Professional Documents
Culture Documents
10、Do Portfolio Manager Contracts Contract Portfolio Management(2019JF)
10、Do Portfolio Manager Contracts Contract Portfolio Management(2019JF)
5 • OCTOBER 2019
ABSTRACT
Most mutual fund managers have performance-based contracts. Our theory predicts
that mutual fund managers with asymmetric contracts and mid-year performance
close to their announced benchmark increase their portfolio risk in the second part
of the year. As predicted by our theory, performance deviation from the benchmark
decreases risk-shifting only for managers with performance contracts. Deviation from
the benchmark dominates incentives from the flow-performance relation, suggesting
that risk-shifting is motivated more by management contracts than by a tournament
to capture flows.
∗ Jung Hoon Lee is at the Freeman School of Business, Tulane University. Charles Trzcinka is
at the Kelley School of Business, Indiana University. Shyam Venkatesan is at the Ivey Business
School, University of Western Ontario. We thank the Editor (Stefan Nagel), the Associate Editor,
an anonymous referee, Vikas Agarwal, Matt Billet, Hsiu-lang Chen, Susan Christofersen (AFA
Discussant), Martijn Cremers, Ted Fee, Craig Holden, Ryan Israelsen, Ron Kaniel, David Lesmond,
Pedro Matos, Veronica Pool, Christopher Schwarz, Clemens Sialm, Laura Starks (FRA discussant),
Matthew Spiegel, Yuehua Tang, and Jun Yang, for their valuable comments. We also benefited from
comments received during presentations at the 2015 Financial Research Association (FRA), 2015
Northern Finance Association (NFA), 2016 Financial Intermediation Research Society (FIRS),
and 2018 American Finance Association (AFA) conferences, as well as at seminars at Indiana
University, Lehigh University, Tulane University, University of Illinois at Chicago, University
of Massachusetts Boston, University of Missouri, University of Notre Dame, and University of
Western Ontario. The authors do not have any potential conflicts of interest to disclose as identified
in the Journal of Finance’s disclosure policy.
DOI: 10.1111/jofi.12823
2543
2544 The Journal of FinanceR
manager and pays the manager a salary. In a recent paper, Ma, Tang, and
Gómez (2019) find that the vast majority of mutual fund managers have vari-
able compensation contracts based on the fund’s performance relative to a
specified benchmark. Moreover, these contracts are asymmetric: the manager
is not penalized if the fund underperforms the benchmark, giving them an
incentive to take additional risk. With the portfolio manager making day-to-
day investment decisions, this asymmetric contract should be an important
determinant of risk-shifting.
We focus on mid-year risk-shifting because of its prominence in financial
economics. Managers can change portfolio risk in numerous ways, making it
difficult to write contracts that prohibit risk-shifting (i.e., Huang, Sialm, and
Zhang (2011)). Traditionally, studies of mid-year risk-shifting motivated by
mutual fund tournaments focus on the contract between advisors and share-
holders (i.e., Brown, Harlow, and Starks (1996)). The advisors’ contract with
shareholders is typically specified as a percentage of the fund’s assets under
management (AUM) and should be symmetric if there is a performance bonus
(a “fulcrum” fee).1 In sharp contrast, the contract with portfolio managers has
an option-like payoff whereby the value of the fund at the end of the year is
the underlying asset of the option and the stochastic benchmark is the strike
price. This paper is the first to examine the contract between the mutual fund
manager and the advisor as an exchange option as in Margrabe (1978) and to
derive the implications of that contract for mid-year risk-shifting.
Using the exchange option model, we show that the vega of the manager’s
contract, that is, the derivative of the option’s price with respect to the volatil-
ity of the portfolio, reaches its maximum value when the distance of the fund’s
return from the benchmark’s return is smallest. This is the point at which
volatility is most valuable to the manager. With this result, we hypothesize
that risk-shifting by the portfolio manager is inversely related to the distance
of the portfolio’s return from the benchmark’s return. However, the impact of
vega will likely be less important as the risk of being fired increases. Kho-
rana (1996) shows that management turnover is empirically related to poor
performance. In the theoretical dynamic investment allocation model of Car-
penter (2000), portfolio volatility converges to infinity as a managed portfolio
approaches bankruptcy. In such instances of extreme poor performance, the
portfolio manager’s desire to preserve their job dominates the vega of the op-
tion contract. Chen and Pennacchi (2009), who model the manager’s contract
relative to a benchmark, also arrive at a similar conclusion. However, their
model assumes that the manager competes in a tournament for flows. Finally,
Khorana (2001) finds evidence that managers engage in risk-shifting before
being replaced. Given these arguments, we hypothesize that there is a positive
relation between risk-shifting and extremely poor performance.
To test these hypotheses, we collect a sample of 3,265 U.S. equity mutual
funds and match them with their announced benchmarks. We use the daily
1See section 205 (a) (1) of the Investment Advisers Act of 1940. See also Deli (2002), Elton,
Gruber, and Blake (2003), and Golec and Starks (2004).
Do Portfolio Manager Contracts Contract Portfolio Management? 2545
fund and benchmark returns to estimate the extent of risk-shifting. Given that
the outliers (or extremely underperforming funds) are clearly influential, we
run a quantile regression model that is robust to outliers. Our baseline results
show that the distance from the benchmark is significantly inversely related
to the ratio of the standard deviations (or risk-shifting ratio), controlling for
variables used in previous studies. This baseline result also holds when we use
the alternative holdings-based measure of Kempf, Ruenzi, and Thiele (2009)
to estimate intended risk-shifting. To ensure that an unlikely mechanical re-
lation is not driving our results, we follow Christoffersen and Simutin (2017)
and use the ratio of the two periods’ portfolio beta as another measure of risk-
shifting. Our results continue to hold. Our results also hold if we define the
evaluation period as a two-year window, the performance period as 1.5 years,
and the risk-shifting period as 0.5 years. Finally, we further corroborate our
findings by fitting a piecewise linear regression, which estimates a separate
slope coefficient for each region of the benchmark-adjusted excess return dis-
tribution. Our estimates clearly indicate that risk-shifting changes along the
excess return distribution.
Next, we hand-collect portfolio manager compensation data from the State-
ment of Additional Information (SAI) to capture the role of benchmarks in
explaining the heterogeneous risk-shifting decisions. We categorize funds into
three subgroups: funds with a clear compensation benchmark, funds with an
unclear compensation benchmark, and funds without a compensation bench-
mark. These subgroups display strikingly different risk-shifting behaviors.
Funds with a clear compensation benchmark shift their portfolio risk the most,
while no such evidence is found among funds that do not have performance-
based compensation. These results are in line with our prediction that risk-
shifting is driven by management contracts. To claim that managerial contracts
have a causal effect on funds’ risk choices, we match funds with performance-
based contracts (treated) to funds without performance-based contracts (con-
trol) on observable characteristics that, we believe, affect the funds’ assignment
to either of these two groups. We then assess the differences in risk-shifting re-
sponse between the two groups assuming that, once matched, the contract type
is uncorrelated with the unobservables. The results confirm our hypothesis
that, on average, compensation contracts, along with mid-year fund perfor-
mance, have a causal effect on the risk-shifting decision.
We also shed light on other perspectives of the contracting environment and
their effects on risk-shifting decisions. First, a manager’s ownership in the fund
significantly reduces risk-shifting. Second, the larger the active share of a fund
(i.e., Cremers and Petajisto (2009)), the greater the risk-shifting. Third, the
tenure of the manager mitigates risk-shifting arising from poor performance
in the previous year. Importantly, the introduction of these measures does not
change the impact of our key variable, Distance.
