Date post: | 05-Apr-2018 |
Category: |
Documents |
Upload: | andra-leitoiu |
View: | 254 times |
Download: | 0 times |
of 38
7/31/2019 proiect de tradus
1/38
Risk Management with Stress Testing: Implications for
Portfolio Selection and Asset Pricing
Gordon J. Alexander
University of Minnesota
Alexandre M. Baptista
The George Washington University
June 22, 2006
7/31/2019 proiect de tradus
2/38
Risk Management with Stress Testing: Implications for
Portfolio Selection and Asset Pricing
Abstract
Stress Testing (ST) is often used by banks and securities firms to set risk exposure
limits. Accordingly, we examine a model with an agent who faces K binding ST constraints
and another who does not. We obtain four results. First, the constrained agents optimal
portfolio exhibits (K+2)-fund separation. Second, the effect of the constraints on the optimal
portfolio is identical to that of an adjustment in the expected returns of the risky securities
that tends to lower them, thereby increasing the optimal portfolios weight in the riskfree
security (or the minimum variance portfolio when this security is not available). Third, the
market portfolio is inefficient. Fourth, a securitys expected return is affected by both its
systematic risk and its idiosyncratic returns in the states used in the constraints. Thus, we
provide further motivation to the literature in which security prices are not solely driven bysystematic risk.
JEL classification: G11; G12; D81
Keywords: Stress Testing; Portfolio Selection; Asset Pricing; Idiosyncratic Returns
7/31/2019 proiect de tradus
3/38
1. INTRODUCTION
Over the past ten years, some investors have suffered huge losses due to extreme events.
For example, Barings Bank failed in 1995, Long Term Capital Management (LTCM) collapsed
in 1998, and Enron went bankrupt in 2001. Furthermore, the terrorist attacks in the U.S.
(2001), Spain (2004), and the U.K. (2005) have tremendously affected financial markets.
Since the occurrence of these events, the importance of risk management has been ex-
tensively recognized by banks and securities firms when deciding the amount of risk they
are willing to take. Moreover, bank regulators now put an emphasis on risk management
practices in attempting to reduce the fragility of banking systems.
Of the risk management tools currently available, Value-at-Risk (VaR) and Stress Testing
(ST) have emerged as two of the most popular. For example, under the Basle Capital Accord,
VaR is used in setting the minimum capital requirement associated with a banks exposure
to market risk. Furthermore, the Committee on the Global Financial System (2005, pp. 1,
15) of the Bank for International Settlements and Scholes (2000) note that ST is often used
by banks and securities firms to set risk exposure limits.
While the previous literature examines the impact of using VaR as a risk management
tool on portfolio selection and asset pricing (see, e.g., Alexander and Baptista (2002)), it
has yet to similarly explore the impact of using ST constraints. Our paper fills this gap
in the literature by providing parsimonious characterizations of (1) optimal portfolios in
the presence of ST constraints and (2) equilibrium security expected returns in a two-agent
economy where one agent faces these constraints and the other does not.
1
7/31/2019 proiect de tradus
4/38
An examination of the impact of ST constraints is of particular interest for several reasons.
First, as noted earlier, financial institutions often use them to set risk exposure limits. Second,
even when a financial institution holds capital in excess of the minimum capital requirements
as determined by the Basle Capital Accord, the ST constraints may still be binding. This
can happen since ST captures extreme events where losses can be very large. Third, since
there is empirical evidence that equity returns have skewness and kurtosis (see, e.g., Harvey
and Siddique (2000) and Dittmar (2002)), an ST constraint is a device that can be utilized to
control portfolio skewness and kurtosis. Fourth, the use of ST constraints can be motivated
by the goal of limiting losses arising from the need to unwind positions in markets that
become illiquid as a result of extreme events such as the LTCM collapse. Finally, since
the return distributions of portfolios can resemble those of certain options strategies (see,
e.g., Merton (1981) and Jagannathan and Korajczyk (1986)), the use of ST constraints can
mitigate the option-like features in these distributions.
In investigating the impact of ST constraints, we use Markowitzs (1952, 1959) mean-
variance model. There are important reasons for doing so. First, this model is the corner-
stone of portfolio theory. Second, it is widely used in practice to (1) determine optimal asset
allocations, (2) measure gains from international diversification, and (3) evaluate portfolio
performance. Third, since a set of ST constraints captures some measure of risk beyond vari-
ance (i.e., the returns in extreme events), a mean-variance objective function is of interest.
Finally, the model has been extensively used in the banking literature.
2
7/31/2019 proiect de tradus
5/38
Regardless on whether a riskfree security is available, we obtain four main results. First,
the constrained agents optimal portfolio exhibits (K + 2)-fund separation, where K is the
number of binding ST constraints. While this result might not seem to be particularly
surprising, Alexander and Baptista (2006a) have shown that under certain conditions a
constrained agents optimal portfolio still exhibits two-fund separation in the presence of
a binding risk management constraint.
Second, the effect of the constraints on the optimal portfolio is identical to that of an
adjustment in the expected returns of all risky securities that tends to lower them. Hence,
the constraints tend to increase the optimal portfolios weight in the riskfree security (or the
minimum variance portfolio when this security is not available).
Third, the market portfolio is inefficient. The reason why this result holds is that only
two of the K+ 2 funds required for the constrained agents optimal portfolio are efficient.
Fourth, a securitys expected return is affected by both its systematic risk (i.e., beta)
and its idiosyncratic returns in each one of those states that are used in the constraints.
Specifically, securities with negative (positive) idiosyncratic returns in these states have rel-
atively high (low) expected returns in equilibrium. The intuition for this result is straight-
forward. Due to the constraints, it is costly (beneficial) for the constrained agent to hold
securities with negative (positive) idiosyncratic returns in these states. Accordingly, the
agent requires them to have relatively high (low) expected returns.
