1、 外文翻译 Predicting Hedge Fund Fraud Material Source:www.american.edu Author:Nicolas P.B. Bollen,Veronika K. Pool Widely publicized cases of hedge fund fraud illustrate different ways that unscrupulous managers can harm investors and have revived calls for greater regulatory oversight of the industry.1
2、 The SECs 2008 complaint against two Bear Stearns hedge fund managers, for example, charges that they misrepresented the performance of the funds, their exposure to subprime mortgages, and the level of redemption requests. These misrepresentations violate Section 17(a) of the Securities Act of 1933
3、and Section 10(b) of the Securities Exchange Act of 1934, which prohibit fraudulent practices when offering securities. More brazenly, the $65 billion Bernard Madoff Ponzi scheme, which lured hedge fund capital through a variety of feeder funds, was built around an entirely fabricated trading strate
4、gy. In an effort to identify fraudulent activity before massive losses are incurred, institutional investors perform due diligence and regulators examine funds, often after receiving complaints from investors. Indicators of hedge fund fraud can be categorized into operational flags and performance f
5、lags. Operational flags are triggered when attributes of a funds operations and organizational structure, such as integration of service providers and infrequent audits, enable fraudulent activity. Brown et al. (2008) report strong evidence that funds run by managers with prior legal trouble are sig
6、nificantly more likely to bear these operational risks, and are more likely to fail. Dimmock and Gerken (2009) find that funds that trigger operational flags are more likely to be charged with SEC violations in subsequent years. Both studies rely on data from SEC Form ADV, which is required annually
7、 under the Investment Advisors Act. Many hedge fund managers are currently exempt from registering as investment advisors, however, hence there is an ongoing effort to eliminate exemptions and require all hedge fund managers to disclose the presence of operational flags. Performance flags are trigge
8、red when attributes of a funds reported returns, shareholder activity, and asset growth are consistent with fraudulent activity. For example, Gregoriou and Lhabitant (2008) describe how a Madoff feeder fund, Fairfield Sentry, reported only ten months of losses in a 215 month history. By their nature
9、, these performance flags can be checked quickly on large-scale databases using quantitative algorithms, suggesting that they may offer a low-cost approach to identifying potentially fraudulent funds. If so, then the SEC could use them as a tool for determining which funds to examine, in the same wa
10、y that the IRS screens tax returns for red flags to determine which to audit. We study five categories of performance flags, each motivated by prior research. These include: (1) a discontinuity in the distribution of a hedge funds returns, (2) low correlation with systematic risk factors, (3) serial
11、 correlation, (4) conditional serial correlation, and (5) a family of data quality indicators. In each case we design appropriate statistical tests for significance. The primary research question we ask is whether funds associated with fraudulent activity or some other regulatory violation are more
12、likely to trigger performance flags than other funds. If so, then the flags could be used to predict future trouble, aiding both the investigative efforts of regulatory authorities as well as the due diligence performed by institutional investors. Consultancies such as Riskdata employ quantitative f
13、ilters, but their information content has not been studied by prior research. To determine whether the performance flags are predictors of fraud, we first construct a sample of funds that have been the subject of SEC enforcement actions, which we label “SEC funds.” Next, we assess whether the perfor
14、mance flags are triggered on each hedge fund in a large sample drawn from the TASS and CISDM databases. Then we measure whether SEC funds are more likely to trigger the performance flags than other funds. For almost all of the performance flags, the frequencies with which they are triggered are stat
15、istically indistinguishable across the two groups. Furthermore, the percentage of SEC funds that triggers a flag ranges from about 7% to 40%. These results suggest that actual examinations may be necessary to identify funds with a heightened risk of fraud, which may only be possible with a significa
16、nt investment in resources for regulators. Alternatively, if additional regulation includes mandatory disclosure of operational risks, of the type studied by Brown et al. (2008) and Dimmock and Gerken (2009), then investors and regulators may be able to predict trouble, as these variables have been
17、linked to premature fund closure and subsequent legal trouble. Our last analysis studies whether investors have any ability already to identify funds at greater risk for fraud by comparing the flow-performance relation in all funds to the flow-performance relation in the SEC sample. After controllin