Finally, previous papers empirically estimates a convexity in the flow-
performance relation, which creates a tournament whereby managers of poorly
performing funds can increase their chances of winning by increasing their
portfolio volatility (i.e., Brown, Harlow, and Starks (1996), Chevalier and
2546 The Journal of FinanceR
2 For example, Busse (2001) argues that tournament effects disappear when daily rather than
monthly returns are used and when the autocorrelations in these returns are considered. Kempf,
Ruenzi, and Thiele (2009) also argue that overall market conditions affect the direction and extent
of managerial risk-taking behavior. In a related context, Basak, Pavlova, and Shapiro (2007)
investigate a fund manager’s risk-shifting incentives by considering a multitude of convex flow-
performance relations. Since the incentive in their model is determined by an assumed flow-
performance relation, the risk-shifting range over which the manager gambles is quite different
from that of the exchange option model. Basak, Pavlova, and Shapiro (2007), however, conclude
that their findings are “in line with Busse (2001), who argues that underperforming managers
do not seem to manipulate their portfolio standard deviations toward year end.” Finally, Schwarz
(2012) argues that the mean-reversion in volatility mechanically gives rise to a tournament effect.
3 However, in Cuoco and Kaniel (2011), the region in which managers shift risk the most dif-
fers slightly from ours. Furthermore, the risk-shifting behavior of extremely poorly performing
funds is not predicted, as their model does not explicitly consider the unemployment risk faced by
portfolio managers.
Do Portfolio Manager Contracts Contract Portfolio Management? 2547
I. Hypothesis
To summarize the discussion above, the main hypothesis of this paper is that
management contracts can be modeled as an exchange option in which the long
position has the option to exchange one risky asset for another. The manager
has the option of exchanging the return of the benchmark for the return of
the portfolio. On expiration, the manager’s bonus is equal to Max(0, PT − BT ),
where PT and BT are the portfolio’s and benchmark’s returns over the evalu-
ation period, T .4 Using the model of Margrabe (1978) to price this option, we
show in the Appendix that the value of this option increases with volatility
and that the vega is highest when the distance between PT and BT is smallest.
Our hypothesis is that this distance is negatively related to the relative risk
of the portfolio in the second half of the year. However, managers in the far
left tail of the excess return distribution are an exception. In this region, we
predict that job security matters more than the vega of the option contract. As
unemployment risk increases, increasing portfolio risk makes sense. Thus, our
second hypothesis is that there is a positive relation between risk-shifting and
extremely poor performance.
A crucial step in identifying the heterogeneity in the risk-shifting response
arising from explicit incentives is to characterize the sample funds by contract
type. In addition, to claim causality, we perform a matched sample analysis
and test whether funds with performance-based contracts (treated) have a
4 The more convex the compensation, the higher the incentive to shift risk. According to Ma,
Tang, and Gómez (2019), even though the Securities and Exchange Commission (SEC) does not
require funds to disclose the relative weights of bonuses and base salary, half of their sample
funds voluntarily release such information. About 35% of such funds report a bonus-to-salary ratio
greater than 200%; about 70% report a ratio greater than or equal to 100%.
2548 The Journal of FinanceR
with respect to 19 indices and assign the index with the lowest active share as the fund’s bench-
mark. This approach is not suitable for us since the only benchmark relevant to a fund manager’s
compensation is the self-designated prospectus benchmark. Sensoy (2009) documents that some
funds pick a benchmark that does not reflect their true investment style. Despite this misleading
assignment, only the prospectus benchmark matters to fund managers since performance-based
bonuses are determined relative to self-declared benchmarks.
6 Despite the use of the MFLINKS file, some share classes are still not mapped to any identifier.
For these remaining observations, we use the CRSP portfolio number to aggregate the different
share classes.
7 For a detailed description of disclosure regarding portfolio managers of registered management
8The Internet Appendix is available in the online version of this article on the Journal of
Finance website.
2550 The Journal of FinanceR
Table I
Summary of the Data
This table provides summary statistics for our sample of funds from January 2000 to December
2013. ColesIncentive Rate is defined as the difference between the last and first marginal compen-
sation rates divided by the effective marginal compensation rate. The effective fee rate is defined
as the compensation rate paid to the advisor on the basis of the current T N A of the fund reported
in the N-SAR filing. RAR is defined as the ratio of the standard deviation of the fund’s excess
return in the second half of the year to the standard deviation of the fund’s excess return in the
(2),int
holdings σi,t
first half of the year. The intended change in portfolio risk, RARi,t = (1) , is the ratio of the
σi,t
standard deviation of tracking errors of the intended portfolio in the second half of the year to
holdings
the realized standard deviation of tracking errors in the first period. RARi,t is computed using
the mutual fund’s holdings information. See equation (6) for more details.
Mean Median SD
the Brown, Harlow, and Starks (1996) risk adjustment ratio (RAR) as follows:
σ2 (r j,t − b j,t )
RAR j,t = , (1)
σ1 (r j,t − b j,t )
where σ1 (r j,t − b j,t ) and σ2 (r j,t − b j,t ) are the standard deviations of fund j’s
return over the benchmark return for the first six months and the second six
months of the year, respectively. These standard deviations are computed using
daily returns and hence provide a much more reliable estimate of managers’
actions regarding fund volatility.
Our hypothesis is that risk-shifting incentives decline as a fund’s perfor-
mance deviates further from its benchmark. We compute the excess return of
each fund over its respective benchmark as the difference between the com-
pounded daily returns of the fund and its benchmark for the duration of the
first six months. For each year, we calculate
where r j,t,n is the daily return for fund j in year t, b j,t,n is the daily return on
the benchmark associated with fund j, and n is the number of days in the first
six months of year t. After computing Exret, we measure the distance of the
Do Portfolio Manager Contracts Contract Portfolio Management? 2551
fund’s return from its benchmark return as the square of Exret, giving equal
importance to returns above and below the benchmark.
Table I provides information on the distribution of excess returns and the
RAR. The median fund’s first-half return is quite close to its benchmark, as
it earns an excess return of 0.1%. In addition, the median RAR is close to
1, suggesting that there is no difference in the relative volatility of the two
periods. However, the standard deviations of 5.6% and 1.005 for excess returns
and RAR, respectively, show that there is considerable variation among funds.
It is worth noting that this is not the ratio of standard deviations first analyzed
by Brown, Harlow, and Starks (1996). Instead, this is the ratio of tracking
errors relative to the fund’s self-selected benchmark.
B. Multivariate Results
We now examine the risk-shifting behavior of fund managers using a re-
gression approach. We begin by estimating the following pooled ordinary least
square (OLS) model:
RAR j,t = at + c1 Distance j,t + c2 Exret j,t + c3 Controls j,t + e j,t . (3)
The dependent variable, RAR j,t , is the change in fund risk relative to a bench-
mark between the first and second half of year t. The key explanatory variable
in equation (3), Distance, is given as the square of the excess return (Exret) and
captures how far the excess return lies from zero. The additional control vari-
ables are the expense ratio (Exp ratio), the turnover ratio (T urnratio), the log
of total AUM (Log size), and the log of the number of years since fund inception
(Log age). The variable Shareclass is a dummy variable that takes the value of
1 if the fund is a multiple share-class fund and 0 otherwise. These variables are
all evaluated at the beginning of the calendar year. Kempf, Ruenzi, and Thiele
(2009) argue that managerial risk-taking changes as a function of the state of
the economy. To account for this temporal variation, all of the specifications
include time fixed effects. Finally, we expect managers in the far left tail of the
excess return distribution to engage in extreme risk-shifting. Because these
potential outliers might unduly influence the model estimates, we winsorize
the data at the top and bottom 1%.