3
7/31/2019 proiect de tradus
6/38
Next, we illustrate our theoretical results with a simple example. In this example, there
are five securities (one of them is riskfree), two agents (each one has half of the wealth in
the economy), and two binding ST constraints. Our main findings are as follows. First,
since there are two constraints, the constrained agents optimal portfolio exhibits four-fund
separation. Second, the portfolios weights in the two inefficient funds are notable, which
cause it to have an efficiency loss of 3.79%. Third, since the market portfolios weights in
these funds are smaller than those of the constrained agents optimal portfolio, the market
portfolio has an efficiency loss of only 0.63%. Fourth, the effect of the constraints on the
optimal portfolio is identical to that of a downward adjustment in the expected returns
of the risky securities ranging from 2.81% to 9.23%. Thus, the weight of the constrained
agents optimal portfolio in the riskfree security is substantially larger than that of the
unconstrained agents optimal portfolio. Fifth, the cost of ST as measured by the reduction
in the certainty equivalent return incurred by selecting the constrained portfolio is 3.16%.
Finally, the component of a securitys expected return that arises from its idiosyncratic
returns can be notable. This component is 2.42% for one of the securities that has negative
idiosyncratic returns in both of the states used in the constraints. However, the component
is 1.25% for one of the other securities since it has positive idiosyncratic returns in these
states. In sum, by exploring the implications of ST constraints, we contribute to the literature
in which security prices are not solely driven by systematic risk.
Previous theoretical papers in this literature have recognized the importance of idiosyn-
cratic risk. For example, Levy (1978), Merton (1987), and Malkiel and Xu (2002) develop
models in which security expected returns depend on both systematic and idiosyncratic risk.
4
7/31/2019 proiect de tradus
7/38
The empirical literature has also examined the importance of idiosyncratic risk. Campbell,
Lettau, Malkiel, and Xu (2001) and Malkiel and Xu (2003) show that the risk of individual
stocks has noticeably increased over time. Goyal and Santa-Clara (2003) provide evidence of
a significant positive relation between (1) a measure of idiosyncratic risk given by the equal-
weighted average stock risk and (2) the market return. However, Bali, Cakici, Yan, and
Zhang (2005) find no evidence of a significant relation between the value-weighted average
stock risk and the market return. Moreover, Guo and Savickas (2006) show that when the
value-weighted average stock risk and aggregate stock market risk are jointly used to forecast
the market return, there is a significant negative relation between the value-weighted average
stock risk and the market return. Ang, Hodrick, Xing, and Zhang (2006a,b) examine the
cross-sectional relation between the idiosyncratic risk of individual stocks and their returns,
and find that U.S. and other developed-market stocks with higher idiosyncratic risk earn
lower average returns.
Two features of our work differ from the idiosyncratic risk literature. First, the reason
why idiosyncratic returns affect expected returns in our model (i.e., the ST constraints) is,
to the best of our knowledge, novel. Second, our model captures the effects of idiosyncratic
returns in some states (i.e., those used in the ST constraints) on asset pricing while the
models in the literature capture the effects of idiosyncratic risk on asset pricing. Hence, the
model is able to generate equilibrium expected returns consistent with the evidence provided
by Ang, Hodrick, Xing, and Zhang (2006a,b) that stocks with high idiosyncratic risk earn
low average returns. For example, consider a security with (1) high idiosyncratic risk and
(2) positive idiosyncratic returns in the states used in the ST constraints. Consistent with
empirical evidence, our model predicts that such a security has a relatively low expected
5
7/31/2019 proiect de tradus
8/38
return.
Since our work derives the effects of a certain measure of risk beyond variance on asset
pricing, it is important to emphasize that there is an extensive literature that examines the
effects of skewness and kurtosis. In a seminal paper, Kraus and Litzenberger (1976) develop
a model that captures the effect of unconditional skewness. Lim (1989) tests their model
using stock returns and provides evidence that skewness is priced. More recently, Harvey and
Siddique (2000) explore a model where conditional skewness is priced and present empirical
evidence that it is helpful in explaining the cross-sectional variation in stock returns. Finally,
Dittmar (2002) develops a framework in which agents are averse to kurtosis.
Our work differs from this literature in two respects. First, the importance of a measure
of risk beyond variance arises in our model due to the existence of ST constraints while it
arises in other models from agents having either a preference for right-skewed portfolios over
left-skewed portfolios or an aversion to kurtosis. Second, our model captures the effects of
the returns in just those states used in the ST constraints on asset pricing while other models
use the returns in all states to capture the effects of skewness and kurtosis.
Also related to our work are papers that investigate portfolio selection when an agent
keeps his or her wealth above a floor. For example, Black and Perold (1992) and Grossman
and Vila (1992) examine the case when the floor is non-stochastic, while Grossman and Zhou
(1993) and Cvitanic and Karatzas (1995) examine the case when the floor is stochastic. Our
work differs from these papers in several important ways. First, we examine the impact of a
set of ST constraints rather than that of a floor. Second, we derive equilibrium results in
6
7/31/2019 proiect de tradus
9/38
the presence of ST constraints, while the aforementioned papers focus on portfolio selection
when an agent keeps his or her wealth above a floor. Finally, we use the mean-variance model
rather than the expected utility maximization continuous-time model.
The paper proceeds as follows. Section 2 characterizes optimal portfolios and equilib-
rium security expected returns when a riskfree security is present. Section 3 examines the
case when a riskfree security is absent. Section 4 provides an example that illustrates our
theoretical results and Section 5 concludes. The Appendix contains the proofs.
2. THE MODEL
2.1. Securities
Consider an economy where uncertainty is represented by S states. There are J risky
securities and a riskfree security with return R . The returns of the risky securities are given
by a J S matrix R, with R denoting the return of security j in state s. Let R be the
J 1 expected return vector and V be the J J variance-covariance matrix associated with
R. Let R [R R ] . Suppose that (1) there is no arbitrage; (2) rank(V) = J
so that there are no redundant securities; and (3) rank([1 R R R ]) = J for
any set ofJ 2 distinct states, s ,...,s , where 1 denotes the J 1 vector [1 1] .
A portfolio is a (J+ 1) 1 vector w with w = 1 w 1, where w [w w ] .Let R denote the return of portfolio w in state s. Let R and denote, respectively, the
expected return and variance of w.
7
7/31/2019 proiect de tradus
10/38
2.2. Agents
There are two agents with a mean-variance objective function U : R R R defined
by
U(R, ) = R
2 , (1)
where > 0. The unconstrained agent has = . The constrained agent has = , and
is restricted to select a portfolio w that satisfies K binding Stress Testing (ST) constraints:
R T , s = s ,...,s ,
where (i) 1 K J 2, (ii) s ,...,s are distinct states, and (iii) T ,...,T are possibly
distinct bounds in states s ,...,s , respectively.
