Basic Investment Models and Their Statistical Analysis

Outline Asset Returns Markowitzs Portfolio Theory CAPM Multifactor Pricing Models Resampling Methods
Basic Investment Models and Their

Statistical Analysis
Haipeng Xing
Haipeng Xing SUNY Stony Brook
Basic investment models and their statistical analysis
Outline
1 Asset Returns
2
Markowitzs Portfolio Theory
3
Capital Asset Pricing Model (CAPM)
4
Multifactor Pricing Models
5 Applications of resampling to portfolio management
Discrete returns
Let P
t
denote the asset price at time t. Suppose the asset does not have
dividends over the period from time t 1 to time t.
The one-period net return on this asset is R
t
= (P
t
P
t1
)/P
t1
,
and the one-period gross return is P
t
/P
t1
= 1 + R
t
.
The gross return over k periods is dened as
1 + R
t
(k) = P
t
/P
tk
=
k1
j=0
(1 + R
tj
),
and the net return over these periods is R
t
(k). In practice, we
usually use years as the time unit. The annualized gross return for
holding an asset over k years is (1 + R
t
(k))
1/k
, and the annualized
net return is (1 + R
t
(k))
1/k
1.
Continuously compounded return (log return)
The logarithmic return or continuously compounded return on an
asset is dened as r
t
= log(P
t
/P
t1
).
One property of log returns is that, as the time step t of a period
approaches 0, the log return r
t
is approximately equal to the net
return:
r
t
= log(P
t
/P
t1
) = log(1 + R
t
) R
t
.
A k-period log return is the sum of k simple single-period log returns
(the additivity of multiperiod returns):
r
t
[k] = log
P
t
P
tk
=
k1
j=0
log(1 + R
tj
) =
k1
j=0
r
tj
.
Adjustment for dividends
Many assets pay dividends periodically. In this case, the denition of
asset returns has to be modied to incorporate dividends. Let D
t
be
the dividend payment between times t 1 and t. The net return and
the continuously compounded return are modied as
R
t
=
P
t
+ D
t
P
t1
1, r
t
= log(P
t
+ D
t
) log P
t1
.
Multiperiod returns can be similarly modied. In particular, a
k-period log return now becomes
r
t
[k] = log
_
k1
j=0
P
tj
+ D
tj
P
tj1
_
=
k1
j=0
log
_
P
tj
+ D
tj
P
tj1
_
.
Excess and portfolio returns
Excess return refers to the dierence r
t
r
t
between the assets log
return r
t
and the log return r
t
on some reference asset, which is
usually taken to be a riskless asset such as a short-term U.S.
Treasury bill.
Suppose one has a portfolio consisting of p dierent assets. Let w
i
be the weight of the portfolios value invested in asset i. Suppose
R
it
and r
it
are the net return and log return of asset i at time t,
respectively. The overall net return R
t
and a corresponding formula
for the log return r
t
of the portfolio are
R
t
=
p
i=1
w
i
R
it
, r
t
= log
_
1 +
p
i=1
w
i
R
it
_
i=1
w
i
r
it
.
Statistical models for asset prices and returns
A commonly used model for risky asset prices P
t
is geometric
Brownian motion (GBM), with volatility and instantaneous rate of
return , dP
t
/P
t
= dt +dw
t
, where {w
t
, t 0} is Brownian
motion. The price process P
t
is called, and has the explicit
representation P
t
= P
0
exp{(