Column (1) of Table II presents the pooled OLS results with the winsorized
data. The specification includes the key variable of interest, Distance, along
with other control variables. The standard errors are clustered by time and
fund to correct for any correlation in the error terms. The negative coefficient
on Distance lends considerable support to our hypothesis that risk-shifting is
strongest in the region in which the fund’s return is close to the benchmark’s
return.9
9 Note that these results are different from those in the literature on flow-driven tournaments
(i.e., Brown, Harlow, and Starks (1996)). Unlike the conventional test of tournaments in mutual
funds, our model includes a squared term, Distance, that uniquely captures the risk-shifting
incentives driven purely from management contracts. This quadratic relation is motivated by our
2552 The Journal of FinanceR
Table II
Relation between Fund Performance and Risk-Taking
This table shows the relation between the fund’s first-half performance and the extent of subse-
quent risk-shifting. The estimates from a pooled OLS are reported in column (1). In column (2), a
quantile regression is estimated, where the conditional median function, Q0.5 (.|.), is specified as
In columns (1) and (2), the dependent variable is the ratio of the standard deviation of the tracking
σ (r −b )
error for the second half of the year to that for the first part of the year, σ2 (r j,t −b j,t ) . The variable
1 j,t j,t
Exret is the fund’s first-half return in excess of its own self-designated benchmark; Distance is the
square of the fund’s return in excess of its benchmark and measures the extent to which the excess
return deviates from zero; Exp ratio is the expense ratio of the fund at the beginning of the year;
T urn ratio is the turnover ratio of the fund at the beginning of the year; Shareclass is a dummy
variable that takes a value of 1 if the fund has multiple share classes; Log age is the log of the
fund’s age; Log size is the log of the fund’s TNA at the beginning of the year; and Flows is the new
T N A j,t+1 −T N A j,t (1+r j,t+1 )
money into fund j, defined as T N A j,t , during the first half of the year. All of the
specifications have time fixed effects. For OLS, standard errors are clustered by time and fund.
For the quantile regression, bootstrapped standard errors are provided in parentheses below the
point estimates. * , ** , and *** indicate whether the results are statistically different from zero at
the 10%, 5%, and 1% significance levels, respectively.
theoretical prediction that the maximal risk-shifting behavior is observed in the neighborhood of
zero excess return.
Do Portfolio Manager Contracts Contract Portfolio Management? 2553
Q0.5 (RAR j,t |It ) = at + c1 Distance j,t + c2 Exret j,t + γ Controls, (4)
where Q0.5 (.|.) is the conditional median function and It is the information set
available at time t. This is the median of the RAR distribution and thus we
examine the response of the median fund manager. All specifications include
time fixed effects. We use the method described by Koenker (2004) to estimate
fixed effects in a quantile regression for panel data. The bootstrapped standard
errors associated with the estimates are reported in parentheses.10
Column (2) of Table II presents results for the quantile regression at the me-
dian. The coefficients of the quantile regression have a similar interpretation
as the OLS coefficients. They represent the marginal effect of the independent
variable of interest on the dependent variable, holding constant the effect of the
other independent variables, except that they are relevant only for the quantile
for which they are estimated. The coefficient on Distance is negative and sta-
tistically significant, suggesting that for the median manager, the portfolio risk
in the second half of the year will decrease as the portfolio’s return deviates
from the benchmark’s return.11 The marginal effects estimated using quan-
tile regression analysis are smaller than those estimated using OLS. However,
given our data and theoretical predictions, the quantile regressions are more
reliable. We therefore use quantile regressions in all subsequent tables.
Overall, the results in Table II support the exchange option model of manage-
ment compensation over competing theories and highlight the region in which
risk-shifting is most prolific.
C. Contracting Environment
C.1. Identification of Contract
To assess the effect of managerial contracts on mid-year risk-shifting, we
adopt the strategy of segmenting the sample by contract type. The contract
standard errors. In the current and subsequent analyses, our quantile regression results are not
clustered for several reasons. First, the literature on clustered standard errors for quantile regres-
sions is still evolving, especially for multiway clusters. Second, following the recent development
in the literature by Parente and Silva (2016), we test for the existence of clusters in our sample.
The tests do not reject the null hypothesis of no cluster at the fund level and weakly reject the null
hypothesis at the 10% level for time-level clustering. Finally, we are concerned about the small
number of time clusters as our contract data from the SAI and the benchmark data allow for six
or 14 time clusters only. Nevertheless, we estimate the quantile regression with time clustering.
The results are qualitatively similar.
11 We also repeat our main specification by using a peer benchmark return instead of the index
benchmark return. The results are qualitatively similar to those provided in the Table II. An
exchange option model is applicable irrespective of which stochastic benchmark—index or peer—
is used.
2554 The Journal of FinanceR
12In addition, Ma, Tang, and Gómez (2019) find no benchmark information or find that the
benchmark information is vague for a similar proportion of observations.
Do Portfolio Manager Contracts Contract Portfolio Management? 2555
Table III
Contract Diversity
We report results from a quantile regression. The dependent variable, RARi,t , is the ratio of the
standard deviation of the tracking error for the second half of the year to that for the first part
σ (r −b )
of the year, σ2 (r j,t −b j,t ) . The variable Exret is the fund’s first-half return in excess of its own self-
1 j,t j,t
designated benchmark; Distance is the square of the fund’s return in excess of its benchmark and
measures the extent to which the excess return deviates from zero; Exp ratio is the expense ratio of
the fund at the beginning of the year; T urn ratio is the turnover ratio of the fund at the beginning
of the year; Shareclass is a dummy variable that takes a value of 1 if the fund has multiple share
classes; Log age is the log of the fund’s age; Log size is the log of the fund’s TNA at the beginning
T N A j,t+1 −T N A j,t (1+r j,t+1 )
of the year; and Flows is the new money into fund j, defined as T N A j,t , during
the first half of the year. Column (1) includes the sample of funds that do not have performance-
based compensation. Column (2) includes the sample of funds that have some performance-based
compensation, but for which the benchmark information is not clear. Column (3) includes the
sample of funds that have performance-based compensation with clear benchmark information.
Column (4) uses the entire hand-collected sample. I{ per f ormance} indicates whether performance-
based compensation exists. I{ per f ormance} turns on irrespective of whether the benchmark is clear.
All of the specifications have time fixed effects, and bootstrapped standard errors are provided in
parentheses. * , ** , and *** indicate whether the results are statistically different from zero at the
10%, 5%, and 1% significance levels, respectively.
clear” group and 0 otherwise.13 In column (4) of Table III, we report results
from the pooled regression. The specification in column (4) interacts the vari-
ables Distance and I{ per f ormance} , as this captures the incremental risk-shifting
undertaken by managers with performance-based compensation. The coeffi-
cient on this interaction term is negative and statistically significant, which is
consistent with our hypothesis of risk-shifting driven by managerial contracts.
