1

A single-factor security market

The single-index model
Estimating the single-index model
Topic 3 (Ch. 8)
Index Models
2
The success of a portfolio selection rule depends
on the quality of the input list (i.e. the estimates of
expected security returns and the covariance
matrix).

e.g. To analyze 50 stocks, the input list includes:
n = 50 estimates of expected returns
n = 50 estimates of variances
(n
2
- n)/2 = 1,225 estimates of covariances
1,325 estimates

A Single-Factor Security Market
3
If n = 3,000 (roughly the number of NYSE stocks),
we need more than 4.5 million estimates.

Errors in the assessment or estimation of
correlation coefficients can lead to nonsensical
results. This can happen because some sets of
correlation coefficients are mutually inconsistent.

e.g.




Asset
Standard
Deviation
(%)
Correlation Matrix
A B C
A 20 1.0 0.9 0.9
B 20 0.9 1.0 0.0
C 20 0.9 0.0 1.0
4
Construct a portfolio with weights: -1.00; 1.00; 1.00,
for assets A; B; C, respectively, and calculate the
portfolio variance.

Portfolio variance = -240!

Covariances between security returns tend to be
positive because the same economic forces affect the
fortunes of many firms (e.g. business cycles, interest
rates, technological changes, etc.).

All these (interrelated) factors affect almost all
firms. Thus, unexpected changes in these variables
cause, simultaneously, unexpected changes in the
rates of return on the entire stock market.
5
Suppose that we summarize all relevant economic
factors by one macroeconomic indicator and
assume that it moves the security market as a whole.

We further assume that, beyond this common effect,
all remaining uncertainty in stock returns is firm
specific (i.e. there is no other source of correlation
between securities).

Firm-specific events would include new inventions,
deaths of key employees, and other factors that
affect the fortune of the individual firm without
affecting the broad economy in a measurable way.
6
We can summarize the distinction between
macroeconomic and firm-specific factors by writing
the holding-period return on security i as:


where E(r
i
): expected return on the security i as
of the beginning of the holding period;
m: impact of unanticipated macro events
on all securities return during the period;
e
i
: impact of unanticipated firm-specific events.

Note: Both m and e
i
have 0 expected values because
each represents the impact of unanticipated events,
which by definition must average out to 0.
i i i
e m r E r + + = ) (
7
Since m and e
i
are uncorrelated, the variance of r
i

arises from two uncorrelated sources, systematic
and firm specific.


Since m is also uncorrelated with any of the firm-
specific surprises, the covariance between any two
securities i and j is
) (
2 2 2
i m i
e o o o + =
2
) , ( ) , (
m j i j i
e m e m Cov r r Cov o = + + =
8
Some securities will be more sensitive than others to
macroeconomic shocks.

We can capture this refinement by assigning each firm
a sensitivity coefficient to macro conditions.

Thus, if we denote the sensitivity coefficient for firm i
by |
i
, we have the following single-factor model:



The systematic risk of security i is determined by its
beta coefficient (|
i
).
i i i i
e m r E r + + = | ) (
9
The variance of the rate of return on each
security includes 2 components:
: variance attributable to the uncertainty
of the common macroeconomic factor (i.e.
systematic risk)
: variance attributable to firm-specific
uncertainty.





2 2
m i
o |
) (
2
i
e o
) (
2 2 2 2
i m i i
e
10
The covariance between any pair of securities
is determined by their betas:

2
m j i
) , ( ) , (
j j i i j i
e m e m Cov r r Cov + + = | |
) , ( m m Cov
j i
11
To make the single-factor model operational,
we use the rate of return on a broad index of
securities (such as S&P 500) as a proxy for the
common macroeconomic factor.

This approach leads to an equation similar to
the single-factor model, which is called the
single-index model, because it uses the market
index to proxy for the common factor.

The Single-Index Model
12
The regression equation of the
single-index model
Denote the market index by M, with excess
return of R
M
= r
M
- r
f
and standard
deviation of
M.

Excess return of a security:
R
i
= r
i
r
f
13
Collect a historical sample of paired observations
and regress R
i
(t) on R
M
(t), where t denotes the date
of each pair of observations.

The regression equation is




Intercept:

i
: the security is expected excess return when the
market excess return is zero.
) ( ) ( ) ( t e t R t R
i M i i i
+ + = | o
14
Slope coefficient:

i
: the security is sensitivity to the market index.

For every + (or -) 1% change in the market excess
return, the excess return on the security will
change by + (or -)
i
%.

Residual:
e
i
is the zero-mean, firm specific surprise in the
security return in time t.
15
The expected return-beta
relationship
) ( ) (
M i i i
R E R E | o + =
part of a securitys
risk premium is due
to the risk premium
of the market index
systematic risk
premium
nonmarket
premium
0 ) ( : =
i
e E Note
16
Risk and covariance in the
single-index model
Recall that we have the following equation:



The variance of the rate of return on each security
includes 2 components:

: variance attributable to the uncertainty of the
market index

2 2
M i
o |
17
: variance attributable to firm-specific uncertainty.



(total risk = systematic risk + firm-specific risk)

Note:

The covariance between R
M
and e
i
is zero because e
i
is
defined as firm specific (i.e. independent of movements
in the market).