2
2
)t +w
t
}.
The discrete-time analog of this price process has returns
r
t
= log(P
t
/P
t1
) that are i.i.d. N(,
2
) with =
2
/2.
Remark 1
The empirical distributions of asset prices and returns are much more
complicated and voluminous than those summarized here, they are
usually characterized by more advanced probabilistic or statistical models.
Portfolio weights
For a single-period portfolio of p assets with weights w
i
, the return of the
portfolio over the period can be represented by R =
p
i=1
w
i
R
i
. The
mean and variance
2
of the portfolio return R are given by
=
p
i=1
w
i
E(R
i
),
2
=
1i,jp
w
i
w
j
Cov(R
i
, R
j
).
With the weights w
i
satisfying the constraints
p
i=1
w
i
= 1 (and w
i
0
if short selling is not allowed).
Remark 2
Such diversication via a portfolio tends to reduce the risk, as
measured by the returns standard deviation, of the risky investment.
Geometry of ecient sets
A
P
1
P
2
= 1
= 1
= 1 = 0.6 = 0.3
Figure 1: Feasible region for two
assets.
Consider the case of p = 2 risky
assets whose returns have means
1
,
2
, standard deviations
1
,
2
,
and correlation coecient . Let
w
1
= and w
2
= 1 ( [0, 1]).
Then the mean return of the
portfolio is () =
1
+ (1 )
2
,
and its volatility () is given by
2
() =
2
2
1
+ 2(1 )
1
2
+(1 )
2
2
2
. Figure 1 plots the
curve {((), ()) : 0 1}
for dierent values of .
Figure 2: Feasible region for p 3
assets.
The set of points in the (, ) plane
that correspond to the returns of
portfolios of the p assets is called a
feasible region. For p 3, the
feasible region is a connected
two-dimensional set. It is also
convex to the left in the sense that
given any two points in the region,
the line segment joining them does
not cross the left boundary of the
region.
Efficient
frontier
Minimum!variance
point
Figure 3: Ecient frontier and
minimum-variance point.
For a given value of the mean
return, the feasible point with the
smallest lies on this left boundary,
which is the minimum-variance
portfolio (MVP). For a given value
of volatility, investors prefer the
portfolio with the largest mean
return, which is achieved at an
upper left boundary point of the
feasible region. The upper portion of
the minimum-variance set is called
the ecient frontier.
Computation of ecient portfolios
Let r = (R
1
, . . . , R
p
)
T
denote the vector of returns of p assets,
1 = (1, . . . , 1)
T
, w = (w
1
, . . . , w
p
)
T
, = (
1
, . . . ,
p
)
T
= (E(R
1
),
. . . , E(R
p
))
T
, and = (Cov(R
i
, R
j
))
1ijp
. Consider the case where
short selling is allowed. Given a target value
for the mean return of

the portfolio, the weight vector w of an ecient portfolio can be
characterized by
w
e
= arg min
w
w
T
w subject to w
T
=
, w
T
1 = 1,
The method of Lagrange multipliers leads to the the explicit solution
w
e
=
_
B
1
1 A
1
+
_
C
1
A
1
1
_
__
D
when is nonsingular, where A =
T
1
1 = 1
T
1
, B =
T
1
,
C = 1
T
1
1, and D = BC A
2
.
The variance of the return on this ecient portfolio is
2
e
=
_
B 2
A +
2
C
__
D.
The
that minimizes
2
e
is given by
minvar
=
A
C
,
which corresponds to the global MVP with variance
2
minvar
= 1/C and
weight vector
w
minvar
=
1
1
_
C.
For two MVPs with mean returns
p
and
q
, their weight vectors are
given by (5) with
=
p
,
q
, respectively. From this it follows that the
covariance of the returns r
p
and r
q
is given by
Cov(r
p
, r
q
) =
C
D
_
A
C
__
A
C
_
+
1
C
.
When short selling is not allowed, we need to add the constraint w
i
0
for all i (denoted by w 0). Hence the optimization problem (??) has
to be modied as
w
e
= arg min
w
w
T
w subject to w
T
=
, w
T
1 = 1, w 0.
Such problems do not have explicit solutions by transforming them to a
system of equations via Lagrange multipliers. Instead, we can use
quadratic programming to minimize the quadratic objective function
w
T
w under linear equality constraints and nonnegativity constraints.
Estimation of and and an example
Table 1 gives the means and covariances of the monthly log returns of six
stocks, estimated from 63 monthly observations during the period August
2000 to October 2005. The stocks cover six sectors in the Dow Jones
Industrial Average: American Express (AXP), Citigroup Inc. (CITI),
Exxon Mobil Corp. (XOM), General Motors (GM), Intel Corp. (INTEL),
and Pzer Inc. (PFE).
Table 1: Estimated mean (in parentheses, multiplied by 10
2
) and covariance
matrix (multiplied by 10
4
) of monthly log returns.
AXP CITI XOM GM INTEL PFE
AXP (0.033) 9.01
CITI (0.034) 5.69 9.64
XOM (0.317) 2.39 1.89 5.25
GM (0.338) 5.97 4.41 2.40 20.2
INTEL (0.701) 10.1 12.1 0.59 12.4 46.0
PFE (0.414) 1.46 2.34 9.85 1.86 1.06 5.33
Estimation of and and an example
0.0178 0.018 0.0182 0.0184 0.0186 0.0188 0.019
!5
0
5
10
15
20
x 10
!4
M
o
n
t
h
ly

