Time-Varying Incentives in the Mutual Fund Industry

1
Time-Varying Incentives in the Mutual Fund
Industry

• Jacques Olivier
• HEC Paris
• Anthony Tay
• Singapore Management University

2
Motivation

• Existing empirical literature on mutual fund
  flows
• Ippolito (1992), Gruber (1996), Chevalier and
  Ellison (1997), Sirri and Tufano (1998), Del
  Guercio and Tkac (2002), Lynch and Musto (2003),
  Barber, Odean and Zheng (2005), Gallaher, Kaniel
  and Starks (2006) and Huang, Wei and Yan (2007)
• Because of fee structure, investor flows shape
  the incentives of mutual fund managers
• Crucial property of flows convex function of
  past performance, which provides incentives for
  strategic risk-shifting
• Chevalier and Ellison (1997), Brown et al. (1996)

3
Preview (1)

• Central result of this paper
• Convexity of the flow-performance relationship
  varies with economic activity
• The stronger is economic activity, the more
  convex is the flow-performance relationship

4
Preview (2)

• 5 issues
• (i) Is the effect economically significant?
• YES
• 1 GDP growth implies (more than) twice as much
  convexity as on average
• -1 GDP growth implies no convexity whatsoever
  (and even some concavity)
• (ii) Is the effect driven by abnormal years?
• NO removing years with deep recessions or strong
  booms leaves the result unchanged

5
Preview (3)

• 5 issues (continued)
• (iii) Through which channel does economic
  activity affect the nature of the
  flow-performance relationship?
• GDP growth, NOT market returns
• Aggregate flows, NOT volatility
• Consistent with consumption smoothing
  disposition effect
• (iv) Does the time variation of the
  flow-performance relationship affect decisions of
  fund managers?
• YES strategic risk-shifting occurs only when GDP
  growth is high
• Effect of GDP growth dominates that of market
  returns

6
Preview (4)

• 5 issues (continued)
• (v) Other reasons why we care about the result
• Rationalizes existing results on mutual fund
  performance over the business cycle
• Methodological aspects
• Time-varying risk premia (more tentative)

7
Data and Methodology (1)

• No-load US domestic equity mutual funds appearing
  in CRSP survival bias free mutual fund database
  between 1980 and 2006
• Exclude multiple classes, index funds, funds of
  funds, funds closed to investors, funds that
  never reached 10M of total net assets
• Flow variablei,t Dollar Flowi,t TNAi,t
  (1ri, t) TNAi, t-1
• Where TNAi,t represent Total Net Assets at the
  end of year t

8
Data and Methodology (2)

• Rank (or relative performance) year-by-year
  ranking of fund managers according to their
  (1-factor) alpha
• Measure between 0 (worse performer) and 1 (best
  performer)
• Following Sirri and Tufano (1998), divide
  performance in three regions
• TOP top quintile (relative performance from 0.8
  to 1)
• MIDDLE middle three quintiles (from 0.2 to 0.8)
• BOTTOM bottom quintile (below 0.2)
• Estimate piecewise linear regression of current
  flows on past performance
• Robustness checks
• Rank managers by their excess returns or by their
  4-factor alphas
• Use 1-factor alphas themselves instead of the
  ranking

9
Data and Methodology (3)

• Standard flow-performance regression (e.g. Sirri
  and Tufano, 1998)
• Where

10
Data and Methodology (4)

• Interpretation of the standard regression
• There is convexity if and only a1 a3 is
  positive and significant
• In other words, if and only if flows react more
  to differences in performance of good performers
  than to differences in performances of lousy
  performers

11
Data and Methodology (5)

• What we test in this paper
• Does the difference a1 a3 vary with economic
  activity?
• Our basic regression

12
Data and Methodology (6)

• Interpretation Business cycle effects measured
  by deviations (in percentage) of real US GDP
  growth from its sample mean
• Year-fixed effects take care of impact of
  business cycle on the intercept
• Slope effects captured by interaction variable

13
Data and Methodology (7)

• Interpretation of coefficient on performance
  impact of performance on flows when US GDP growth
  is equal to its sample mean
• Interpretation of coefficient on interaction
  variable how does a 1 deviation of GDP growth
  change the (total) impact of performance on
  growth

