We find a significant discontinuity in the pooled distribution of reported hedge fund returns: the number of small gains far exceeds the number of small losses. The discontinuity is present in live funds, defunct funds, and funds of all ages, suggesting that it is not caused by database biases. The discontinuity is absent in the three months culminating in an audit, funds that invest in liquid assets, and hedge fund risk factors, suggesting that it is generated neither by the skill of managers to avoid losses nor by nonlinearities in hedge fund asset returns. A remaining explanation is that hedge fund managers avoid reporting losses to attract and retain investors.Hedge funds are currently attracting a great deal of attention from investors, academics, and regulators for a number of reasons, but primarily due to the returns that hedge fund managers report. Investors want to share in the riches, academics want to understand the underlying risk factors, and regulators are concerned about the potential for fraud. Some members of the SEC support additional regulation of hedge funds, and championed an amendment to the Investment Advisors Act to force more hedge fund managers to register.1 Others argue that the low number of hedge fund fraud cases indicates that there is no need for greater oversight. 2 Though the number of fraud cases is modest, violations of the law may be widespread but undetected. In particular, the discretion with which managers voluntarily submit returns to databases may permit purposeful misreporting to attract and retain investors. We conduct a simple test for misreporting that measures discontinuities in the pooled cross-sectional, time series distribution of monthly hedge fund returns. In particular, we examine the histogram of returns to determine whether certain categories, e.g. those just below zero, appear systematically underrepresented. Our analytical framework has been used in prior research linking asymmetric incentives around a fixed hurdle with breakpoints in the empirical distribution of an outcome. Examples include the frequency of corporate earnings just below and just above zero (Burgstahler and Dichev (1997)), the winning percentage of sumo wrestlers in critical bouts (Duggan and Levitt (2002)), and the ability of management to sponsor shareholder resolutions that receive just enough votes for approval (Listokin (2007)). Our test is also related to Abdulali's (2006) bias ratio, which compares the number of positive returns to the number of negative returns within one standard deviation of zero. 4 An unusually high bias ratio is suggestive of manipulated returns, although it is unclear what levels are expected under the null hypothesis of distortion-free returns. In contrast, the null hypothesis for our test is based on the simple assumption that the distribution of returns is smooth. returns. The interpretation in both papers that incentives lead to performance relies on the assumption that some managers are skillful. The classic method of distinguishing luck from ski...