lo
g

r
e
t
u
r
n
Monthly standard deviation
Figure 4: Estimated ecient frontier of portfolios that
consist of six assets.
Figure 4 shows the
plug-in ecient
frontier for these six
stocks allowing short
selling. By plug-in
we mean that the
mean and
covariance in (5)
are substituted by the
estimated values
and

given in Table
1.
The CAPM
The capital asset pricing model, introduced by Sharpe (1964) and Lintner
(1965), builds on Markowitzs portfolio theory to develop economy-wide
implications of the trade-o between return and risk, assuming that there
is a risk-free asset and that all investors have homogeneous expectations
and hold mean-variance-ecient portfolios.
Suppose the market has a risk-free asset with return r
f
(interest rate)
besides n risky assets. If both lending and borrowing of the risk-free asset
at rate r
f
are allowed, the feasible region is an innite triangular region.
The ecient frontier is a straight line that is tangent to the original
feasible region of the n risky assets at a point M, called the tangent
point; see Figure 5. This tangent point M can be thought of as an index
fund or market portfolio.
The CAPM
M
r
f
Efficient
frontier
M
r
f
Efficient
frontier
Figure 5: Minimum-variance portfolios of risky assets and a risk-free asset.
Left panel: short selling is allowed. Right panel: short selling is not allowed.
The CAPM
One Fund Theorem
There is a single fund M of risky assets such that any ecient portfolio
can be constructed as a linear combination of the fund M and the
risk-free asset.
When short selling is allowed, the minimum-variance portfolio (MVP)
with the expected return
can be computed by solving the optimization

problem
min
w
w
T
w subject to w
T
+ (1 w
T
1)r
f
=
.
The problem has an explicit solution for w when is nonsingular:
w
e
=
(
r
f
)
( r
f
1)
T
1
( r
f
1)
1
( r
f
1)

1
( r
f
1)/[1
T
1
( r
f
1)](The fund)
Sharpe ratio and the capital market line
For a portfolio whose return has mean and variance
2
, its Sharpe
ratio is ( r
f
)/, which is the expected excess return per unit of
risk.
The straight line joining (0, r
f
) and the tangent point M in Figure
5, which is the ecient frontier in the presence of a risk-free asset, is
called the capital market line and given by
= r
f
+