18、g for known determinants of fund flow, we find a convex relation in all funds, which is consistent with prior literature. Superior performance is rewarded with large inflows whereas inferior performance is only slightly penalized by outflows. More importantly, outflows in poor performers are substan
19、tially higher for the SEC sample, providing managers of these funds with a strong incentive to avoid reporting losses. This result indicates that investors have some ability to detect fraud, but only exit these funds when their performance is weak. The rest of the paper is organized as follows. In S
20、ection I we review related literature. In Section II we discuss the set of quantitative screens used in the study and provide implementation details. Section III describes the data. Section IV presents the results. Section V offers concluding remarks. Prior research Two strands of prior research are
21、 closely related to our study and help motivate our analysis. In subsection A we discuss methods of predicting fund failures. In subsection B we explain the incentive for managers to misreport returns and empirical evidence that this practice is not uncommon. A. Fund failures Brown et al. (2001) stu
22、dy the attrition rate of hedge funds and commodity trading advisors in the TASS database, augmented with hand collected data, using data from 1989 to 1998. Hedge funds are dropped from the database at a rate of 15% per year, resulting in a half-life of 30 months. Even more extreme, the CTA attrition
23、 rate is 20% per year, resulting in a half-life of only 24 months. Using both probit and Cox semiparametric hazard rate analyses, Brown et al. find that low returns and high volatility increase the probability of termination. This result suggests that career concerns offset the performance-based inc
24、entives for managers to increase risk. Gregoriou (2002) studies fund failures in the Zurich Capital Markets database from 1990 to 2001 using several statistical techniques including the Kaplan-Meier estimator of the survival function. He reports that hedge funds have a half-life of 5.5 years and fun
25、ds of funds have a half-life of 7.5 years, significantly longer than found by Brown et al. (2001). Failure rates peak at age two or three years, and then decline with fund age. He shows that poor performance accelerates in the six months preceding fund failures, and that larger funds and those with
26、higher incentive fees have lower failure rates than other funds. Grecu et al. (2006) test a variety of parametric models of the hazard function, concluding that the log-logistic distribution fits the data best. This distribution results in an inverted U-shaped hazard function, and empirical estimate
27、s suggest that failure rates increase until age 6. Liang and Park (2008) argue that prior estimates of failure rates are overstated because funds may stop reporting, and so be listed as defunct funds, for reasons other than poor performance. For example, superior funds which have exhausted capacity
28、may close to new investment and so no longer need to report returns to attract new partners. Since we focus on hedge fund managers charged with violation SEC law, we avoid this type of ambiguity. Liang and Park (2008) distinguish funds which stop reporting due to failure from funds which stop report
29、ing for other reasons by examining performance and capital flows leading up to the date on which reporting ceases. The resulting “real” attrition rate is 3.1% per year. Liang and Park (2008) find that downside risk measures dominate volatility as predictors of failure. Consistent with this result, w
30、e show in Section III that defunct funds that are the subject of SEC lawsuits feature dramatic negative skewness. Brown et al. (2008) focus not on performance risk measures, but rather on operational risk measures, which relate to “the risks of failure of the internal operational, control, and accou
31、nting systems; failure of the compliance and internal audit systems; and failure of personnel oversight systems, that is, employee fraud and misconduct.” The authors gather data from a one-time surge in filings of Form ADV by hedge fund managers in response to the SECs short-lived reinterpretation o
32、f the 1940 Investment Advisor Act. Data include revelation of any past legal or regulatory problems on the part of management or related advisors, as well as other items related to conflicts of interest, ownership structure, and use of leverage. The authors first separate funds into “problem funds”
33、and “non-problem funds” as determined by whether the manager admitted to any prior legal or regulatory problem. There are 368 problem funds in their sample of 2,299. Next, the authors report how problem funds differ from non-problem funds along a variety of dimensions related to external conflicts o