Overall, the fact that we are able to categorize individual funds according to
whether their compensation contracts have clear, fuzzy, or no benchmarks, as
well as the fact that the risk-shifting results line up exactly as we would expect,
further supports our hypothesis.14
Segmenting the sample by contract type enables us to differentiate our hy-
pothesis from other alternatives discussed in Section I. First, our results clearly
help us distinguish intentional risk-shifting from reversion of tracking error
to the mean (i.e., Schwarz (2012)) or any mechanical relation between track-
ing error and fund performance in the first half of the year. If any of these
alternative stories is true, we should observe similar effects across funds with
different contract types. Similarly, if our Distance measure captures only flow-
driven implicit incentives, then the importance of the coefficient on Distance
should again be uniform across contract types. Our evidence, however, runs
contrary to either of these possibilities.15
Importantly, even though the literature about the convex flow-performance
relation seems questionable (i.e., Spiegel and Zhang (2013)), we observe mid-
year risk-shifting in the mutual fund industry. The results from Table III show
that this phenomenon stems from portfolio managers’ compensation contract.
Essentially, when the compensation contract does not offer risk-enhancing in-
centives, we do not see managers gaming funds’ volatility as the year pro-
gresses. Overall, a focus on mutual fund managers and their portfolio con-
tracts explains mid-year risk increases more reliably than using a potentially
unstable flow-performance relation.
13 For the funds in the “performance unclear” sample, the relevant benchmark used for perfor-
mance evaluation is unclear. However, this is unclear only to the econometricians reading the SAI,
not necessarily to the manager making the portfolio decisions.
14 Consistent with Ma, Tang, and Gómez (2019), our extensive reading of the compensation
contracts in the SAI provides reasonable grounds to assume that managers have an asymmetric
option-like contract, with the benchmark being the strike price. Multiple kink points in the contract
are a possibility. This would mean that managers have an incentive to shift risk somewhere
other than at the benchmark. However, if this were true, this would work against us finding any
evidence of risk-shifting at the prespecified benchmark. Nevertheless, we find evidence supporting
our hypothesis.
15 Our results hold irrespective of whether the fund is subadvised. When funds are subadvised,
we collect data on managerial compensation at the fund subadvisor level. In that sample, we find
that risk-shifting exists in both types of funds (advised and subadvised). Additionally, we find that
risk-shifting is accentuated in subadvised funds, especially when there are multiple subadvisors.
Do Portfolio Manager Contracts Contract Portfolio Management? 2557
D. Variation in Incentives
Our key claim is that the manager’s incentive to risk-shift is not uniform, but
rather, it changes along the benchmark-adjusted excess return distribution. In
particular, equation (A.8) in the Appendix shows that the maximum vega oc-
curs near the benchmark (slightly above it), depending on the level of volatility.
Our results thus far are consistent with the prediction of an exchange option
model for management contracts and contrast sharply with theories that pre-
dict a monotonic (Chen and Pennacchi (2009)) or U -shaped (Hu et al. (2011))
16 Huang, Sialm, and Zhang (2011) show that, on average, funds that risk-shift have significantly
lower ex post performance. Our back-of-the-envelope calculation (available upon request) shows
that the economic consequence of risk-shifting is an average annual loss of $24.99 billion. Despite
this distortion in risk incentives, including an incentive fee in the compensation contract can still
be optimal since it motivates the manager to expend increased effort. For instance, Li and Tiwari
(2009) show that due to the feedback effect of risk incentives on effort incentives, contracts in
which the fee is linearly related to fund returns lead to an underinvestment of effort.
2558 The Journal of FinanceR
Table IV
Matched Sample and Causal Inference
We report the matched sample quantile regression results. Funds with performance-based compen-
sation (treated sample) are matched to funds without performance-based compensation (control
sample) on a variety of dimensions. We match the funds in the treated and control samples based
on size of the fund, size of the fund family, age of the fund, expense ratio, turnover ratio, fund flows,
previous-year fund return, and previous-year annual return standard deviation. In addition, we
require that the treated fund and the matched control fund be in the same year and have the same
fund style. Figure IA.1 in the Internet Appendix displays the balance of the sample postmatch-
ing. In the regression, the dependent variable, RARi,t , is the ratio of the standard deviation of
σ2 (r j,t −b j,t )
the tracking error for the second half of the year to that for the first part of the year, σ1 (r j,t −b j,t )
.
The variable Exret is the fund’s first-half return in excess of its own self-designated benchmark;
Distance is the square of the fund’s return in excess of its benchmark and measures the extent
to which the excess return deviates from zero; Exp ratio is the expense ratio of the fund at the
beginning of the year; T urn ratio is the turnover ratio of the fund at the beginning of the year;
Shareclass is a dummy variable that takes a value of 1 if the fund has multiple share classes; Log
age is the log of the fund’s age; Log size is the log of the fund’s TNA at the beginning of the year;
T N A j,t+1 −T N A j,t (1+r j,t+1 )
Flows is the new money into fund j, defined as T N A j,t , during the first half of
the year; and I{ per f ormance} indicates whether the manager has performance-based compensation.
All of the specifications have time fixed effects, and bootstrapped standard errors are provided in
parentheses. * , ** , and *** indicate whether the results are statistically different from zero at the
10%, 5%, and 1% significance levels, respectively.
RARi,t RARi,t
(1) (2)
relation. However, due to employment risk, an exception may occur in the case
of extremely negative returns. Carpenter (2000), Chen and Pennacchi (2009),
and Hu et al. (2011) predict that the worst-performing funds will increase their
portfolio risk the most.
To provide further evidence on this relation, we estimate a piecewise linear
regression, as this approach allows us to separately calculate the risk-shifting
response for each performance region of interest. We create three performance
regions for our analysis: “region1,” “region2,” and “region3.”17 Funds whose
mid-year excess returns are more than two standard deviations below 0 (r j,t <
−2σ ) are in region1; funds whose mid-year excess returns are between two
standard deviations below 0 and 0.01 standard deviation (−2σ ≤ r j,t ≤ 0.01σ )
are in region2; and funds whose mid-year excess returns are greater than 0.01
standard deviations (r j,t > 0.01σ ) are in region3.18 These standard deviations
are based on the annual excess return distribution, and hence the threshold
is different for each year. We run the following specification for the piecewise
regression:
RAR j,t = at + c1 Exret j,t + c2 Midper f j,t + c3 Highper f j,t + c4 Control j,t + e j,t . (5)
The variables Midper f and Highper f are given by I{−2σ ≤Exret} × (Exret −
(−2σ )) and I{0.01σ <Exret} × (Exret − (0.01σ )). The variable Midper f takes the
value of 0 if the excess return of the fund is below the threshold of −2σ ; other-
wise, it is the difference between the excess return (Exret) and −2σ . Similarly,
Highper f takes the value of 0 if the excess return of the fund is below the
threshold of 0.01σ ; otherwise, it is the difference between the excess return
(Exret) and 0.01σ . The coefficient c1 is the slope for region1. The slope for re-
gion2 is captured by c1 + c2 and the slope for region3 is captured by c1 + c2 + c3 .