2.3. Portfolio Selection Implications
Next, we characterize the optimal portfolios of unconstrained and constrained agents.
2.3.1. Unconstrained Agent
A portfolio is on the mean-variance boundary if there is no other portfolio with the same
expected return and a smaller variance. A portfolio is efficient if it lies on this boundary and
has an expected return equal to or greater than R . Otherwise, the portfolio is inefficient.
Let A 1 V R, B R V R, and C 1 V 1. Let w [0 0 1] and
w 0 denote the riskfree and tangency portfolios. Portfolios w and ware useful to characterize the unconstrained agents optimal portfolio.
8
7/31/2019 proiect de tradus
11/38
Theorem 1. The unconstrained agents optimal portfolio is
w = (1 )w + w , (2)
where .
Theorem 1 says that the unconstrained agents optimal portfolio w exhibits two-fund
separation. Since these funds are w and w , w is efficient. Note that if R is smaller
(larger) than A/C, then w is efficient (inefficient) and is positive (negative). Figure 1
illustrates Theorem 1 when > 0. Note that w , w , and w , represented by, respectively,
points f, t, and u, lie on the line representing the efficient frontier.
2.3.2. Constrained Agent
Let F 1 V R and w
0
where s = s ,...,s . Portfolios
w , ...,w are useful to characterize the constrained agents optimal portfolio.
Theorem 2. The constrained agents optimal portfolio is
w =
1
w + w +
w , (3)
where , for s = s ,...,s , and ,..., are Lagrange multipliers
associated with the ST constraints.
Theorem 2 says that the constrained agents optimal portfolio w exhibits (K+ 2)-fund
separation. Since these funds are w , w , and w ,...,w where the last K funds are
inefficient, w is also inefficient.
Figure 1 illustrates Theorem 2 when K = 1 and < 0. Note that w and w ,
represented by, respectively, points s and c, lie below the efficient frontier. Using Equation
9
7/31/2019 proiect de tradus
12/38
(3), we have
w = (1 )w + w , (4)
where + and w . Note that w , represented by point x, lies on the
dashed hyperbola representing combinations of w and w . As Equation (4) implies, w
lies on the dotted line representing combinations of w and w .
To isolate the impact of the ST constraints on the optimal portfolio from that of, assume
that = . Using Theorems 1 and 2, we have = . Thus, the constraints do not affect
the optimal portfolios weight in fund w . Furthermore, since the fund weights of each of the
optimal portfolios sum to one, we have = (1 ) 1 . That is,the sum of w s weights in funds w ,...,w is equal to the difference between the weights
of w and w in fund w .
The following result further explores the effect of the constraints on the optimal portfolio.
Theorem 3. Suppose that = . The constrained agents optimal portfolio when the
expected return vector is R coincides with the unconstrained agents optimal portfolio when
the expected return is R R +
(R 1R ).
Using Theorem 3, the effect of the ST constraints on the optimal portfolio is identical
to that of an adjustment in the expected return vector R. Since > 0 for s = s ,...,s ,
security js adjusted expected return R is smaller (larger) than its expected return R if
R is smaller (larger) than R for s = s ,...,s . In practice, a state is chosen to be in an
ST constraint only if there are some securities with negative returns in the state. Hence, the
case when R R for j = 1,...,J and s = s ,...,s is not plausible.
In the case when R < R for j = 1,...,J and s = s ,...,s , the expected returns
of all risky securities are adjusted downward. Thus, fund w is relatively more attractive.
10
7/31/2019 proiect de tradus
13/38
Consequently, the optimal portfolios weight in w tends to be larger in the presence of the
constraints.
Consider now the case when R < R < R for some distinct securities j , j
{1,...,J} and states s, s {s ,...,s }. In this case, (1) the expected return of at least
one risky security can be adjusted downward and (2) the expected return of at least one
risky security can be adjusted upward. Thus, fund w can be either relatively more or less
attractive. Consequently, there is no clear tendency for the optimal portfolios weight in w
in the presence of the constraints.
2.4. Asset Pricing Implications
Letting m R denote the market portfolio and denote the covariance between
security j and m, security js beta is / . Furthermore, let be the constrained
agents fraction of the wealth in the economy, where 0 < < 1.
Theorem 4. In equilibrium, the market portfolio is inefficient. Furthermore, security js
expected return is
R = R + (R R )
[(R R ) (R R )] , (5)
where
/
(1 )/ + /, k = 1,...,K, (6)
is a positive constant.
Theorem 4 says that in equilibrium the market portfolio is inefficient. The intuition for
this result is straightforward. Since (1) the unconstrained agents optimal portfolio is efficient,
(2) the constrained agents optimal portfolio is inefficient, and (3) the market portfolio is a
11
7/31/2019 proiect de tradus
14/38
combination of these optimal portfolios, the market portfolio is inefficient. Figure 1 illustrates
that when K = 1, the market portfolio, represented by point m, lies below the efficient frontier
on the dashed hyperbola representing combinations of w and w .
Equation (5) indicates that security js expected return depends on K + 2 terms. The
first two terms, R and (R R ), are identical to those contained in the CAPM.
However, the last K terms, which are subtracted from the sum of the first two terms, are not
present in the CAPM. Each one of these K terms is given by the product of: (1) security js
idiosyncratic return in state s , (R R ) (R R ), and (2) the risk premium on
the idiosyncratic return in state s , , where k {1,...,K}. We refer to the sum of these
K terms as the idiosyncratic return adjustment.
Since > 0 for k = 1,...,K, whether security js expected return is larger than, equal
to, or smaller than that in the CAPM depends on its idiosyncratic returns in states s ,...,s .
First, if security js idiosyncratic return is zero in states s ,...,s , then its expected return
is equal to that in the CAPM. Second, if security js idiosyncratic return is (1) either zero or
positive in states s ,...,s , and (2) positive for some state s {s ,...,s }, then its expected
return is smaller than that in the CAPM. Third, if security js idiosyncratic return is (1)
either zero or negative in states s ,...,s , and (2) negative for some state s {s ,...,s },
then its expected return is larger than that in the CAPM. Finally, if security js idiosyncratic
return is negative in some state s {s ,...,s } and positive in some state s {s ,...,s },
then its expected return can be smaller than, equal to, or larger than that in the CAPM.