M
r
f
M
;
i.e., the Sharpe ratio of any ecient portfolio is the same as that of
the market portfolio.
Beta and the security market line
The beta, denoted by
i
, of risky asset i that has return r
i
is dened
by
i
= Cov(r
i
, r
M
)/
2
M
. The CAPM relates the expected excess
return (also called the risk premium)
i
r
f
of asset i to its beta
via
i
r
f
=
i
(
M
r
f
),
which is referred to as the security market line.
The above linear relationship can be rewritten as
r
i
r
f
=
i
(r
M
r
f
) +
i
,
in which E(
i
) = 0 and Cov(
i
, r
M
) = 0, it follows that
2
i
=
2
i
2
M
+ Var(
i
),
decomposing the variance
2
i
of the ith asset return as a sum of the
systematic risk
2
i
2
M
, which that is associated with the market, and
the idiosyncratic risk, which is unique to the asset and uncorrelated
with market movements.
Investment implications
Using as a measure of risk, the Treynor index is dened by
( r
f
)/.
The Jensen index is the in the generalization of CAPM to
r
f
= +(
M
r
f
). An investment with a positive is
considered to perform better than the market.
Jensen (1968) perform an empirical study using the regression model
r
f
= +(
M
r
f
) +. His ndings support the ecient
market hypothesis, according to which it is not possible to
outperform the market portfolio in an ecient market.
Estimation and testing
Let y
t
be a q 1 vector of excess returns on q assets and let x
t
be the
excess return on the market portfolio (or, more precisely, its proxy) at
time t. The CAPM can be associated with the null hypothesis
H
0
: = 0 in the regression model
y
t
= + x
t
+
t
, 1 t n, (1)
where E(
t
) = 0, Cov(
t
) = V, and E(x
t
t
) = 0.
Letting x = n
1
n
t=1
x
t
and y = n
1
n
t=1
y
t
, the ordinary least
squares (OLS) estimates of and are given by
n
t=1
(x
t
x)y
t
n
t=1
(x
t
x)
2
, = y x
. (2)
The maximum likelihood estimate of V is
V = n
1
n
t=1
(y
t

x
t
)(y
t

x
t
)
T
. (3)
The properties of OLS can be used to establish the asymptotic normality
of

and , yielding the approximations
N
_
,
n
t=1
(x
t
x)
2
_
, N
_
,
_
1
n
+
x
2
n
t=1
(x
t
x)
2
_
V
_
,
from which approximate condence regions for and can be
constructed.
When
t
are i.i.d. normal and are independent of the market excess
returns x
t
, ,

, and

V are the maximum likelihood estimates of , ,
and V. Furthermore, the conditional distribution of ( ,
,

V) given
(x
1
, . . . , x
n
) are expressed as
N
_
,
_
1
n
+
x
2
n
t=1
(x
t
x)
2
_
V
_
, (4)
N
_
,
V
n
t=1
(x
t
x)
2
_
, n
V W
q
(V, n 2),
with

V independent of ( ,
). Moreover, we can show that, under H

0
,
_
n q 1
q
_

T
V
1

__
1 +
x
2
n
1
n
t=1
(x
t
x)
2
_
F
q,nq1
. (5)
Note that (5) still holds approximately without the normality assumption
when n q 1 is moderate or large.
An illustrative example
We illustrate the statistical analysis of CAPM with the monthly returns
data of the six stocks in the previous section using the Dow Jones
Industrial Average index as the market portfolio M and the 3-month U.S.
Treasury bill as the risk-free asset. These data are used to estimate
quantities in Table 2.
Table 2: Performance of six stocks from August 2000 to October 2005.
AXP CITI XOM GM INTEL PFE
10
3
0.87 0.81 2.23 2.41 4.31 5.21
p-value
0.72 0.76 0.40 0.59 0.52 0.06

1.23 1.20 0.52 1.44 2.28 0.46
2
M
10
4
5.77 5.48 1.04 7.91 19.8 0.80
10
4
3.50 4.18 4.22 12.0 26.7 4.64
Sharpe10
2
5.49 5.36 5.05 12.1 13.2 26.4
Treynor10
3
1.35 1.38 2.22 3.74 3.96 13.4
refers to the p-value of the t-test of H
0
:
i
= 0.
Empirical literature on the CAPM
Since the development of CAPM in the 1960s, a large body of literature
has evolved on empirical evidence for or against the model. The early
evidence was largely positive, but in the late 1970s, less favorable
evidence began to appear.
Basu (1977) reported the priceearnings-ratio eect: Firms with
low priceearnings ratios have higher average returns, and rms with
high priceearnings ratios have lower average returns than the values
implied by CAPM.
Banz (1981) noted the size eect, that rms with low market
capitalizations have higher average returns than under CAPM.
Empirical literature on the CAPM
Fama and French (1992, 1993) have found that beta cannot explain
the dierence in returns between portfolios formed on the basis of
the ratio of book value to market value of equity.
Jegadesh and Titman (1995) have noted that a portfolio formed by
buying stocks whose values have declined and selling stocks whose
values have risen has a higher average return than predicted by
CAPM.
Remark 3
The empirical study for or against CAPM might involves the issues of
data snooping, selection bias, and proxy bias.
Arbitrage pricing theory (APT)
Multifactor pricing models generalize CAPM by embedding it in a
regression model of the form
r
i
=
i
+
T
i
f +
i
, i = 1, , p, (6)
in which the r
i
is the return on the ith asset,
i
and
i
are
unknown regression parameters, f = (f
1
, . . . , f
k
)
T
is a regression
vector of factors, and
i
is an unobserved random disturbance that
has mean 0 and is uncorrelated with f .
Ross (1976) introduced the APT which allows multiple risk factors
for asset returns. Unlike the CAPM, APT does not require
identication of the market portfolio and relates the expected return
i
of the ith asset to the risk-free return, or to a more general
parameter
0
without assuming the existence of a risk-free asset,
and to a k 1 vector of risk premiums:
i