The appropriate standard errors can be calculated using the standard delta
method. The specification we run includes time fixed effects, and the standard
errors are clustered by both time and fund. Column (1) of Table V reports
results for the piecewise regression. The slope coefficients are statistically sig-
nificant in the second and the third regions of our empirical setup. Despite
the lack of statistical significance in the first region, the coefficient patterns
clearly show that poorly performing funds increase portfolio risk the most. We
continue to observe an increase in risk-shifting as we move closer to the mid-
dle of the excess return distribution, and then a decrease in risk-shifting as
we move to funds that outperform their benchmarks. The standard error for
region1 is high because there are only 600 observations in this region. Overall,
the results clearly reflect variations in risk-shifting across the different re-
gions of fund performance. Our results support Cuoco and Kaniel (2011), who
show analytically how risk-shifting is influenced by performance relative to the
17 Based on our predictions in Section I, the three regions are not symmetrically defined.
18 We truncate the bottom 0.1% of our sample. We use the excess return distribution to eliminate
funds, as we consider them to be observational anomalies. Returns for these omitted funds are at
least 3,000 basis points lower than their benchmark returns. Using −2.25σ and −1.75σ thresholds
yields qualitatively similar results to those presented here.
2560 The Journal of FinanceR
Table V
Variation in Incentives
This table reports evidence on variation in the relation between the fund’s first-half performance
and subsequent risk-shifting. The dependent variable is the ratio of the standard deviation of
σ (r −b )
the tracking error for the second half of the year to that for the first part of the year, σ2 (r j,t −b j,t ) .
1 j,t j,t
The variable Exret is the fund’s first-half return in excess of its own self-designated benchmark;
Distance is the square of the fund’s return in excess of its benchmark and measures the extent
to which the excess return deviates from zero; Exp ratio is the expense ratio of the fund at the
beginning of the year; T urn ratio is the turnover ratio of the fund at the beginning of the year;
Shareclass is a dummy variable that takes a value of 1 if the fund has multiple share classes; Log
age is the log of the fund’s age; Log size is the log of the fund’s TNA at the beginning of the year;
T N A j,t+1 −T N A j,t (1+r j,t+1 )
and Flows is the new money into fund j, defined as T N A j,t , during the first half
of the year. In column (1), the following piecewise linear regression is estimated:
RAR j,t = at + c1 Exret j,t + c2 Midper f j,t + c3 Highper f j,t + c4 Controls j,t + e j,t .
Midper f and Highper f are defined as I{Exret≥(−2σ )} × (Exret − (−2σ )) and I{Exret>(0.01σ )} × (Exret −
(0.01σ )), respectively. The slope for Region2 ((−2σ ) ≤ Exret ≤ (0.01σ )) and Region3 (Exret >
(0.01σ )) are c1 + c2 and c1 + c2 + c3 , respectively. In the table, we report the slope of each region. In
column (2), we estimate a quantile regression at the 95th percentile of the RAR distribution. All of
the specifications have time fixed effects. For piecewise regression in column (1), standard errors
are clustered by time and fund. For the quantile regression in column (2), bootstrapped standard
errors are provided in parentheses below the point estimates. We use the standard delta method
to compute the standard errors for Region2 and Region3. * , ** , and *** indicate whether the results
are statistically different from zero at the 10%, 5%, and 1% significance levels, respectively.
Distance −0.120
(0.449)
Exret –8.166 −0.407***
(8.562) (0.118)
Region2 0.715**
(0.341)
Region3 −0.391*
(0.236)
Exp ratio −1.367 −0.078
(1.395) (0.655)
Turn ratio −0.006 0.026***
(0.008) (0.006)
Shareclass −0.002 0.013
(0.011) (0.010)
Log size −0.002 −0.009***
(0.003) (0.003)
Log age −0.010 0.002
(0.015) (0.007)
Flows −0.001* −0.001
(0.001) (0.001)
Observations 27,113 27,141
R2 0.14 0.56
Do Portfolio Manager Contracts Contract Portfolio Management? 2561
benchmark portfolio. Our results are also consistent with their prediction that
the decline in the incentive to increase tracking error is nonsymmetric with
respect to over- and underperformance.
Next, we perform a quantile regression at the 95th percentile instead of the
median. This is the 95th percentile of the RAR distribution (i.e., funds that
risk-shift the most). The quantile regression at the 95th percentile comple-
ments the piecewise regression above since we are not able to reliably examine
the risk-shifting response in region1 due to the small number of observations.
Column (2) of Table V shows that for the extreme risk-shifters, Distance no
longer matters. Instead, the most significant explanatory variable is excess re-
turn (Exret). The highly negative coefficient implies that extreme risk-shifters
increase their portfolio risk further in response to poor performance despite the
already severe deviation from their performance benchmark. We also perform
a statistical test to check for differences in the coefficients across the quantile
regressions at the 50th and 95th percentiles. This test confirms that excess re-
turn is more important in explaining risk-shifting at the 95th percentile than
the 50th . These results support the general arguments of Kempf, Ruenzi, and
Thiele (2009), who posit that risk-shifting for compensation is traded off with
employment risk.
benchmark is critically important to management. (See Harvard Case #9-211-025 “AQR’s Momen-
tum Funds.”) The Form N-1A filed on January 29, 2015 states that the compensation of portfolio
managers who are not principals is based on a fixed salary and a discretionary bonus. The bonus is
“not based on any specific fund’s or strategy’s performance but is affected by the overall performance
of the firm.”
2562 The Journal of FinanceR
20 Cremers and Petajisto (2009) argue that tracking error and active share are distinct active
management measures in that one can choose tracking error as a proxy for factor bets and active
share for stock selection. We use active share as a proxy for the degree of freedom that managers
have in their investment decisions. We thank Veronica Pool for graciously sharing the ownership
data and Martijn Cremers for the updated active share data.
Do Portfolio Manager Contracts Contract Portfolio Management? 2563
Table VI
Managerial Ownership, Active Share, and Tenure
A quantile regression is estimated where the conditional median function, Q0.5 (.|.), is specified as
RARi,t is the ratio of the standard deviation of the tracking error for the second half of the year
σ2 (r j,t −b j,t ) holdings
to that for the first part of the year, σ1 (r j,t −b j,t )
. RARi,t is the intended change in portfolio risk
holdings
computed using holdings of the fund. RARi,t is the ratio of the standard deviation of tracking
error of the intended portfolio in the second half of the year to the realized standard deviation of
tracking error for the first period. See equation (6) for more details. The variable Exret is the fund’s
first-half return in excess of its own self-designated benchmark; Distance is the square of the fund’s
return in excess of its benchmark and measures the extent to which the excess return deviates
from zero; Exp ratio is the expense ratio of the fund at the beginning of the year; T urn ratio is the
turnover ratio of the fund at the beginning of the year; Shareclass is a dummy variable that takes
a value of 1 if the fund has multiple share classes; Log age is the log of the fund’s age; Log size is
the log of the fund’s TNA at the beginning of the year; Flows is the new money into fund j, defined
T N A j,t+1 −T N A j,t (1+r j,t+1 )
as T N A j,t , during the first half of the year; Max ownership is the maximum of the
dollar amount (in millions) invested by all managers; High ownership dummy is a dummy variable
that takes the value of 1 if Max ownership is greater than 1 million; Mediumownership dummy is
a dummy variable that takes the value of 1 if Max ownership is greater than 0 but is less than
1 million; LagExret is the previous-year return of the fund in excess of its own self-designated
benchmark; T enure is the demeaned value of the log of the number of years of service of the
manager in the fund; and LagExret ∗ T enure is the interaction between LagExret and T enure.