The intuition for why securities with negative (positive) idiosyncratic returns in states
s ,...,s have relatively high (low) expected returns in equilibrium is straightforward. Due
12
7/31/2019 proiect de tradus
15/38
to the constraints, it is costly (beneficial) for the constrained agent to hold securities with
negative (positive) idiosyncratic returns in these states. Accordingly, the agent requires them
to have relatively high (low) expected returns.
Using Equation (6), the risk premium on the idiosyncratic return in state s depends on
two constants: (1) and (2) . The first constant measures the constrained
agents marginal cost (in terms of utility) arising from marginally decreasing the bound T
(i.e., tightening the ST constraint in state s ). The second constant measures the constrained
agents fraction of the risk-aversion-adjusted wealth in the economy. Note that if = ,
then this constant is equal to .
3. ABSENCE OF A RISKFREE SECURITY
Suppose now that there is no riskfree security. Let w and w denote, re-
spectively, the minimum variance portfolio and the portfolio on the mean-variance boundary
with expected return B/A.
3.1. Portfolio Selection Implications
Next, we characterize the optimal portfolios of unconstrained and constrained agents.
3.1.1. Unconstrained Agent
In the absence of a riskfree security, a portfolio is efficient if it lies on the mean-variance
boundary and has an expected return equal to or greater than A/C (i.e., that of w ).
Theorem 5. The unconstrained agents optimal portfolio is
w = (1 )w + w , (7)
where .
Theorem 5 says that the unconstrained agents optimal portfolio w exhibits two-fund
13
7/31/2019 proiect de tradus
16/38
separation. Since these funds are w and w , w is efficient. Figure 2 illustrates Theorem 5.
Note that w , w , and w , represented by, respectively, points a, b, and u, lie on the upper
part of the thick hyperbola representing the mean-variance boundary.
3.1.2. Constrained Agent
Let w for s = s ,...,s . Portfolios w ,...,w are useful to characterize the
constrained agents optimal portfolio.
Theorem 6. The constrained agents optimal portfolio is
w = 1 w + w + w , (8)where , for s = s ,...,s , and ,..., are Lagrange multipliers associ-
ated with the ST constraints.
Theorem 6 says that the constrained agents optimal portfolio w exhibits (K+ 2)-fund
separation. Since these funds are w , w , and w ,...,w where the last K funds are
inefficient, w is also inefficient.
Figure 2 illustrates Theorem 6 when K = 1 and < 0. Note that w and w ,
represented by, respectively, points s and c, lie below the efficient frontier. Using Equation
(8), we have
w = (1 )w + w , (9)
where + and w . Note that w , represented by point x, lies on the
dashed hyperbola representing combinations of w and w . As Equation (9) implies, w
lies on the dotted hyperbola representing combinations of w and w .
To isolate the impact of the ST constraints on the optimal portfolio from that of, assume
that = . Using Theorems 5 and 6, we have = . Thus, the constraints do not affect
14
7/31/2019 proiect de tradus
17/38
the optimal portfolios weight in fund w . Furthermore, since the fund weights of each of the
optimal portfolios sum to one, we have
= (1 )
1
. That is,
the sum of w s weights in funds w ,...,w is equal to the difference between the weights
of w and w in fund w .
The following result further explores the effect of the constraints on the optimal portfolio.
Theorem 7. Suppose that = . The constrained agents optimal portfolio when the
expected return vector is R coincides with the unconstrained agents optimal portfolio when
the expected return is R R + R .Using Theorem 7, the effect of the ST constraints on the optimal portfolio is identical
to that of an adjustment in the expected return vector R. Since > 0 for s = s ,...,s ,
security js adjusted expected return R is smaller (larger) than its expected return R if
R is smaller (larger) than zero for s = s ,...,s . Note that the case when R 0 for
j = 1,...,J and s = s ,...,s is not plausible since in practice a state is chosen to be in an
ST constraint only if there are some securities with negative returns in the state.
In the case when R < 0 for j = 1,...,J and s = s ,...,s , the expected returns of
all risky securities are adjusted downward. Thus, fund w is relatively more attractive.
Consequently, the optimal portfolios weight in w tends to be larger in the presence of the
constraints.
Consider now the case when R < 0 < R for some distinct securities j , j {1,...,J}
and states s, s {s ,...,s }. In this case, (1) the expected return of at least one security can
be adjusted downward and (2) the expected return of at least one security can be adjusted
upward. Thus, fund w can be either relatively more or less attractive. Consequently, there
is no clear tendency for the optimal portfolios weight in w in the presence of the constraints.
15
7/31/2019 proiect de tradus
18/38
3.2. Asset Pricing Implications
We now characterize the equilibrium.
Theorem 8. In equilibrium, the market portfolio is inefficient. Furthermore, security js
expected return is
R = R + (R R) R R , (10)where
R 1 (1 )A/ A/
F /
(1 )/ + / 1
C(11)
and is a positive constant as defined in Equation (6).
Theorem 8 says that the market portfolio is inefficient. The intuition for this result is
similar to that of Theorem 4. Figure 2 illustrates that the market portfolio, represented by
point m, lies below the efficient frontier on the thin hyperbola representing combinations of
wand
w.
Equation (10) indicates that security js expected return depends on K+ 2 terms. The
first two terms are related to those in the Black model. However, the last K terms, which
are subtracted from the sum of the first two terms, are not present in the Black model. Each
one of these K terms is given by the product of: (1) the idiosyncratic return of security j
in state s , R R , and (2) the risk premium on the idiosyncratic return in state s ,
, where k {1,...,K}.
16
7/31/2019 proiect de tradus
19/38
Since > 0 for k = 1,...,K, securities with negative (positive) idiosyncratic returns in
states s ,...,s have relatively high (low) expected returns in equilibrium. The intuition for
this result is similar to that presented for Theorem 4.
4. EXAMPLE
In this section, we illustrate our theoretical results using a simple example.
