An Institutional Theory of Momentum and Revealsal Rev. Financ. Stud.-2013-Vayanos-1087-145

[14:49 8/4/2013 RFS-hht014.
tex] Page: 1087 10871145

An Institutional Theory of Momentum and
Reversal
Dimitri Vayanos
London School of Economics, CEPR, and NBER
Paul Woolley
London School of Economics
We propose a theory of momentum and reversal based on ows between investment
funds. Flows are triggered by changes in fund managers efciency, which investors either
observe directly or infer from past performance. Momentum arises if ows exhibit inertia,
and because rational prices underreact to expected future ows. Reversal arises because
ows push prices away from fundamental values. Besides momentum and reversal, ows
generate comovement, lead-lag effects, and amplication, with these being larger for high
idiosyncratic risk assets. Acalibration of our model using evidence on mutual fund returns
and ows generates sizeable Sharpe ratios for momentum and value strategies. (JEL D82,
G11, G12, G14, G23)
Two of the most prominent nancial market anomalies are momentum
and reversal. Momentum is the tendency of assets with good (bad) recent
performance to continue overperforming (underperforming) in the near future.
Reversal concerns predictability based on a longer performance history:
Assets that performed well (poorly) over a long period tend to subsequently
underperform (overperform). Closely related to reversal is the value effect,
whereby the ratio of an assets price relative to book value is negatively related
to subsequent performance. Momentum and reversal have been documented
extensively and for a wide variety of assets.
1
For helpful comments, we thank Nick Barberis, Jonathan Berk, Bruno Biais, Pierre Collin-Dufresne, Peter
DeMarzo, Xavier Gabaix, John Geanakoplos, Jennifer Huang, Ravi Jagannathan, Jakub Jurek, Peter Kondor,
Arvind Krishnamurthy, Dong Lou, Toby Moskowitz, Anna Pavlova, Lasse Pedersen, Christopher Polk, Matthew
Pritzker, Tarun Ramadorai, Tano Santos, Jeremy Stein, Pietro Veronesi, Luigi Zingales, seminar participants
at Amsterdam, Bergen, Berkeley, Brunel, Chicago, Columbia, Lausanne, Leicester, LSE, MIT, Munich,
Northwestern, NYU, Oslo, Oxford, Paris, Stanford, Sydney, Toulouse, Zurich, and participants at the American
Economic Association 2010, American Finance Association 2010, CREST-HEC 2010, CRETE 2009, EFMA
2011, Gerzensee 2008, and NBER Asset Pricing 2009 conferences. Send correspondence to Dimitri Vayanos,
Department of Finance OLD 3.41, London School of Economics, London WC2A 2AE, UK; telephone:
+44-20-7955-6382. E-mail: d.vayanos@lse.ac.uk.
1
Jegadeesh and Titman (1993) document momentum for individual U.S. stocks, predicting returns over horizons
of 312 months by returns over the past 312 months. DeBondt and Thaler (1985) document the reversal,
predicting returns over horizons of up to ve years by returns over the past 35 years. Fama and French (1992)
document the value effect. This evidence has been extended to stocks in other countries (Fama and French 1998;
The Author 2013. Published by Oxford University Press on behalf of The Society for Financial Studies.
All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.
doi:10.1093/rfs/hht014 Advance Access publication March 27, 2013

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1088 10871145
The Review of Financial Studies / v 26 n 5 2013
Momentum and reversal are viewed as anomalies because they are
hard to explain within the standard asset-pricing paradigm with a rational
representative agent. The prevalent explanations of these phenomena are
behavioral.
2
In this paper we show that momentum and reversal can result
from ows between investment funds in markets in which fund investors and
managers are rational. Besides momentum and reversal, fund ows generate
comovement, lead-lag effects, and amplication, with all these being larger for
assets with high idiosyncratic risk. Acalibration of our model using evidence on
mutual fund returns and ows generates sizeable Sharpe ratios for momentum
and value strategies.
Our explanation of momentum and reversal is as follows. Suppose that a
negative shock hits the fundamental value of some assets. Investment funds
holding these assets realize low returns, triggering outows by investors who
update negatively about the efciency of the managers running these funds.
As a consequence of the outows, funds sell assets they own, and this further
depresses the prices of the assets hit by the original shock. Momentum arises
if the outows are gradual, and if they trigger a gradual price decline and a
drop in expected returns. Reversal arises because outows push prices below
fundamental values, and so expected returns eventually rise. Gradual outows
can be the consequence of investor inertia or institutional constraints, and we
simply assume them. We explain, however, why gradual outows can trigger
a gradual decline in rational prices and a drop in expected returns. This result,
key to momentum, is surprising. Indeed, why do rational investors absorb the
outows, buying assets whose expected returns have decreased?
Rational investors in our model buy assets whose expected returns have
decreased because of what we term the bird-in-the-hand effect. Assets that
experience a price drop and are expected to continue underperforming in the
short run are those held by investment funds expected to experience outows.
The anticipation of outows causes these assets to be underpriced and to
guarantee investors an attractive return (bird in the hand) over a long horizon.
Investors could earn an even more attractive return on average (two birds in the
bush) by buying these assets after the outows occur. This, however, exposes
them to the risk that the outows might not occur, in which case the assets
would cease to be underpriced.
3
Rouwenhorst 1998), industry-level portfolios (Grinblatt and Moskowitz 1999), country indices (Asness, Liew,
and Stevens 1997; Bhojraj and Swaminathan 2006), bonds (Asness, Moskowitz, and Pedersen 2013), currencies
(Bhojraj and Swaminathan 2006) and commodities (Gorton, Hayashi, and Rouwenhorst 2013).
2
See, for example, Barberis, Shleifer, and Vishny (1998), Daniel, Hirshleifer, and Subrahmanyam (1998), Hong
and Stein (1999), and Barberis and Shleifer (2003).
3
The bird-in-the-hand effect can be illustrated in the following simple example. An asset is expected to pay off
100 in period 2. If outows do not occur in period 1, the price will be 100, but if they occur, the price will drop
to 80. Each scenario is equally likely. Buying the asset in period 0 at 92 earns an investor a two-period expected
capital gain of 8. Buying in period 1 earns an expected capital gain of 20 if outows occur and 0 if they do not. A
risk-averse investor might prefer earning 8 rather than 20 or 0 with equal probabilities, even though the expected
1088

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1089 10871145
An Institutional Theory of Momentum and Reversal
The bird-in-the-hand effect, and its implication that rational prices underreact
to expected future ows, extend beyond institutional ows; all that is needed is
that ows are gradual, uncertain, and can push prices away from fundamental
values. Provided that ows have these properties, momentum and reversal can
arise even in a model with one risky asset and no delegated fund management.
4
Thus, our theory has two distinct parts: First, show the bird-in-the-hand effect
and rational momentum, and second do this in a model with multiple risky
assets and delegation. Multiple assets allowus to derive additional phenomena,
such as comovement and lead-lag effects. Delegation provides us with a driver
of ows: investor response to changes in fund manager efciency. It also
allows us to tie the pricing phenomena that we derive to a growing empirical
literature on institutional ows, summarized later in this section. At the same
time, delegation gives rise to a career-concern effect, which works against
the bird-in-the-hand effect and eliminates momentum for some parameter
values: Because fund managers care about fund ows in addition to investment
performance, they hedge against the risk that outows might occur and that
mispricings increase.
Section 1 presents our model. We consider an innite-horizon economy with
multiple risky assets, which we refer to as stocks, and one riskless asset. A
competitive investor can access the stocks only through two investment funds.
We assume that one of the funds tracks mechanicallya market index, soportfolio
optimization concerns only the other fund, which we refer to as the active fund.
To ensure that the active fund can add value over the index fund, we assume
that the market index differs from the true market portfolio characterizing
equilibrium asset returns. Our assumptions on the index and the active fund
are meant to capture in a stylized and parsimonious manner a setting in which
investors have access to multiple funds with different portfolios. Flows between
the active and the index fund could be interpreted more generally as between
active funds pursuing different strategies.
The active fund is run by a competitive manager, who can also invest his
personal wealth through the fund. The latter assumption is for parsimony: In
addition to generating a simple objective that the manager maximizes when
choosing the funds portfolio, it ensures that the manager acts as a trading
counterparty to the investors ows. The manager can be interpreted more
generally as an aggregate of all agents absorbing the ows, and this eliminates
the need to introduce additional agents into the model. Both investor and
manager are risk averse. To ensure that the investor has a motive to move
across funds, and so generate ows, we assume that she suffers a time-varying
cost from holding the active fund. The interpretation of the cost that best ts
capital gain between periods 0 and 1 is negative. The bird-in-the-hand effect can be reinterpreted in the formal
language of Mertons ICAPM: The investor buys an underpriced asset even though the price is expected to drop
even further in the short run, to hedge against a reduction in the mispricing.
4
See Vayanos and Woolley (2013).
1089

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1090 10871145
our model is as a managerial perk, although other interpretations, such as
managerial ability, are possible within more complicated versions of the model.
Section 2 solves for equilibriumin the case of symmetric information, where
the cost of holding the active fund is observable by both the investor and the
manager. Section 3 considers the more complicated case of asymmetric infor-
mation, where the investor does not observe the cost and must infer it fromfund
returns. Asymmetric information is more realistic, because investors typically
do not observe managerial perks or ability, and yields a richer set of results.
Momentum and reversal arise even under symmetric information. For
example, following an increase in the cost, the investor ows out of the
active and into the index fund, effectively selling stocks that the active fund
overweights relative to the index. Because ows are gradual, the bird-in-the-
hand effect implies that the price of these stocks declines gradually, yielding
momentum. Flows also generate comovement and lead-lag effects, that is,
cross-asset predictability. Because, for example, outows from the active
fund lower the prices of stocks that it overweights and raise those that it
underweights, they increase comovement within each group, while reducing
comovement across groups. Moreover, because a price drop of an overweighted
stock is correlated with outows, it forecasts low expected returns of other
overweighted stocks in the short run and high returns in the long run.
The new key feature of asymmetric information is that fund ows not only
cause stock returns, as under symmetric information, but are also caused by
them. For example, a negative cash-ow shock to a stock that the active fund
overweights lowers the active funds performance relative to the index fund.
The investor theninfers that the cost has increasedandows out of the active and
into the index fund. This lowers the stocks price, amplifying the effect of the
original shock. Amplication generates new channels of momentum, reversal,
and comovement. For example, momentum and reversal arise conditional not
only on past returns, as under symmetric information, but also on past cash-ow
shocks. Moreover, a new channel of comovement is that a cash-ow shock to
one stock induces fund ows that affect the prices of other stocks.
Momentum, reversal, lead-lag effects, and comovement are larger for stocks
with high idiosyncratic risk. This result holds under both symmetric and
asymmetric information, with the intuition being different in the two cases.
For example, in the case of asymmetric information, a cash-ow shock to a
stock with high idiosyncratic risk generates a large discrepancy between the
performance of the active and of the index fund. This causes large fund ows
and price effects.
Our models implications for return predictability map naturally into
implications for active portfolio management. We sketch some of them in
Section 4, where we calibrate our model using evidence on mutual fund returns
and ows. Assuming a Sharpe ratio of 30%for the market index, we nd Sharpe
ratios of 40% for a cross-sectional momentum strategy and 26% for a cross-
sectional value strategy. Thus, a signicant fraction of these strategies actual
1090

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1091 10871145
Sharpe ratios can perhaps be explained based only on institutional ows and
rational behavior.
Empirical support for our theory comes from Lou (2012), who examines
more generally the implications of predictable fund ows. Lou forecasts ows
in or out of mutual funds based on the funds past performance, and imputes
these fund-level ows into ows in or out of individual stocks held by the
funds. Consistent with our theory, the stock-level ows explain a signicant
part of stock-level momentum, especially for large stocks and in recent data
in which mutual funds are more prevalent. Other predictions of our theory are
also supported in the data. For example, Coval and Stafford (2007) nd that
mutual funds experiencing large outows engage in the distressed selling of
their stock portfolios, and this generates signicant price pressure and return
predictability. Anton and Polk (2012) and Greenwood and Thesmar (2011) nd
that comovement between stocks is larger when these are held by many mutual
funds in common, controlling for style characteristics. Jotikasthira, Lundblad,
and Ramadorai (2012) nd price pressure and comovement in an international
context.
Behavioral models of momentum and reversal include Barberis, Shleifer,
and Vishny (1998), Daniel, Hirshleifer, and Subrahmanyam (1998), Hong and
Stein (1999), and Barberis and Shleifer (BS; 2003). BS are closest to our work.
They assume that stocks belong to styles and are traded between switchers and
fundamental investors. Following a stocks bad performance, switchers assume
that the correspondingstyle will performpoorlyinthe future, andswitchtoother
styles. Switching is assumed to be gradual, and leads to momentum because
fundamental investors are assumed to not anticipate the switchers predictable
ows. Switching also generates comovement of stocks within a style, lead-lag
effects, and amplication. We show that these effects can be consistent with
rational behavior. We also study the effects of idiosyncratic risk and career
concerns, neither of which are examined by BS. Rational models of momentum
include Berk, Green, and Naik (1999), Johnson (2002), and Shin (2006), who
assume symmetric information, and Albuquerque and Miao (2012) and Cespa
and Vives (2012), who assume asymmetric information. In these papers, a risky
assets expected return decreases following a low return, typically because
volatility or asset supply decreases.
The equilibrium implications of delegated portfolio management are the
subject of a growing body of literature. In Shleifer and Vishny (1997), fund
ows are an exogenous function of the funds past performance, and amplify
the effects of cash-ow shocks. In He and Krishnamurthy (2012a, 2012b) and
Brunnermeier and Sannikov (2012), the equity stake of fund managers must
exceed a lower bound because of optimal contracting under moral hazard, and
amplication effects can again arise.
5
In Dasgupta, Prat, and Verardo (2011),
5
Amplication effects can also arise when agents face margin constraints or have wealth-dependent risk aversion.
See the survey by Gromb and Vayanos (2010) and the references therein.
1091

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1092 10871145
reputation concerns cause managers to herd, and this generates momentum
under the additional assumption that the market makers trading with the
managers are either monopolistic or myopic. In Basak and Pavlova (2012),
ows by investors benchmarked against an index cause stocks in the index to
comove.
6
We contribute a number of new results to this literature, for example,
momentum with competitive and rational agents, and larger effects for high
idiosyncratic risk assets. We also bring the analysis of delegation and fund ows
within a exible normal-linear framework that yields closed-form solutions.
1. Model
1.1 Assets
Time t is continuous and goes from zero to innity. There are N risky assets
and a riskless asset. We refer to the risky assets as stocks, but they could also be
interpreted as industry-level portfolios or asset classes. The riskless asset has
an exogenous, continuously compounded return r. The stocks pay dividends
over time, and their prices are determined endogenously in equilibrium. We
denote by D
nt
the cumulative dividend per share of stock n=1, ... ,N, by S
nt
the stocks price, and by
n
the stocks supply in terms of number of shares.
The exact specication of the dividend process has little bearing on our results,
so we consider a simple i.i.d. specication. We assume that the vector D
t
(D
1t
,..,D
Nt
)
follows the process

dD
t
=

Fdt +dB
D
t
, (1)
where

F is a constant vector, is a constant matrix of diffusion coefcients,
B
D
t
is a d-dimensional Brownian motion, and v
is the transpose of the vector v.

Thus, the instantaneous dividends dD
t
paidbetweent andt +dt are independent
over time and identically distributed, with expectation

Fdt and covariance
matrix dt
dt .
1.2 Investment funds
A competitive investor can invest in the riskless asset and in the stocks. The
investor, however, faces the friction that she can access the stocks only through
two investment funds. This could reect the cost of learning about individual
stocks and trading them. We assume that one investment fund is passively
managed and tracks mechanically a market index. This is for simplicity, so
that portfolio optimization concerns only the other fund, which we term the
active fund. We assume that the market index includes a xed number,
n
, of
6
Other models exploring equilibrium implications of delegated portfolio management include Brennan (1993),
Vayanos (2004), Dasgupta and Prat (2008), Petajisto (2009), Cuoco and Kaniel (2011), Malliaris and Yan (2011),
Guerreri and Kondor (2012), and Kaniel and Kondor (2012). See also Berk and Green (2004), in which fund ows
are driven by fund performance because investors learn about managers ability, and feed back into performance
because of decreasing returns to managing a large fund.
1092

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1093 10871145
shares of stock n. Thus, if the vectors (
1
,..,
N
) and (
1
,..,
N
) are
collinear, the market index is capitalization-weighted and coincides with the
market portfolio.
To ensure that the active fund can add value over the index fund, we assume
that the market index differs from the true market portfolio characterizing
equilibrium asset returns. This can be because the market index does not
include some stocks, and so the vectors and are not collinear. Alternatively,
the market index can coincide with the market portfolio, but unmodeled buy-
and-hold investors, such as rms managers or founding families, can hold a
portfolio different from the market portfolio. That is, buy-and-hold investors
hold
n
shares of stock n, and the vectors and (
1
,..,
N
) are not collinear.
To nest the two cases, we dene a vector (
1
,..,
N
) to coincide with in
the rst case and in the second. The vector represents the residual
supply left over from buy-and-hold investors, and is the true market portfolio
characterizing equilibrium asset returns. We assume that is not collinear with
the market index , and set
)
2
>0.
Our assumptions on the index and the active fund are meant to capture in
a stylized and parsimonious manner a setting in which investors have access
to multiple funds with different portfolios. The main intuitions coming out of
our model are likely to carry over to alternative settings, for example, no index
fund but two active funds that specialize in different segments of the market.
1.3 Agents
The investor determines how to allocate her wealth between the riskless
asset, the index fund, and the active fund. She maximizes expected utility of
intertemporal consumption. Utility is exponential, that is,
E
_

0
exp(c
t
t )dt, (2)
where is the coefcient of absolute risk aversion, c
t
is consumption, and is
the discount rate. The investors control variables are consumption c
t
and the
number of shares x
t
and y
t
of the index and active fund, respectively.
The active fund is run by a competitive manager, who can also invest his
personal wealth in the fund. The manager determines the active portfolio and the
allocation of his wealth between the riskless asset and the fund. He maximizes
expected utility of intertemporal consumption. Utility is exponential, that is,
E
_