0
+
T
i
, i = 1, . . . , p. (7)
Arbitrage pricing theory (APT)
While APT provides an economic theory underlying multifactor
models of asset returns, the theory does not identify the factors.
Approaches to the choice of factors can be broadly classied as
economic and statistical.
The economic approach species (i) macroeconomic and nancial
market variables that are thought to capture the systematic risks of
the economy or (ii) characteristics of rms that are likely to explain
dierential sensitivity to the systematic risks, forming factors from
portfolios of stocks based on the characteristics.
The statistical approach uses factor analysis or PCA (principal
component analysis) to estimate the parameters of model (6) from a
set of observed asset returns.
Factor analysis
Letting r = (r
1
, . . . , r
p
)
T
, = (
1
, . . . ,
p
)
T
, = (
1
, . . . ,
p
)
T
, and B
to be the p k matrix whose ith row vector is
T
i
, we can rewrite the
multifactor pricing model (6) as r = +Bf + with E = Ef = 0 and
E(f
T
) = 0. Note that the regressor f is unobservable.
Let r
t
, t = 1, . . . , n, be independent observations from the model so that
r
t
= +Bf
t
+
t
and E
t
= Ef
t
= 0, E(f
t
T
t
) = 0, Cov(f
t
) = , and
Cov(
t
) = V. Then
E(r
t
) = , Cov(r
t
) = BB
T
+V. (8)
The decomposition of the covariance matrix of r
t
in (8) is the essence
of factor analysis, which separates variability into a systematic part due
to the variability of certain unobserved factors, represented by BB
T
,
and an error (idiosyncratic) part, represented by V.
The factor analysis Identiability
In standard factor analysis, V is assumed to be diagonal; i.e.,
V = diag(v
1
, . . . , v
p
). Since B and are not uniquely determined by
= BB
T
+V, the orthogonal factor model assumes that = I so
that B is unique up to an orthogonal transformation and
r = +B + with Cov(f ) = yields
Cov(r, f ) = E{(r )f
T
} = B = B, (9)
Var(r
i
) =
k
j=1
b
2
ij
+ Var(
i
), 1 i p, (10)
Cov(r
i
, r
j
) =
k
l=1
b
il
b
jl
, 1 i, j p. (11)
The factor analysis MLE
Assuming the observed r
t
to be independent N(, ), the likelihood
function is
L(, ) = (2)
pn/2
(det )
n/2
exp
_
1
2
n
t=1
(r
t
)
T
1
(r
t
)
_
,
with constrained to be of the form = BB
T
+ diag(v
1
, . . . , v
p
), in
which B is p k. The MLE of is r := n
1
n
t=1
r
t
, and we can
maximize
1
2
nlog det()
1
2
tr(W
1
) over of the form above,
where W =
n
t=1
(r
t
r)(r
t
r)
T
. Iterative algorithms can be used to
nd the maximizer

and therefore also

B and v
1
, . . . , v
p
.
The factor analysis factor rotation
In factor analysis, the entries of the matrix

B are called factor loadings.
Since

B is unique only up to orthogonal transformations, the usual
practice is to multiply