4.1. Securities
There are four risky securities (j = 1, 2, 3, 4) and a riskfree security (j = 5). The risky
securities are assumed to be in positive net supply, while the riskfree security is assumed
to be in zero net supply. Panel (a) of Table 1 presents their expected returns and standard
deviations. For simplicity, assume that the correlation coefficient between the returns of each
pair of distinct risky securities is equal to 0.4. As explained shortly, the constrained agent
faces two ST constraints which use states s and s . Panel (b) of Table 1 provides the security
returns in these states. These returns are notably negative, but nevertheless plausible. For
example, the Nasdaq index declined by 27.1% and 22.9% in, respectively, October 1987 and
November 2000. It is important to emphasize that the qualitative results in our example do
not depend on the assumptions that are imposed on (1) the existence of a riskfree security,
(2) the net supply of the riskfree security, (3) the distribution of security returns, and (4)
the number of ST constraints.
4.2. Agents
The unconstrained and constrained agents have the objective function defined by Equa-
17
7/31/2019 proiect de tradus
20/38
tion (1) with = = = 2. The constrained agent faces two ST constraints:
R T , s = s , s ,
where T = 3% and T = 4%. The reason why we have T > T is that the returns of
three out of the four risky securities in state s are larger in absolute terms than those in
state s . Nevertheless, examples with similar results can be constructed when T T .
4.3. Portfolio Selection Implications
Table 2 presents the optimal portfolios of unconstrained and constrained agents.
4.3.1. Unconstrained Agent
The first row of panel (a) indicates that the unconstrained agents optimal portfolio w
is characterized by weights = 36.93% and = 136.93% in, respectively, funds w and
w . The first row of panel (b) shows that w s weights in the risky securities range from
28.50% (security 4) to 41.42% (security 1). The first row of panel (d) says that w s expected
return and standard deviation are, respectively, 16.32% and 25.07%. Moreover, w s returns
in states s and s , at 21.02% and 25.74%, are notably negative.
4.3.2. Constrained Agent
The second row of panel (a) indicates that the constrained agents optimal portfolio w
is characterized by weights = 36.93%, = 136.93%, = 35.13%, and = 38.74%
in, respectively, funds w , w , w , and w . Note that the weights of w and w in fund w
are identical. Furthermore, w s weight in fund w is not only notably larger than w s, it
is also positive. The intuition for this result is simple. Since w s returns in states s and
s are positive, a relatively large weight in w is useful to meet the ST constraints. Further
18
7/31/2019 proiect de tradus
21/38
intuition can be seen in panel (c), which provides the adjustment in expected returns with the
same effect on the optimal portfolio as the constraints. Observe in line 2 that the expected
return of all risky securities are notably adjusted downward. Since w is now relatively more
attractive, the constraints cause the optimal portfolios weight in w to increase substantially.
The second row of panel (b) shows that w s weights in the risky securities range from
13.13% (security 1) to 44.52% (security 2). Note that w s weights in securities 1, 3, and 4
are smaller than those of w , while that in security 2 is larger. The intuition for these results
is simple. The returns of securities 1, 3, and 4 in states s and s are larger in absolute terms
than those of security 2. Thus, the ST constraints force a reduction of the weights in securities
1, 3, and 4. Panel (c) provides further intuition. The largest adjustment in the expected
return occurs for securities 1 (9.23%), 3 (5.24%), and 4 (3.31%). Consequently, the
ST constraints lead to a reduction of the weights in these securities. Since security 1 has a
relatively small adjusted expected return (5.77%), w s weight in this security is negative.
In contrast, since security 2 has a relatively large adjusted expected return (10.39%), w s
weight in this security is positive.
The second row of panel (d) says that w s expected return and standard deviation are,
respectively, 8.79% and 13.84%. Thus, they are notably smaller than those of w . It can be
seen that w s returns in states s and s meet the ST constraints. Since w is inefficient,
of particular interest is its efficiency loss, denoted by , which we measure in terms of
standard deviation. That is, w s efficiency loss is the increase in standard deviation arising
from selecting w instead of the efficient portfolio with the same expected return. Note that
w s efficiency loss is = 3.79%.
19
7/31/2019 proiect de tradus
22/38
4.3.3. Cost of Stress Testing
Next, we assess the cost of ST. In doing so, we measure the cost of ST by the reduction
in the certainty equivalent return incurred by the agent who selects the constrained opti-
mal portfolio instead of the unconstrained optimal portfolio. Since the certainty equivalent
returns of the unconstrained and constrained optimal portfolios are, respectively, 10.03%
and 6.87%, the cost of ST is 3.16%. That is, the protection provided by imposing the ST
constraints is procured at the cost of a reduction in the certainty equivalent return of 3.16%.
4.4. Asset Pricing Implications
Suppose that the constrained agents fraction of the wealth in the economy is = 50%.
4.4.1. Market Portfolio
The third row of panel (a) indicates that the market portfolio m is characterized by
weights = 136.93%, = 17.56%, and = 19.37% in, respectively, funds w , w ,
and w . The third row of panel (b) shows that ms weights in the risky securities range
from 14.15% (security 1) to 37.58% (security 2). The third row of panel (d) says that ms
expected return and standard deviation are, respectively, 12.55% and 18.20%. Thus, they
are smaller (larger) than those of w (w ). Moreover, ms returns in states s and s are
smaller (larger) in absolute terms than those of w (w ). Finally, ms efficiency loss is only
= 0.63%. The reason why we obtain this small efficiency loss is that ms weights in funds
w and w are substantially smaller than that in fund w .
4.4.2. Security Expected Returns
Panel (a) of Table 3 shows the risk premia. First, the CAPMs risk premium is R R =
8.81%. Second, the risk premium of the idiosyncratic return in state s is = 8.69% so
20
7/31/2019 proiect de tradus
23/38
that a securitys expected return increases by 8.69 basis points per percentage point of its
idiosyncratic return in state s . Third, the risk premium of the idiosyncratic return in state
s is = 7.35% so that a securitys expected return increases by 7.35 basis points per
percentage point of its idiosyncratic return in state s .
Panel (b) presents characteristics of the risky securities. The first row shows that their
betas range from 0.82 (security 4) to 1.22 (security 2). The second and third rows pro-
vide their idiosyncratic returns in states s and s . While securities 1 and 3 have negative
idiosyncratic returns in both states, security 2 has positive ones. Security 4 has a negative
idiosyncratic return in state s , but a positive one in state s .