0
exp( c
t

t )dt, (3)
where is the coefcient of absolute risk aversion, c
t
is consumption, and

is the discount rate. The managers control variables are consumption c

t
, the
1093

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1094 10871145
number of shares y
t
of the active fund, and the active portfolio z
t
(z
1t
,..,z
Nt
),
where z
nt
denotes the number of shares of stock n included in one share of the
active fund.
The assumption that the manager can invest his personal wealth in the
active fund is for parsimony: It generates a simple objective that the manager
maximizes when choosing the funds portfolio and ensures that the manager
acts as a trading counterparty to the investors ows.
7
Under the alternative
assumption that the manager must invest his wealth in the riskless asset, we
would need to introduce two newelements into the model: a performance fee to
provide the manager with incentives for portfolio choice and an additional set of
smart-money agents with the expertise to invest in individual stocks, identify
the investors ows, and act as counterparty to them. This would complicate
the model without changing the main intuitions. The manager in our model can
be interpreted as an aggregate of all smart-money agents.
1.4 Cost of active fund
Under the assumptions introduced so far, and in the absence of other frictions,
the equilibriumtakes a simple form. As we showin Section 2, the investor holds
stocks only through the active fund because its portfolio dominates the index
portfolio. As a consequence, the active fund holds the true market portfolio ,
and there are no ows between the two funds.
To generate fund ows, we assume that the investor suffers a time-varying
cost from investing in the active fund. Empirical evidence on the existence
of such a cost is provided in a number of papers. For example, Grinblatt and
Titman (1989), Wermers (2000), and Kacperczyk, Sialm, and Zheng (KSZ;
2008) study the return gap, dened as the difference between a mutual funds
return over a given quarter and the return of a hypothetical portfolio invested
in the stocks that the fund holds at the beginning of the quarter. The return
gap varies signicantly across funds and over time. It is also persistent, with a
half-life of about 2.5 years according to KSZ. The high persistence indicates
that the return gap is linked to underlying fund characteristics, and indeed such
a correlation exists in the data.
We model the return gap in a simple manner: We assume that the investors
return from the active fund is equal to the gross return, made of the dividends
and capital gains of the stocks held by the fund, net of a time-varying cost.
Empirical studies typically attribute the return gap to agency costs, operational
costs, and managerial stock-picking ability. The interpretation that best ts our
model is agency costs: The cost is a perk that the manager can extract from the
7
Restricting the manager not to invest his personal wealth in the index fund is also in the spirit of generating
a simple objective. Indeed, in the absence of this restriction, the active portfolio would be indeterminate: The
manager could mix a given active portfolio with the index, and make that the newactive portfolio, while achieving
the same personal portfolio through an offsetting short position in the index. Note that restricting the manager not
to invest in the index only weakly constrains his personal portfolio because he can always modify the portfolio
of the active fund and his stake in that fund.
1094

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1095 10871145
investor.
8
Operational costs and stock-picking ability t better with the more
complicated version of our model, where the manager must invest his wealth
in the riskless asset and other smart-money agents absorb the investors ows.
Indeed, when the manager also invests in the active fund, it would be natural
to assume that his investment is affected by changes in operational costs or
stock-picking ability. We are assuming, however, that the manager does not
incur the cost on his investment.
We assume that the index fund entails no cost, so its gross and net returns
coincide. This is for simplicity, but also ts the interpretations of the return
gap. Indeed, managing an index fund involves no stock-picking ability, and
operational and agency costs are smaller than for active funds.
We model the cost as a ow (i.e., the cost between t and t +dt is of order dt ),
and assume that the ow cost is proportional to the number of shares y
t
that the
investor holds in the active fund. We denote the coefcient of proportionality
by C
t
and assume that it follows the process
dC
t
=
_
CC
t
_
dt +sdB
C
t
, (4)
where is a mean-reversion parameter,

C is a long-run mean, s is a positive
scalar, and B
C
t
is a Brownian motion independent of B
D
t
. The mean-reversion
of C
t
is not essential for momentumand reversal, which occur even when =0.
To remain consistent with the managerial-perk interpretation of the cost, we
allow the manager to derive a benet from the investors participation in the
active fund. We model the benet in the same manner as the cost, that is, a ow
which is proportional to the number of shares y
t
that the investor holds in the
active fund. If the cost is a perk that the manager can extract efciently, then
the coefcient of proportionality for the benet is C
t
. We allow more generally
the coefcient of proportionality to be C
t
, where 0 is a constant that can
be interpreted as the efciency of perk extraction. Varying generates a rich
specication of the managers objective. When =0, the manager cares about
fund performance only through his personal investment in the fund, and his
objective is similar to the fund investors. When instead >0, the manager has
career concerns, that is, cares about the risk that the investor might reduce her
participationinthe fund. As we showinSection2.3, momentumarises whenis
not too large, that is, perk extraction is inefcient. Momentum would also arise
under nonperk interpretations of the cost (operational costs or stock-picking
ability) because would then be zero.
The cost and benet are assumed proportional to y
t
for analytical
convenience. At the same time, these variables are sensitive to how shares of
8
Examples of perks in a delegated portfolio management context are late trading and soft-dollar commissions.
Managers engaging in late trading use their privileged access to the fund to buy or sell fund shares at stale prices.
Late trading was common in many funds and led to the 2003 mutual fund scandal. Soft-dollar commissions
is the practice of inating funds brokerage commissions to pay for services that mainly benet managers, for
example, promote the fund to new investors.
1095

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1096 10871145
the active fund are dened (e.g., they change with a stock split). We dene one
share of the fund by the requirement that its market value equals the equilibrium
market value of the entire fund. Under this denition, the number of fund shares
held by the investor and the manager in equilibrium sum to one, that is,
y
t
+ y
t
=1. (5)
We dene one share of the index fund to coincide with the market index .
We assume that the investor can adjust her active fund holdings, y
t
, to new
information only gradually. Gradual adjustment can result from contractual
restrictions or institutional decision lags. For simplicity, we model these
frictions as a ow cost (dy
t
/dt )
2
/2 that the investor must incur when
changing y
t
.
9
The manager observes all the variables in the model. The investor observes
the returns and share prices of the index and active funds, but not the same
variables for the individual stocks. We study both the case of symmetric
information, where the investor observes the cost C
t
, and that of asymmetric
information, where C
t
is observable only by the manager.
2. Symmetric Information
In this section, we study the case of symmetric information, where the investor
observes the cost C
t
. We look for an equilibrium in which stock prices S
t
(S
1t
,..,S
Nt
)
take the form

S
t
=
F
r
(a
0
+a
1
C
t
+a
2
y
t
), (6)
where (a
0
,a
1
,a
2
) are constant vectors. The rst term is the present value of
expected dividends, discounted at the riskless rate r, and the second term is
a risk discount linear in (C
t
,y
t
). The risk discount moves in response to fund
ows, as we showlater in this section. The rate v
t
dy
t
/dt at which the investor
changes her active fund holdings in our conjectured equilibrium is
v
t
=b
0
b
1
C
t
b
2
y
t
, (7)
where (b
0
,b
1
,b
2
) are constants. We expect (b
1
,b
2
) to be positive, that is, the
investor disinvests faster from the active fund when C
t
and y
t
are large. We
refer to an equilibrium satisfying (6) and (7) as linear.
9
An example of a contractual restriction is lockup periods, often imposed by hedge funds, which require investors
to not withdraw capital for a prespecied time period. Institutional decision lags can arise for investors such as
pensionfunds, foundations, or endowments, where decisions are made byboards of trustees that meet infrequently.
We employ the quadratic specication (dy
t
/dt )
2
/2 to model these frictions because it preserves the linearity
of the model.
1096

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1097 10871145
2.1 Managers optimization
The manager chooses the active funds portfolio z
t
, the number y
t
of fund shares
that he owns, and consumption c
t
. The managers budget constraint is
dW
t
=rW
t
dt + y
t
z
t
(dD
t
+dS
t
rS
t
dt )+C
t
y
t
dt c
t
dt. (8)
The rst term is the return from the riskless asset; the second term is the
return from the active fund in excess of the riskless asset; the third term is
the managers benet from the investors participation in the fund; and the
fourth term is consumption. To compute the return from the active fund, we
note that because one share of the fund corresponds to z
t
shares of the stocks,
the managers effective stock holdings are y
t
z
t
shares. These holdings are
multiplied by the vector dR
t
dD
t
+dS
t
rS
t
dt of the stocks excess returns
per share (referred to as returns, for simplicity).
The managers optimization problem is to choose controls ( c
t
, y
t
,z
t
) to
maximize the expected utility (3) subject to the budget constraint (8) and the
investors holding policy (7). The active funds portfolio z
t
satises, in addition,
the normalization
z
t
S
t
=( x
t
)S
t
. (9)
This is because one share of the active fund is dened so that its market value
equals the equilibrium market value of the entire fund. Moreover, the latter
is ( x
t
)S
t
because in equilibrium the active fund holds the true market
portfolio minus the investors holdings x
t
of the index fund. We conjecture
that the managers value function is
V(W
t
,

X
t
) exp
_
_
r W
t
+ q
0
+( q
1
, q
2
)
X
t
+
1
2
X
t
__
, (10)
where

X
t
(C
t
,y
t
)
, ( q
0
, q
1
, q
2
) are constants, and

Q is a constant symmetric
22 matrix. The Bellman equation is
max
c
t
, y
t
,z
t
_
exp( c
t
)+D

V
_
=0, (11)
where D
V is the drift of the process

V under the controls ( c
t
, y
t
,z
t
).
Proposition 1. The value function (10) satises the Bellman equation (11) if
( q
0
, q
1
, q
2
,

Q) satisfy a system of six scalar equations.
In the proof of Proposition 1, we show that the optimization over ( c
t
, y
t
,z
t
)
can be reduced to optimization over the managers consumption c
t
and effective
stock holdings z
t
y
t
z
t
. Given z
t
, the decomposition between y
t
and z
t
is
determined by the normalization (9). The rst-order condition with respect
to z
t
is
E
t
(dR
t
) =r Cov
t
(dR
t
, z
t
dR
t
)+( q
1
+ q
11
C
t
+ q
12
y
t
)Cov
t
(dR
t
,dC
t
). (12)
Equation (12) links expected stock returns to the risk faced by the manager. The
expected return that the manager requires from a stock depends on the stocks
1097

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1098 10871145
covariance with the managers portfolio z
t
(rst term in the right-hand side),
and on the covariance with changes to the cost C
t
(second term). The latter
effect reects a hedging demand by the manager. We derive the implications
of (12) for the cross-section of expected returns later in this section.
2.2 Investors optimization
The investor chooses a number of shares x
t
in the index fund and y
t
in the
active fund, and consumption c
t
. The investors budget constraint is
dW
t
=rW
t
dt +x
t
dR
t
+y
t
(z
t
dR
t
C
t
dt )
1
2
v
2
t
dt c
t
dt. (13)
The rst three terms are the returns from the riskless asset, the index fund,
and the active fund (net of the cost C
t
), respectively. The fourth term is
the cost of adjustment, and the fth term is consumption. The investors
optimization problemis to choose controls (c
t
,x
t
,y
t
) to maximize the expected
utility (2) subject to the budget constraint (13). The investor takes the active
funds portfolio z
t
as given and equal to its equilibrium value x
t
. We
study this optimization problem in two steps. In a rst step, we optimize over
(c
t
,x
t
), assuming that v
t
is given by (7). We solve this problem using dynamic
programming, and conjecture the value function
V(W
t
,X
t
) exp
_
_
rW
t
+q
0
+(q
1
,q
2
)X
t
+
1
2
X
t
QX
t
__
, (14)
where X
t
(C
t
,y
t
)
, (q
0
,q
1
,q
2
) are constants, and Q is a constant symmetric
22 matrix. The Bellman equation is
max
c
t
,x
t
[exp(c
t
)+DV V] =0, (15)
where DV is the drift of the process V under the controls (c
t
,x
t
). In a second
step, we derive conditions under which the control v
t
given by (7) is optimal.
(q
0
,q
1
,q
2
,Q) satisfy a system of six scalar equations. The control v
t
given by
(7) is optimal if (b
0
,b
1
,b
2
) satisfy a system of three scalar equations.
2.3 Equilibrium
In equilibrium, the active funds portfolio z
t
is equal to x
t
, and the shares
held by the manager and the investor sum to one. Combining these equations
with the rst-order conditions and value-function equations (Propositions 1
and 2), yields a system of equations characterizing a linear equilibrium.
Proposition 3 shows that a unique linear equilibrium exists when the diffusion
coefcient s of C
t
is small. This is done by computing explicitly the linear
equilibriumfor s =0 and applying the implicit function theorem. Our numerical
solutions for general values of s in Section 4 seem to also generate a unique
1098

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1099 10871145
linear equilibrium. Moreover, the properties that we derive analytically for
small s in the rest of this section seem to hold for general values of s.
10
Proposition 3. For small s, there exists a unique linear equilibrium. The
constants (b
1
,b
2
) are positive, and the vectors (a
1
,a
2
) are given by
a
i
=
i
p
f
, (16)
where
1
is a positive constant,
2
is a negative constant, and
p
f

(17)
is the owportfolio. Equation (16) holds in any linear equilibriumfor general
values of s.
Proposition 3 can be specialized to the benchmark case of costless delegation,
where the investors cost C
t
of investing in the active fund is constant over time
and equal to zero. This case can be derived by setting C
t
, as well as its long-run
mean

C and diffusion coefcient s, to zero.
Corollary 1 (costless delegation). When C
t
=

C=s =0, the investor adjusts
her holdings of the active fund to the steady-state value lim
t
y
t
= /(+ )
and those of the index fund to lim
t
x
t
=0. Stocks expected returns in the
steady state are given by the one-factor model
lim
t
E
t
(dR
t
) =
r
+
dt =
r
+
Cov
t
(dR
t
,dR
t
), (18)
with the factor being the true market portfolio .
Corollary1concerns a steadystate reachedfor t because the adjustment
cost prevents the investor from adjusting instantaneously to her optimal
holdings. In the steady state, the investor holds only the active fund because that
fund offers a more superior portfolio than the index fund at no cost. The relative
shares of the investor and the manager in the active fund are determined by their
risk-aversion coefcients, according to optimal risk sharing. Stocks expected
returns are determined by the covariance with the true market portfolio. The
intuitionfor the latter result is that since the indexfundreceives zeroinvestment,
the true market portfolio coincides with the active portfolio z
t
, which is also
the portfolio held by the manager. Because the manager determines the cross-
section of expected returns through the rst-order condition (12), and there is
no hedging demand because C
t
is constant, the true market portfolio is the only
pricing factor.
10
This applies to b
1
>0, b
2
>0,
1
>0,
2
<0 and to Corollaries 2, 5, 6 (with a different threshold
R
) and 7.
1099

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1100 10871145
When the cost C
t
is stochastic, the investors fund holdings and the stocks
expected returns are stochastic in the steady state. We next determine how
shocks to C
t
affect fund ows, stock prices, and expected returns.
Following an increase in C
t
, the investor ows out of the active and into
the index fund. Because of the adjustment cost, this ow does not occur
instantaneously at time t , but only gradually for t
>t . Corollary 2 computes

the implied change in the number of shares of each stock that the investor holds
through the aggregate of the two funds. We consider the expectation as of time
t of stock holdings at time t
>t , to isolate the effect of the time-t shock from

those of subsequent shocks. Because of the linearity of our model, this amounts
to setting the realized values of the subsequent shocks to zero.
Corollary 2 (fund ows). The change in the investors expected stock
holdings at time t
>t , caused by a change in C

t
at time t , is proportional
to the ow portfolio p
f
:
E
t
(x
t
+y
t
z
t
)
C
t
=
b
1
_
e
(t
t )
e
b
2
(t
t )
_
b
2
p
f
. (19)
For small s, the coefcient of p
f
is negative.
The change in the investors stock holdings is proportional to the ow
portfolio p
f
, dened in (17). This portfolio consists of the true market portfolio
, plus a position in the market index that renders the covariance with the
index equal to zero. Long positions in p
f
correspond to large components of
the vector relative to , and hence to stocks that the active fund overweights
relative to the index fund. Conversely, short positions correspond to stocks that
the active fund underweights. In owing out of the active and into the index
fund, the investor is selling a slice of the ow portfolio, thus selling stocks that
the active fund overweights and buying stocks that it underweights.
The effect of C
t
on stock prices derives from that on fund ows, and is
computed in Corollary 3. Following an increase in C
t
, the investor gradually
sells a slice of the ow portfolio. Because the manager takes the other side of
this transaction, he becomes increasingly averse to holding the ow portfolio
and stocks covarying positively with it. Therefore, the expected returns of these
stocks increase, and their price decreases. Moreover, the price decreases not
only when the manager acquires the ow portfolio, after time t , but also in
anticipation of this happening, at time t . Conversely, the time-t price of stocks
covarying negatively with the ow portfolio increases.
Corollary 3 (prices). The change in stock prices at time t , caused by a
contemporaneous change in C
t
, is proportional to stocks covariance with the
ow portfolio p
f
:
S
t
C
t
=
1
p
f
=

1
1+
s
2
2
1

Cov
t
(dR
t
,p
f
dR
t
) =

1
1+
s
2
2
1

Cov
t
(d
t
,p
f
d
t
).
(20)
1100

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1101 10871145
where d
t
(d
1t
,..,d
Nt
)
denotes the residual from a regression of stock

returns dR
t
on the market-index return dR
t
.
Which characteristics of a stock determine its covariance with the ow
portfolio, and hence its sensitivity to fund ows? A stocks relative weight
in the active and index funds inuences the sign of the covariance. Indeed,
stocks that the active fund overweights are likely to covary positively with the
ow portfolio because they receive positive weight in that portfolio, whereas
stocks that the active fund underweights are likely to covary negatively.
A stocks idiosyncratic risk inuences the magnitude of the covariance:
Stocks with high idiosyncratic risk have higher covariance with the ow port-
folio in absolute value, and are thus more sensitive to fund ows. The intuition
is that changes in the cost C
t
of investing in the active fund cause the investor
to rebalance between the active and the index fund, but do not affect her overall
exposure to the index. Indeed, because investing in the index fund is costless,
the investors willingness to bear risk that correlates perfectly with the index
remains constant, and only her willingness to bear orthogonal risk changes.
Therefore, changes in C
t
, and the fund ows they trigger, do not affect the
expected return and price of the index and of stocks that correlate perfectly with
the index.
11
They affect stocks that carry orthogonal, that is, idiosyncratic, risk.
Because changes in C
t
, and the fund ows they trigger, affect prices,
they contribute to comovement between stocks. Corollary 4 decomposes the
covariance matrix of stock returns into a fundamental covariance, driven purely
by cash ows, and a nonfundamental covariance, introduced by fund ows.
Corollary 4 (comovement). The covariance matrix of stock returns is
Cov
t
(dR
t
,dR
t
) =
_
+s
2
2
1
p
f
p
f
_
dt, (21)
the sum of a fundamental covariance dt , driven purely by cash ows, and a
nonfundamental covariance s
2
2
1
p
f
p
f
dt , introduced by fund ows. The
nonfundamental covariance is positive for stock pairs whose covariance with
the ow portfolio has the same sign and is negative otherwise.
The nonfundamental covariance between a pair of stocks is proportional
to the product of the covariances between each stock in the pair and the
ow portfolio. It is thus large in absolute value when the stocks have high
idiosyncratic risk, because they are more affected by changes in C
t
. Moreover,
it can be positive or negative: positive for stock pairs whose covariance with
the ow portfolio has the same sign, and negative otherwise. Intuitively, two
stocks move in the same direction in response to fund ows if they are both
overweighted or both underweighted by the active fund, but move in opposite
directions if one is overweighted and the other underweighted.
11
Stocks that correlate perfectly with the index, and hence have no idiosyncratic risk, have zero covariance with
the ow portfolio, as the last equality in Corollary 3 conrms.
1101

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1102 10871145
Corollary 5 derives the cross-section of stocks expected returns. A stocks
expected return is determined by the stocks covariance with two risk factors:
the market index and the ow portfolio. The covariance with the index is
driven purely by cash ows. The covariance with the ow portfolio is instead
inuenced by the stocks relative weight in the active and index funds, and by
the stocks idiosyncratic risk. It determines both the stocks sensitivity to fund
ows (Corollary 3) and the stocks expected return (Corollary 5). Our model
implies that stock characteristics, such as relative weight across investment
funds and idiosyncratic risks should be considered alongside cash-ow risk in
empirical studies of expected returns.
Corollary 5 (expected returns). Stocks expected returns are given by the
two-factor model
E
t
(dR
t
) =
r
+
Cov
t
(dR
t
,dR
t
)+
t
Cov
t
(dR
t
,p
f
dR
t
), (22)
with the factors being the market index and the ow portfolio. The factor risk
premium
t
associated to the ow portfolio is
t
=r +
1
1+
s
2
2
1

