrasts with Sirri and Tufano s (1998) evidence of a convex 15

16 flow/performance relation for mutual funds. More importantly, consistent with our Prediction 1, the positive and significant coefficient estimate on crp_perf_low indicates that fund flows are even more responsive to bad performance when corruption perceptions are high. In Panel C of Table 3, we report the results from including fund fixed effects in our flow/performance analysis. In these specifications, several control variables (e.g., high watermark and lockup period) are dropped since they do not vary over time. Therefore, we include four specifications using raw return in (1) and (2) and style adjusted return in (3) and (4), and present the linear specification in (1) and (3) and nonlinear specification in (2) and (4). Overall, we continue to observe that corruption perceptions are associated with a greater investor flow response, especially when funds have bad performance Alternative Measures of Corruption Perception We acknowledge that corruption and investor protection are broad concepts. Therefore, we consider additional proxies for investor protection and corruption to show that our results are robust using alternative measures. First, the prior literature finds that countries with Common Law tradition offer better investor protection (La Porta, Lopez-de-Silanes and Shleifer, 1998). We repeat our analysis in Panel D using an indicator variable that is equal to one if the country follows the Civil law tradition, and zero otherwise. The first two panels use the raw return as performance measure, while the last two use the style adjusted return. Consistent with the idea that Common law countries offer better investor protection, the results in Panel D indicate that investor flows in Civil law countries are more responsive to past poor performance. Second, we use an alternative measure of corruption from the World Bank s Control of Corruption index and repeat the flow-performance analysis. Although the CPI and Control of Corruption index use different methodologies, the results from this analysis (reported in Table 16

17 A1 of the Appendix) again show a greater sensitivity of investor flows to poor performance when corruption perceptions are high. This further indicates that our findings are not sensitive to the choice of the corruption measure. Third, we confine our analysis to hedge funds managed in the United States and exploit state-by-state variation in levels of corruption perceptions. Specifically, we follow prior literature (e.g. Butler, Fauver and Mortal (2009), and Dass, Nanda and Xiao (2016) among others) and obtain a state level corruption measure (stcorrupt) by computing the number of corruption-related convictions per one million populations in a given state for a given year. The state level measure not only allows us to explore within country variations by controlling for the country fixed effect, but also provides an objective measure of corruption (versus corruption perceptions) because it is based on actual convictions. The results are reported in Table 4 and show that, among U.S. hedge funds, those managed in a state with more corruption convictions face more monitoring from investors, especially when past performance is poor Asset Liquidity In our baseline results, we find that investors exhibit more run-like behaviors in more corrupt countries. A natural question is whether such behavior is driven by the illiquidity of fund investments. The flow-performance relation can exhibit concavity due to asset illiquidity and fund runs (Chen, Goldstein and Jiang, 2010; Goldstein, Jiang and Ng, 2015). Facing poor fund performance, investors are more likely to redeem their shares when funds invest in illiquid assets, since they are more concerned about the fire sale costs generated by the other investors redemptions. We conduct several sets of tests to show that our finding on the effect of corruption perceptions is not driven by country-level variation in asset liquidity. First, we construct a sample of twin hedge funds that have similar returns but are sold in different countries. Within each fund company, we convert the fund returns into U.S. 17

18 dollars and compute the correlations of returns for all possible fund pairs. The pair is classified as twin fund candidates if the return correlation for the two funds in the pair is over 99%. Then, we check the funds management offices and make sure they are located in two different countries. Finally, we require that the two funds assets under management are sufficiently different, i.e. the difference in funds asset growth from last quarter is over 5%. After obtaining a sample of twin funds, for each twin we take the differences in their quarterly flows (dflow) and the corruption perception measures (dcoutcrpt). We also compute the interactions between dcoutcrpt and the corresponding performance variables. Panel A of Table 5 shows that our main variable of interest, dcrp_retlow is positive and highly significant, which is consistent with our previous finding on the concavity of flow-performance relation. This set of results suggest that our previous finding is not driven by the differences in the hedge funds asset liquidity. 10 Second, we use the return autocorrelation as an alternative measure of fund liquidity (Getmansky, Lo and Makarov, 2004; Jorion and Schwarz, 2014) to see if the results are also robust for the entire sample of hedge funds after controlling for fund liquidity. Specifically, we construct an indicator variable autocorr that is equal to one when the autocorrelation of monthly returns is positive and we reject the null hypothesis at the 10% level (Aragon and Nanda, 2015). We allow autocorr to separately affect both the linear and nonlinear flowperformance relation in Panel B of Table 5, and find that the effect of corruption is robust. Third, as shown above, we observe a stronger sensitivity of flows to poor performance among U.S. funds that are managed in states with a greater incidence of corruption convictions. The fact that our main finding holds up within a single country (i.e., United States) makes it 10 Using a sample of domestic mutual funds, Evans and Fahlenbrach (2012) show that the institutional flows are more sensitive to poor performance compared with the flows from their retail twin funds. Although the separation of institutional and retail share classes is prevalent in mutual funds, such structure is rare in hedge funds. 18