B by a suitably chosen orthogonal matrix Q,
called a factor rotation, so that the factor loadings have a simple
interpretable structure. Letting

B
=

BQ, a popular choice of Q is that
which maximizes the varimax criterion
C = p
1
k
j=1
_
p
i=1
b
4
ij

_
p
i=1
b
2
ij
_
2
_
p
_
j=1
Var
_
squared loadings of the jth factor
_
.
The factor analysis factor scores
Since the values of the factors f
t
, 1 t n, are unobserved, it is often
of interest to impute these values, called factor scores, for model
diagnostics. From the model r = Bf + with Cov() = V, the
generalized least squares estimate of f when B, V, and are known is
f = (B
T
V
1
B)
1
B
T
V
1
(r
t
).
Bartlett (1937) therefore suggested estimating f
t
by
f
t
= (
B
T

V
1
B)
1
B
T

V
1
(r
t
r). (12)
The factor analysis the number of factors
The theory underlying multifactor pricing models and factor analysis
assumes that the number k of factors has been specied and does
not indicate how to specify it.
When the r
t
are independent N(, ), we may consider a formal
hypothesis testing that the k-factor model indeed holds. The null
hypothesis H
0
is that = BB
T
+V with V diagonal and B being
p k.
The generalized likelihood ratio statistic that tests the H
0
against
unconstrained is of the form
= n
_
log det
_
B
T
+

V
_
log det
_
_
_
, (13)
where

= n
1
n
t=1
(r
t
r)(r
t
r)
T
is the unconstrained MLE of
.
The factor analysis the number of factors
Under H
0
, is approximately
2
with
1
2
p(p + 1)
_
p(k + 1)
1
2
k(k 1)
_
=
1
2
_
(p k)
2
p k
_
degrees of freedom.
Bartlett (1954) has shown that the
2
approximation to the
distribution of (13) can be improved by replacing n in (13) by
n 1 (2p + 4k + 5)/6, which is often used in empirical studies.
The PCA approach
The fundamental decomposition =
1
a
1
a
T
1
+ +
p
a
p
a
T
p
in
PCA (see Section 2.2.2) can be used to decompose into
= BB
T
+ V. Here
1

p
are the ordered eigenvalues of
, a
i
is the unit eigenvector associated with
i
, and
B = (
_
1
a
1
, . . . ,
_
k
a
k
), V =
p
l=k+1
l
a
l
a
T
l
. (14)
PCA is particularly useful when most eigenvalues of are small in
comparison with the k largest ones, so that k principal components
of r
t
r account for a large proportion of the overall variance. In
this case, we can use PCA to determine k.
The Fama-French three-factor model
Fama and French (1993, 1996) propose a three-factor model which
has the form
E(r
i
) r
f
=
i
+
T
i
(r
M
r
f
, r
S
r
L
, r
H
r
L
)
T
.
The factor r
M
r
f
is the only factor in the CAPM. The factor
r
S
r
L
captures the risk factor in returns related to size. Here
small and large refer to the market value of equity. The factor
r
H
r
L
, which captures the risk factor in returns related to the
book-to-market equity.
Fama and French (1992, 1993) argue that their three-factor model
removes most of the pricing anomalies with CAPM. Because the
factors in the Fama-French model are specied, one can use standard
regression analysis to test the model and estimate its parameters.
Bootstrap estimate of the CAPM
We illustrate how bootstrap resampling can be applied to estimate
the CAPM in Table 2 based on the monthly excess log returns of six
stocks from August 2000 to October 2005. The and in the
table are estimated by applying OLS to the regression model
r
i
r
f
=
i
+
i
(r
M
r
f
) +
i
, in which the market portfolio M is
taken to be the Dow Jones Industrial Average index and r
f
is the
annualized rate of the 3-month U.S. Treasury bill.
Let x
t
= r
M,t
r
f,t
and y
i,t
= r
i,t
r
f,t
. We draw B = 500
bootstrap samples {(x
t
, y
i,t
), 1 t n = 63} from the observed
sample {(x
t
, y
i,t
), 1 t n = 63} and compute the OLS estimates