R
1
C
t
+
R
2
y
t
1
s
2
q
1
_
, (23)
where (
R
1
,
R
2
) are constants. For small s, the constant
R
1
is negative if
<
R

2(+ )+

2
_
r +(r +2)
_
1+
4(+ )
r
_, (24)
and is positive otherwise, and the constant
R
2
is negative.
The factor risk premium of the index is constant over time. The factor risk
premium
t
of the ow portfolio, however, is time-varying and depends on
fund ows. For example, outows from the active fund raise
t
, hence raising
the expected returns of stocks that covary positively with the ow portfolio.
The time-variation of
t
is closely related to momentum and reversal.
Consider an increase in the cost, C
t
, of investing in the active fund.
Corollary 3 shows that the prices of stocks that covary positively with the
ow portfolio decrease at time t , in anticipation of the future outows from
the active fund. Corollary 5 implies that when
R
1
<0, the expected returns of
these stocks also decrease at time t . The simultaneous decrease in prices and
expected returns can appear puzzling because a decrease in expected returns
is typically accompanied by a price increase. The explanation is that although
expected returns decrease in the short run, they increase in the long run, in
response to the gradual outows triggered by the increase in C
t
. It is the latter
increase that causes the price decrease at time t .
The short-run decrease in expected returns is puzzling. Indeed, in absorbing
the investors outows, the manager (who as we point out in Section 1.3 can
1102

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1103 10871145
be viewed as an aggregate of all smart-money agents) buys stocks that the
active fund overweights. Why is he willing to buy these stocks, even though
their expected return has decreased? The intuition is that the manager prefers
to guarantee a bird in the hand. Indeed, the anticipation of future outows
causes active overweights to become underpriced and offer an attractive return
over a long horizon. The manager could earn an even more attractive return, on
average, by buying these stocks after the outows occur. This, however, exposes
him to the risk that the outows might not occur, in which case the stocks
would cease to be underpriced. Thus, the manager might prefer to guarantee an
attractive long-horizon return (bird in the hand) and pass on the opportunity to
exploit an uncertain short-run price drop (two birds in the bush). Note that in
seeking to guarantee the long-horizon return, the manager is, in effect, causing
the short-run drop. This is because his buying pressure prevents the price in the
short run from dropping to a level that fully reects the future outows, that is,
from which a short-run drop is not expected.
The bird-in-the-hand effect can be seen formally in the managers rst-order
condition (12). Following an increase in C
t
, the expected return of a stock that
covaries positively with the ow portfolio decreases, lowering the left-hand
side of (12). Therefore, the manager remains willing to hold the stock only if
its risk, described by the right-hand side of (12), also decreases. The decrease
in risk is not caused by a lower covariance between the stock and the managers
portfolio z
t
(rst term in the right-hand side). Indeed, because the adjustment
cost prevents the investor from adjusting instantaneously, z
t
remains constant
immediately following the increase in C
t
. The decrease in risk is instead driven
by the managers hedging demand (second term in the right-hand side), which
means that a stock covarying positively with the ow portfolio becomes a
better hedge for the manager when C
t
increases. The intuition is that when C
t
increases, mispricing becomes severe, and the manager has attractive invest-
ment opportunities. Hedging against a reduction in these opportunities requires
holding stocks that perform well when C
t
decreases, and these are the stocks
covarying positively with the ow portfolio. Holding such stocks guarantees
the manager an attractive long-horizon returnthe bird-in-the-hand effect.
The response of expected returns to changes in C
t
causes returns to
be predictable based on past returns. We characterize this predictability in
Corollary 6. As in the rest of our analysis, we evaluate returns over an
innitesimal time period dt . We compute the covariance between the vector of
returns at time t , i.e., between t and t +dt , and the same vector at time t
>t ,
that is, between t
and t
+dt .
Corollary 6 (return predictability). The covariance between stock returns at
time t and those at time t
>t is
Cov
t
(dR
t
,dR
) =
_
1
e
(t
t )
+
2
e
b
2
(t
t )
_
p
f
p
f
dt dt
, (25)
where (
1
,
2
) are constants. For small s, the term in the square bracket of
(25) is positive if t
t u, for a threshold u which

1103

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1104 10871145
is positive if <
R
and zero if >
R
. Astocks return predicts positively the
stocks subsequent return for t
t u (long-run reversal). It predicts in the same manner the subsequent

return of another stock when the covariance between each stock in the pair and
the ow portfolio has the same sign and in the opposite manner otherwise.
This autocovariance matrix of stock returns is equal to the nonfundamental
(contemporaneous) covariance matrix times a scalar, which is negative for
long lags but can be positive for short lags. Thus, stocks can exhibit positive
autocovariance for short lags and negative for long lags, that is, short-run
momentum and long-run reversal. This is because changes in C
t
can move
prices and short-run expected returns in the same direction, but long-run
expected returns in the opposite direction.
The nondiagonal elements of the autocovariance matrix characterize lead-lag
effects, that is, whether the past return of one stock predicts the future return of
another. Lead-lag effects have the same sign as autocovariance for stock pairs
whose covariance with the ow portfolio has the same sign. This is because
changes in C
t
inuence both stocks in the same manner.
Momentum arises when the parameter that characterizes the managers
career concerns is not too large (<
R
). Corollary 7 derives more generally
the effects of career concerns on stock prices.
Corollary 7 (career concerns). For small s, an increase in the parameter
that characterizes the managers career concerns
raises
1
and thus increases the nonfundamental volatility and covariance
of stock returns, and
lowers
1
+
2
and u and thus reduces the size of momentum and the
horizon over which it occurs.
A career-concerned manager seeks to hedge against outows. Hedging
requires holding a portfolio close to the market index because outows do
not affect the index price. Moreover, the managers demand to hedge against
outows becomes stronger following outows. This is because outows are
triggered by increases in the cost C
t
, which raise the managers perk and hence
his marginal benet from investor participation.
The increase in the managers hedging demand following outows
exacerbates stocks nonfundamental volatility and covariance and reduces
momentum. For example, an increase in C
t
causes stocks that the active fund
overweights to drop because of the anticipation of future outows (Corollary 3).
The drop is exacerbated because the increase in C
t
makes the manager more
willing to hedge against outows, and hence less willing to deviate from the
market index and overweight some stocks. This career concern effect also
works against the bird-in-the-hand effect, which causes momentum, because
the latter makes it attractive for the manager to increase his holdings of active
1104

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1105 10871145
overweights following increases in C
t
. When exceeds the threshold
R
, the
career concern effect dominates and momentum does not arise.
3. Asymmetric Information
In this section, we study the case of asymmetric information, where the investor
does not observe the cost C
t
and seeks to infer it from the returns and share
prices of the active and index funds. To prevent share prices fromfully revealing
C
t
, as they would under (6), we introduce time-variation in stocks expected
dividends. We replace (1) by
dD
t
=F
t
dt +dB
D
t
, (26)
and assume that F
t
follows the process
dF
t
=(

F F
t
)dt +dB
F
t
, (27)
where the mean-reversion parameter is the same as for C
t
for simplicity,

F
is a long-run mean, is a positive scalar, and B
F
t
is a d-dimensional Brownian
motion independent of (B
D
t
,B
C
t
). The diffusion matrices of D
t
and of F
t
are proportional for simplicity. We assume that only the manager observes F
t
,
so F
t
is reected in prices and prevents the investor from inferring C
t
. We set
f 1+
2
/(r +)
2
.
We look for an equilibriumwith the following characteristics. The investors
conditional distribution of C
t
is normal with mean

C
t
. The variance of the
conditional distribution is, in general, a deterministic function of time, but we
focus on the steady state reached for t , where it is constant. Stock prices
take the form
S
t
=
F
r
+
F
t

F
r +
(a
0
+a
1

C
t
+a
2
C
t
+a
3
y
t
), (28)
where (a
0
,a
1
,a
2
,a
3
) are constant vectors. The rst two terms are the present
value of expected dividends discounted at the riskless rate r, and the third
term is a risk discount linear in (
C
t
,C
t
,y
t
). We conjecture that the effects of
(
C
t
,C
t
,y
t
) depend on the covariance with the ow portfolio, as is the case for
(C
t
,y
t
) under symmetric information. That is, there exist constants (
1
,
2
,
3
)
such that for i =1,2,3,
a
i
=
i
p
f
. (29)
The rate v
t
dy
t
/dt at which the investor changes her active fund holdings in
our conjectured equilibrium is
v
t
=b
0
b
1

C
t
b
2
y
t
, (30)
where (b
0
,b
1
,b
2
) are constants. We refer to an equilibrium satisfying (28)(30)
as linear.
1105

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1106 10871145
3.1 Investors inference
The investor seeks to infer the cost C
t
from fund returns and share prices. She
observes the net-of-cost return z
t
dR
t
C
t
dt of the active fund, the return dR
t
of the index fund, the price z
t
S
t
of the active fund, and the price S
t
of the
index fund. Because she observes prices, she also observes capital gains, and
therefore can deduce net dividends (i.e., dividends minus C
t
).
In equilibrium, the active funds portfolio z
t
is equal to x
t
. Because the
investor knows x
t
, observing the price and net dividends of the active and index
funds is informationally equivalent to observing the price and net dividends of
the index fund and of a hypothetical fund holding the true market portfolio .
Therefore, we can take the investors information to be the net dividends of the
true market portfolio dD
t
C
t
dt , the dividends of the index fund dD
t
, the
price of the true market portfolio S
t
, and the price of the index fund S
t
.
12
We
solve the investors inference problem using recursive (Kalman) ltering.
Proposition 4. The mean

C
t
of the investors conditional distribution of C
t
evolves according to the process
d

C
t
=(
C

C
t
)dt
1
_
p
f
[dD
t
E
t
(dD
t
)](C
t

C
t
)dt
_
2
p
f
_
dS
t
+a
1
d

C
t
+a
3
dy
t
E
t
(dS
t
+a
1
d

C
t
+a
3
dy
t
)
_
, (31)
where
1
T
_
1(r +k)

2
, (32)
s
2
2
(r+)
2
+
s
2
2
2

, (33)
and T denotes the distributions steady-state variance. The variance T is the
unique positive solution of the quadratic equation
T
2
_
1(r +)

2
_
2
+2T
s
2
2
(r+)
2
2
(r+)
2
+
s
2
2
2

=0. (34)
The term in
1
in (31) represents the investors learning from net dividends.
Recalling the denition (17) of the ow portfolio, we can write this term as
1
_
dD
t
C
t
dt E
t
(dD
t
C
t
dt )
[dD
t
E
t
(dD
t
)]
_
. (35)
12
We are assuming that the investors information is the same in and out of equilibrium, that is, the manager cannot
manipulate the investors beliefs by deviating fromhis equilibriumstrategy and choosing a portfolio z
t
= x
t
.
This is consistent with the assumption of a competitive manager. Indeed, one interpretation of this assumption is
that there exists a continuum of managers, each with the same C
t
. A deviation by one manager would then not
affect the investors beliefs about C
t
because these would depend on averages across managers.
1106

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1107 10871145
The investor lowers her estimate of the cost, C
t
, if the net dividends of the true
market portfolio, dD
t
C
t
dt , are above expectations. Of course, net dividends
can be high not only because C
t
is low but also because gross dividends are
high. The investor adjusts for this by comparing with the dividends D
t
of the
index fund. The adjustment is made by computing the regression residual of
dD
t
C
t
dt on D
t
, which is the term in curly brackets in (35).
The term in
2
in (31) represents the investors learning from prices. The
investor lowers her estimate of C
t
if the price of the true market portfolio is
above expectations. Indeed, the price can be high because the manager knows
privately that C
t
is low, and anticipates that the investor will increase her
participation in the fund, causing the price to rise, as she learns about C
t
. As
with dividends, the investor needs to account for the fact that the price of the
true market portfolio can be high not only because C
t
is low but also because
the manager expects future dividends to be high (F
t
small). She adjusts for this
by comparing with the price of the index fund. Note that if F
t
is constant (=0),
learning from prices is perfect: (34) implies that the conditional variance T is
zero.
Because the investor compares the performance of the true market portfolio,
and hence of the active fund, to that of the index fund, she is effectively using
the index as a benchmark. Note that benchmarking is not part of an explicit
contract tying the managers compensation to the index. Compensation is tied to
the index only implicitly: If the active fund outperforms the index, the investor
infers that C
t
is low and increases her participation in the fund.
3.2 Optimization
The manager chooses controls ( c
t
, y
t
,z
t
) to maximize the expected utility (3)
subject to the budget constraint (8), the normalization (9), and the investors
holding policy (30). Because stock prices depend on (
C
t
,C
t
,y
t
), the same is
true for the managers value function. We conjecture that the value function is
V(W
t
,

X
t
) exp
_
_
r W
t
+ q
0
+( q
1
, q
2
, q
3
)
X
t
+
1
2
X
t
__
, (36)
where

X
t
(
C
t
,C
t
,y
t
)
, ( q
0
, q
1
, q
2
, q
3
) are constants, and

Q is a constant
symmetric 33 matrix.
( q
0
, q
1
, q
2
, q
3
,

Q) satisfy a system of ten scalar equations.
The investor chooses controls (c
t
,x
t
,v
t
) to maximize the expected utility
(2) subject to the budget constraint (13) and the managers portfolio policy
z
t
= x
t
. We study this optimization problem in two steps: rst optimize
over (c
t
,x
t
), assuming that v
t
is given by (30), and then derive conditions under
which (30) is optimal. We solve the rst problem using dynamic programming,
and conjecture the value function (14), where X
t
(
C
t
,y
t
)
, (q
0
,q
1
,q
2
) are
constants, and Q is a constant symmetric 22 matrix.
1107

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1108 10871145
(q
0
,q
1
,q
2
,Q) satisfy a system of six scalar equations. The control v
t
given by
(30) is optimal if (b
0
,b
1
,b
2
) satisfy a system of three scalar equations.
3.3 Equilibrium
Proposition 7 shows that a unique linear equilibrium exists when the diffusion
coefcient s of C
t
is small. Our numerical solutions for general values of s in
Section 4 seem to also generate a unique linear equilibrium, with properties
similar to those derived in the rest of this section for small s.
13
Proposition 7. For small s, there exists a unique linear equilibrium. The
constants (b
1
,b
2
,
1
,
2
) are positive and the constant
3
is negative.
Asymmetric information gives rise to amplication: Cash-ow shocks
trigger fund ows, which amplify the effects of cash-ow shocks on stock
returns. Suppose, for example, that a stock experiences a negative cash-ow
shock. If the stock is overweighted by the active fund, then the shock lowers
the return of the active fund more than of the index fund. As a consequence, the
investor infers that C
t
has increased, and ows out of the active and into the
index fund. Because the active fund overweights the stock, the investors ows
cause the stock to be sold and push its price down. Conversely, if the stock is
underweighted, then the investor infers that C
t
has decreased, and ows out of
the index and into the active fund. Because the active fund underweights the
stock, the investors ows again cause the stock to be sold and push down its
price.
Amplication is related to comovement. Recall that under symmetric
information, fund ows generate comovement between a pair of stocks because
they affect the expected return of each stock in the pair. This channel of
comovement, which we refer to as ER/ER (where ER stands for expected
return), is also present under asymmetric information. Asymmetric information
introduces an additional channel involving fund ows, which we refer to as
CF/ER (where CF stands for cash ow). This is that cash-ow news of one
stock in a pair triggers fund ows that affect the expected return of the other
stock. The CF/ER channel is related to amplication.
Although the ER/ER and CF/ER channels are conceptually distinct, they
have similar effects: The covariance matrix generated by CF/ER is equal to
that generated by ER/ER times a positive scalar. Thus, if ER/ER generates
positive covariance between a pair of stocks, so does CF/ER, and if the former
covariance is large, so is the latter. Consider, for example, two stocks that
the active fund overweights. Because outows from the active fund (triggered
13
This applies to b
1
>0, b
2
>0,
1
>0,
2
>0,
3
<0, and to Corollaries 811. The only exception is that for large
values of s, the short-run momentum derived in Corollary 11 for all values of arises only when is not too
large.
1108

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1109 10871145
by, e.g., a cash-ow shock to a third stock) push down the prices of both
stocks, ER/ER generates positive covariance. Moreover, because a negative
cash-owshock to one stock triggers outows fromthe active fund, which push
down the price of the other stock, CF/ER also generates positive covariance.
The former covariance is large if the two stocks have high idiosyncratic risk
because this makes them more sensitive to fund ows. But, high idiosyncratic
risk also renders the latter covariance large: cash-ow shocks to stocks having
low correlation with the index generate a large discrepancy between the active
and the index return, hence triggering large fund ows.
The proportionality of the covariance matrices generated by ER/ER and
CF/ER also implies proportionality of the nonfundamental covariance matrix
under asymmetric information (ER/ER and CF/ER) to that under symmetric
information (ER/ER). The proportionality coefcient in the latter relationship is
larger than one for small s. Thus, the nonfundamental volatility of each stock is
larger under asymmetric information, and so is the absolute value of the nonfun-
damental covariance between any pair of stocks. This is because of the ampli-
cation channel CF/ER, which is present only under asymmetric information.
Corollary 8 (comovement and amplication). The covariance matrix of
stock returns is
Cov
t
(dR
t
,dR
t
) =
_
f +kp
f
p
f
_
dt, (37)
where k is a positive constant. The fundamental covariance f dt is driven
purely by cash ows and is equal to its symmetric-information counterpart.
14
The nonfundamental covariance kp
f
p
f
dt is introduced by fund ows and
is equal to its symmetric-information counterpart times a positive scalar, which
is larger than one for small s.
Stocks expected returns are determined by the covariance with the market
index and the ow portfolio, as under symmetric information. Moreover, the
risk premium
t
of the ow portfolio is time-varying, and its variation is
closely related to momentum and reversal. Consider, for example, a cash-ow
shock that is negative and hits a stock that the active fund overweights. The
shock raises

C
t
, the investors estimate of C
t
. This lowers the time-t prices and
expected returns of stocks covarying positively with the ow portfolio (
1
>0
and
R
1
<0) and raises their expected returns in the long run.
Corollary 9 (expected returns). Stocks expected returns are given by the
two-factor model (22), with the factors being the market index and the ow
portfolio. The factor risk premium
t
associated to the ow portfolio is
t
=r +
1
f +
k
R
1

C
t
+
R
2
C
t
+
R
3
y
t
k
1
q
1
k
2
q
2
_
, (38)
14
The fundamental covariance dt in Corollary 4 needs to be multiplied by f to account for the effects of F
t
. See
the proof of the corollary.
1109

