Failure Risk and the Cross‐Section of Hedge Fund Returns

• Date: 01 December 2016
• Author: Jung‐Min Kim
• Published date: 01 December 2016
• DOI: http://doi.org/10.1111/fima.12101

Failure Risk and the Cross-Section

of Hedge Fund Returns

Jung-Min Kim∗

Modeling a hedge fund’s probability of failure with a dynamic logit regression, I find that the

probability of a fund’s failure has a significantly negative effect on the fund’s future returns.

A quintile portfolio with the highest failure probability underperforms a quintile portfolio with

the lowest failure probability by 5% to 6% per year from 1997 to 2012. The results are robust

to the definition of hedge fund failure and controlling for a large set of risk factors and fund

characteristics. Moreover, the negative effect of failure probability on future fund returns is

stronger for funds with weak sharerestrictions.

Hedge funds are largely unregulated and their operations lack transparency. As a result, both

regulators and hedge fund investors are concerned about hedge funds’ potential failure. Prior

literature (Brown, Goetzmann, and Park, 2001; Baquero, Horst, and Verbeek, 2005; Malkiel and

Saha, 2005; Chan et al., 2006; Grecu, Malkiel, and Saha, 2007; Liang and Park, 2010) focuses

on the preconditions of hedge fund failure. I extend the previous literature by systematically

investigating how the probability of fund failure affects future hedge fund returns.

Hedge funds with a high probability of failure could earn higher future fund returns. These

hedge funds with high failure risk could have greater exposure to macrolevel systematic risks

that, on average, lead to greater future fund returns due to a positive link between exposure

to macrolevel systematic risks and future hedge fund returns documented in Bali, Brown, and

Caglayan (2011, 2012, 2014). Furthermore, fund managers in funds with high failure probability

may have an incentive to take on extreme risk, which could also lead to higher expected fund

returns.

Alternatively, hedge funds with a high probability of failure could earn lower future fund

returns. Hedge fund failures are often characterized by poor performance over time following

investor withdrawals (Liang and Park, 2010). As a hedge fund becomes increasingly concerned

with investors withdrawing their capital due to poor performance, it may be forced to change its

investment policy. In particular, it must increase its positions in liquid assets in order to meet

rising redemption requests. To become more liquid, a fund may be forced to sell off some illiquid

assets at fire sale prices. This, in turn, can be costly and reduce fund performance. Therefore,

I am grateful for helpful comments froman anonymous referee, Marc Lipson (Editor), Jack Bao,Paul Borochin, Byoung-

Hyoun Hwang, Kewei Hou, Bong-Chan Kho, Robert Kimmel, Bing Liang, Roger Loh, Andrew Karolyi, Alain Krapl,

Ronnie Sadka, Melvyn Teo,Sheridan Titman, Russ Wermers, Scott Yonker, and especially Ren´

e Stulz. I am also grateful

to seminar participants at the Bank of Korea,Korea Development Institute, Korea Institute of Finance,Korea University,

Loyola Marymount University,Nanyang Technological University,Sungkyunkwan University, The Ohio State University,

UC Riverside, University at Albany-SUNY, University of Connecticut, University of Dayton, University of Michigan-

Dearborn, University of Seoul, the 2008 FMA Doctoral Consortium, the 2009 FMA Annual Meeting,and the 2014 Korea

International Economic Association Annual Meeting for providing many useful suggestions. Research support from the

Dice Center for Financial Economics and financial support from the Korea-AmericaFinancial Association and Shinhan

Bank are gratefullyacknowledged.

∗Jung-Min Kim is an Assistant Professorat the University of Seoul, Seoul, South Korea.

Financial Management •Winter 2016 •pages 845 – 876

846 Financial Management rWinter 2016

this paper examines empirically how failure probability and future fund returns are linked in the

cross-section of hedge funds.

In this paper, I use the TASS hedge fund database. TASS assigns each defunct hedge fund,

who stops reporting to the database, to one of seven drop reasons: (1) liquidated, (2) no longer

reporting to TASS, (3) TASS has been unable to contact the manager for updated information,

(4) closed to new investment, (5) merged into another entity, (6) dormant, and (7) unknown.

Among the seven drop reasons, “liquidation” seems to be the best candidate for defining hedge

fund failure. However, prior literature suggests that liquidation does not necessarily mean failure

and some of the other defunct funds should be considered as failures (Liang and Park, 2010).

Therefore, I consider two different definitions of hedge fund failure. The first def inition of failure

is liquidation. The second definition follows Liang and Park (2010) (hereafter, LP-failure) where

failure is defined as follows: (1) once listed in a database, but stopped reporting, (2) negative

average rate of return for the last six months, and (3) decreased assets under management (AUM)

for the last 12 months. Interestingly, only 541 funds (of 1,205 liquidated funds or of 1,041 failed

funds by LP-failure) are classified as failed funds by both definitions. Since one def inition of fund

failure does not include the other definition of fund failure, the application of both def initions of

fund failure would be a good compromise for testing robustness.

First, I examine the predictability of hedge fund failure by employing a dynamic logitmodel in

a manner similar to Shumway (2001) and Campbell, Hilscher,and Szilagyi (2008), who apply the

model to forecast corporate failure based on accounting and stock market variables. Motivated by

the prior literature and economic intuition, I propose several covariates for the failure prediction

model (e.g., past performance, fund flows, risks, fund size, and share restrictions). Overall,

poor past performance, fund outflows, greater risk-taking, smaller fund size, and weak share

restrictions increase the probability of hedge fund failure.