34、f interest, internal conflicts, and ownership or capital structure. In all cases a significantly larger percentage of the problem funds trigger these other operational risk flags than the non-problem funds. Since the U.S. Court of Appeals overturned the SEC rule changes requiring many managers to re
35、gister as advisers, and submitting the Form ADV, it is unclear whether operational risk variables will be widely available in the future. For this reason, Brown et al. (2009) use canonical correlation analysis to develop an instrument for operational risk, labeled the .-score, using commonly observa
36、ble data on fund performance, size, age, and fees. The .-score is a linear combination of these variables that is maximally correlated with the Form ADV information. In Brown et al. (2009), the authors show that the .-score tends to increase the failure rate of funds in the context of a Cox hazard r
37、ate analysis, where fund failure is defined as when the fund stops reporting to the database and subsequently is classified as defunct. While Brown et al. (2008) show that past legal trouble is correlated with the presence of conflicts of interest and concentrated ownership, Dimmock and Gerken (2009
38、) find that past legal trouble is correlated with subsequent SEC violations. In the same spirit of these papers, we map observable fund information into the likelihood of fund failure. The primary difference is that we search for patterns in fund returns that indicate purposeful misreporting, rather
39、 than the incidence of past legal trouble, and use the presence of these anomalies as predictors o f subsequent fraudulent activity. B. Misreporting returns Prior hedge fund research motivates our study by showing that managers have an incentive to report the most attractive return series possible,
40、and that managers have discretion over trading strategies and reporting practices to affect the shape of the distribution of reported returns. Managers have an incentive to report attractive return series because their compensation is tightly related to the level of assets under management, and inve
41、stors are performance-sensitive. Goetzmann et al. (2003) find a positive relation between capital flow and lagged return at the annual frequency for the worst-performing quintile and a negative relation for the best-performing quintile. This result can be explained by investors fleeing bad funds and
42、 managers restricting access to good funds. Ding et al. (2008) study the impact of share restrictions, including subscription waiting periods and closure to new investment, on the flow-performance relation in hedge funds. They find that in the absence of restrictions, the relation is convex, whereas
43、 in the presence of restrictions, it is concave, consistent with the results in Goetzmann et al. (2003). Goetzmann et al. (2007) point out that if investors use scalar performance measures, such as the Sharpe ratio or Jensens alpha, to select managers, then managers have an incentive to take actions
44、 to enhance these measures, including manipulation of the performance measures through “information-free” trading activities. The authors show how traditional measures can be distorted through simple manipulation strategies, and devise a performance measure that is not subject to the same type of ma
45、nipulation. Note that the authors do not consider misreporting or other types of misrepresentation they instead focus on trading strategies that are conducted to affect the return distribution of a fund in specific ways to game standard performance measures. In addition to manipulating returns throu
46、gh trading, managers can affect the distribution of reported returns by deciding when to report returns to a database. Ackermann et al. (1999), Brown et al. (1999), and Liang (1999) show how the reporting decision can generate biases in hedge fund databases, including survivorship bias and backfill
47、bias. In contrast to all of these studies, we examine patterns in reported returns that suggest they are purposefully misreported. Liang (2003) examines the quality of self-reported hedge fund return data by comparing returns of funds that report to both the TASS Management Ltd. database and the U.S
48、. Offshore Fund Directory. He finds an average absolute difference in annual returns of 8.83% for funds with incomplete audit information and 4.91% for funds with complete audit information, suggesting that regular audits increase the quality of hedge fund data. Liang (2003) does not study whether e
49、rrors are more pronounced when managers have higher incentives to misreport; nonetheless his results show that there is significant opportunity for returns to be manipulated. Prior research has uncovered some evidence that some managers purposefully misreport returns, presumably in an effort to attract and maintain their investor base. Several of these studies introduce some of the quantitative filters which we discuss in more detail in the next section. Getmansky et al. (2004) show that managers who report moving averages of fund returns generate artificially l