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1110 10871145
where (
R
1
,
R
2
,
R
3
,k
1
,k
2
) are constants. For small s, the constants (
R
1
,
R
3
) are
negative and the constant
R
2
is positive.
The response of expected returns to cash-ow shocks causes returns to
be predictable based on these shocks. We characterize this predictability in
Corollary 10, where we compute the covariance between the vectors (dD
t
,dF
t
)
of cash-ow shocks at time t and the vector of returns at time t
>t . Both
covariance matrices are equal to the nonfundamental covariance matrix times
a scalar that is positive for short lags and negative for long lags. Thus, cash-ow
shocks generate short-runmomentumandlong-runreversal inreturns. Note that
predictability based on cash ows arises only under asymmetric information
because only then will cash-ow shocks trigger fund ows.
Corollary 10 (return predictability based on cashows). The covariance
between cashow shocks (dD
t
,dF
t
) at time t and stock returns at time t
>t
is given by
Cov
t
(dD
t
,dR
) =

1
(r +)Cov
t
(dF
t
,dR
2
=
_
D
1
e
(+)(t
t )
+
D
2
e
b
2
(t
t )
_
p
f
p
f
dt dt
,
(39)
where (
D
1
,
D
2
,) are constants. For small s, the term in the square bracket of
t u
D
, for a threshold u
D
>0.
A stocks cash-ow shocks predict positively the stocks subsequent return
for t
t u
D
(long-run
reversal). They predict in the same manner the subsequent return of another
stock when the covariance between each stock in the pair and the owportfolio
has the same sign and in the opposite manner otherwise.
We nallyexamine predictabilitybasedonpast returns rather thancashows.
This predictability is driven both by cash-ow shocks and by shocks to C
t
and
has the same form as under symmetric information.
15
Corollary 11 (return predictability). The covariance between stock returns
at time t and those at time t
>t is
Cov
t
(dR
t
,dR
) =
_
1
e
(+)(t
t )
+
2
e
(t
t )
+
3
e
b
2
(t
t )
_
p
f
p
f
dt dt
,
(40)
where (
1
,
2
,
3
) are constants. For small s, the term in the square bracket of
t u for a threshold u>0. Given

u, predictability is as in Corollary 6.
15
The only difference is that short-run momentum arises for all values of . This result, however, relies on the
assumption that s is small; for large values of s, short-run momentum arises only when is not too large, as
under symmetric information.
1110

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1111 10871145
4. Momentum and Value Strategies
In this section we compute the magnitude of momentum and value effects
that our model generates. The magnitude of these effects is typically measured
through the performance of trading strategies. We rst construct a measure of
performance and then compute it in a calibration of our model.
4.1 Performance measure
Consider a trading strategy consisting of a vector of weights w
t
(w
1t
,..,w
Nt
),
where w
nt
is the number of shares invested in stock n at time t . Part of the
strategys expected return is compensation for bearing risk that correlates with
the market index. We focus on the remainder by index-adjusting the strategy,
that is, combining it with a position in the index such that the covariance
between the overall position and the index is zero. The index-adjusted
strategy is
w
t
w
t
Cov
t
(w
t
dR
t
,dR
t
)
Var
t
(dR
t
)
. (41)
Note that the position in the index can be time-varying, reecting possible time-
variation in the covariance between the strategy and the index. We measure the
performance of the strategy w
t
by the Sharpe ratio of its index-adjusted version
w
t
.
16
The Sharpe ratio is the ratio of expected return to standard deviation.
We also divide by
dt to express the Sharpe ratio in annualized terms, given

that returns are evaluated over an innitesimal period dt . The Sharpe ratio
corresponding to the strategy w
t
is thus
SR
w
E( w
t
dR
t
)
_
Var( w
t
dR
t
)dt
. (42)
Proposition 8 computes the Sharpe ratio under the prices in the asymmetric-
information equilibriumof Section 3 and in the steady state reached for t .
It also determines the strategy maximizing the Sharpe ratio.
Proposition 8. The Sharpe ratio corresponding to the strategy w
t
is
SR
w
=
_
f +
k
_
E
_
t
w
t
p
f
_
_
f
_
E(w
t
w
t
)
E[(w
t
)
2
]
_
+kE[(w
t
p
f
)
2
]
, (43)
and is maximized for w
t
=
t
p
f
.
16
Empirical studies that compute Sharpe ratios of momentum and value strategies typically consider long-short
portfolios with zero initial investment; that is, they require the dollar weights to sumto zero. Our index-adjustment
is in a similar spirit: The weights that sum to zero are the number of shares times the covariance between one
share and the index rather than times the dollar value of one share. We dene weights differently because this
preserves linearity and simplies the algebra.
1111

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1112 10871145
The intuition for the optimal strategy can be derived from the two-factor
model of Corollary 9. Astrategys expected return consists of a compensation
for bearing risk that correlates with the index, and a compensation for bearing
risk that correlates with the ow portfolio. Index adjustment isolates the latter
component. Maximizing that component per unit of risk requires holding
the ow portfolio because this eliminates uncompensated risk. Moreover,
investment in the ow portfolio should be larger when the premium
t
associated to that risk factor is high. Because time-variation in
t
is caused
by fund ows, past and anticipated, the optimal strategy effectively exploits
mispricing generated by ows.
Momentum and value strategies exploit aspects of the ow-generated
mispricing, and are imperfect approximations of the optimal strategy. We
consider the following simple implementation of cross-sectional momentum
and value strategies:
_
w
M
t
_
_
t
t
dR
u
, (44)
_
w
V
t
_
F
r
+
F
t

F
r +
S
t
, (45)
respectively. A stocks momentum weight increases linearly in the stocks
cumulative past return over the window [t ,t ] for some >0. A stocks
value weight increases linearlyinthe difference betweenthe present value of the
stocks expected dividends discounted at the riskless rate, and the stocks price.
Substituting (44) and (45) into (43) yields closed-form solutions for the
Sharpe ratios of momentum and value strategies. We omit these calculations
because of space constraints. The calculations are inVayanos andWoolley(VW;
2012), who also compute closed-form Sharpe ratios achieved by alternative
implementations of momentum and value strategies, by optimal combinations
of these strategies, and for general investment horizons.
4.2 Calibration of model parameters
We set the riskless rate r to 4%. We assume that there are N=10 stocks, which
we interpret as industry sectors. We assume that the market index includes
one share of each stock, that is, =1, where 1(1,..,1), and that the true market
portfolioincludes one share of eachstockonaverage, that is,

N
n=1
n
/N=1.
These are normalizations because we can redene one share of each stock
and of the index, leaving Sharpe ratios unchanged. We assume that stocks
are symmetric in the sense that they all have the same standard deviation of
dividends and the same pairwise correlations. (Our closed-form solutions for
Sharpe ratios, however, do not require any symmetry.) Hence, the covariance
matrix of dividends is =
2
(I +1
1), where I is the identity matrix and ( ,)

are scalars. We calibrate using the Sharpe ratio SR
of the market index .

Closed-formsolutions for SR
and for all other quantities used in the calibration

1112

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1113 10871145
are in VW. We express SR
in annualized terms and set it to 30%. This is equal

to the Sharpe ratio of the S&P500 index, assuming an annual expected excess
return of 4.5% and a standard deviation of 15%. The implied value of is
0.22.
17
We calibrate using the correlation between industry sectors and the
market. Ang and Chen (2002) nd that the average correlation between the
returns of an industry sector and of a broad market index is 87% across the
thirteen sectors that they consider. The implied value of is seven. We set
to 0.3. This parameter determines the size of shocks to the expected dividend
rate F
t
relative to dividends D
t
and has small effects on our calibration results.
VWshowthat the only characteristic of the true market portfolio that affects
Sharpe ratios when stocks are symmetric is ()
_
N
n=1
(
n
)
2
. This is the
standard deviation across stocks of the number of shares included in and must
be strictly positive so that differs from the market index . We calibrate ()
using the average deviation between the weight that an active fund gives to an
industry sector and the sectors weight in a broad market index. Kacperczyk,
Sialm, and Zheng (KSZ1; 2005) nd that the sum of squared deviations across
the ten sectors that they consider is 4.36% for the median fund, implying an
average deviation of 6.6% (106.6%
2
=4.36%). To map this into a value for
(), we adjust for the fact that is the sumof active- and index-fund holdings.
The holdings of active funds are about ten times those of index funds in KSZ1s
sample period, so the average deviation for a combined active and index fund
(which is what represents) is 6%. The implied value of () is 0.6.
To calibrate the diffusion coefcient s of the cost C
t
, we recall the costs
interpretation as minus the return gap. Kacperczyk, Sialm, and Zheng (KSZ;
2008) nd that the top decile of mutual funds in terms of lagged one-year return
gap earns a monthly CAPM alpha of 0.273%, whereas the bottom decile earns
0.431%.
18
Because in our model there is only one active fund, we interpret the
differential between deciles in a time-series rather than a cross-sectional sense.
The implied value of s is 1.6. We set the persistence parameter of the cost to
0.3. This is consistent with KSZs nding that shocks to the return gap shrink to
about one-thirdof their size withinfour years (log(3)/0.3=3.7). We set the long-
run mean

C of the cost to zero, consistent with KSZs nding that the average
return gap in the cross-section is zero. With

C=0, negative values of the cost are
equallylikelyas positive values, whichmeans that the cost cannot be interpreted
solely as a managerial perk. Hence, we emphasize again the managerial-ability
interpretation, and for consistency we set the parameter to zero.
We calibrate the adjustment-cost parameter using the empirical response
of fund ows to performance. Coval and Stafford (2007) nd that a positive
shock to a funds return generates ows into the fund in each of the next four
17
Note that is a volatility per share rather than per dollar because this is how returns are expressed in our model.
We calibrate using Sharpe ratios because these are comparable for per-share and per-dollar returns.
18
KSZ derive two different sets of estimates; we focus on those derived using a back-testing procedure that reduces
estimation noise.
1113

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1114 10871145
quarters, with the effect dying off in the fth. We set =1.2, which ensures that
following a positive shock to the active funds return, the investors holdings
y
t
in the fund increase in the next four quarters and start decreasing afterwards.
We set the investors coefcient of absolute risk aversion to one. This is
a normalization because we can redene the units of the consumption good,
leaving Sharpe ratios unchanged. To calibrate the risk aversion of the manager,
we recall that he can be interpreted as an aggregate of all smart-money agents
with the expertise to exploit mispricings. We are interested in the capital that
these experts own, rather than in the capital they might manage on behalf of
outsiders, because only the former can be used to exploit mispricings generated
by outsiders ows. Because most of the nancial expertise lies within the
nancial industry, the capital of experts can be linked to that industrys GDP
share. Philippon (2008) reports that the GDPshare of the nance and insurance
industry was 5.5% on average during 19602007 in the United States. We
view this as an upper bound because only part of that industry concerns asset
markets, and we set the managers coefcient of absolute risk aversion to 30.
This means that the manager accounts for 3.2%(=1/(30+1)) of aggregate risk
tolerance.
19
As an independent check for our choices of s and , we compute two
additional quantities: the turnover and the return variance generated by fund
ows. Lou (2012) nds that the standard deviation of a stocks quarterly
turnover generated by fund ows is 0.7%. Because the funds in Lous sample
account for about 10% of market capitalization, the standard deviation of
ow-generated volume is 7% of assets managed by these funds; we nd
7.8%. Greenwood and Thesmar (GT; 2011) nd that fund ows explain 8%
of stock-return variance; we nd 16%. GTs sample, however, includes less
than half of all professionally managed wealth. Accounting for that, and for
possible measurement noise, could well produce a number even larger than
16%. Raising s and to match such a larger number would raise the Sharpe
ratios of momentum and value strategies that we nd in the next section.
4.3 Calibration results
The maximum Sharpe ratio across all strategies (Proposition 8) is 61%. The
value strategy (45) achieves a Sharpe ratio of 26%. Figure 1 plots the Sharpe
ratio of the momentumstrategy (44) as a function of the length of the window
over which past returns are calculated. This Sharpe ratio is positive for windows
of less than three years and then turns negative. Thus, a strategy based on
short-run momentum is protable, and so is one based on long-run reversal.
The highest Sharpe ratio of momentum is achieved using a window of four
months, and is 40%. Moreover, windows from one to eleven months yield
Sharpe ratios larger than 30%, the ratio of the market index. The Sharpe ratio
19
Risk tolerance in our model is independent of capital because of exponential utility. Our choice of is based on
the notion that risk tolerance is proportional to capital, which is true under power utility.
1114

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1115 10871145
0 1 2 3 4 5 6 7 8 9 10
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
tau (years)
S
R
Figure 1
Sharpe ratio of the momentum strategy (44) as a function of the length of the window over which past
returns are calculated
The windowlengthis measuredinyears. The parameters for whichthe gure is drawnare describedinSection 4.2.
of momentum converges to zero as the window length goes to zero because
very recent performance is a very noisy signal of future fund ows.
VWdecompose momentumprots into three sources. The rst is the positive
short-run response of expected returns to shocks (Corollaries 6 and 11): Stocks
hit by positive shocks receive high weight in the momentum strategy, and are
expected to do well in the short run. The second is the time-series variation of
expected returns (regardless of how these respond to shocks): Stocks whose
conditional expected returns are higher than their unconditional averages
receive high expected weight in the momentum strategy, and are expected
do well in the short run because conditional expected returns are persistent.
The third, emphasized by Conrad and Kaul (1998), is the cross-sectional
variation of unconditional expected returns: Stocks with high unconditional
expected returns receive high expected weight in the momentum strategy and
are expected to do well going forward. The rst source of prots is dominant
in our calibration: For example, 62% of the maximum Sharpe ratio in Figure 1
is generated by the rst source, 36% by the second, and 2% by the third.
Asness, Moskowitz, and Pedersen (AMP; 2013) present evidence on Sharpe
ratios of momentum and value strategies across a variety of markets and asset
classes. Strategies exploiting momentum of individual stocks within a market
yield an average Sharpe ratio of 70%across the four markets that AMPconsider
(the United States, the United Kingdom, Continental Europe, and Japan), and
strategies exploiting value yield 36%. Strategies exploiting momentum of
country-level stockindices yield34%, andsodostrategies exploitingvalue. Our
Sharpe ratios are somewhat smaller on average. This is not surprising because
1115

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1116 10871145
we focus only on ows between investment funds and ignore other types of
ows, for example, those generated by individuals holding stocks directly. Such
ows would likely increase the Sharpe ratios relative to our calibration. Our
calibration shows, however, that even by focusing only on institutional ows
and restricting parameters based on evidence from the mutual-fund literature,
we can generate sizeable Sharpe ratios.
5. Conclusion
We propose a theory of momentum and reversal based on delegated portfolio
management. Flows between investment funds are triggered by changes in
fund managers efciency, which investors either observe directly or infer
from past performance. Momentum arises if fund ows exhibit inertia, and
because rational prices do not fully adjust to reect future ows. Reversal arises
because ows push prices away from fundamental values. Besides momentum
and reversal, ows generate comovement, lead-lag effects, and amplication,
with these being larger for high-idiosyncratic-risk assets. Although our model
focuses on institutional ows, the mechanisms that we uncover are broader and
can apply to other types of ows, for example, those generated by individuals
holding stocks directly.
Our models implications for return predictability map naturally into
implications for active portfolio management. We sketch some of these
implications in Section 4, where we compute the Sharpe ratios of momentum
and value strategies. Much more can be done, however, and in a tractable
closed-form manner. For example, although we evaluate momentum and value
strategies inisolation, these canbe combined. Asness, Moskowitz, andPedersen
(2013) nd empirically a negative correlation between these strategies, and
hence large gains in combining them. Our model also yields a negative
correlation, as shown in Vayanos and Woolley (VW; 2012). Perhaps more
intriguing is the Sharpe ratio of the optimal momentum-value combination,
which is signicantly smaller than the maximum possible Sharpe ratio. Thus,
momentumand value strategies can be improved, possibly by using information
on fund ows. Finally, although momentum delivers a higher Sharpe ratio than
value over a short horizon, VW show that the comparison reverses over a long
horizon. Hence, the portfolio allocation of a long-horizon investor between
momentum and value can be very different than of a short-horizon investor.
Appendix
A. Symmetric Information
Proof of Proposition 1. Equations (1), (4), (6), and (7) imply that the vector of returns is
dR
t
=dD
t
+dS
t
rS
t
dt
=
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
dt +dB
D
t
sa
1
dB
C
t
, (A1)
1116

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1117 10871145
where
a
R
1
(r +)a
1
+b
1
a
2
,
a
R
2
(r +b
2
)a
2
.
Equations (4), (7), (8), and (A1) imply that
d
_
r W
t
+ q
0
+( q
1
, q
2
)
X
t
+
1
2
t

Q
X
t
_
=

Gdt +r z
t
dB
D
t
s
_
r z
t
a
1

f
1
(
X
t
)
_
dB
C
t
, (A2)
where
Gr
_
rW
t
+ z
t
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
+C
t
y
t
c
t
_
+

f
1
(
X
t
)(
CC
t
)+

f
2
(
X
t
)v
t
+
1
2
s
2
q
11
,
f
1
(
X
t
) q
1
+ q
11
C
t
+ q
12
y
t
,
f
2
(
X
t
) q
2
+ q
12
C
t
+ q
22
y
t
,
and q
ij
denotes the (i,j)th element of

Q. Equations (10) and (A2) imply that
D
V =
V
_
G
1
2
(r )
2
z
t
z
1
2
s
2
_
r z
t
a
1

f
1
(
X
t
)
_
2
_
. (A3)
Substituting (A3) into (11), we can write the rst-order conditions with respect to c
t
and z
t
as
exp( c
t
)+r

V =0, (A4)
h(
X
t
) =r (+s
2
a
1
a
1
) z
t
, (A5)
respectively, where
h(
X
t
) ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
+s
2
a
1

f
1
(
X
t
). (A6)
Equation (A5) is equivalent to (12) because of (4), (A1), and
Cov
t
(dR
t
,dR
t
) =
_
+s
2
a
1
a
1
_
dt. (A7)
(Equation (A7) follows from (A1).) Using (A3) and (A4), we can simplify (11) to
G
1
2
(r )
2
z
t
(+s
2
a
1
a
1
) z
t
+r s
2
z
t
a
1

f
1
(
X
t
)
1
2
s
2

f
1
(
X
t
)
2
+

r =0. (A8)
Equations (10) and (A4) imply that
c
t
=rW
t
+
1

_
q
0
+( q
1
, q
2
)
X
t
+
1
2
t

Q
X
t
log(r)
_
. (A9)
Substituting (A9) into (A8) the terms in W
t
cancel, and we are left with
r z
t
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
+r C
t
y
t
r
_
q
0
+( q
1
, q
2
)
X
t
+
1
2
t

Q
X
t
_
+

f
1
(
X
t
)(
CC
t
)+

f
2
(
X
t
)v
t
+
1
2
s
2
q
11
+

r +r log(r)
1
2
(r )
2
z
t
(+s
2
a
1
a
1
) z
t
+r s
2
z
t
a
1

f
1
(
X
t
)
1
2
s
2

f
1
(
X
t
)
2
=0. (A10)
1117

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1118 10871145
The terms in (A10) that involve z
t
can be written as
r z
t
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
1
2
(r )
2
z
t
(+s
2
a
1
a
1
) z
t
+r s
2
z
t
a
1

f
1
(
X
t
)
=r z
t
h(
X
t
)
1
2
(r )
2
z
t
(+s
2
a
1
a
1
) z
t
=
1
2
r z
t
h(
X
t
)
=
1
2
h(
X
t
)
(+s
2
a
1
a
1
)
1
h(
X
t
), (A11)
where the rst step follows from (A6) and the last two from (A5). Substituting (A11) into (A10),
we nd
1
2
h(
X
t
)
(+s
2
a
1
a
1
)
1
h(
X
t
)+r C
t
y
t
r
_
q
0
+( q
1
, q
2
)
X
t
+
1
2
t