19 less likely that our findings from the cross-country analysis are driven by a tendency for managers to hold less liquid assets in funds that are managed in countries with high corruption perceptions. Finally, in Section we use lockup and restriction periods as proxies for fund liquidity, and allow them to have different impacts on the flow-performance relation by controlling for the interactions between these measures and fund performance. We find that our inference on the flow-performance sensitivity is unchanged after this robustness check Heterogeneity in Hedge Fund Regulations Cumming and Dai (2009) show that country level hedge fund regulations on fund distribution channels, such as wrappers, distribution via investment managers, and fund distribution companies are related to the flow-performance relation. To control for these differences, we repeat our analysis on subsamples of funds based on the presence or absence of these regulations in the countries in which they are managed. The results are reported in Appendix Table A2 and show that, in all subsamples, investors withdraw their capital more intensively facing poor performance in corrupt countries. In other words, the effect of corruption perceptions is over and above the differences in country level hedge fund regulations documented in Cumming and Dai (2009) Region of Asset Investment Our measure of corruption perception is constructed based on the management office location country of the hedge funds. One may wonder if our results on investor flows are driven by the characteristics of hedge funds asset investments, rather than investors perception on the corruption risk of the fund operations. All of our four datasets report the region of investments for our hedge funds. To merge such information across the four datasets, we broadly classify the region of investments into four categories: Global, North America (including US and Canada), Europe and Emerging Markets (including Asia excluding Japan, South Africa, etc.). For each category, we construct a subsample of the funds investing only in 19

20 the corresponding region and repeat our flow-performance analysis. We report the results in Appendix Table A3. The coefficient estimates on crp_perf_low for the Global, North America and Europe subsamples indicate that our inference on the flow-performance relation is not driven by the region of fund investment. The results on crp_perf_low for the Emerging Market are not consistent using our two performance measures, suggesting that we may need additional adjustments for risk factors to better evaluate the performance of funds investing in emerging markets Instrumental Variable Test In this section, we use the Ethnolinguistic Fractionalization (ELF) index as an instrument for corruption and repeat our analysis for the flow-performance relation. Ethnolinguistic Fractionalization is the probability that two randomly selected people from the same country speak different languages. Previous literature suggests that ethnically diverse societies are more fractionalized and likely to have more independent bribe-takers (Shleifer and Vishny, 1993; Easterly and Levine, 1997). Mauro (1995) uses ELF as an instrumental variable for corruption and shows that corruption has a negative impact on economic growth. The strong association between ELF and corruption has been confirmed in a number of subsequent studies such as Easterly and Levine (1997) and Svensson (2000). Meanwhile, there is no economic reason to believe that the ELF index measured in 1985 has an effect on the country s hedge fund flow-performance relation other than through corruption. We use the Ethnolinguistic Fractionalization data compiled by Roeder (2001) for the year 1985 as our instrument for corruption. 11 Panel A, Table A4 shows the first stage regression results of corruption on the ELF index (eth_fact). Since both the corruption perception index and ELF are measured at the country level, and the ELF index does not change over time for a 11 Roeder, G Philip, 2001, Ethnolinguistic Fractionalization (ELF) Indices, 1961 and 1985, retrieved from http//weber.ucsd.edu\~proeder\elf.htm, November 08,

21 given country, we cluster the standard errors at the country level. Consistent with prior literature, we find that ELF is significantly related to the corruption index, even after clustering the standard errors at the country level. Panel B and C show the second stage linear and nonlinear specifications for the flow-performance analysis, respectively. The second stage results use the predicted value of corruption from the first stage, as well as the interactions between the predicted value and the performance measures. Despite a significant shrink in sample size due to the data availability on ELF index, we continue to find a stronger flowperformance relation in more corrupt countries in the linear specification, and a more concave relation in the nonlinear specification Additional Robustness We also conduct a number of additional robustness checks on the relation between flow-performance and corruption. First, we cluster the standard errors at the country instead of the fund level for robustness since the corruption is measured at the country level. Second, our results are robust to alternative measures of fund performance, including the Fung and Hsieh (2004) seven-factor alpha. Third, our results are robust when we control fund risk taking using the standard deviation of the previous one year s fund returns. Finally, our inference is unchanged if we de-mean the fund returns using country-quarter mean return to control for the time-varying country specific characteristics that can affect the fund performance in a given country. These results are not reported to conserve space and are available upon request from authors. 4.2 Manager Behavior Strategic return reporting of fund managers We next investigate whether funds that are managed in countries with greater corruption perceptions display stronger evidence of returns management. In specification (1) of Table 6, we focus on the first return flags index1, which is the sum of all return flags during 21