All of the specifications have time fixed effects, and bootstrapped standard errors are provided in
parentheses. * , ** , and *** indicate whether the results are statistically different from zero at the
10%, 5%, and 1% significance levels, respectively.
holdings
RARi,t RARi,t RARi,t RARi,t RARi,t RARi,t
(1) (2) (3) (4) (5) (6)
Table VI—Continued
holdings
RARi,t RARi,t RARi,t RARi,t RARi,t RARi,t
(1) (2) (3) (4) (5) (6)
Table VII
Placebo Test
This table summarizes results from a placebo test via a bootstrapping exercise. The bootstrapping
exercise randomly assigns each fund to one of 57 benchmarks, and a total of 500 different random-
ization trials are performed. For each of the 500 random assignments, quantile regression at the
median and pooled OLS regression are performed. These regression specifications are the same as
those used in columns (1) and (2) of Table II. We then provide the 5th and 95th percentiles of the
point estimates associated with Distance from the 500 random benchmark assignments exercise.
We also provide the coefficient estimate of Distance from our baseline quantile and pooled OLS
regressions.
not hold for randomly selected benchmarks, again suggesting that it is not a
product of a “sorting bias” (i.e., Schwarz (2012)) or mechanical relation. Portfo-
lio managers clearly respond more to their own benchmark than to a randomly
selected benchmark.
21 Elton, Gruber, and Blake (2003) make a similar distinction by stating that “our definition of
incentive-fee funds does not include those funds where investors pay a fixed fee (to advisors) but
the portfolio manager has an incentive contract.”
2566 The Journal of FinanceR
H. Robustness
In this section, we explore the robustness of our results. We test whether our
findings hold using alternative risk-shifting measures, are stable over time, are
driven by sorting bias, or are explained by any known managerial behaviors.
22 From option theory perspective, increasing risk has greatest value when the current price
of the asset is near the option’s strike price. Therefore, in the flat section of the convex flow-
performance curve, the incentive to increase risk should be quite low. We thank an anonymous
referee for pointing this out.
23 Regarding the advisory fee, different marginal rates of compensation apply to different levels
of a fund’s AUM. Typically, the marginal advisory compensation rate is a nonincreasing function of
the fund’s AUM. Therefore, the marginal rate of AUM fee charged by the advisor on an incremental
dollar decreases for concave contracts.
24 Table I shows that the average value of ColesIncentive Rate is −0.103. This is very close to
portfolio returns based on the actual portfolio weights in the second half of the
year and stock returns and benchmark returns from the first half of the year.
(2),int
The term σ j,t is the standard deviation of this daily time series.25 Finally,
we calculate the intended risk ratio by taking the ratio of the intended risk in
the second half of the year to the realized risk in the first half of the year:
(2),int
holdings σi,t
RARi,t = (1)
. (6)
σi,t
Note that using a ratio may inflate the magnitude of risk-shifting in a non-
trivial way if the denominator becomes small. To circumvent this issue, we next
use the difference between the numerator and the denominator. As columns
(1) and (2) of Table VIII show, the coefficient on Distance is negative and sta-
tistically significant regardless of how we measure risk-shifting.26
Third, one might be concerned that, given how the variables are defined, RAR
may be mechanically related to Distance. To mitigate this concern, we use the
ratio of the portfolio beta of the two periods to capture changes in portfolio risk,
as suggested by Christoffersen and Simutin (2017). Managers wanting to beat
their benchmark will do so by increasing their exposure to high-beta stocks.
Christoffersen and Simutin (2017) show that funds with a higher percentage of
defined contribution assets invest in high-beta stocks to beat their benchmark.
We use daily return data to compute beta. This estimation is repeated for each
of the two six-month periods separately. Beta is the slope coefficient from the
regression of the fund’s excess return on the benchmark’s excess return. Our
measure of risk-shifting is the beta of the second period divided by that of the
first period. Column (3) of Table VIII reports the relevant results. Given the
magnitude and sign of the coefficient on Distance, we are fairly confident that
our results are not driven by a mechanical relation.27
Fourth, our main hypothesis assumes that the portfolio manager is evaluated
on an annual calendar basis and that the risk of a portfolio cannot be shifted
over a very short period of time. Ma, Tang, and Gómez (2019) report that most
funds have multiple evaluation windows simultaneously, ranging from one-
quarter to 10 years, with a median minimum evaluation period of one year.
Our investigation of the SAI reveals that for more than 90% of compensation
contracts, a significant part of the bonus is based on an annual evaluation.
However, fund manager compensation can be influenced by performance over
a horizon longer than a year. Hodder and Jackwerth (2007) and Aragon and
Nanda (2012) predict that a relatively long investment horizon discourages
excessive risk-taking by fund managers. To address this issue, we divide the
25 Kempf, Ruenzi, and Thiele (2009) use weekly returns rather than daily returns. We believe
daily returns provide a better measure of standard deviation and are more consistent with our
measure of RAR, which is computed with daily returns.
26 The sample size for the specification using the holdings-based measure of risk-shifting is
Table VIII
Robustness: Fund Performance and Risk-Taking
This table shows the relation between the fund’s first-half performance and the extent of subse-
quent risk-shifting. A quantile regression is estimated, where the conditional median function,
Q0.5 (.|.), is specified as
In column (1), the dependent variable is the difference between the standard deviation of tracking
error for the second half of the year and that for the first part of the year, σ2 (r j,t − b j,t ) − σ1 (r j,t −
b j,t ). In column (2), the dependent variable is the intended change in portfolio risk computed using
(2),int
holdings σi,t
holdings of the fund. The intended change in portfolio risk, RARi,t = (1) , is the ratio of the
σi,t
standard deviation of tracking error of the intended portfolio in the second half of the year to the
realized standard deviation of tracking error for the first half of the year. See equation (6) for more
details. In column (3), the dependent variable is the ratio of the fund-level beta from the second
half of the year to that from the first part of the year. The fund beta is estimated by regressing
the excess fund return on the excess benchmark return. In column (4), we divide the two-year
horizon into an evaluation period and a response period by using a break point of 1.5 years to
investigate the case of a multi-year evaluation period. The variable Exret is the fund’s first-half
return in excess of its own self-designated benchmark; Distance is the square of the fund’s return
in excess of its benchmark and measures the extent to which the excess return deviates from zero;
Exp ratio is the expense ratio of the fund at the beginning of the year; T urn ratio is the turnover
ratio of the fund at the beginning of the year; Shareclass is a dummy variable that takes a value
of 1 if the fund has multiple share classes; Log age is the log of the fund’s age; Log size is the log
of the fund’s TNA at the beginning of the year; and Flows is the new money into fund j, defined
T N A j,t+1 −T N A j,t (1+r j,t+1 )
as T N A j,t , during the first half of the year. All of the specifications have time fixed
effects. Bootstrapped standard errors for quantile regressions are provided in parentheses below
the point estimates. * , ** , and *** indicate whether the results are statistically different from zero
at the 10%, 5%, and 1% significance levels, respectively.
holdings
RARi,t Difference β : Ratio Multi-year
(1) (2) (3) (4)
two-year horizon into an evaluation period and a response period using a break
point of 1.5 years. At the end of the first half of each year, we compute the
historical 18-month benchmark-adjusted return. We use this historical return
as Exret and the square of it as Distance. Column (4) of Table VIII reports the
results of the multi-year case. Even for the multi-year evaluation, we continue
to find evidence of risk-shifting. The coefficient is negative and statistically
significant. However, it is one-third the size of the coefficient in column (2)
of Table II. The smaller coefficient supports the predictions of Hodder and
Jackwerth (2007) and Aragon and Nanda (2012).
the findings of Kempf, Ruenzi, and Thiele (2009), we find that management
contract-driven risk-shifting is not time-varying.
from the benchmark return (or higher Distance), managers should decrease
the portfolio weights in high-risk securities. Table IA.V in the Internet Ap-
pendix presents the OLS and quantile regression results. OLS coefficients are
computed using Fama and MacBeth (1973), and standard errors are computed
using the Newey-West method with three lags. We also winsorize the data at
the top 1% and 99% to ensure that extreme risk-shifters do not influence our
results. We find that the results in Table IA.V in the Internet Appendix are con-
sistent with our priors and that funds that are further from their benchmarks
decrease their weights in high-standard-deviation stocks. Overall, Table IA.V
of the Internet Appendix provides additional support for our earlier results and
helps us rule out mean-reversion in volatility as being the main driver of our
results.