Q
X
t
_
+

f
1
(
X
t
)(
CC
t
)+

f
2
(
X
t
)v
t
+
1
2
s
2
_
q
11

f
1
(
X
t
)
2
_
+

r +r log(r) =0. (A12)
Equation (A12) is quadratic in

X
t
. Identifying quadratic, linear, and constant terms yields six
scalar equations in ( q
0
, q
1
, q
2
,

Q). We defer the derivation of these equations until the proof of
Proposition 3 (see (A48)(A50)).
Proof of Proposition 2. Suppose that the investor optimizes over (c
t
,x
t
) but follows the control
v
t
given by (7). Equations (4), (7), (13), (14), and (A1) imply that
d
_
rW
t
+q
0
+q
1
C
t
+
1
2
q
11
C
2
t
_
=Gdt +r(x
t
+y
t
z
t
)dB
D
t
s[r(x
t
+y
t
z
t
)a
1
f
1
(X
t
)]dB
C
t
,
(A13)
where
Gr
_
rW
t
+(x
t
+y
t
z
t
)
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
y
t
C
t
1
2
v
2
t
c
t
_
+f
1
(X
t
)(
CC
t
)+f
2
(X
t
)v
t
+
1
2
s
2
q
11
,
f
1
(X
t
) q
1
+q
11
C
t
+q
12
y
t
,
f
2
(X
t
) q
2
+q
12
C
t
+q
22
y
t
,
and q
ij
denotes the (i,j)th element of Q. Equations (14) and (A13) imply that
DV =V
_
G
1
2
(r)
2
(x
t
+y
t
z
t
)(x
t
+y
t
z
t
)
1
2
s
2
[r(x
t
+y
t
z
t
)a
1
f
1
(X
t
)]
2
_
. (A14)
Substituting (A14) into (15), we can write the rst-order conditions with respect to c
t
and x
t
as
exp(c
t
)+rV =0, (A15)
h(X
t
) =r(+s
2
a
1
a
1
)(x
t
+y
t
z
t
)
, (A16)
respectively, where
h(X
t
) ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
+s
2
a
1
f
1
(X
t
). (A17)
1118

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1119 10871145
Solving for c
t
, and proceeding as in the proof of Proposition 1, we can simplify (15) to
r(x
t
+y
t
z
t
)
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
ry
t
C
t
1
2
rv
2
t
r
_
q
0
+(q
1
,q
2
)X
t
+
1
2
X
t
QX
t
_
+f
1
(X
t
)(
CC
t
)+f
2
(X
t
)v
t
+
1
2
s
2
q
11
+r +r log(r)
1
2
(r)
2
(x
t
+y
t
z
t
)(+s
2
a
1
a
1
)(x
t
+y
t
z
t
)
+rs
2
(x
t
+y
t
z
t
)a
1
f
1
(X
t
)
1
2
s
2
f
1
(X
t
)
2
=0.
(A18)
The terms in (A18) that involve x
t
+y
t
z
t
can be written as
r(x
t
+y
t
z
t
)
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
1
2
(r)
2
(x
t
+y
t
z
t
)(+s
2
a
1
a
1
)(x
t
+y
t
z
t
)
+rs
2
(x
t
+y
t
z
t
)a
1
f
1
(X
t
)
=r(x
t
+y
t
z
t
)h(X
t
)
1
2
(r)
2
(x
t
+y
t
z
t
)(+s
2
a
1
a
1
)(x
t
+y
t
z
t
)
=ry
t
h(X
t
)
1
2
(r)
2
y
2
t
(+s
2
a
1
a
1
)
+rx
t
(1y
t
)
_
h(X
t
)
1
2
r(+s
2
a
1
a
1
)[x
t
(1y
t
)+2y
t
]
, (A19)
where the rst step follows from (A17) and the second from the equilibrium condition z
t
= x
t
.
Using z
t
= x
t
, we can write (A16) as
h(X
t
) =r(+s
2
a
1
a
1
)[x
t
(1y
t
)+y
t
]
x
t
(1y
t
) =
h(X
t
)ry
t
(+s
2
a
1
a
1
)
r(+s
2
a
1
a
1
)
. (A20)
Equation (A20) implies that
rx
t
(1y
t
)
_
h(X
t
)
1
2
r(+s
2
a
1
a
1
)[x
t
(1y
t
)+2y
t
]
=
1
2
[rx
t
(1y
t
)]
2
(+s
2
a
1
a
1
)
=
1
2
_
h(X
t
)ry
t
(+s
2
a
1
a
1
)
_
2
(+s
2
a
1
a
1
)
. (A21)
Substituting (A19) and (A21) into (A18), we nd
ry
t
h(X
t
)
1
2
(r)
2
y
2
t
(+s
2
a
1
a
1
)
+
1
2
_
h(X
t
)ry
t
(+s
2
a
1
a
1
)
_
2
(+s
2
a
1
a
1
)
ry
t
C
t
1
2
rv
2
t
r
_
q
0
+(q
1
,q
2
)X
t
+
1
2
X
t
QX
t
_
+f
1
(X
t
)(
CC
t
)+f
2
(X
t
)v
t
+
1
2
s
2
_
q
11
f
1
(X
t
)
2
_
+r +r log(r) =0. (A22)
Because v
t
in (7) is linear in X
t
, (A22) is quadratic in X
t
. Identifying quadratic, linear, and constant
terms yields six scalar equations in (q
0
,q
1
,q
2
,Q). We defer the derivation of these equations until
the proof of Proposition 3 (see (A52)(A54)).
1119

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1120 10871145
We next study optimization over v
t
, and derive a rst-order condition under which the control
(7) is optimal. We use a perturbation argument, which consists in assuming that the investor follows
the control (7), except for an innitesimal deviation over an innitesimal internal.
20
Suppose that
the investor adds d to the control (7) over the interval [t,t +d] and subtracts d over the
interval [t +dt d,t +dt ], where the innitesimal d >0 is o(dt ). The increase in adjustment cost
over the rst interval is v
t
(d)
2
and over the second interval is v
t +dt
(d)
2
. These changes
reduce the investors wealth at time t +dt by
v
t
(d)
2
(1+rdt )v
t +dt
(d)
2
=(d)
2
(rv
t
dt dv
t
) =(d)
2
(rv
t
dt +b
1
dC
t
+b
2
dy
t
)
=(d)
2
_
(r +b
2
)v
t
dt +b
1
_
(
CC
t
)dt +sdB
C
t
__
, (A23)
where the second step follows from (7) and the third from (4). The change in the investors wealth
between t and t +dt is derived from (13) and (A1), by subtracting (A23) and replacing y
t
by
y
t
+(d)
2
:
dW
t
=G
dt (d)
2
b
1
_
(
CC
t
)dt +sdB
C
t
_
+
_
x
t
+
_
y
t
+(d)
2
_
z
t
_
_
dB
D
t
sa
1
dB
C
t
_
,
(A24)
where
G
rW
t
+
_
x
t
+
_
y
t
+(d)
2
_
z
t
_
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
_
y
t
+(d)
2
_
C
t
v
2
t
2
c
t
(d)
2
(r +b
2
)v
t
.
The investors position in the active fund at t +dt is the same under the deviation as under no
deviation. Therefore, the investors expected utility at t +dt is given by the value function (14)
with the wealth W
t +dt
determined by (A24). The drift DV corresponding to the change in the value
function between t and t +dt is given by the following counterpart of (A14):
DV =V
_
G
1
2
(r)
2
_
x
t
+
_
y
t
+(d)
2
_
z
t
_
_
x
t
+
_
y
t
+(d)
2
_
z
t
_
1
2
s
2
_
r
_
x
t
+
_
y
t
+(d)
2
_
z
t
_
a
1
f
1
(X
t
)
_
2
_
, (A25)
where
GrG
+f
1
(X
t
)(
CC
t
)+f
2
(X
t
)v
t
+
1
2
s
2
q
11
,
f
1
(X
t
) f
1
(X
t
)r(d)
2
b
1
.
The drift is maximum for =0, and this yields the rst-order condition
z
t
h(X
t
)rb
1
s
2
(x
t
+y
t
z
t
)a
1
C
t
=rz
t
(+s
2
a
1
a
1
)(x
t
+y
t
z
t
)
+h
(X
t
), (A26)
where
h
(X
t
) (r +b
2
)v
t
+b
1
(
CC
t
)b
1
s
2
f
1
(X
t
).
20
The perturbation argument is simpler than the dynamic programming approach, which assumes that the investor
can follow any control v
t
over the entire history. Indeed, under the dynamic programming approach, the state
variable y
t
, which describes the investors holdings in the active fund, must be replaced by two state variables:
the holdings out of equilibrium and the holdings in equilibrium. This is because the latter affects the equilibrium
price, which the investor takes as given.
1120

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1121 10871145
Using (A16) and the equilibrium condition z
t
= x
t
h(X
t
)rb
1
s
2
[x
t
(1y
t
)+y
t
]a
1
C
t
=r(+s
2
a
1
a
1
)[x
t
(1y
t
)+y
t
]
+h
(X
t
).
(A27)
Using (A20), we can write (A27) as
_
h(X
t
)rb
1
s
2
y
t
a
1
_
rb
1
s
2
h(X
t
)ry
t
(+s
2
a
1
a
1
)
r(+s
2
a
1
a
1
)
a
1
C
t
=r(+s
2
a
1
a
1
)
_
y
t
+
h(X
t
)ry
t
(+s
2
a
1
a
1
)
r(+s
2
a
1
a
1
)
+h
(X
t
). (A28)
Equation (A28) is linear in X
t
. Identifying linear and constant terms, yields three scalar equations
in (b
0
,b
1
,b
2
). We defer the derivation of these equations until the proof of Proposition 3 (see
(A43)(A45)).
Proof of Proposition 3. We rst impose market clearing and derive the constants
(a
0
,a
1
,a
2
,b
0
,b
1
,b
2
) as functions of ( q
1
, q
2
,

Q,q
1
,q
2
,Q). For these derivations, as well as for
many later proofs, we use the following important properties of the ow portfolio:
p
f
=0,
p
f
=p
f
p
f
=

.
Setting z
t
= x
t
and y
t
=1y
t
h(
X
t
) =r (+s
2
a
1
a
1
)(1y
t
)( x
t
)
. (A29)
Premultiplying (A29) by , dividing by r , and adding to (A20) divided by r, we nd
_
h(X
t
)
r
+
h(
X
t
)
r
_
=(+s
2
a
1
a
1
)
. (A30)
Equation (A30) is linear in (C
t
,y
t
). Identifying terms in C
t
and y
t
, we nd
_
r + +s
2
q
11
r
+
r + +s
2
q
11
r
_
a
1
+
b
1
(+ )
r
a
2
=0, (A31)
_
s
2
q
12
r
+
s
2
q
12
r
_
a
1
+
(r +b
2
)(+ )
r
a
2
=0, (A32)
respectively. Equations (A31) and (A32) imply that
a
1
=a
2
=0. (A33)
Identifying constant terms in (A30), and using (A33), we nd
a
0
=

+
. (A34)
Substituting (A34) and (A33) into (A20), we nd
r
+
=r[x
t
(1y
t
)+y
t
]
x
t
=

+
y
t
1y
t
. (A35)
Substituting (A35) into (A29), we nd
h(
X
t
) =r (+s
2
a
1
a
1
)
_

+
+(1y
t
)p
f
_
. (A36)
1121

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1122 10871145
Equation (A36) is linear in

X
t
. Identifying terms in C
t
and y
t
, we nd
_
r + +s
2
q
11
_
a
1
+b
1
a
2
=0, (A37)
s
2
q
12
a
1
+(r +b
2
)a
2
=r
_
p
f
+s
2
a
1
p
f
a
1
_
, (A38)
respectively. Therefore, (a
1
,a
2
) are collinear to the vector p
f
, as in (16). Substituting (16) into
(A37) and (A38), we nd
_
r + +s
2
q
11
_
1
+b
1
2
=0, (A39)
s
2
q
12
1
+(r +b
2
)
2
=r
_
1+
s
2
2
1

_
, (A40)
respectively. Identifying constant terms in (A36), and using (16), we nd
a
0
=

+
+
_
1
(

Cs
2
q
1
)+b
0
2
r
+
_
1+
s
2
2
1

__
p
f
. (A41)
Using (A35), we can write (A27) as
h(X
t
)rb
1
s
2
_

+
+y
t
p
f
_
a
1
C
t
=r(+s
2
a
1
a
1
)
_

+
+y
t
p
f
_
+h
(X
t
)
h(X
t
)rb
1
s
2
y
t
C
t
=
r
+
(
)
2
+r
_
1+
s
2
2
1

y
t
+h
(X
t
),
(A42)
where the second step follows from (16). Equation (A42) is linear in (C
t
,y
t
). Identifying terms in
C
t
and y
t
, and using (7) and (16), we nd
_
(r + +s
2
q
11
)
1
+b
1
2
_

1=b
1
(r + +b
2
+s
2
q
11
), (A43)
_
(r +b
2
)
2
+(q
12
rb
1
)s
2
1
_

=r
_
1+
s
2
2
1

_
(r +b
2
)b
2
+b
1
s
2
q
12
_
,
(A44)
respectively. Identifying constant terms, and using (7), (16), and (A41), we nd
_
s
2
1
(q
1
q
1
)+r
_
1+
s
2
2
1

__

=
_
(r +b
2
)b
0
+b
1
(

Cs
2
q
1
)
_
. (A45)
The system of equations characterizing equilibrium is as follows. The endogenous variables are
(a
0
,a
1
,a
2
,b
0
,b
1
,b
2
,
1
,
2
, q
0
, q
1
, q
2
,

Q,q
0
,q
1
,q
2
,Q). The equations linking themare (16), (A39)
(A41), (A43)(A45), the six equations derived from (A12) by identifying quadratic, linear, and
constant terms, and the six equations derived from (A22) through the same procedure. To simplify
the system, we note that the variables ( q
0
,q
0
) enter only in the equations derived from (A12)
and (A22) by identifying constants. Therefore, they can be determined separately, and we need
to consider only the equations derived from (A12) and (A22) by identifying linear and quadratic
terms. We next simplify these equations, using implications of market clearing.
1122

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1123 10871145
Using (A36), we nd
1
2
h(
X
t
)
(+s
2
a
1
a
1
)
1
h(
X
t
) =
r
2
2

2
(
)
2
2(+ )
2
+
1
2
r
2

2
(1y
t
)
2
_
1+
s
2
2
1

. (A46)
We next substitute (A46) into (A8), and identify terms. Identifying terms in C
2
t
, C
t
y
t
and y
2
t
, we nd
1
2
t
_

Q

R
2

Q+

Q

R
1
+

R
1

Q

R
0
_

X
t
=0, (A47)
where
R
2
_
s
2
0
0 0
_
,
R
1
_
r
2
+ 0
b
1
r
2
+b
2
_
,
R
0
0 r
r r
2

2
_
1+
s
2
2
1

.
Equation (A47) must hold for all

X
t
. Because the square matrix in (A47) is symmetric, it must
equal zero, and this yields the algebraic Riccati equation
Q

R
2

Q+

Q

R
1
+

R
1

Q

R
0
=0. (A48)
We next identify terms in C
t
and y
t
, which yield
_
r + +s
2
q
11
_
q
1
+b
1
q
2

C q
11
b
0
q
12
=0, (A49)
(r +b
2
) q
2
+s
2
q
1
q
12
+r
2

2
_
1+
s
2
2
1

C q
12
b
0
q
22
=0, (A50)
respectively. Using (16) and (A35), we can write (A20) as
h(X
t
) =
r
+
. (A51)
Using (7), (16), (A42), and (A51), we nd that the equation derived from (A22) by identifying
terms in C
2
t
, C
t
y
t
, and y
2
t
is
QR
2
Q+QR
1
+R
1
QR
0
=0, (A52)
where
R
2
_
s
2
0
0 0
_
,
R
1
_
r
2
+ rb
1
s
2
b
1
r
2
+b
2
_
,
R
0
rb
2
1
rb
1
(r + +2b
2
)
rb
1
(r + +2b
2
) r
2
2
_
1+
s
2
2
1

+2r
2
2
b
1
s
2
rb
2
(2r +3b
2
)
,
and the equations derived by identifying terms in C
t
and y
t
are
(r + +s
2
q
11
)q
1
+b
1
q
2
rb
0
b
1

Cq
11
b
0
q
12
=0, (A53)
(r +b
2
)q
2
+s
2
(q
12
+rb
1
)q
1
r
_
(r +2b
2
)b
0
+b
1

C
_

Cq
12
b
0
q
22
=0, (A54)
respectively.
1123

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1124 10871145
Solving for equilibrium amounts to solving the system of (16), (A39)(A41), (A43)(A45),
(A48)(A50), and (A52)-(A54) in the unknowns (a
0
,a
1
,a
2
,b
0
,b
1
,b
2
,
1
,
2
, q
1
, q
2
,

Q,q
1
,q
2
,Q).
This reduces to solving the system of (A39), (A40), (A43), (A44), (A48), and (A52) in the
unknowns (b
1
,b
2
,
1
,
2
,

Q,Q): Given (b
1
,b
2
,
1
,
2
,

Q,Q), (a
1
,a
2
) can be determined from (16),
(b
0
, q
1
, q
2
,q
1
,q
2
) from the linear system of (A45), (A49), (A50), (A53), and (A54), and a
0
from
(A41). We replace the system of (A39), (A40), (A43), (A44), (A48), and (A52) by the equivalent
system of (A39), (A40), (A48), (A52),
b
1
(r + +b
2
+s
2
q
11
) =1+s
2
1
( q
11
q
11
)

, (A55)
_
(r +b
2
)b
2
+b
1
s
2
q
12
_
rb
1
s
2
=r(+ )
_
1+
s
2
2
1

+s
2
1
( q
12
q
12
)

.
(A56)
Equations (A55) and (A56) follow by substituting (A39) and (A40) into (A43) and (A44).
For s =0, (A39), (A40), (A48), (A52), (A55), and (A56) become
(r +)
1
+b
1
2
=0, (A57)
(r +b
2
)
2
=r , (A58)
Q

R
0
1
+

R
0
1

Q

R
0
0
=0, (A59)
QR
0
1
+R
0
1
QR
0
0
=0, (A60)
b
1
(r + +b
2
) =1, (A61)
(r +b
2
)b
2
=r(+ )

, (A62)
respectively, where
R
0
1
=R
0
1
_
r
2
+ 0
b
1
r
2
+b
2
_
,
R
0
0
_
0 r
r r
2

2
_
,
R
0
0
_
rb
2
1
rb
1
(r + +2b
2
)
rb
1
(r + +2b
2
) r
2
rb
2
(2r +3b
2
)
_
.
Equation (A62) is quadratic and has a unique positive solution b
2
.
21
Given b
2
, b
1
is determined
uniquely from (A61),
2
from (A58),
1
from (A57),

Q from (A59) (which is linear in

Q), and Q
from (A60) (which is linear in Q). We denote this solution by (b
0
1
,b
0
2
,
0
1
,
0
2
,