22 the year. We regress index1 on coutcrpt and control variables, including fund flows and contract features. We find a positive and significant coefficient on coutcrpt, suggesting that suspicious patterns in reported returns are more prevalent when corruption perceptions are high. A one standard deviation change of corruption corresponds to a change of 0.047*1.35 = for the return manipulation index1. 12 Further, we see from specification (2) that our inference is robust to the alternative index measure index2, which is the first principal component of return flags. We note that suspicious patterns in hedge fund returns may reflect factors other than deliberate misreporting, such as asset illiquidity (Getmansky, Lo, and Makarov, 2004; Cassar and Gerakos, 2011) and asymmetries in the manager s incentive fees (Jorion and Schwarz, 2014). For robustness, we exclude two flags from the original eleven return flags: the discontinuity of return distribution around zero and the monthly return autocorrelation. We continue to find that corruption perception is significantly and positively related to suspicious patterns in reported returns. These results are not reported to conserve space and are available upon request. Aragon and Nanda (2015) provide evidence that hedge fund managers strategically delay the reporting of fund performance figures to commercial databases. Therefore, we next examine whether strategic timing behavior is more apparent among funds with greater coutcrpt. We report the disclosure timing results in specification (3) of Table 6. The variable replag is the style-adjusted reporting lag for the subsample of TASS funds. The reporting lag for a fund is the number of days between the date the fund reports its monthly return to the database and the date the monthly return was realized. Overall, we find a positive and significant relation between coutcrpt and replag specifically, a one standard deviation change in our measure of 12 In untabulated results, we find the economic magnitude to be even stronger for countries in the top quartile in terms of corruption, e.g. around for the return manipulation index1. 22

23 corruption perceptions corresponds to a 1.80 (=1.33*1.35) increase in reporting lag. This represents 12.6% of one standard deviation of the reporting lag measure (see Table 2). Taken together, the evidence in Table 6 makes a connection between country-level corruption perceptions and evidence of strategic delay and management of reported returns. This helps explain our findings of a heightened sensitivity of investor flows to poor performance when corruption perceptions are high: Investors view poor performance as being more informative and, in turn, withdraw their capital more aggressively. In the extreme, if poor performance is indeed more informative about the financial situation of the fund, then we might expect poor performance in corrupt environments to also be a better predictor about fund failures. In Table 7, we model the determinants of fund failure in the subsequent year using concurrent returns, corruption and their interactions. The dependent variable is an indicator variable liq that is equal to one if the fund is liquidated in the next year, and zero otherwise. 13 The first two specifications use ordinary least squares while the last two use the Cox hazard model for robustness. We see that the coefficients on fund returns are negative, suggesting that worse performance increases the likelihood of fund liquidation. In addition, the interaction between return and coutcrpt is negative, suggesting that the predictive power of return is even stronger among funds that are managed in countries that are perceived to be more corrupt. This finding provides additional evidence that, in such environments, poor performance figures are indeed more informative about funds real performance and operations Lockup and Restriction Periods 13 We use fund-year instead of fund-quarter observations since major decisions such as fund closures are likely to be based on long-term performance. 23

24 We next investigate whether funds in countries with higher perceived corruption are associated with a greater incidence of redemption restrictions, like lockup and redemption notice periods. In anticipation of a greater sensitivity of flows to performance, managers may choose to impose these restrictions to limit investor outflows. Therefore, we examine whether these contract features are, indeed, useful to reduce the greater flow-performance sensitivity that result when corruption perceptions are high. In Panel A of Table 8, we model the determinants of the use of lockup (lockup) and restriction periods (restrict). Since the choices of such contract features are made at fund inception, we only keep the first fund-year observation for each fund. The results suggest that country level corruption has a strong effect on the choice of lockup and restriction periods: a one standard deviation increase of coutcrpt leads to a collective increase of 1.6 months for the total change of lockup and restriction periods. In Panel B of Table 8 we report the results testing whether redemption restrictions help reduce the flow-performance sensitivity when corruption perceptions are high. This is captured by the triple interaction variable that combines coutcrpt, performance, and the restriction variables. We find a negative and significant coefficient on the triple interaction variable. This suggests that contract variables, such as lockup and restrictions, are effective at reducing the greater flow-performance sensitivity in these environments. Further, we also find in Panel C that the lockup and redemption restrictions help attenuate the sensitivity of investor flows to poor performance when corruption perceptions are high. Our main variables of interest crp_perflow_lock and crp_perflow_restr are both highly significant. Although the share restrictions can help reduce the flow-performance sensitivity, the cost associated with imposing 24