Panel (c) decomposes security expected returns in four terms. The first three rows provide
the two terms contained in the CAPM and their sum, i.e., the CAPM expected return. As
expected, this sum is higher when the security beta is higher. The next three rows provide
the two terms that are not present in the CAPM and their sum, i.e., the idiosyncratic return
adjustment. Since securities 1 and 3 have negative idiosyncratic returns in states s and
s , the idiosyncratic return adjustment is negative for these securities. The idiosyncratic
return adjustment for security 1 (2.42%) is larger in absolute terms than that for security
3 (0.76%) because the idiosyncratic returns of the former are also larger in absolute terms
than those of the latter as shown in panel (b). Since security 2 has positive idiosyncratic
returns in states s and s , the idiosyncratic return adjustment for this security is positive
(1.25%). Finally, since security 4 has a negative idiosyncratic return in state s , but a positive
one in state s , the idiosyncratic return adjustment for this security is close to zero (0.13%).
In sum, the idiosyncratic return adjustments can be notably positive or negative.
21
7/31/2019 proiect de tradus
24/38
5. CONCLUSION
Over the past ten years, some investors have suffered huge losses due to extreme events
(e.g., the Barings Bank, LTCM, and Enron failures). In order to prevent such losses, ST
is now commonly utilized as a risk management tool. For example, the Committee on the
Global Financial System (2005, pp. 1, 15) of the Bank for International Settlements and
Scholes (2000) note that ST is often used by banks and securities firms to set risk exposure
limits. Accordingly, our paper examines a simple model to explore the portfolio selection
and asset pricing implications of ST constraints.
Regardless on whether a riskfree security is available, we obtain four main results. First,
the constrained agents optimal portfolio exhibits (K + 2)-fund separation, where K is the
number of binding ST constraints. Second, the effect of the constraints on the optimal
portfolio is identical to that of an adjustment in the expected returns of all risky securities
that tends to lower them. Hence, the constraints tend to increase the optimal portfolios
weight in the riskfree security (or the minimum variance portfolio when this security is not
available). Third, the market portfolio is inefficient. Fourth, a securitys expected return is
affected by both its systematic risk (i.e., beta) and its idiosyncratic returns in each one of
those states that are used in the constraints. Specifically, securities with negative (positive)
idiosyncratic returns in these states have relatively high (low) expected returns in equilibrium.
Next, we illustrate our theoretical results with a simple example. In this example, there
are five securities (one of them is riskfree), two agents (each of them has half of the wealth
in the economy), and two binding ST constraints. Our main findings are as follows. First,
since there are two constraints, the constrained agents optimal portfolio exhibits four-fund
separation. Second, the portfolios weights in the two inefficient funds are notable, which
22
7/31/2019 proiect de tradus
25/38
cause it to have an efficiency loss of 3.79%. Third, since the market portfolios weights in
these funds are smaller than those of the constrained agents optimal portfolio, the market
portfolio has an efficiency loss of only 0.63%. Fourth, the effect of the constraints on the
optimal portfolio is identical to that of a downward adjustment in the expected returns
of the risky securities ranging from 2.81% to 9.23%. Thus, the weight of the constrained
agents optimal portfolio in the riskfree security is substantially larger than that of the
unconstrained agents optimal portfolio. Fifth, the cost of ST as measured by the reduction
in the certainty equivalent return incurred by selecting the constrained portfolio is 3.16%.
Finally, the component of a securitys expected return that arises from its idiosyncratic
returns can be notable. This component is 2.42% for one of the securities that has negative
idiosyncratic returns in both of the states used in the constraints. However, the component
is 1.25% for one of the other securities since it has positive idiosyncratic returns in these
states. In sum, by exploring the implications of ST constraints, we provide further motivation
to the literature in which security prices are not solely driven by systematic risk.
23
7/31/2019 proiect de tradus
26/38
APPENDIX
Proof of Theorem 1. Observe that w solvesmax R +
w
(R
1
R )
2w Vw
. (12)
A first-order condition for w to solve problem (12) isR 1R Vw = 0. (13)
Since rank(V) = J, Equation (13) implies that
w =V (R 1R )
. (14)
Using the definition of w and Equation (14), we have Equation (2).
Proof of Theorem 2. Note that the ST constraints are assumed to bind. Hence, w solvesmax R + w (R 1R )
2w Vw (15)
s.t. w (R 1R ) = T R , s = s ,...,s . (16)
A first-order condition for w to solve problem (15) subject to constraints (16) isR 1R Vw + (R 1R ) = 0, (17)
where ,..., are Lagrange multipliers associated with these constraints. Since rank(V) =
J, Equation (17) implies that
w = V (R 1R ) + V (R 1R ) . (18)Using the definition of w , w ,...,w , and Equation (18), we have Equation (3).
We now find ,..., . Premultiplying Equation (18) by (R 1R ) and using Equa-
tion (16), we have
T R =I +
I
, s = s ,...,s , (19)
24
7/31/2019 proiect de tradus
27/38
where I (R 1R ) V (R 1R ) and I (R 1R ) V (R 1R ). It follows
from Equation (19) that
I = [I + (T + R ) ] , s = s ,...,s . (20)Let X [ ] , U [R 1R R 1R ], Y U V U , and
Z [[I + (T + R ) ] [I + (T + R ) ]] . Using Equation (20), we have
Y X =Z . Hence, X =Y Z .
Proof of Theorem 3. The desired result follows from Equations (14) and (18).
Proof of Theorem 4. We begin by showing that the market portfolio is inefficient. Using
Equations (2) and (3), we have
m =
1
w + w , (21)
where
=
(1 ) +
+
, s = s ,...,s . (22)
Since (1) = 0 for s = s ,...,s , (2) portfolio w is inefficient for s = s ,...,s , and (3)
rank([1 R R R ]) = K+ 2, Equation (21) implies that m is also inefficient.
Next, we show that Equation (5) holds. Using Equation (21) and the definitions of wand
w , we have
[ ] = Vm =
1
R 1R
A CR
+
R 1R
F CR
.
(23)
It follows from Equation (23) that
= m Vm =
1
R R
A CR
+
R R
F CR
. (24)
25
7/31/2019 proiect de tradus
28/38
The fact that = and Equations (23) and (24) imply that
R = R + (R R )
[(R R ) (R R )]
1
A CR
F CR
.
(25)
Using Equations (22) and (25), we have Equation (5). Note that > 0 since > 0.