Q
0
,Q
0
).
To showthat the systemof (A39), (A40), (A48), (A52), (A55), and (A56) has a solution for small
s, we apply the implicit function theorem. We move all terms in each equation to the left-hand side,
and stack all left-hand sides into a vector F, in the order (A56), (A55), (A40), (A39), (A48), and
21
The positive solution is the relevant one. Indeed, because the negative solution satises r +2b
2
<0, (A59) implies
that q
22
<0. Therefore, the managers certainty equivalent would converge to at the rate y
2
t
, when |y
t
| goes
to and C
t
is held constant. The manager can, however, achieve higher certainty equivalent by not investing
in the active fund.
1124

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1125 10871145
(A52). Treated as a function of (b
1
,b
2
,
1
,
2
,

Q,Q,s), F is continuously differentiable around the
point A(b
0
1
,b
0
2
,
0
1
,
0
2
,

Q
0
,Q
0
,0) and is equal to zero at A. To show that the Jacobian matrix
of F with respect to (b
1
,b
2
,
1
,
2
,

Q,Q) has nonzero determinant at A, we note that F has a
triangular structure for s =0: F
1
depends only on b
2
, F
2
only on (b
1
,b
2
), F
3
only on (b
2
,
2
), F
4
only on (b
1
,
1
,
2
), F
5
only on (b
1
,b
2
,

Q), and F
6
only on (b
1
,b
2
,Q). Therefore, the Jacobian
matrix of F has nonzero determinant at A if the derivatives of F
1
with respect to b
2
, F
2
with
respect to b
1
, F
3
with respect to
2
, and F
4
with respect to
1
are nonzero, and the Jacobian
matrices of F
5
with respect to

Q and F
6
with respect to Q have nonzero determinants. These
results follow from (A57)(A62) and the positivity of (b
0
1
,b
0
2
). Therefore, the implicit function
theorem applies, and the system of (A39), (A40), (A48), (A52), (A55), and (A56) has a solution for
small s. This solution is unique in a neighborhood of (b
0
1
,b
0
2
,
0
1
,
0
2
,

Q
0
,Q
0
), which corresponds
to the unique equilibrium for s =0. Because b
0
1
>0, b
0
2
>0,
0
1
>0,
0
2
<0, continuity implies that
b
1
>0, b
2
>0,
1
>0,
2
<0 for small s.
Proof of Corollary 1. Equation (7) implies that when C
t
=0 for all t , lim
t
y
t
=b
0
/b
2
. Moreover,
when

C=s =0,
b
0
b
2
=
r

(r +b
2
)b
2
=

(+ )
,
where the rst stepfollows from(A45) andthe secondfrom(A62). Equations (A35) andlim
t
y
t
=
/(+ ) imply that lim
t
x
t
=0. The rst equality in (18) follows from (A1), (A5), (A6), and
lim
t
z
t
= lim
t
y
t
lim
t
z
t
=(1 lim
t
y
t
)( lim
t
x
t
) =

+
,
and the second equality follows from (A7).
Proof of Corollary 2. The investors stock holdings are
x
t
+y
t
z
t
=x
t
+y
t
( x
t
) =y
t
p
f
+

+
,
where the second step follows from (A35). Therefore,
E
t
(x
t
+y
t
z
t
)
C
t
=
E
t
(y
t
)
C
t
p
f
. (A63)
To determine how a shock to C
t
affects (E
t
(C
t
),E
t
(y
t
)), we use the linearity of (4) and (7) and
solve the impulse-response dynamics
dC
t
=C
t
dt, (A64)
dy
t
=(b
1
C
t
+b
2
y
t
)dt, (A65)
with the initial conditions
C
t
=1,
y
t
=0.
The solution is
C
t
=e
(t
t )
, (A66)
y
t
=
b
1
_
e
(t
t )
e
b
2
(t
t )
_
b
2
. (A67)
The derivative E
t
(y
t
)/C
t
is equal to (A67). Substituting into (A63), we nd (19). Because
b
1
>0 for small s, the coefcient of p
f
in (19) is negative.
1125

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1126 10871145
Proof of Corollary 3. The rst equality in (20) follows from (6) and (16). The second equality
follows from (16) and (A7). To derive the third equality, we note from (A7) that
Cov
t
(dR
t
,p
f
dR
t
) =0.
Therefore, if denotes the regression coefcient of dR
t
on dR
t
, then
Cov
t
(dR
t
,p
f
dR
t
) =Cov
t
_
dR
t
dR
t
,p
f
dR
t
_
=Cov
t
_
d
t
,p
f
dR
t
_
=Cov
t
_
d
t
,p
f
(dR
t
dR
t
)
_
=Cov
t
_
d
t
,p
f
d
t
_
,
where the second and fourth steps follow from the denition of d
t
, and the third step follows
because d
t
is independent of dR
t
.
Proof of Corollary 4. The corollary follows by substituting (16) into (A7).
Proof of Corollary 5. Stocks expected returns are
E
t
(dR
t
) =
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
dt
=
_
r
+
+
_
R
1
C
t
+
R
2
y
t
+r
_
1+
s
2
2
1

1
s
2
q
1
_
p
f
_
dt
=
_
r
+
(+s
2
a
1
a
1
)
+
t
(+s
2
a
1
a
1
)p
f
_
dt, (A68)
where
R
1
(r +)
1
+b
1
2
,
R
2
(r +b
2
)
2
.
The rst step in (A68) follows from (A1), the second from (16) and (A41), and the third from (16)
and (23). Equation (A68) is equivalent to (22) because of (A7).
Equation (A39) implies that
R
1
has the opposite sign of
1
q
11
. For small s,
1
>0 and q
11
has
the same sign as its value q
0
11
for s =0. Equation (A59) implies that
q
0
11
=
2b
0
1
q
0
12
r +2
=
2b
0
1
(r +2)(r + +b
0
2
)
_
r b
0
1
q
0
22
_
,
=
2r b
0
1
(r +2)(r + +b
0
2
)
_
r b
0
1
(r +2b
0
2
)
_
,
=
2r b
0
1
(r +2)(r + +b
0
2
)
_
r
(r + +b
0
2
)(r +2b
0
2
)
_
, (A69)
where the last step follows from (A61). Using (A62), we nd
(r + +b
0
2
)(r +2b
0
2
) =2r(+ )

+
_
(r +2)b
0
2
+r(r +)
_
=2r(+ )

+
r
2
_
r +(r +2)
_
1+
4(+ )f
r
_
. (A70)
Equations (A69) and (A70) imply that q
0
11
is positive if (24) holds, and is negative otherwise.
Therefore, for small s,
R
1
is negative if (24) holds, and is positive otherwise. Moreover,
R
2
<0
because b
2
>0 and
2
<0.
1126

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1127 10871145
Proof of Corollary 6. The autocovariance matrix is
Cov
t
(dR
t
,dR
) =Cov
t
_
dB
D
t
sa
1
dB
C
t
,
__
a
R
1
C
t
+a
R
2
y
t
_
dt
+dB
D
t
sa
1
dB
C
t
_
=Cov
t
_
dB
D
t
sa
1
dB
C
t
,
_
a
R
1
C
t
+a
R
2
y
t
dt
_
=Cov
t
_
sa
1
dB
C
t
,
_
a
R
1
C
t
+a
R
2
y
t
dt
_
=s
1
Cov
t
_
dB
C
t
,
R
1
C
t
+
R
2
y
t
_
p
f
p
f
dt
, (A71)
where the rst step follows by using (A1) and omitting quantities known at time t ; the second step
follows because the increments (dB
D
t
,dB
C
t
) are independent of information up to time t
; the third
step follows because of (4), (7), and the independence of (B
D
t
,B
C
t
); and the fourth step follows
from (16). Using (4) and (7), we can express (C
t
,y
t
) as a function of their time t values and the
Brownian shocks dB
C
u
for u[t,t
]. The covariance (A71) depends only on how the Brownian

shock dB
C
t
impacts (C
t
,y
t
). To compute this impact, we solve the impulse-response dynamics
(A64) and (A65) with the initial conditions
C
t
=sdB
C
t
,
y
t
=0.
The solution is
C
t
=e
(t
t )
sdB
C
t
, (A72)
y
t
=
b
1
_
e
(t
t )
e
b
2
(t
t )
_
b
2
sdB
C
t
. (A73)
Substituting into (A71), we nd (25) with
1
s
2
1
_
b
1
R
2
b
2
R
1
_
=s
2
(r +)
1
_
b
1
2
b
2
1
_
, (A74)
s
2
b
1
R
2
b
2
=
s
2
(r +b
2
)b
1
2
b
2
. (A75)
The function (u)
1
e
u
+
2
e
b
2
u
can change sign only once, is equal to s
2
R
1
when u=0,
and has the sign of
1
if b
2
> and of
2
if b
2
< when u goes to . For small s,
R
1
is negative
if (24) holds, and is positive otherwise. The opposite is true for (0) because
1
>0. Because, in
addition, b
1
>0, b
2
>0 and
2
<0, (A74) and (A75) imply that
1
<0 if b
2
> and
2
<0 if b
2
<.
Therefore, there exists a threshold u0, which is positive if (24) holds and is zero otherwise, such
that (u) >0 for 0<u u.
Proof of Corollary 7. Equations (A57), (A58), (A61), and (A62) imply that (b
0
1
,b
0
2
,
0
1
,
0
2
), the
terms in (b
1
,b
2
,
1
,
2
) of order zero in s
2
, are independent of . Hence, to determine how affects
(b
1
,b
2
,
1
,
2
), we must consider the terms of order one. We set
b
1
=b
0
1
+s
2
b
1
1
+o(s
2
), (A76)
b
2
=b
0
2
+s
2
b
1
2
+o(s
2
), (A77)
1
=
0
1
+s
2
1
1
+o(s
2
), (A78)
2
=
0
2
+s
2
1
2
+o(s
2
). (A79)
1127

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1128 10871145
Substituting (A76)(A79) into (A39), (A40), (A55), and (A56), and using (A57), (A58), (A61) and
(A62) to eliminate terms of order zero, we nd
(r +)
1
1
+
0
2
b
1
1
+b
0
1
1
2
=
0
1
q
0
11
, (A80)
(r +b
0
2
)
1
2
+
0
2
b
1
2
=
0
1
q
0
12
r (
0
1
)
2

, (A81)
_
b
0
1
b
1
2
+(r + +b
0
2
)b
1
1
_
=
0
1
( q
0
11
q
0
11
)

b
0
1
q
0
11
, (A82)
(r +2b
0
2
)b
1
2
=r(+ )
_

0
1

_
2
+
0
1
( q
0
12
q
0
12
)

b
0
1
q
0
12
+rb
0
1
0
1
.
(A83)
Differentiating (A80)(A83) with respect to , and noting that (b
0
1
,b
0
2
,
0
1
,
0
2
,q
0
11
,q
0
12
) are
independent of , we nd
1
1
=
0
1
_
b
0
1
r +b
0
2
_
1+
(2r +2 +b
0
2
)
0
2
(r + +b
0
2
)(r +2b
0
2
)
_
q
0
12
_
1+
0
2
(r + +b
0
2
)
_
q
0
11
_
, (A84)
1
2
=

0
1
r +b
0
2
_
1+
0
2
(r +2b
0
2
)
_
q
0
12
, (A85)
b
1
1
0
1

_
q
0
11

b
0
1
r+2b
0
2
q
0
12
_
(r + +b
0
2
)
, (A86)
b
1
2
0
1

q
0
12
(r +2b
0
2
)
. (A87)
Substituting
0
2
from (A58), and using (A62), we nd that the terms in square brackets in (A84) are
positive. Moreover, (b
0
1
,b
0
2
,
0
1
) are positive, and (A69) implies that q
0
11
/<0 and q
0
12
/>0.
Therefore,
1
1
/>0, and so
1
/>0. Equations (A74) and (A75) imply that
1
+
2
=s
2
R
1
=s
4
2
1
q
11
=s
4
(
0
1
)
2
q
0
11
+o(s
4
),
where the second step follows from (A39). Because q
0
11
/<0, (
1
+
2
)/<0. The value u
for the function (u) is equal to zero is
u=
1
b
2
log
_
1
(b
2
)
R
1
b
1
R
2
_
=
1
b
2
log
_
1+
s
2
(b
2
)
1
q
11
b
1
(r +b
2
)
2
_
=
s
2
0
1
q
0
11
b
0
1
(r +b
0
2
)
0
2
+o(s
2
),
where the second step follows from (A39). Because
0
2
<0 and q
0
11
/<0, u/<0.
B. Asymmetric Information
Proof of Proposition 4. We use Theorem 10.3 of Liptser and Shiryaev (LS; 2000). The investor
learns about C
t
, which follows the process (4). She observes the following information:
The net dividends of the true market portfolio D
t
C
t
dt . This corresponds to the process
1t
D
t
_
t
0
C
s
ds.
1128

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1129 10871145
The dividends of the index fund dD
t
. This corresponds to the process
2t
D
t
.
The price of the true market portfolio S
t
. Given the conjecture (28) for stock prices, this
is equivalent to observing the process
3t
(S
t
+a
1

C
t
+a
3
y
t
).
The price of the index portfolio S
t
. This is equivalent to observing the process
4t
(S
t
+a
1

C
t
+a
3
y
t
).
The dynamics of
1t
are
d
1t
=(F
t
dt +dB
D
t
)C
t
dt
=
_
(r +)a
0

F
r
+(r +)
3t
+(r +)a
2
C
t
C
t
_
dt +dB
D
t
=
_
(r +)a
0

F
r
+(r +)
3t
_
1
(r +)
2
_
C
t
_
dt +dB
D
t
, (B1)
where the rst step follows from (26), the second from (28), and the third from (29). Likewise, the
dynamics of
2t
are
d
2t
=
_
(r +)a
0

F
r
+(r +)
4t
_
dt +dB
D
t
. (B2)
The dynamics of
3t
are
d
3t
=d
_
_

F
r
+
F
t

F
r +
(a
0
+a
2
C
t
)
__
=
_
(

F F
t
)dt +dB
F
t
r +
a
2
_
(
CC
t
)dt +sdB
C
t
_
_
=
_
_

F
r
a
0
a
2

C
_
3t
_
dt +
dB
F
t
r +
sa
2
dB
C
t
=
_

F
r
a
0

3t
_
dt +
dB
F
t
r +
s
2
dB
C
t
, (B3)
where the rst step follows from (28), the second from (4) and (27), and the fourth from (29).
Likewise, the dynamics of
4t
are
d
4t
=
_

F
r
a
0
4t
_
dt +
dB
F
t
r +
. (B4)
The dynamics (4) and (B1)(B4) map into the dynamics (10.62) and (10.63) of LSby setting
t
C
t
,
t
(
1t
,
2t
,
3t
,
4t
)
, W
1t
_
B
D
t
B
F
t
_
, W
2t
B
C
t
, a
0
(t )

C, a
1
(t ) , a
2
(t ) 0, b
1
(t ) 0,
b
2
(t ) s,
t
T ,
A
0
(t )
(r +)a
0

F
r
(r +)a
0

F
r

F
r
a
0

F
r
a
0
_
,
A
1
(t )
1
(r+)
2
0
0
0
,
1129

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1130 10871145
A
2
(t )
0 0 r + 0
0 0 0 r +
0 0 0
0 0 0
,
B
1
(t )
0
0
0

r+
0

r+
,
B
2
(t )
0
0
s
2
.
The quantities (bb)(t ), (bB)(t ), and (BB)(t ), dened in LS (10.80) are
(bb)(t ) =s
2
,
(bB)(t ) =
_
0 0
s
2
0
_
,
(BB)(t ) =
0 0
0 0
0 0

2
(r+)
2
+
s
2
2
2

2
(
)
2
(r+)
2
0 0

2
(r+)
2
(r+)
2
.
Theorem 10.3 of LS (rst subequation of (10.81)) implies that
d

C
t
=(
C

C
t
)dt
1
_
d
1t
_
(r +)a
0

F
r
+(r +)
3t
_
1
(r +)
2
_

C
t
_
dt
_
d
2t
_
(r +)a
0

F
r
+(r +)
4t
_
dt
__
2
_
d
3t

F
r
a
0

3t
_
dt
_
d
4t

F
r
a
0
4t
_
dt
__
. (B5)
Equation (31) follows from(B5) by noting that the termin dt after each d
it
, i =1,2,3,4, is E
t
(d
it
).
In subsequent proofs, we use a different form of (31), where we replace each d
it
, i =1,2,3,4, by
its value in (B1)(B4):
d

C
t
=(
C

C
t
)dt
1
_
p
f
dB
D
t

_
1
(r +)
2
_
(C
t

C
t
)dt
_
2
_
p
f
dB
F
t
r +
s
2
dB
C
t
_
.
(B6)
Equation (34) follows from Theorem 10.3 of LS (second subequation of (10.81)).
1130

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1131 10871145
Proof of Proposition 5. Equations (4), (26)(30), and (B6) imply that the vector of returns is
dR
t
=
_
ra
0
+
_
R
1

C
t
+
R
2
C
t
+
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
dt +
_
+
1
1
p
f
p
f
_
dB
D
t
+

r +
_
+
2
1
p
f
p
f
_
dB
F
t
s
2
_
1+

2
_
p
f
dB
C
t
, (B7)
where
R
1
(r + +)
1
+b
1
3
,
R
2
(r +)
2
1
,
R
3
(r +b
2
)
3
,
and

1
_
1
(r +)
2
_
. (B8)
Equations (4), (8), (30), (B6), and (B7) imply that
d
_
r W
t
+ q
0
+( q
1
, q
2
, q
3
)
X
t
+
1
2
t

Q
X
t
_
=

Gdt +
_
r z
t
_
+
1
1
p
f
p
f
1

f
1
(
X
t
)p
f
_
dB
D
t
+

r +
_
r z
t
_
+
2
1
p
f
p
f
2

f
1
(
X
t
)p
f
_
dB
F
t
s
_
r
2
_
1+

2
_
z
t
p

f
1
(
X
t
)

f
2
(
X
t
)
_
dB
C
t
, (B9)
where
Gr
_
rW
t
+ z
t
_
ra
0
+
_
R
1

C
t
+
R
2
C
t
+
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
+C
t
y
t
c
t
_
+

f
1
(
X
t
)
_
(
C

C
t
)+(C
t

C
t
)
_
+

f
2
(
X
t
)(
CC
t
)+

f
3
(
X
t
)v
t
+
1
2
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
11
+
s
2
2
q
12
+
1
2
s
2
q
22
,
f
1
(
X
t
) q
1
+ q
11

C
t
+ q
12
C
t
+ q
13
y
t
,
f
2
(
X
t
) q
2
+ q
12

C
t
+ q
22
C
t
+ q
23
y
t
,
f
3
(
X
t
) q
3
+ q
13

C
t
+ q
23
C
t
+ q
33
y
t
.
Equations (36) and (B9) imply that
D
V =
V
_
G
1
2
(r )
2
f z
t
z
1
2
1
_
r
1
z
t
p

f
1
(
X
t
)
_
_
r
_
2+

1
_
z
t
p

f
1
(
X
t
)
1
2
2
(r +)
2
_
r
1
z
t
p

f
1
(
X
t
)
_
_
r
_
2+

2
_
z
t
p

f
1
(
X
t
)
1
2
s
2
_
r
2
_
1+

2
_
z
t
p

f
1
(
X
t
)

f
2
(
X
t
)
_
2
_
. (B10)
1131

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1132 10871145
Substituting (B10) into (11), we can write the rst-order conditions with respect to c
t
and z
t
as
(A4) and
h(
X
t
) =r (f +kp
f
p
f
) z
t
, (B11)
respectively, where
h(
X
t
) ra
0
+
_
R
1