i
and

i
for the regression model y
i,t
=
i
+
i
x
t
+
t
. We then
report in Table 3 the average values of
i
,

i
, Sharpe index and
Treynor index of the B bootstrap samples, and their standard
deviations (given in parentheses).
Bootstrap estimate of the CAPM
Table 3: Bootstrapping CAPM.
10
3
Sharpe 10
2
Treynor 10
3
AXP 1.00 (0.28) 1.23 (0.02) 4.51 (1.61) 1.14 (0.40)
CITI 0.84 (0.32) 1.20 (0.02) 5.16 (1.60) 1.36 (0.41)
XOM 2.24 (0.33) 0.53 (0.02) 4.69 (1.59) 2.27 (0.75)
GM 1.99 (0.58) 1.44 (0.04) 12.2 (1.65) 4.02 (0.57)
Intel 4.33 (0.74) 2.29 (0.05) 13.0 (1.44) 3.95 (0.44)
Pzer 5.23 (0.33) 0.45 (0.02) 26.2 (1.62) 13.5 (4.05)
Michauds resampled ecient frontier
As pointed out before, the estimated (plug-in) ecient frontier
based on the sample mean and covariance matrix

diers from
the true ecient frontier. Frankfurter, Phillips, and Seagle (1971)
and Jobson and Korkie (1980) have found that portfolios thus
constructed may perform worse than the equally weighted portfolio.
Michaud (1989) proposes to use instead of w the average of
bootstrap weights
w = B
1
B
b=1
w
b
,
where w
b
is the estimated optimal portfolio weight vector based on
the bth bootstrap sample {r
b1
, . . . , r
bn
} drawn with replacement
from the observed sample {r
1
, . . . , r
n
}. Thus, Michauds resampled
ecient frontier corresponds to the plot
_
w
T
w versus w
T
r =
for a ne grid of
values, as shown in Figure 6 (in which we have

used B = 1000) for the six stocks considered in Figure 4.
0.0178 0.018 0.0182 0.0184 0.0186 0.0188 0.019
!5
0
5
10
15
20
x 10
!4
M
o
n
t
h
ly

lo
g

r
e
t
u
r
n
Estimated efficient frontier
Resampled efficient frontier
Figure 6: The estimated ecient frontier (solid curve) and
the resampled ecient frontier (dotted curve) of six U.S.
stocks.
0.0178 0.018 0.0182 0.0184 0.0186 0.0188 0.019
!5
0
5
10
15
20
x 10
!4
M
o
n
t
h
ly

lo
g

r
e
t
u
r
n
Estimated efficient frontier
Resampled efficient frontier
Figure 6: The estimated ecient frontier (solid curve) and
the resampled ecient frontier (dotted curve) of six U.S.
stocks.
Although
Michaud claims
that w provides
an improvement
over w, there
have been no
convincing
theoretical
developments and
simulation studies
to support the
claim.
Bootstrap estimates of performance
Whereas simulation studies of performance require specic
distributional assumptions on r
t
, it is desirable to be able to assess
performance nonparametrically and the bootstrap method provides a
practical way to do so.
The bootstrap samples {r
b1
, . . . , r
bn
; r
b
}, 1 b B, can be used
to estimate the means E(w
T
P
r) and variances Var(w
T
P
r) of various
portfolios P whose weight vectors w
P
may depend on the observed
data (for which E(w
T
P
r) can no longer be written as w
T
P
E(r) since
w
P
is random). Details and illustrative examples are given in Lai,
Xing, and Chen (2011).

Basic Investment Models and Their Statistical Analysis

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Basic Investment Models and Their Statistical Analysis

Uploaded by

Copyright:

Available Formats

Outline Asset Returns Markowitzs Portfolio Theory CAPM Multifactor Pricing Models Resampling Methods

Basic Investment Models and Their

for the mean return of

can be computed by solving the optimization

). Moreover, we can show that, under H

0.72 0.76 0.40 0.59 0.52 0.06

values, as shown in Figure 6 (in which we have

You might also like