14
Data and Methodology (8)

• Tests of convexity
• Flow-performance relationship is convex on
  average if and only if
• a1 is (significantly) larger than a3
• A 1 increase of GDP growth rate increases
  convexity if and only if
• a4 is (significantly) larger than a6

15
Data and Methodology (9)

• Unbalanced Panel Data
• Year fixed effects though year dummy variables
• Standard errors clustered by funds
• Methodological remark
• Usual methodology used in mutual fund flows
  literature Fama-Mac Beth regressions
• Assumes that slope coefficients in each annual
  regression drawn from the same distribution
• Not valid if systematic time variation at
  business cycle frequency in slope coefficients
• Comparable to point made 10 years ago in the
  asset pricing literature (conditional vs.
  unconditional CAPM)

16
Basic Results (1)
17
Basic Results (2)
18
Basic Results (3)

• Interpretation
• Flow-performance relationship convex on average
• Stronger reaction of flows to good performance
  when economic activity is strong
• Stronger convexity of the flow-performance
  relationship when economic activity is strong
• Order of magnitude a /- 1 change of GDP growth
  (more than) doubles / eliminates the convexity in
  the flow-performance relationship

19
Robustness Checks
20
Economic Interpretation (1)
21
Economic Interpretation (2)

• Candidate 1 flow composition effect
• Step 1 convexity of new inflows and of portfolio
  rebalancing flows
• Investors look for positive alpha funds
• Positive alpha funds concentrated in upper tail
  of the distribution
• Step 2 outflows are a flat or concave function
  of performance
• Concentration of portfolios short-sale
  constraints
• Disposition effect
• Step 3 more outflows when economic activity is
  weak
• Consumption smoothing

22
Economic Interpretation (3)

• Candidate 2 volatility effect
• Step 1 convexity driven by investors looking for
  positive alpha funds
• Step 2 volatility is countercyclical
• Step 3 performance is less informative about
  skill when volatility is high (more noise)

23
Economic Interpretation (4)
24
Implications (1)

• Tournament Hypothesis
• Brown et al. (1996), Chevalier and Ellison
  (1997) convexity of flow-performance
  relationship provides incentives for poor
  mid-year performers to take on more risk
• Empirical evidence on risk-shifting very mixed
  depending on samples
• Kempf et al. (2008) cost of switching jobs imply
  more risk-shifting under good than under bad
  market conditions
• 2 issues
• No direct estimate of cost of switching jobs and
  relative magnitude compared to high-powered
  incentives in the industry
• Could go either way (foregone bonuses)

25
Implications (2)

• Conditional Tournament Hypothesis
• When the flow-performance relationship is convex,
  then poor mid-year performers have incentives to
  increase the risk of their portfolios
• Thus, more risk-shifting when economic activity
  is strong
• If risk-shifting mostly driven by the
  flow-performance relationship then no impact of
  market conditions on risk-shifting once business
  cycle effects are accounted for

26
Implications (3)

• Conditional Tournament Hypothesis (continued)
• Negative coefficient of interaction variable
  poor performers increase their risk even more
  when GDP growth is high
• Year fixed effect and fund clustered standard
  errors

27
Implications (4)
28
Implications (6)

• Conclusion
• Behavior of fund managers is consistent with
  time-series properties of the flow-performance
  relationship
• Reconciles insights of seminal papers in the
  field with conflicting empirical evidence
• Once time-varying nature of incentives are
  accounted for, only mild support for impact of
  employment risk
• Some evidence in favor of market timing by fund
  managers

29
Other Reasons to Care About the Result

• Kosowski (2006) Funds have significantly larger
  alphas during recessions than during booms
• This paper provides a possible rationale for the
  result more distortion of incentives of mutual
  fund managers during booms
• Mechanism supported by Huang et al. (2008)
  risk-shifting destroys value
• Asset pricing literature Non constant discount
  factors
• This paper provides a (very) specific example
  where business cycle variations generate
  endogenously shifts to risk aversion of agents
  (fund managers)