C
t
+
R
2
C
t
+
R
3
y
t
(
1
+
2
)
Cb
0
3
+k
1

f
1
(
X
t
)+k
2

f
2
(
X
t
)
_
p
f
, (B12)
k
1
1
_
2+

1
_
+

2
1
(r +)
2
_
2+

2
_
+s
2
2
2
_
1+

2
_
2
, (B13)
k
1
1
_
1+

1
_
+

2
2
(r +)
2
_
1+

2
_
+
s
2
2
2

_
1+

2
_
, (B14)
k
2
s
2
2
_
1+

2
_
. (B15)
Proceeding as in the proof of Proposition 1, we nd the following counterpart of (A12):
1
2
h(
X
t
)
(f +kp
f
p
f
)
1
h(
X
t
)+r C
t
y
t
r
_
q
0
+( q
1
, q
2
, q
3
)
X
t
+
1
2
t

Q
X
t
_
+

f
1
(
X
t
)
_
(
C

C
t
)+(C
t

C
t
)
_
+

f
2
(
X
t
)(
CC
t
)+

f
3
(
X
t
)v
t
+
1
2
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
11
+
s
2
2
q
12
+
1
2
s
2
q
22
+

r +r log(r)
1
2
_
2
1
+

2
2
2
(r +)
2
_

f
1
(
X
t
)
2

1
2
s
2
_

f
1
(
X
t
)
+

f
2
(
X
t
)
_
2
=0. (B16)
Equation (B16) is quadratic in

X
t
. Identifying quadratic, linear, and constant terms yields ten
scalar equations in ( q
0
, q
1
, q
2
, q
3
,

Q). We defer the derivation of these equations until the proof of
Proposition 7 (see (B40)(B43)).
Proof of Proposition 6. Dynamics under the investors ltration can be deduced fromthose under
the managers by replacing C
t
by the investors expectation

C
t
. Equation (B6) implies that the
dynamics of

C
t
are
d

C
t
=(
C

C
t
)dt
1
p
f
d

B
D
t

2
_
p
f
dB
F
t
r +
s
2
dB
C
t
_
, (B17)
where

B
D
t
is a Brownian motion under the investors ltration. Equation (B7) implies that the
net-of-cost return of the active fund is
z
t
dR
t
C
t
dt =z
t
_
ra
0
+
_
(g
R
1
+g
R
2
)
C
t
+g
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
dt

C
t
dt
+z
t
_
+
1
1
p
f
p
f
_
d

B
D
t
+z
t
r +
_
+
2
1
p
f
p
f
_
dB
F
t
s
2
_
1+

2
_
z
t
p
f
dB
C
t
,
(B18)
1132

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1133 10871145
and the return of the index fund is
dR
t
=
_
ra
0
+
_
(g
R
1
+g
R
2
)
C
t
+g
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
dt
+
_
+
1
1
p
f
p
f
_
d

B
D
t
+
r +
_
+
2
1
p
f
p
f
_
dB
F
t
s
2
_
1+

2
_
p
f
dB
C
t
.
(B19)
Suppose that the investor optimizes over (c
t
,x
t
) but follows the control v
t
given by (30).
Equations (13), (30), (B17), (B18), and (B19) imply that
d
_
rW
t
+q
0
+(q
1
,q
2
)X
t
+
1
2
X
t
QX
t
_
=Gdt +
_
r(x
t
+y
t
z
t
)
_
+
1
1
p
f
p
f
1
f
1
(X
t
)p
f
_
d

B
D
t
+

r +
_
r(x
t
+y
t
z
t
)
_
+
2
1
p
f
p
f
2
f
1
(X
t
)p
f
_
dB
F
t
s
_
r
2
_
1+

2
_
(x
t
+y
t
z
t
)p
2
f
1
(X
t
)
_
dB
C
t
, (B20)
where
Gr
_
rW
t
+(x
t
+y
t
z
t
)
_
ra
0
+
_
(g
R
1
+g
R
2
)
C
t
+g
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
y
t

C
t
v
2
t
2
c
t
_
+f
1
(X
t
)(
C

C
t
)+f
2
(X
t
)v
t
+
1
2
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
11
,
f
1
(X
t
) q
1
+q
11

C
t
+q
12
y
t
,
f
2
(X
t
) q
2
+q
12

C
t
+q
22
y
t
.
Equations (14) and (B20) imply that
DV =V
_
G
1
2
(r)
2
f (x
t
+y
t
z
t
)(x
t
+y
t
z
t
)
1
2
1
_
r
1
(x
t
+y
t
z
t
)p
f
f
1
(X
t
)
_
_
r
_
2+

1
_
(x
t
+y
t
z
t
)p
1
f
1
(X
t
)
1
2
2
(r +)
2
_
r
1
(x
t
+y
t
z
t
)p
f
f
1
(X
t
)
_
_
r
_
2+

2
_
(x
t
+y
t
z
t
)p
2
f
1
(X
t
)
1
2
s
2
_
r
2
_
1+

2
_
(x
t
+y
t
z
t
)p
2
f
1
(X
t
)
_
2
_
. (B21)
Substituting (B21) into (15), we can write the rst-order conditions with respect to c
t
and x
t
as
(A15) and
h(X
t
) =r(f +kp
f
p
f
)(x
t
+y
t
z
t
)
, (B22)
respectively, where
h(X
t
) ra
0
+
_
(g
R
1
+g
R
2
)
C
t
+g
R
3
y
t
(
1
+
2
)
Cb
0
3
+k
1
f
1
(X
t
)
_
p
f
. (B23)
1133

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1134 10871145
Proceeding as in the proof of Proposition 2, we nd the following counterpart of (A22):
ry
t
h(X
t
)
1
2
(r)
2
y
2
t
(f +kp
f
p
f
)
+
_
h(X
t
)rfy
t
_
2
2f
ry
t

C
t
1
2
rv
2
t
r
_
q
0
+(q
1
,q
2
)X
t
+
1
2
X
t
QX
t
_
+f
1
(X
t
)(
C

C
t
)+f
2
(X
t
)v
t
+
1
2
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
[q
11
f
1
(X
t
)
2
]
+r +r log(r) =0. (B24)

Because v
t
in (7) is linear in X
t
, (B24) is quadratic in X
t
. Identifying quadratic, linear and constant
terms yields six scalar equations in (q
0
,q
1
,q
2
,Q). We defer the derivation of these equations until
the proof of Proposition 7 (see (B44)(B46)).
We next study optimization over v
t
, using the same perturbation argument as in the proof of
Proposition 2. The counterparts of (A27) and (A28) are
_
h(X
t
)rb
1
k
1
y
t
p
f
_

C
t
=r(f +kp
f
p
f
)[x
t
(1y
t
)+y
t
]
+h
(X
t
), (B25)
_
h(X
t
)rb
1
k
1
y
t
p
f
_

C
t
=r(f +kp
f
p
f
)
_
y
t
+
h(X
t
)ry
t
f
rf
+h
(X
t
), (B26)
respectively, where
h
(X
t
) (r +b
2
)v
t
+b
1
(
C

C
t
)b
1
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
f
1
(X
t
)
.
Equation (B26) is linear in X
t
. Identifying linear and constant terms, yields three scalar equations
in (b
0
,b
1
,b
2
). We defer the derivation of these equations until the proof of Proposition 7 (see
(B36)(B38)).
Proof of Proposition 7. We rst impose market clearing and derive the constants (a
0
,b
0
,
b
1
,b
2
,
1
,
2
,
3
) as functions of ( q
1
, q
2
, q
3
,

Q,q
1
,q
2
,Q). Setting z
t
= x
t
and y
t
=1y
t
, we
can write (B11) and (B22) as
h(
X
t
) =r (f +kp
f
p
f
)(1y
t
)( x
t
)
, (B27)
h(X
t
) =r(f +kp
f
p
f
)[x
t
(1y
t
)+y
t
]
, (B28)
respectively. Premultiplying (B27) by , dividing by r , and adding to (B28) divided by r, we nd
_
h(X
t
)
r
+
h(
X
t
)
r
_
=(f +kp
f
p
f
)
. (B29)
Equation (B29) is linear in (
C
t
,C
t
,y
t
). The terms in

C
t
, C
t
, and y
t
are zero because p
f
=0.
Identifying constant terms, we nd (A34). Substituting (A34) into (B28), we nd (A35).
Substituting (A35) into (B27), we nd
h(
X
t
) =r (f +kp
f
p
f
)
_

+
+(1y
t
)p
f
_
. (B30)
1134

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1135 10871145
Equation (B30) is linear in

X
t
. Identifying terms in

C
t
, C
t
, and y
t
, we nd
(r + +)
1
+b
1
3
+k
1
q
11
+k
2
q
12
=0, (B31)
(r +)
2
1
+k
1
q
12
+k
2
q
22
=0, (B32)
(r +b
2
)
3
+k
1
q
13
+k
2
q
23
=r
_
f +
k
_
, (B33)
respectively. Identifying constant terms, we nd
a
0
=
f
+
+
_
(
1
+
2
)
C+b
0
3
k
1
q
1
k
2
q
2
r
+
_
f +
k
__
p
f
. (B34)
Using (A35), we can write (B26) as
h(X
t
)rb
1
k
1
y
t

C
t
=r(f +kp
f
p
f
)
_

+
+y
t
p
f
_
+h
(X
t
)
h(X
t
)rb
1
k
1
y
t

C
t
=
r f
+
(
)
2
+r
_
f +
k
y
t
+h
(X
t
).
(B35)
Equation (B35) is linear in (
C
t
,y
t
). Identifying terms in

C
t
and y
t
, and using (30), we nd
[(r +)(
1
+
2
)+b
1
3
+k
1
q
11
]

1=b
1
_
r + +b
2
+
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
11
_
,
(B36)
[(r +b
2
)
3
+(q
12
rb
1
)k
1
]

=r
_
f +
k
_
(r +b
2
)b
2
+b
1
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
12
_
, (B37)
respectively. Identifying constant terms, and using (30) and (B34), we nd
_
k
1
(q
1
q
1
)k
2
q
2
+r
_
f +
k
__

=
_
(r +b
2
)b
0
+b
1

Cb
1
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
1
_
. (B38)
The system of equations characterizing equilibrium is as follows. The endogenous variables
are (a
0
,b
0
,b
1
,b
2
,
1
,
2
,
3
,
1
,
2
,T, q
1
, q
2
, q
3
,

Q,q
1
,q
2
,Q). (As in Proposition 3, we can drop
( q
0
,q
0
).) The equations linking them are (32)(34), (B31)(B34), (B36)(B38), the nine equations
derived from (B16) by identifying linear and quadratic terms, and the ve equations derived from
(B24) by identifying linear and quadratic terms. We next simplify the latter two sets of equations,
using implications of market clearing.
Using (B30), we nd
1
2
h(
X
t
)
(f +kp
f
p
f
)
1
h(
X
t
) =
r
2
2

2
f (
)
2
2(+ )
2
+
1
2
r
2

2
(1y
t
)
2
_
f +
k
.
(B39)
1135

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1136 10871145
We next substitute (B39) into (B16), and identify terms. Quadratic terms yield the algebraic Riccati
equation
Q

R
2

Q+

Q

R
1
+

R
1

Q

R
0
=0, (B40)
where
R
2
2
1
+

2
2
2
(r+)
2
+
s
2
2
2
2
2

s
2
0
s
2
s
2
0
0 0 0
R
1
r
2
+ + 0
0
r
2
+ 0
b
1
0
r
2
+b
2
R
0
0 0 0
0 0 r
0 r r
2

2
_
f +
k
.
Terms in

C
t
, C
t
, and y
t
yield
(r + +) q
1
+b
1
q
3
+
_
2
1
+

2
2
2
(r +)
2
_
q
1
q
11
+s
2
_
2
q
1
+ q
2
__
2
q
11
+ q
12
_

C( q
11
+ q
12
)b
0
q
13
=0, (B41)
(r +) q
2
q
1
+
_
2
1
+

2
2
2
(r +)
2
_
q
1
q
12
+s
2
_
2
q
1
+ q
2
__
2
q
12
+ q
22
_

C( q
12
+ q
22
)b
0
q
23
=0, (B42)
(r +b
2
) q
3
+
_
2
1
+

2
2
2
(r +)
2
_
q
1
q
13
+s
2
_
2
q
1
+ q
2
__
2
q
13
+ q
23
_
+r
2

2
_
f +
k

C( q
13
+ q
23
)b
0
q
33
=0, (B43)
respectively. Using (A35), we can write (B28) as (A51). Using (30), (A51), and (B35), we nd that
the equation derived from (B24) by identifying quadratic terms is
QR
2
Q+QR
1
+R
1
QR
0
=0, (B44)
where
R
2
2
1
+

2
2
2
(r+)
2
+
s
2
2
2
2
2

0
0 0
,
R
1
r
2
+ rb
1
_
2
1
+

2
2
2
(r+)
2
+
s
2
2
2
2
2

b
1
r
2
+b
2
,
R
0
_
rb
2
1
rb
1
(r + +2b
2
)
rb
1
(r + +2b
2
) r
2
2
_
f +
k
+2r
2
2
b
1
k
1

rb
2
(2r +3b
2
)
_
,
1136

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1137 10871145
and the equations derived by identifying terms

C
t
and y
t
are
(r +)q
1
+b
1
q
2
+
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
1
q
11

Cq
11
b
0
q
12
rb
0
b
1
=0, (B45)
(r +b
2
)q
2
+
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
(q
12
+rb
1
)q
1

Cq
12
b
0
q
22
r
_
b
0
(r +2b
2
)+b
1

C
_
=0, (B46)
respectively.
Solving for equilibrium amounts to solving the system of (32)(34), (B31)(B34), (B36)
(B38), (B40)(B43), and (B44)(B46) in the unknowns (a
0
,b
0
,b
1
,b
2
,
1
,
2
,
3
,
1
,
2
,T, q
1
,
q
2
, q
3
,

Q,q
1
,q
2
,Q). This reduces to solving the system of (32)(34), (B31)(B33), (B36), (B37),
(B40), and (B44) in the unknowns (b
1
,b
2
,
1
,
2
,
3
,
1
,
2
,T,

Q,Q): Given (b
1
,b
2
,
1
,
2
,
3
,
1
,
2
,T,

Q,Q), (b
0
, q
1
, q
2
, q
3
,q
1
,q
2
) can be determined from the linear system of (B38), (B41)
(B43), (B45), (B46), and a
0
from (B34). We replace the system of (32)(34), (B31)(B33), (B36),
(B37), (B40), and (B44) by the equivalent system of (32)(34), (B31)(B33), (B40), (B44),
b
1
_
r + +b
2
+
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
11
_
=1+[k
1
( q
11
+ q
12
)+k
2
( q
12
+ q
22
)k
1
q
11
]

, (B47)
_
(r +b
2
)b
2
+b
1
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_
q
12
_
rb
1
k
1
=r(+ )
_
f +
k
+(k
1
q
13
+k
2
q
23
k
1
q
12
)

. (B48)
For s =0, the unique nonnegative solution of (34) is T =0. Equations (32), (33), (B8), and (B13)
(B15) imply that
1
=
2
= =k =k
1
=k
2
=0. Equations (B31)(B33), (B40), (B44), (B47), and
(B48) become
(r +)
1
+b
1
3
=0, (B49)
(r +)
2
=0, (B50)
(r +b
2
)
3
=r f, (B51)
(A59), (A60), (A61), and (A62), respectively, where
R
0
1
r
2
+ 0 0
0
r
2
+ 0
b
1
0
r
2
+b
2
R
0
0 0 0
0 0 r
0 r r
2

2
f

,
and (R
0
1
,R
0
0
) are as under symmetric information (Proposition 3). Given the unique positive
solution b
2
of (A62), (b
1
,
3
,
1
,

Q,Q) are determined uniquely from (A61), (B51), (B49),
(A59) and (A60), respectively, and (B50) implies that
2
=0. We denote the solution for
1137

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1138 10871145
s =0 by (b
0
1
,b
0
2
,
0
1
,
0
2
,
0
3
,
0
1
,
0
2
,T
0
,

Q
0
,Q
0
). The variables (b
0
1
,b
0
2
,
0
1
,
0
3
,Q
0
) coincide with
(b
0
1
,b
0
2
,
0
1
,
0
2
,Q
0
) under symmetric information. Proceeding as in the proof of Proposition 3,
we can apply the implicit function theorem and show that the system of (32)(34), (B31)(B33),
(B40), (B44), (B47), and (B48) has a solution for small s. Moreover, this solution is unique
in a neighborhood of (b
0
1
,b
0
2
,
0
1
,
0
2
,
0
3
,
0
1
,
0
2
,T
0
,

Q
0
,Q
0
), which corresponds to the unique
equilibrium for s =0. Because b
0
1
>0, b
0
2
>0,
0
1
>0,
0
3
<0, continuity implies that b
1
>0, b
2
>0,
1
>0,
3
<0 for small s. Because
0
2
=0, continuity does not establish the sign of
2
for small s, so
we need to study the asymptotic behavior of the solution. Equations (34), (32), and (33) imply that
T =
s
2
2
+o(s
2
), (B52)
1
=

2
s
2
+o(s
2
)

0
1
s
2
+o(s
2
), (B53)
2
=o(s
2
), (B54)
respectively, where
o(s
2
)
s
2
converges to zero when s goes to zero. Equations (B8) and (B13)(B15)
imply that
=

0
1
s
2
+o(s
2
), (B55)
k =2
0
1
0
1
s
2
+o(s
2
), (B56)
k
1
=

0
1
s
2
+o(s
2
), (B57)
k
2
=o(s
2
), (B58)
respectively, and (B40) implies that
Q
0
=
2r
2

2
(b
0
1
)
2
f
(r+2)(r++b
0
2
)(r+2b
0
2
)

r b
0
1
(r+2)(r++b
0
2
)

r
2

2
b
0
1
f
(r++b
0
2
)(r+2b
0
2
)
r b
0
1
(r+2)(r++b
0
2
)
0
r
r++b
0
2
r
2

2
b
0
1
f
(r++b
0
2
)(r+2b
0
2
)
r
r++b
0
2
r
2

2
f
(r+2b
0
2
)
. (B59)
Equations (B32), (B55), (B57), (B58), and (B59) imply that
2
=
0
1
r +
_
0
1
+
r b
0
1
(r +2)(r + +b
0
2
)
_
s
2
+o(s
2
).
Therefore,
2
>0.
Proof of Corollary 8. Equation (B7) implies that the covariance matrix of stock returns is
Cov
t
(dR
t
,dR
t
) =
_
+
1
1
p
f
p
f
__
+
1
1
p
f
p
f
+

2
(r +)
2
_
+
2
1
p
f
p
f
__
+
2
1
p
f
p
f
+s
2
2
2
_
1+

2
_
2
p
f
p
f
,
which is equal to (37) because of (B13). To compare with symmetric information, we need to also
introduce F
t
in that case. Replacing (1) by (26) and (27), we nd that the vector of returns under
1138