As a next step, I examine howa hedge fund’s failure probability affects its future returns. Forthe

performance test, I remove the last two monthly returns of failed funds to mitigate a concern that

a potential negative link betweenfailure probability and future hedge fund retur ns could be driven

by large negativeretur ns of failed funds in failure months. In order to obtain the predicted failure

probability, I estimate the failure prediction model every month using a rolling-window approach

based on only prior information, instead of using the parameter estimates obtained from the full

sample results. This allows me to perform an out-of-sample prediction analysis and implement

a trading strategy. To determine whether a fund’s failure probability is helpful in predicting the

fund’s future returns, I sort hedge funds into quintiles by their predicted failure probability. By

examining the monthly time-series returns of quintile portfolios from 1997 to 2012, I find that a

quintile group with the highest failure probability significantly underperforms a quintile group

with the lowest failure probability by 0.49% per month with a t-stat =5.50 (wheref ailure means

liquidation) or 0.41% per month with t-stat =3.98 (where failure is defined by LP-failure) after

adjusting for exposure to more than ten hedge fund risk factors.

To be robust, I also estimate a panel regression with two-way (fund and month) clustered errors

to determine how a hedge fund’s probability of failure affects its future returns. A panel regression

allows me to control for any effect of a hedge fund’s fund-level variables on its future returns.

In addition, the two-way clustered errors suggested by Petersen (2009) ensure a robust standard

error. The results confirm a significantly negative link between failure probability and future

fund returns even after controlling for more than ten fund-level variables. Further tests suggest

that the negative effect of failure probability on future fund returns is not explained by the effect

of any single variable included in the failure prediction model.

Why do hedge funds perform poorly when their failure risk is high? According to the results

from the failure prediction model, a high probability of hedge fund failure is characterized by

Kim rFailure Risk and the Cross-Section of Hedge Fund Returns 847

fund outflows due to poor performance. To meet increasing redemption requests, the fund is

forced to sell off some assets, potentially at fire sale prices thereby decreasing fund performance.

Thus, if there is a negative link between failure probability and future hedge fund returns, the

negative effect of failure probability on future hedge fund returns should be heightened for funds

with weak share restrictions. Using double sorts, I find that a return spread by failure probability

within the funds with weak share restrictions is three to four times higher than a return spread by

failure probability within funds with strong share restrictions.

Robustness tests suggest that: (1) the negative link betweenfailure probability and future hedge

fund returns is robust to macroeconomic uncertainty beta, (2) the negative link is robust to several

specifications of the failure prediction model, (3) the negative link is robust to all three investment

styles (directional, semi-directional, or nondirectional) and the negative link within the directional

funds appears to be relatively strong, (4) the negative link is found in both onshore and offshore

funds, (5) the negative link is robust to a new sample excluding small hedge funds, and (6) the

negative link is robust to two subsample periods (1997–2005 vs. 2006–2012).

The rest of the paper is organized as follows. Section I describes the sample construction,

proposes fund-level variablesincluded in a failure prediction model, and conducts a t-test on fund-

levelvariables between failure and nonfailure groups across two definitions of hedge fund failures.

Section II models a fund’s probability of failure using a dynamic logit regression and reports the

regression results for the full sample. Section III presents the out-of-sample predictability of the

failure prediction model, documents a robust negative link between failure probability and future

hedge fund returns, and demonstrates that the negative effect of failure probability on future fund

returns is sharply higher when hedge funds have weak share restrictions. Section IV reports the

results from several robustness tests, while Section V provides my conclusions.

I. The Characteristics of Hedge Fund Failures

A. Sample Construction

In this paper, I employ the TASS database that provides monthly returns, monthly assets under

management (AUM), and several fund characteristics at the individual fund level. The sample

period includes data from January 1994 to December 2012.

There are two major data biases documented in the hedge fund literature. First, a survivorship

bias naturally occurs if a sample includes only live funds, whilethe tr ue population includes both

live and defunct (so-called dead) funds. TASS maintains two separate databases: (1) Live and

(2) Graveyard databases. Hedge funds in the Live database are considered to be live funds. Once

a hedge fund stops reporting to the database, the fund is transferred to the Graveyard database

and is considered to be a defunct fund. Since TASS created the Graveyard database in 1994, my

sample period starts from January 1994 in order to be free of survivorship bias. In addition, a

backfill bias can occur as many hedge funds bring successful past fund performance in order to

attract potential investors when they enter the database. To avoid the backfill bias, I follow Fung

and Hsieh (2000) and delete the first 12-month return histories of all funds.1

For further data screening, funds are dropped if they do not report net-of-fee (vs. gross) returns,

do not report monthly (vs. quarterly) returns, or do not report returns and assets under management

in US dollars. I also require each fund to have at least a 24-month consecutive return history

1Although TASS records the date on whicha hedge fund enters the database, Fung and Hsieh (2009) f ind that 1) not all

of the data prior to the entering date are “backfill-biased” and 2) a substantial loss of information occurs if we delete each

fund’sreturns before its entry date.