Proof of Theorem 5. Observe that w solves
max w R
2w Vw (26)
s.t. w 1 = 1. (27)
A first-order condition for w to solve to problem (26) subject to constraint (27) is
R Vw + 1 = 0, (28)
where is the Lagrange multiplier associated with this constraint. Since rank(V) = J,
Equation (28) implies that
w =V R + V 1
. (29)
Using the definition of w and w , and Equation (29), we have Equation (7).
We now find . Premultiplying Equation (29) by 1 and using constraint (27), we have
1 = . Hence, = .
Proof of Theorem 6. Note that the ST constraints are assumed to bind. Hence, w solves
max w R
2w Vw (30)
s.t. w 1 = 1 (31)
w R = T , s = s ,...,s . (32)
26
7/31/2019 proiect de tradus
29/38
A first-order condition for w to solve problem (30) subject to constraints (31) and (32) is
R Vw + 1 +
R = 0, (33)
where , ,..., are Lagrange multipliers associated with these constraints. Since rank(V) =
J, Equation (33) implies that
w =V R + V 1 +
V R
. (34)
Using the definition of w , w , w ,...,w and Equation (34), we have Equation (8).
We now find , ,..., . Premultiplying Equation (34) by 1 and using constraint
(31), we have
1 =A + C+
F
. (35)
It follows from Equation (35) that
C+
F = A. (36)
Premultiplying Equation (34) by R and using constraint (32), we have
T =L + F +
L
, s = s ,...,s , (37)
where L R V R and L R V R . It follows from Equation (37) that
F + L = T L , s = s ,...,s . (38)Let X [ ] , U [1 R R ], Y U V U , and
Z [ A (T + L ) (T + L )] . Using Equations (36) and (38), we
have Y X =Z . Hence, X =Y Z .
Proof of Theorem 7. The desired result follows from Equations (29) and (34).
27
7/31/2019 proiect de tradus
30/38
Proof of Theorem 8. We begin by showing that the market portfolio m is inefficient.
Using Equations (7) and (8), we have
m = 1 w + w + w , (39)where
= (1 ) + , (40)
= , s = s ,...,s . (41)
Since (1) = 0 for s = s ,...,s , (2) portfolio w is inefficient for s = s ,...,s , and (3)
rank([1 R R R ]) = K+ 2, Equation (39) implies that m is also inefficient.
Next, we show that Equation (10) holds. Using Equation (39) and the definitions of w ,
w , and w , we have
[ ] = Vm =
1
1
C+
R
A+
R
F. (42)
It follows from Equation (42) that
= m Vm =
1
1
C+
R
A+
R
F. (43)
The fact that = and Equations (42) and (43) imply that
R = 1
A
C+ R
1
A
C
R R
A
F. (44)
Using Equations (40), (41), and (44), we have Equation (10). Note that > 0 since
> 0.
28
7/31/2019 proiect de tradus
31/38
7/31/2019 proiect de tradus
32/38
Francis, J.C., 1978, Portfolio Analysis of Asset and Liability Management in Small-, Medium-and Large-Sized Banks, Journal of Monetary Economics 4, 459480.
Goyal, A., Santa-Clara, P. 2003, Idiosyncratic Risk Matters!, Journal of Finance 58, 9751007.
Grossman, S.J., Vila, J.-L., 1992, Optimal Dynamic Trading with Leverage Constraints,Journal of Financial and Quantitative Analysis 27, 151168.
Grossman, S.J., Zhou, Z., 1993, Optimal Investment Strategies for Controlling Drawdowns,Mathematical Finance 3, 241276.
Guo, H., Savickas, R., 2006, Idiosyncratic Volatility, Stock Market Volatility, and ExpectedStock Returns, Journal of Business and Economic Statistics 24, 4356.
Hart, O.D., Jaffee, M.D., 1974, On the Application of Portfolio Theory to Depository Finan-cial Intermediaries, Review of Economic Studies 41, 129147.
Harvey, C.R., Siddique, A., 2000, Conditional Skewness in Asset Pricing Tests, Journal ofFinance 55, 12631295.
Huang, C., Litzenberger, R.H., 1988, Foundations for Financial Economics, Prentice-Hall,Englewood Cliffs, NJ.
Hull, J.C., 2006, Options, Futures, & Other Derivatives, Pearson Prentice Hall, Upper SaddleRiver, NJ.
Jagannathan, R., Korajczyk, R.A., 1986, Assessing the Market Training Performance ofManaged Portfolios, Journal of Business 59, 217235.
Jorion, P., 2002, How Informative Are Value-at-Risk Disclosures?, Accounting Review 77,911931.
Kim, D., Santomero, A.M., 1988, Risk in Banking and Capital Regulation, Journal of Fi-nance 43, 12191233.
Koehn, M., Santomero, A.M., 1980, Regulation of Bank Capital and Portfolio Risk, Journalof Finance 35, 12351244.
Levy, H., 1978, Equilibrium in an Imperfect Market, American Economic Review 68, 643658.
Lim, K.-G., 1989, A New Test of the Three-Moment Capital Asset Pricing Model, Journalof Financial and Quantitative Analysis 24, 205216.
Lintner, J., 1965, The Valuation of Risk Assets and the Selection of Risky Investments inStock Portfolios and Capital Budgets, Review of Economics and Statistics 47, 1337.
Kraus, A., Litzenberger, R.H. 1976, Skewness Preference and the Valuation of Risk Assets,Journal of Finance 31, 10851100.
30
7/31/2019 proiect de tradus
33/38
Malkiel, B.G., Xu, Y., 2002, Idiosyncratic Risk and Security Returns, working paper, Uni-versity of Texas at Dallas.
Malkiel, B.G., Xu, Y., 2003, Investigating the Behavior of Idiosyncratic Volatility, Journalof Business 76, 613644.
Markowitz, H.M., 1952, Portfolio Selection, Journal of Finance 7, 7791.
Markowitz, H.M., 1959, Portfolio Selection: Efficient Diversification of Investments, YaleUniversity Press, New Haven, CT.
Markowitz, H.M., 1987, Mean-Variance Analysis in Portfolio Choice and Capital Markets,Basil Blackwell, New York.
Markowitz, H.M., van Dijk, E.L., 2003, Single-Period MeanVariance Analysis in a ChangingWorld, Financial Analysts Journal 59, 3044.
