Payoff Complementarities and Financial Fragility: Evidence from Mutual Fund Outflows

1
Payoff Complementarities and Financial Fragility
Evidence from Mutual Fund Outflows

• Qi Chen (Duke)
• Itay Goldstein (Wharton)
• Wei Jiang (Columbia)

2
Payoff Complementarities Theory and Evidence

• Payoffs to take an action increases when others
  do the same.
• Optimizing agents could end up in an inferior
  equilibrium.
• Extreme cases Bank runs (Diamond and Dybvig,
  1983) Currency attacks (Morris and Shin, 1998).
• Key parameters
• Degree of complementarities affects the level of
  fragility.
• Ease of internalization/coordination affects the
  equilibrium.
• Empirical evidence?
• Hard to show (data constraints/missing
  variables).
• This paper is the first attempt.

3
Payoff complementarities in mutual funds
redemption
Day 1
Day 3
Day 4
Day 2
At 359pm, investor i submits redemption
Mutual fund trades to raise the cash or to
restore cash balance.
NAV determined by the closing price at 400pm

• Remaining shareholders bear most of the cost of
  redemption.
• Direct commissions, bid-ask spread, price
  impact.
• Indirect forced deviation from desired
  portfolio liquidity-based trading.
• Tax?
• The costs are higher when the underlying assets
  are less liquid.

4
Extent of the problem

• Edelen (1999) On average, 76 of gross outflows
  lead to forced sell 2.2 lowered return per unit
  of forced trading.
• Calibrate to data from Christofferson, Evans, and
  Musto (2005) With large redemption (95th to
  99th percentile) and illiquid assets, damage
  amounts to 50 bps to more than 100 bps in a
  month.
• Alexander, Cici, and Gibson (2007) Stocks sold
  for liquidity outperform by 1.55 annually.
• Our story applies mostly to the marginal investor
  making a redemption decision, not to the average
  investor.
• Most investors do not trade much (Johnson (2006)
  and Agnew, Pierluigi, and Sunden (2003)) .
• Total runs on mutual funds are very uncommon
  (there have been a few cases since 2006).

5
Remedies of the problem

• By funds
• Attempts to predict flows (usually difficult)
• Cash reserves (costly to performance)
• Restriction on redemption frequency (compromising
  liquidity to investors)
• Emergency rules suspension of redemption
  redemption in kind(seldom used)
• By the market
• Reflow (very recent)
• By SEC
• Redemption fee formalized in 2005.
• Summary These are mitigating measures that do
  not eliminate the problem.

6
Simple setup

• Parameters
• Returns R1 and R2. NAV(t1) R1.
• Proportion of redeemers 0 N lt 1.
• Liquidity need to sell (1 ?) in order to raise
  1.
• Payoff at t 2 R1 R2 1-(1 ?)N/(1-N).
• With inflows I(R1)
• R1 R2 1-(1 ?)max0,(N-I(R1))
• 1-max0,(N-I(R1))

7
Designing the Empirical Tests from Theory

• Two Premises
• Complementarities arise when funds experience
  outflows.
• Complementarities are stronger when funds hold
  more illiquid assets.
• Based on a global-game model (Morris and Shin,
  1998 Goldstein and Pauzner, 2005)
• H1 Conditional on low performance, funds that
  hold illiquid assets will experience more
  outflows.
• i.e., fragility increases in complementarities.
• Sharpen the test (based on Corsetti et al.,
  2004)
• H2 Pattern weakens when fund is held by large
  investors.
• i.e., large investors internalize the externality.

8
Everybody observes a (correlated) private signal
about fundamentals
More complementarities
Low
High
Never run
Always run
Run
Do not run
MaybeDepend on what others do
9
Empirical implications flow-to-performance
relation
Net Flows
Flows w/o complementarities (liquid assets)
Past Performance
Institutional shares (illiquid assets)
Flows w/ complementarities (illiquid assets)
10
Data

• CRSP/Morningstar 4,393 actively-managed equity
  funds from 1995-2005. Main analysis conducted on
  fund-share basis.
• Distinguish between liquid funds and illiquid
  funds.
• Illiquid funds are small-cap mid-cap equity
  funds (domestic or international), or
  single-country funds excluding US, UK, Japan and
  Canada. Altogether 1,227 funds.
• Or, portfolio weighted average of Amihud (2002)
  measure.
• Distinguish between large and small shareholders.
• Institutional vs. retail share class. About 22
  of all fund shares are institutional.
• Minimum initial purchase.

11
Regression Specification
12
Liquidity and OutflowsHypothesis 1
13
The Effect of Large Investors (all fund shares)
Hypothesis 2
  Institutional-oriented funds with Institutional-oriented funds with Institutional-oriented funds with Institutional-oriented funds with Retail-oriented funds with Retail-oriented funds with Retail-oriented funds with Retail-oriented funds with
  Large Investorgt75 Large Investorgt75 Large Investorgt75 Large Investorgt75 Large Investorlt25 Large Investorlt25 Large Investorlt25 Large Investorlt25
Inst Share Class Inst Share Class Min. Purgt250k Min. Purgt250k Inst Share Class Inst Share Class Min. Purgt250k Min. Purgt250k
COEF T-STAT COEF T-STAT COEF T-STAT COEF T-STAT
Alpha1 0.27 1.66 0.43 2.26 0.24 3.36 0.25 3.68
IlliqAlpha1 0.02 0.18 0.06 0.33 0.20 2.91 0.16 2.71

• Illiquid funds see more sensitive flows only when
  they are retail-oriented
• Support our Hypothesis 2 (about
  internalization/coordination).
• Not supported by the missing variable
  hypothesis.
• Can it be supported by the information story?