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1139 10871145
symmetric information changes from (A1) to
dR
t
=
_
ra
0
+a
R
1
C
t
+a
R
2
y
t
a
1

Cb
0
a
2
_
dt +
_
dB
D
t
+
dB
F
t
r +
_
sa
1
dB
C
t
. (B60)
Hence, the covariance matrix under symmetric information changes from (21) to
Cov
t
(dR
t
,dR
t
) =(f +s
2
2
1
p
f
p
f
)dt. (B61)
Equations (37) and (B61) imply that the fundamental covariance under asymmetric information
is equal to that under symmetric information, and the nonfundamental covariance is proportional.
Moreover, the proportionality coefcient is larger than one if k >s
2
2
1sym
, where
1sym
denotes the
value of
1
under symmetric information. Rearranging (B13), we nd
k =2
_
1
+
2
_

2
(r +)
2
+
s
2
2
2

__
1
+s
2
2
2
+
_
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

_

2
1

. (B62)
Rearranging (33), we nd
2
_

2
(r +)
2
+
s
2
2
2

_
=s
2
2
, (B63)
and rearranging (34), we nd
T
2
_
1(r +k)

2
_
2
+
s
4
2
2

2
(r+)
2
+
s
2
2
2

=s
2
2T
2
1
+

2
2
2
(r +)
2
+
s
2
2
2
2
2

=s
2
2T, (B64)
where the second step follows from (32) and (33). Substituting (B63) and (B64) into (B62),
we nd
k =2
1
1
+s
2
(
1
+
2
)
2
2T
2
1
=s
2
(
1
+
2
)
2
+2T
1
_
1
(r +)
2
_
, (B65)
where the second step follows from (32).
Equations (B52), (B65), and
0
2
=0 imply that for small s,
k =s
2
(
0
1
)
2
+
s
2
0
1
0
1
_
+o(s
2
). (B66)
The variables (b
0
1
,b
0
2
) are identical under symmetric and asymmetric information. Moreover, (A57),
(A58), (B49), and (B51) imply that the same is true for
0
1
. Therefore, k >s
2
2
1sym
for small s if
0
1
>0
r f b
0
1
(r +)(r +b
0
2
)
>0
_
1
b
0
2
(r +)(+ )(r + +b
0
2
)
_
>0, (B67)
where the second step follows from (B49) and (B51), and the third step follows from (A61) and
(A62). Because b
0
2
>0, (B67) holds.
1139

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1140 10871145
Proof of Corollary 9. Stocks expected returns are
E
t
(dR
t
) =
_
ra
0
+
_
R
1

C
t
+
R
2
C
t
+
R
3
y
t
(
1
+
2
)
Cb
0
3
_
p
f
_
dt
=
_
r f
+
+
_
R
1

C
t
+
R
2
C
t
+
R
3
y
t
+r
_
f +
k
_
k
1
q
1
k
2
q
2
_
p
f
_
dt
=
_
r
+
_
f +kp
f
p
f
+
t
_
f +kp
f
p
f
_
p
f
_
dt, (B68)
where the rst step follows from (B7), the second from (B34), and the third from (38). Equation
(B68) is equivalent to (22) because of (37).
Equations (B31) and (B32) imply that
R
1
and
R
2
have the opposite sign of k
1
q
11
+k
2
q
12
and
k
1
q
12
+k
2
q
22
, respectively. Equations (B57) and (B58) imply that for small s, the latter variables
have the same sign as q
0
11
and q
0
12
, respectively. Because b
0
1
>0 and b
0
2
>0, (B59) implies that
q
0
11
>0 and q
0
12
<0. Therefore, for small s,
R
1
<0 and
R
2
>0. Moreover,
R
3
<0 because b
2
>0
and
3
<0.
Proof of Corollary 10. Using (B7) and proceeding as in the derivation of (A71), we nd
Cov
t
(dD
t
,dR
) =Cov
t
_
dB
D
t
,
R
1

C
t
+
R
2
C
t
+
R
3
y
t
_
p
f
dt
, (B69)
Cov
t
(dF
t
,dR
) =Cov
t
_
dB
F
t
,
R
1

C
t
+
R
2
C
t
+
R
3
y
t
_
p
f
dt
. (B70)
The covariances (B69) and (B70) depend only on how the Brownian shocks dB
D
t
and dB
F
t
,
respectively, impact (
C
t
,C
t
,y
t
). To compute the impact of these shocks, as well as of dB
C
t
for
the next corollary, we solve the impulse-response dynamics
dC
t
=C
t
dt, (B71)
d

C
t
=
_

C
t
+(C
t

C
t
)
_
dt, (B72)
dy
t
=
_
b
1

C
t
+b
2
y
t
_
dt, (B73)
with the initial conditions
C
t
=sdB
C
t
,
C
t
=
1
p
f
dB
D
t

2
_
p
f
dB
F
t
r +
s
2
dB
C
t
_
,
y
t
=0.
The solution is (A72),
C
t
=e
(t
t )
sdB
C
t
e
(+)(t
t )
_
1
p
f
dB
D
t
+

2
p
f
dB
F
t
r +
+s
_
1
_
dB
C
t
_
,
(B74)
y
t
=
b
1
b
2
_
e
(t
t )
e
b
2
(t
t )
_
sdB
C
t
+
b
1
b
2

_
e
(+)(t
t )
e
b
2
(t
t )
_
_
1
p
f
dB
D
t
+

2
p
f
dB
F
t
r +
+s
_
1
_
dB
C
t
_
. (B75)
1140

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1141 10871145
Substituting (A72), (B74), and (B75) into (B69) and (B70), and using the mutual independence of
(dB
D
t
,dB
F
t
,dB
C
t
), we nd (39) with
D
1

1
_
b
1
R
3
b
2

R
1
_
=(r + +)
1
_
b
1
3
b
2

1
_
, (B76)
D
2

b
1
R
3
b
2

=
(r +b
2
)b
1
3
b
2

. (B77)
The function
D
(u)
D
1
e
(+)u
+
D
2
e
b
2
u
can change sign only once, is equal to
1
R
1
when
u=0, and has the sign of
1
if b
2
> + and of
2
if b
2
< + when u goes to . For small s,
(0) >0 because
R
1
<0. Because, in addition, b
1
>0, b
2
>0,
1
>0,
3
<0 and >0, (B76) and
(B77) imply that
1
<0 if b
2
> + and
2
<0 if b
2
< +. Therefore, there exists a threshold
u
D
>0 such that (u) >0 for 0<u u
D
.
Proof of Corollary 11. Using (B7) and proceeding as in the derivation of (A71), we nd
Cov
t
(dR
t
,dR
) =
_
+
1
1
p
f
p
f
_
Cov
t
_
dB
D
t
,
R
1

C
t
+
R
2
C
t
+
R
3
y
t
_
p
f
dt
+

r +
_
+
2
1
p
f
p
f
_
Cov
t
_
dB
F
t
,
R
1

C
t
+
R
2
C
t
+
R
3
y
t
_
p
f
dt
s
2
_
1+

2
_
Cov
t
_
dB
C
t
,
R
1

C
t
+
R
2
C
t
+
R
3
y
t
_
p
f
p
f
dt
.
(B78)
Substituting (A72), (B74), and (B75) into (B78), and using (33) and the mutual independence of
(dB
D
t
,dB
F
t
,dB
C
t
), we nd (40) with
D
1
_
1+

1
_
, (B79)
2
s
2
2
_
b
1
R
3
b
2
R
1

R
2
_
_
1+

2
_
=s
2
(r +)
2
_
b
1
3
b
2
2
__
1+

2
_
, (B80)
3
b
1
R
3
_

1
b
2

_
1+

1
_
+
s
2
2
b
2
_
1+

2
__
=(r +b
2
)b
1
3
_

1
b
2

_
1+

1
_
+
s
2
2
b
2
_
1+

2
__
. (B81)
The function (u)
1
e
(+)u
+
2
e
u
+
3
e
b
2
u
has the same sign as (u)
1
e
u
+
2
+
3
e
(b
2
)u
. The latter function is equal to
R
1
_
1+

1
_
s
2
2
(
R
1
+
R
2
)
_
1+

2
_
when u=0, and has the sign of
2
if b
2
> and >0 and of
3
if b
2
< and >0 when u goes to
. Moreover, its derivative
(u) =
1
e
u
3
(b
2
)e
(b
2
)u
is equal to
3
(b
2
) =
1
(
R
1
+b
1
R
3
)
_
1+

1
_
+s
2
b
1
R
3
_
1+

2
_
(B82)
when u=0. For small s, (0) >0 because
R
1
<0 and s
2
2
/
1
=o(1). Because, in addition, b
1
>0,
b
2
>0,
1
>0,
2
>0,
3
<0,
R
3
<0, and >0, (B80) and (B81) imply that
2
<0 if b
2
> and
1141

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1142 10871145
3
<0 if b
2
<, and (B82) implies that
(0) <0. Because
(u) can change sign only once, it

is either negative or negative and then positive. Therefore, (u) is positive and then negative.
The same is true for (u), which means that there exists a threshold u>0 such that (u) >0 for
0<u u.
C. Momentum and Value Strategies
Proof of Proposition 8. Using (37), we can write (41) as
w
t
w
t
w
t
. (C1)
The expected return of the index-adjusted strategy is
E( w
t
dR
t
) =E
_
E
t
( w
t
dR
t
)
_
=E
_
w
t
E
t
(dR
t
)
_
=E
__
w
t
w
t

__
r
+
_
f +kp
f
p
f
+
t
_
f +kp
f
p
f
_
p
f
_
dt
_
=E
__
w
t
w
t

__
r
+
+
t
_
f +
k
_
p
f
_
dt
_
=
_
f +
k
_
E
_
t
w
t
p
f
_
dt, (C2)
where the third step follows from (B68) and (C1). The variance of the index-adjusted strategy is
Var( w
t
dR
t
) =E
_
Var
t
( w
t
dR
t
)
_
+Var
_
E
t
( w
t
dR
t
)
_
=E
_
Var
t
( w
t
dR
t
)
_
=E
__
w
t
w
t

_
_
f +kp
f
p
f
_
_
w
t
w
t
_
dt
=
_
f
_
E(w
t
w
t
)
E
_
(w
t
)
2
_
_
+kE
_
(w
t
p
f
)
2
_
_
dt, (C3)
where the second step follows because E[Var
t
(w
t
dR
t
)] is of order dt and Var[E
t
(w
t
dR
t
)] of
order (dt )
2
, and the third step follows from (37) and (C1). Equation (43) follows from (42), (C2),
and (C3).
To showthat the Sharpe ratio is maximized for w
t
=
t
p
f
, we write, for any given t , the strategy
w
t
as a linear combination of the market index, the ow portfolio, and an orthogonal component,
that is,
w
t
=
1t
+
2t
p
f
+ w
t
, (C4)
where w
t
=p
f
w
t
=0. Substituting w
t
from (C4), we can write (43) as
SR
w
=
_
f +
k
E(
t
2t
)
_
_
f +
k
E(
2
2t
)+f E( w
t
w
t
)
. (C5)
The Sharpe ratio is maximized for w
t
=0. Substituting into (C5), we nd
SR
w
=
E(
t
2t
)
_
E(
2
2t
)
_
_
f +
k
. (C6)
1142

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1143 10871145
The Cauchy-Schwarz inequality implies that the term
E(
t
2t
)
_
E(
2
2t
)
is maximized when
2t
is proportional to
t
. Therefore, the Sharpe ratio is maximized by the
strategy
t
p
f
.
References
Asness, C., J. Liew, and R. Stevens. 1997. Parallels between the cross-sectional predictability of stock and country
returns. Journal of Portfolio Management 23:7987.
Asness, C., T. Moskowitz, and L. Pedersen. 2013. Value and momentumeverywhere. Journal of Finance. Advance
Access published January 30, 2013, 10.1111/jo.12021.
Albuquerque, R., and J. Miao. Forthcoming 2012. Advance information and asset prices. Journal of Economic
Theory.
Ang, A., and J. Chen. 2002. Asymmetric correlations of equity portfolios. Journal of Financial Economics
63:44394.
Anton, M., and C. Polk. Forthcoming 2012. Connected stocks. Journal of Finance.
Barberis, N., and A. Shleifer. 2003. Style investing. Journal of Financial Economics 68:16199.
Barberis, N., A. Shleifer, and R. Vishny. 1998. A model of investor sentiment. Journal of Financial Economics
49:30743.
Basak, S., and A. Pavlova. Forthcoming 2012. Asset prices and institutional investors. American Economic
Review.
Berk, J., and R. Green. 2004. Mutual fund ows and performance in rational markets. Journal of Political
Economy 112:126995.
Berk, J., R. Green, and V. Naik. 1999. Optimal investment, growth options, and security returns. Journal of
Finance 54:1553607.
Bhojraj, S., and B. Swaminathan. 2006. Macromomentum: returns predictability in international equity indices.
Journal of Business 79:42951.
Brennan, M. 1993. Agency and asset pricing. Working Paper, University of California at Los Angeles.
Brunnermeier, M., and Y. Sannikov. 2012. A macroeconomic model with a nancial sector. Working Paper,
Princeton University.
Cespa, G., and X. Vives. 2012. Expectations, liquidity, and short-term trading. Working Paper, Cass Business
School.
Conrad, J., and G. Kaul. 1998. An anatomy of trading strategies. Review of Financial Studies 11:489520.
Coval, J., and E. Stafford. 2007. Asset re sales (and purchases) in equity markets. Journal of Financial Economics
86:479512.
Cuoco, D., and R. Kaniel. 2011. Equilibrium prices in the presence of delegated portfolio management. Journal
of Financial Economics 101:26496.
Daniel, K., D. Hirshleifer, andA. Subrahmanyam. 1998. Atheory of overcondence, self-attribution, and security
market under and over-reactions. Journal of Finance 53:183985.
Dasgupta, A., and A. Prat. 2008. Information aggregation in nancial markets with career concerns. Journal of
Economic Theory 143:83113.
1143

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1144 10871145
Dasgupta, A., A. Prat, and M. Verardo. 2011. The price impact of institutional herding. Review of Financial
Studies 24:892925.
DeBondt, W., and R. Thaler. 1985. Does the stock market overreact? Journal of Finance 40:793805.
Fama, E., and K. French. 1992. The cross-section of expected stock returns. Journal of Finance 47:42765.
. 1998. Value versus growth: The international evidence. Journal of Finance 53:197599.
Gorton, G., F. Hayashi, and G. Rouwenhorst. 2013. The fundamentals of commodity futures returns. Review of
Finance 17:35105.
Greenwood, R., and D. Thesmar. 2011. Stock price fragility. Journal of Financial Economics 102:47190.
Grinblatt, M., and S. Titman. 1989. Mutual fund performance: An analysis of quarterly portfolio holdings. Journal
of Business 62:393416.
Grinblatt, M., and T. Moskowitz. 1999. Do industries explain momentum? Journal of Finance 54:124990.
Gromb, D., and D. Vayanos. 2010. Limits of arbitrage. Annual Review of Financial Economics 2:25175.
Guerreri, V., and P. Kondor. 2012. Fund managers, career concerns and asset price volatility. American Economic
Review 102:19862017.
He, Z., and A. Krishnamurthy. 2012a. Amodel of capital and crises. Review of Economic Studies 79:73577.
. Forthcoming 2012b. Intermediary asset pricing. American Economic Review.
Hong, H., and J. Stein. 1999. A unied theory of underreaction, momentum trading, and overreaction in asset
markets. Journal of Finance 54:214384.
Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers: Implications for stock market
efciency. Journal of Finance 48:6591.
Johnson, T. 2002. Rational momentum effects. Journal of Finance 57:585608.
Jotikasthira, P., C. Lundblad, and T. Ramadorai. 2012. Asset re sales and purchases and the international
transmission of funding shocks. Journal of Finance 67:201550.
Kacperczyk, M., C. Sialm, and L. Zheng. 2005. Industry concentration of actively managed equity mutual funds.
Journal of Finance 60:19832011.
. 2008. Unobserved actions of mutual funds. Review of Financial Studies 21:2379416.
Kaniel, R., and P. Kondor. 2012. The delegated Lucas tree. Reviewof Financial Studies. AdvanceAccess published
December 20, 2012, 10.1093/rfs/hhs126.
Liptser, R., and A. Shiryaev. 2000. Statistics of random processes I, 2nd ed. New York: Springer.
Lou, D. 2012. Aow-based explanation for return predictability. Review of Financial Studies 25:345789.
Malliaris, S., and H. Yan. 2011. Reputation concerns and slow-moving capital. Working Paper, Yale University.
Petajisto, A. 2009. Why do demand curves for stocks slope down? Journal of Financial and Quantitative Analysis
44:101344.
Philippon, T. 2008. The evolution of the US nancial industry from1860 to 2007: Theory and evidence. Working
Paper, New York University.
Rouwenhorst, G. 1998. International momentum strategies. Journal of Finance 53:26784.
Shin, H. 2006. Disclosure risk and price drift. Journal of Accounting Research 44:35179.
Shleifer, A., and R. Vishny. 1997. The limits of arbitrage. Journal of Finance 52:3555.
Vayanos, D. 2004. Flight to quality, ight to liquidity, and the pricing of risk. Working Paper, London School of
Economics.
1144

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

[14:49 8/4/2013 RFS-hht014.tex] Page: 1145 10871145
Vayanos, D., and P. Woolley. 2012. A theoretical analysis of momentum and value strategies. Working Paper,
London School of Economics.
. 2013. The bird-in-the-hand effect and asset momentum. Working Paper, London School of Economics.
Wermers, R. 2000. Mutual fund performance: An empirical decomposition into stock-picking talent, style,
transaction costs, and expenses. Journal of Finance 55:1655703.
1145

a
t

T
h
a
m
m
a
s
a
t

U
n
i
v
e
r
s
i
t
y
,
F
a
c

o
f

C
o
m
m
e
r
c
e

&

A
c
c
o
u
n
t
a
n
c
y

L
i
b
r
a
r
y

o
n

S
e
p
t
e
m
b
e
r

1
,

2
0
1
3
h
t
t
p
:
/
/
r
f
s
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

An Institutional Theory of Momentum and Revealsal Rev. Financ. Stud.-2013-Vayanos-1087-145

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

An Institutional Theory of Momentum and Revealsal Rev. Financ. Stud.-2013-Vayanos-1087-145

Загружено:

Авторское право:

Доступные форматы

[14:49 8/4/2013 RFS-hht014.

tex] Page: 1087 10871145

follows the process

is the transpose of the vector v.

is the discount rate. The managers control variables are consumption c

take the form

V is the drift of the process

>t . Corollary 2 computes

>t , to isolate the effect of the time-t shock from

>t , caused by a change in C

denotes the residual from a regression of stock

t < u and negative if t

t > u, for a threshold u which

t < u (short-run momentum) and negatively

t > u (long-run reversal). It predicts in the same manner the subsequent

t < u and negative if t

t > u for a threshold u>0. Given

dt to express the Sharpe ratio in annualized terms, given

1), where I is the identity matrix and ( ,)

of the market index .

and for all other quantities used in the calibration

in annualized terms and set it to 30%. This is equal

) are independent of information up to time t

]. The covariance (A71) depends only on how the Brownian

+r +r log(r) =0. (B24)

(0) <0. Because

(u) can change sign only once, it

Вам также может понравиться