Merton, R.C., 1972, An Analytic Derivation of The Efficient Portfolio Frontier, Journal of
Financial and Quantitative Analysis 7, 18511872.
Merton, R.C., 1981, On Market Timing and Investment Performance. I. An EquilibriumTheory of Value for Market Forecasts, Journal of Business 54, 363406.
Merton, R.C., 1987, A Simple Model of Capital Market Equilibrium with Incomplete Infor-mation, Journal of Finance 42, 483510.
Mossin, J., 1966, Equilibrium in a Capital Asset Market, Econometrica 34, 768783.
Rochet, J.-C., 1992, Capital Requirements and the Behaviour of Commercial Banks, Euro-pean Economic Review 36, 11371178.
Roll, R., 1992, A Mean/Variance Analysis of Tracking Error, Journal of Portfolio Manage-ment 18, 13-22.
Scholes, M.S., 2000, Crisis and Risk Management, American Economic Review 90, 1721.
Sharpe, W.F., 1964, Capital Asset Prices: A Theory of Market Equilibrium Under Conditionsof Risk, Journal of Finance 19, 425442.
Sharpe, W.F., 1987, Asset Allocation Tools, The Scientific Press, Palo Alto, CA.
Shleifer, A., Vishny, R.W., 1997, The Limits of Arbitrage, Journal of Finance 52, 3555.
31
7/31/2019 proiect de tradus
34/38
Figure 1. Optimal portfolios and the market portfolio in the presence of a
riskfree security
This figure illustrates the unconstrained and constrained agents optimal portfolios (w and
w
) and the market portfolio (m
), represented by, respectively, points u, c, and m, inthe presence of a riskfree security. Also shown are w , w , w , and w , represented by,
respectively, points f, t, s , and x. The line represents combinations of w and w (i.e.,
the efficient frontier). The dashed hyperbola represents combinations ofw and w . The
dotted line represents combinations ofw and w .
f
t
s
u
c
mx
R
1
32
7/31/2019 proiect de tradus
35/38
Figure 2. Optimal portfolios and the market portfolio in the absence of a
riskfree security
This figure illustrates the unconstrained and constrained agents optimal portfolios (w and
w
) and the market portfolio (m
), represented by, respectively, points u, c, and m, inthe absence of a riskfree security. Also shown are w , w , w , and w , represented by,
respectively, points a, b, s , and x. The thick hyperbola represents combinations of w and
w (i.e., the mean-variance boundary). The dashed hyperbola represents combinations ofw
and w . The dotted hyperbola represents combinations ofw and w . The thin hyperbola
represents combinations ofw and w .
a
bs
u
c
m
x
R
1
33
7/31/2019 proiect de tradus
36/38
Table 1. Parameters used in the example
This table presents the parameters used in the example of Section 6. There are four risky
securities (j = 1, 2, 3, 4) and a riskfree security (j = 5). Panel (a) shows their annualized
expected returns (R , j = 1, ..., 5) and standard deviations ( , j = 1, ..., 5). Panel (b) showsthe monthly returns in the states used in stress testing (R and R , j = 1, ..., 5). All
numbers are reported in percentage terms.
(a) Annualized expected returns and standard deviations
j 1 2 3 4 5
R 15.00 13.20 12.00 10.80 3.74
27.71 25.98 22.52 20.78 0.00
(b) Monthly returns in the states used in stress testing
R 28.00 8.00 11.00 10.00 0.31
R 29.00 9.00 22.00 10.00 0.31
34
7/31/2019 proiect de tradus
37/38
Table 2. Optimal portfolios and the market portfolio
This table presents the unconstrained and constrained agents optimal portfolios (w andw )
and the market portfolio (m). Panel (a) provides the portfolios fund weights. The weights
in funds w and w are denoted by, respectively, and . The weights in funds w and
w are denoted by, respectively, and . Panel (b) provides the portfolios security
weights. Panel (c) provides the adjustment in expected returns with the same effect on the
optimal portfolio as the ST constraints and the adjusted expected returns (R , j = 1,..., 4).
This adjustment depends on the Lagrange multipliers associated with the constraints (
and ). Panel (d) provides summary statistics on the optimal portfolios and the market
portfolio. Portfolio ws efficiency loss is denoted by . The rest of the notation is defined
in Table 1. All numbers are reported in percentage terms.
(a) Fund weights
w 36.93 136.93
w 36.93 136.93 35.13 38.74
m 136.93 17.56 19.37
(b) Security weights
j 1 2 3 4 5
w
41.42 30.65 36.36 28.50
36.93w 13.13 44.52 6.24 25.44 36.93
m 14.15 37.58 21.30 26.97
(c) Adjustment in expected returns
j 1 2 3 4
R 15.00 13.20 12.00 10.80
(R R ) + (R R ) 9.23 2.81 5.24 3.31
R 5.77 10.39 6.76 7.49
(d) Summary statistics
j R R R
w 16.32 25.07 21.02 25.74
w 8.79 13.84 3.00 4.00 3.79
m 12.55 18.20 12.01 14.87 0.63
35
7/31/2019 proiect de tradus
38/38
Table 3. Risk premia, security characteristics, and decomposition of
security expected returns
Panel (a) provides the risk premia. The risk premia on idiosyncratic returns in states s
and s are denoted by, respectively, and . Panel (b) presents security characteristics.
The security betas are denoted by , j = 1, ..., 4. The rest of the notation is defined in
Tables 1 and 2. Panel (c) decomposes security expected returns. With the exception of the
idiosyncratic returns in states s and s , all numbers are annualized. With the exception of
the betas, all numbers are reported in percentage terms.
(a) Risk premia
R R 8.81
8.69
7.35
(b) Security characteristics
j 1 2 3 4
1.00 1.22 0.85 0.82
(R R ) (R R ) 15.95 6.66 0.83 0.26
(R R ) (R R ) 14.08 9.14 9.40 2.07
(c) Decomposition of security expected returns
j 1 2 3 4
R 3.74 3.74 3.74 3.74
(R R ) 8.84 10.71 7.50 7.19
CAPM expected return 12.58 14.45 11.24 10.93
[(R R ) (R R )] 1.39 0.58 0.07 0.02
[(R R ) (R R )] 1.03 0.67 0.69 0.15
Idiosyncratic return adjustment
2.42 1.25
0.76 0.13
R 15.00 13.20 12.00 10.80