28 Note that we do not include Distance as an explanatory variable here because, for Subsam-
ple1 and Subsample2, the variable Exret uniquely captures the distance of the fund return from
the benchmark.
2572 The Journal of FinanceR
Table IX
Robustness: Different Independent and Dependent Variables
Panel A reports results of a quantile regression at the median given in equation (4). Column (1)
considers only the subsample of funds with zero or positive excess returns. Column (2) considers
only the subsample of funds with negative excess returns. Column (3) introduces a new explanatory
variable (AbsExret), which is the absolute value of the excess returns. Column (4) uses the months
until July as the first half of the year. In column (5), we drop the observations of the last two months
of the year to compute the dependent variable (RAR). In column (6), we use the residuals from
the moving average process to compute the dependent variable (see equation (9)). The dependent
variable in all specifications is the ratio of the standard deviation of tracking error for the second
half of the year to that for the first half of the year. The variable Exret is the fund’s first-half
return in excess of its own self-designated benchmark; Distance is the square of the fund’s return
in excess of its benchmark and measures the extent to which the excess return deviates from zero;
Exp ratio is the expense ratio of the fund at the beginning of the year; T urn ratio is the turnover
ratio of the fund at the beginning of the year; Shareclass is a dummy variable that takes a value
of 1 if the fund has multiple share classes; Log age is the log of the fund’s age; Log size is the log
of the fund’s TNA at the beginning of the year; and Flows is the new money into fund j, defined
T N A j,t+1 −T N A j,t (1+r j,t+1 )
as T N A j,t , during the first half of the year. All of the specifications have time fixed
effects, and bootstrapped standard errors are reported in parentheses. Panel B reports results of a
quantile regression at the median for funds with different investment styles. Only the coefficient
on Distance is reported in Panel B. * , ** , and *** indicate whether the results are statistically
different from zero at the 10%, 5%, and 1% significance levels, respectively.
Window Return
Above Below Absolute SevenMonths Dressing Correlation
(1) (2) (3) (4) (5) (6)
with our expectation. This result shows that it is the vega of the management
contract that is driving the result and not the particular definition of Distance.
Third, a cutoff of June is arbitrary but widely used in the literature (i.e.,
Brown, Harlow, and Starks (1996), Busse (2001), Kempf, Ruenzi, and Thiele
(2009), Schwarz (2012)) to make the point that a manager’s choice of risk is
conditional on prior performance. In a robustness test, we change the mid-
year point from June to July. Evidence reported in column (4) suggests that
changing the mid-year point does not affect the key results: we continue to find
supportive evidence for risk-shifting among fund managers and for our priors
regarding the region in which it is strongest.
Fourth, in an attempt to mislead investors about their skills, mutual fund
managers often change their portfolios by disclosing large positions in win-
ner stocks and small positions in loser stocks, as shown by Lakonishok et al.
(1991). This practice, referred to as “window dressing,” is more pronounced for
underperforming funds and may lead managers to decrease their holdings in
high-risk securities to make their portfolios appear less risky as they get closer
to the fiscal year-end. In addition, Gibson, Safieddine, and Titman (2000) show
that to minimize the taxable distribution, funds tend to trade more as they
approach the end of the year. Such activities could distort our measure of risk-
shifting. Since these incentives drive managerial actions toward the end of the
year, we drop the return data in November and December to compute RAR.
The results from this revised approach, presented in column (5) of Table IX,
continue to support our hypothesis.
Fifth, Busse (2001) argues that intra-year changes in daily return autocor-
relations can cause unintended changes to a mutual fund’s intra-year risk. To
address this bias, we follow Busse (2001) and model the fund’s returns as a
moving average (MA(1)) process. We estimate the moving average process for
each of the two halves of the year as follows:
where r j1,t and r j2,t represent the return of fund j on date t in the first and
second halves of the year, respectively. We estimate the above process for
each fund-year in our sample. We then compute the risk-shifting measure as
follows:
σ (ε j2,t − b j,t )
RARM A = . (10)
σ (ε j1,t − b j,t )
Column (6) of Table IX presents results when RARM A is used as the mea-
sure of risk-shifting. The coefficient on Distance continues to be negative and
statistically significant.
Finally, we test whether risk-shifting behavior is exhibited only by a specific
style of funds, or whether this behavior is pervasive across fund styles. To do
so, we sort our sample into subsamples based on the objective code provided by
2574 The Journal of FinanceR
CRSP. We divide the sample into cap-based funds, growth funds, income funds,
and funds that focus on both growth and income. The category “Cap-Based”
includes large-, mid-, small-, and micro-cap funds. Panel B of Table IX presents
results from a quantile regression for each group. We estimate the full model
shown in our baseline results with RAR as the dependent variable, but for
brevity we only report the coefficient on Distance. The strong negative coef-
ficients across the different categories suggest that the risk-shifting behavior
is not concentrated within a few styles but rather is prevalent across broad
investment objectives.
IV. Conclusion
Our results suggest that the motives for risk-shifting should include the ex-
plicit incentives within the compensation contract provided by the investment
advisor to the portfolio manager. A critical feature of this contract is an option-
like payoff, with the manager receiving higher compensation if the fund’s return
is higher than the benchmark. We find that the funds whose return is close to
the benchmark risk-shift the most, with the exception of very poorly performing
funds. For the managers of these funds, the probability of being fired outweighs
the incentives in the contract. These results strongly support our hypothesis
that portfolio managers who are compensated by performance-based contracts
shift the volatility of the fund to maximize the value of their compensation.
However, risk-shifting based on management contracts is not mutually ex-
clusive relative to risk-shifting from tournaments. Nevertheless, given our re-
sults, any study of tournaments needs to recognize the role of management
contracts. Moreover, we are agnostic on the optimality of existing compensation
contracts. Actual contracts are incomplete and cannot cover all contingencies.
Therefore, we can view portfolio managers as having residual control rights
to the assets of fund, and we can view the loss associated with risk-shifting
as the costs of incomplete contracts. The gains of an incentive contract will
undoubtedly vary across funds and managers and are the subject of future
research.
dP = μ p Pdt + σ p PdW p,
Do Portfolio Manager Contracts Contract Portfolio Management? 2575
dB = μb Bdt + σb BdWb,
and the Brownian motion driving the two asset prices is correlated, that is,
dW pdWb = ρ. The price of this exchange option, EO, is given by the following
equation:
EO = P N(d1 ) − BN(d2 ),
where
P
+σ T √
2
ln
d1 = B
√ 2 , d2 = d1 − σ T . (A.1)
σ T
In the above formula,
σ = σ p2 + σb2 − 2σbσ pρ
x
1 s2
N(x) = √ e− 2 ds and
2π −∞
1 x2
N (x) = √ e− 2 .