14
Predictability of fund return (in the absence of
extreme outflows)
Lag Performance Quintiles Alpha1 Alpha4
  Liquid Funds Liquid Funds
Q1 -0.007 -0.005
Q2 -0.003 -0.002
Q3 -0.001 -0.002
Q4 0.000 -0.001
Q5 0.003 0.001
Q5-Q1 0.011 0.006
t-stat 4.032 2.053
  Illiquid Funds Illiquid Funds
Q1 -0.005 -0.003
Q2 -0.001 -0.002
Q3 0.001 -0.001
Q4 0.003 0.000
Q5 0.005 0.004
Q5-Q1 0.012 0.007
t-stat 3.073 1.779
  Difference Difference
Liq(Q5-Q1) - Illiq(Q5-Q1) -0.001 -0.001
t-stat -0.231 -0.285
15
A cleaner (and out-of-sample) test offered by
closed-end funds (1988-2004)
Lag Performance Quintiles Alpha1 Alpha4
  Liquid Funds  
Q1 -0.005 -0.002
Q2 -0.001 0.000
Q3 0.000 -0.001
Q4 0.000 0.000
Q5 0.002 0.002
Q5-Q1 0.007 0.004
t-stat 2.144 1.643
  Illiquid Funds  
Q1 -0.001 -0.002
Q2 -0.002 0.000
Q3 -0.001 -0.005
Q4 -0.002 -0.003
Q5 -0.002 -0.005
Q5-Q1 -0.001 -0.003
t-stat -0.190 -0.570
  Difference  
Liq(Q5-Q1) - Illiq(Q5-Q1) 0.008 0.007
t-stat 1.268 1.295
16
Large Investors Only The Effect of Clientele
  Institutional-oriented funds with Institutional-oriented funds with Institutional-oriented funds with Institutional-oriented funds with Retail-oriented funds with Retail-oriented funds with Retail-oriented funds with Retail-oriented funds with
  Large Investorgt75 Large Investorgt75 Large Investorgt75 Large Investorgt75 Large Investorlt25 Large Investorlt25 Large Investorlt25 Large Investorlt25
Inst Share Class Inst Share Class MinPurgt250k MinPurgt250k Inst Share Class Inst Share Class MinPurgt250k MinPurgt250k
COEF T-STAT COEF T-STAT COEF T-STAT COEF T-STAT
Alpha1 0.42 4.97 0.52 3.21 0.32 2.79 0.16 1.13
IlliqAlpha -0.03 -0.28 -0.22 -1.03 0.34 1.69 0.50 1.94
Large investors (the same clientele) behave
differently in institutional- and retail-oriented
funds.
17
Alternative and out-of-sample tests

• Using refined holding-data liquidity measures on
  the sub-sample of U.S. equity funds, we obtain
  similarly supportive results for both hypotheses.
• Outflows negatively affect funds future
  performance (controlling for serial correlations
  of returns). The effect is 20 bps higher for
  bottom-quartile liquidity funds after a month of
  5 or higher net outflows.
• Funds attempt to accommodate such effects. More
  illiquid funds are more likely to
• hold larger cash reserve,
• implement redemption fee, with more stringent
  conditions (higher fee and/or longer restriction
  period after 2005.

18
Alternative Measures of Liquidity
  Ln(trade_vol) Ln(trade_vol) Ln(trade_vol) Ln(trade_vol) Amihud liquidity Amihud liquidity Amihud liquidity Amihud liquidity
All observations All observations INSTgt75 INSTgt75 All observations All observations INSTgt75 INSTgt75
COEF T-STAT COEF T-STAT COEF T-STAT COEF T-STAT
Alpha 0.24 2.61 0.71 4.89 0.32 3.72 0.70 5.02
Liq_Holding Alpha -0.13 -5.78 -0.02 -0.43 -0.33 -4.6 -0.15 0.05
More illiquid funds experience more outflow after
poor performance Less so in the presence of
large institutional investors. The above results
also hold in the subsample of illiquid funds.
19
Marginal Liquidity
  Amihud liquidity (most liquid quartile) Amihud liquidity (most liquid quartile) Amihud liquidity (most liquid quartile) Amihud liquidity (most liquid quartile)
All observations All observations INSTgt75 INSTgt75
COEF T-STAT COEF T-STAT
Alpha 0.26 2.56 0.68 4.35
Liq_Holding Alpha -0.09 -2.69 0.03 0.48
Funds can sell the most liquid portion of the
portfolio facing outflows. Marginal liquidity
maintains the same prediction.
20
Outflow, Liquidity, and Performance
Outflow negatively affects funds future
performance above and beyond the momentum effect,
only significantly for illiquid funds.
21
Fund Policies

• More illiquid funds are more likely to
• hold larger cash reserve,
• implement redemption fee, with more stringent
  conditions (higher fee and/or longer restriction
  period).

22
Main Contributions

• Mutual fund literature.
• e.g., Brown, Harlow, and Starks (1996), Chevalier
  and Ellison (1997), Sirri and Tufano (1998), and
  Zheng (1999).
• Show that interaction among investors matters for
  mutual fund flows.
• Financial fragility literature.
• Demonstrate that payoff complementarities
  contribute to financial fragility.
• Show the vulnerability of open-end financial
  institutions (Stein, 2005 Cherkes, Sagi, and
  Stanton, 2006).
• Policy implications.
• Testing predictions from models with strategic
  complementarities.
• e.g., Manski (1993), Glaeser, Sacerdote, and
  Scheinkman (2003), and Matvos and Ostrovsky
  (2006).
• Show that global games prove to be a useful tool
  to provide empirical implications that can be
  taken to the data.

23
Liquidity and OutflowsSemi-Parametric Approach