2π
The first object of interest is the sensitivity of the exchange option to the
volatility of the risky portfolio (σ p). One should also note that the following
expressions are true:
∂ 1 2
σ T = T × (σ p − σbρ), (A.2)
∂σ p 2
∂ 1
(σ ) = (σ p − σbρ), (A.3)
∂σ p σ
∂ 1 −1
= 3 (σ p − σbρ), (A.4)
∂σ p σ σ
∂ (σ p − σbρ) √ d1
(N(d1 )) = N (d1 ) T− , (A.5)
∂σ p σ σ
and
∂ (σ p − σbρ) d1
(N(d2 )) = N (d2 ) − . (A.6)
∂σ p σ σ
Based on these expressions, we can compute the response of the exchange
option to the change in the volatility of the manager’s portfolio:
∂ EO (σ p − σbρ) √ d1 (σ p − σbρ) d1
= P N (d1 ) T− − B N (d2 ) − .
∂σ p σ σ σ σ
2576 The Journal of FinanceR
N (d2 )B − N (d1 )P = 0.
Using the above expressions, we obtain the “vega” of the option as follows:
∂ EO (σ p − σbρ) √
υ= = P N (d1 ) T. (A.7)
∂σ p σ
Vega maximization: Equation (A.7) shows that the value of the manager’s
option increases as the manager increases the value of her portfolio volatility.
The derivative of vega(υ) with respect to the portfolio price leads to the following
first-order condition:
∂υ ∂ ∂ EO (σ p − σbρ) √ ∂
= = T× (P N (d1 )) = 0.
∂P ∂ P ∂σ p σ ∂P
Solving the above equation yields ln( PB ) = σ2 T . Therefore, the value of the
2
REFERENCES
Aragon, George O., and Vikram Nanda, 2012, Tournament behavior in hedge funds: High-water
marks, fund liquidation, and managerial stake, Review of Financial Studies 25, 937–974.
Basak, Suleyman, Anna Pavlova, and Alexander Shapiro, 2007, Optimal asset allocation and risk
shifting in money management, Review of Financial Studies 20, 1583–1621.
Brown, Keith C., W. Van Harlow, and Laura T. Starks, 1996, Of tournaments and temptations: An
analysis of managerial incentives in the mutual fund industry, Journal of Finance 51, 85–110.
Busse, Jeffrey A., 2001, Another look at mutual fund tournaments, Journal of Financial and
Quantitative Analysis 36, 53–73.
Carpenter, Jennifer N., 2000, Does option compensation increase managerial risk appetite? Jour-
nal of Finance 55, 2311–2331.
Chen, Hsiu-lang, and George G. Pennacchi, 2009, Does prior performance affect a mutual fund’s
choice of risk? Theory and further empirical evidence, Journal of Financial and Quantitative
Analysis 44, 745–775.
Chen, Qi, Itay Goldstein, and Wei Jiang, 2010, Payoff complementarities and financial fragility:
Evidence from mutual fund outflows, Journal of Financial Economics 97, 239–262.
Chevalier, Judith, and Glenn Ellison, 1997, Risk taking by mutual funds as a response to incentives,
Journal of Political Economy 105, 1167–1200.
Christoffersen, Susan E. K., and Mikhail Simutin, 2017, On the demand for high-beta stocks:
Evidence from mutual funds, Review of Financial Studies 30, 2596–2620.
Coles, Jeffrey L., Jose Suay, and Denise Woodbury, 2000, Fund advisor compensation in closed-end
funds, Journal of Finance 55, 1385–1414.
Martijn Cremers, K. J., and Antti Petajisto, 2009, How active is your fund manager? A new measure
that predicts performance, Review of Financial Studies 22, 3329–3365.
Cuoco, Domenico, and Ron Kaniel, 2011, Equilibrium prices in the presence of delegated portfolio
management, Journal of Financial Economics 101, 264–296.
Deli, Daniel N., 2002, Mutual fund advisory contracts: An empirical investigation, Journal of
Finance 57, 109–133.
Do Portfolio Manager Contracts Contract Portfolio Management? 2577
Elton, Edwin J., Martin J. Gruber, and Christopher R. Blake, 2003, Incentive fees and mutual
funds, Journal of Finance 58, 779–804.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests,
Journal of Political Economy 81, 607–636.
Gibson, Scott, Assem Safieddine, and Sheridan Titman, 2000, Tax-motivated trading and price
pressure: An analysis of mutual fund holdings, Journal of Financial and Quantitative Analysis
35, 369–386.
Golec, Joseph, and Laura Starks, 2004, Performance fee contract change and mutual fund risk,
Journal of Financial Economics 73, 93–118.
Hart, Oliver, 2017, Incomplete contracts and control, American Economic Review 107, 1731–1752.
Hodder, James E., and Jens Carsten Jackwerth, 2007, Incentive contracts and hedge fund man-
agement, Journal of Financial and Quantitative Analysis 42, 811–826.
Hu, Ping, Jayant R. Kale, Marco Pagani, and Ajay Subramanian, 2011, Fund flows, performance,
managerial career concerns, and risk taking, Management Science 57, 628–646.
Huang, Jennifer, Clemens Sialm, and Hanjiang Zhang, 2011, Risk shifting and mutual fund per-
formance, Review of Financial Studies 24, 2575–2616.
Kempf, Alexander, Stefan Ruenzi, and Tanja Thiele, 2009, Employment risk, compensation in-
centives, and managerial risk taking: Evidence from the mutual fund industry, Journal of
Financial Economics 92, 92–108.
Khorana, Ajay, 1996, Top management turnover: An empirical investigation of mutual fund man-
agers, Journal of Financial Economics 40, 403–427.
Khorana, Ajay, 2001, Performance changes following top management turnover: Evidence from
open-end mutual funds, Journal of Financial and Quantitative Analysis 36, 371–393.
Koenker, Roger, 2004, Quantile regression for longitudinal data, Journal of Multivariate Analysis
91, 74–89.
Lakonishok, Josef, Andrei Shleifer, Richard Thaler, and Robert Vishny, 1991, Window dressing by
pension fund managers, Technical report, National Bureau of Economic Research.
Wei Li, C., and Ashish Tiwari, 2009, Incentive contracts in delegated portfolio management, Review
of Financial Studies 22, 4681–4714.
Ma, Linlin, Yuehua Tang, and Juan-Pedro Gómez, 2019, Portfolio manager compensation in the
U.S. mutual fund industry, The Journal of Finance 74, 587–638.
Margrabe, William, 1978, The value of an option to exchange one asset for another, Journal of
Finance 33, 177–186.
Massa, Massimo, and Rajdeep Patgiri, 2009, Incentives and mutual fund performance: Higher
performance or just higher risk taking? Review of Financial Studies 22, 1777–1815.
Parente, Paulo M. D. C., and João M. C. Santos Silva, 2016, Quantile regression with clustered
data, Journal of Econometric Methods 5, 1–15.
Schwarz, Christopher G., 2012, Mutual fund tournaments: The sorting bias and new evidence,
Review of Financial Studies 25, 913–936.
Sensoy, Berk A., 2009, Performance evaluation and self-designated benchmark indexes in the
mutual fund industry, Journal of Financial Economics 92, 25–39.
Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual fund flows, Journal of Finance
53, 1589–1622.
Spiegel, Matthew, and Hong Zhang, 2013, Mutual fund risk and market share-adjusted fund flows,
Journal of Financial Economics 108, 506–528.
Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:
Appendix S1: Internet Appendix.
Replication Code.