) (
2
i
e o
) (
2 2 2 2
i M i i
e o o | o + =
18
The covariance between the rates of return on 2
securities:



Note:
Since o
i
and o
j
are constants, their covariance with any
variable is zero.
Further, the firm-specific terms (e
i
, e
j
) are assumed
uncorrelated with the market and with each other.



Covariance = Product of betas Market index risk


) , ( ) , ( ) , (
j M j j i M i i j i j i
e R e R Cov R R Cov r r Cov + + + + = = | o | o
2
) , ( ) , (
M j i M j M i j i
R R Cov r r Cov o | | | | = =
19
The covariance between the return on stock i and the
market index:





Notes:
We can drop o
i
from the covariance terms because o
i

is a constant and thus has zero covariance with all
variables.

The firm-specific or nonsystematic component is
independent of the marketwide or systematic
component (i.e. Cov(e
i
, R
M
) = 0).
2
) , ( ) , (
) , ( ) , ( ) , (
M i
M i M M i
M i M i M i M i
R e Cov R R Cov
R e R Cov R R Cov r r Cov
o |
|
|
=
+ =
+ = =
20
The correlation coefficient between the rates of return
on 2 securities:





j i
j i
j i j i
R R Cov
R R Corr r r Corr
o o
) , (
) , ( ) , ( = =
M j
M j
M i
M i
M j M i
M j M i
j i
M j i
o o
o |
o o
o |
o o o o
o | o |
o o
o | |
2
2
2 2 2
= = =
M j
M j
M i
M i
) r , r ( Cov
) r , r ( Cov
o o o o
=
) , ( ) , (
M j M i
r r Corr r r Corr =
(product of correlations with the market index)
21
If we have:
n estimates of the extra-market expected excess returns,
i
n estimates of the sensitivity coefficients,
i
n estimates of the firm-specific variances,
2
(e
i
)
1 estimate for the market risk premium,
1 estimate for the variance of the (common)
macroeconomic factor,
M
2

then these (3n + 2) estimates will enable us to prepare
the input list for this single-index security universe.


The set of estimates needed
for the single-index model
22
For n = 50: need 152 estimates (not 1,325 estimates).

n = 3,000: need 9,002 estimates (not 4.5 million).
23
Suppose that we choose an equally weighted portfolio
of n securities (I.e. w
i
= 1/n).

The excess rate of return on each security is:


The excess return on the portfolio of securities:




Note:



i M i i i
e R R + + = | o

+ + =

+ +

= = =
= = =
= = =
n
i
i
n
i
n
i
M i i
n
i
i M i i
n
i
n
i
i i i p
e
n
R
n n
e R
n
R
n
R w R
1 1 1
1 1 1
1
)
1
(
1
) (
1 1
| o
| o
P M P P P
e R R + + = | o
The index model and diversification
24
The portfolio has a sensitivity to the market given by:
(the average of the individual |
i
s)

It has a nonmarket return component of a constant
(intercept):

(the average of the individual alphas)

It has a zero mean variable:


(the average of the firm-specific components)

=
=
n
i
i p
n
1
1
| |

=
=
n
i
i p
n
1
1
o o

=
=
n
i
i p
e
n
e
1
1
25
The portfolios variance is:

The systematic risk component of the portfolio variance
(the component that depends on marketwide
movements) is and depends on the sensitivity
coefficients of the individual securities.

This part of the risk depends on portfolio beta and ,
and will persist regardless of the extent of portfolio
diversification.

No matter how many stocks are held, their common
exposure to the market will be reflected in portfolio
systematic risk.
2 2
M P
o |
2
M
o
) (
2 2 2 2
p M p p
e o o | o + =
26
In contrast, the nonsystematic component of the
portfolio variance is o
2
(e
P
) and is attributable to firm-
specific components e
i
.
Because the e
i
s are uncorrelated, we have:


where : the average of the firm-specific variances.
Because this average is independent of n, when n
gets large, o
2
(e
P
) becomes negligible.

Thus, as more and more securities are added to the
portfolio, the firm-specific components tend to cancel
out, resulting in ever-smaller nonmarket risk.
) (
2
e o
) (
2
e o
) (
1
) (
1
) (
2 2
1
2
2
e
n
e
n
e
i
n
i
P
o o o =

|
.
|

\
|
=
=
27
28
Summary:

As more and more securities are combined into a
portfolio, the portfolio variance decreases because of
the diversification of firm-specific risk.

However, the power of diversification is limited.

Even for very large n, part of the risk remains because
of the exposure of virtually all assets to the common, or
market, factor.

Therefore, this systematic risk is said to be
nondiversifiable.
29
The single-index model


suggests how we might go about actually measuring
market and firm-specific risk.

Suppose that we observe the excess return on the
market index and a specific asset over a number of
holding periods.

We use as an example monthly excess returns on the
S&P 500 index and GM stock for a one-year period.
i M i i i
e R R + + = | o
Estimating the Single-Index Model
30
31
We can summarize the results for a sample period in a scatter diagram:
32
The single-index model states that the relationship
between the excess returns on GM and the S&P 500 is
given by the following regression equation:



In this single-variable regression equation, the
dependent variable plots around a straight line with an
intercept o and a slope |.

The deviations from the line (e) are assumed to be
mutually uncorrelated and uncorrelated with the
independent variable.


) ( ) ( ) ( t e t R t R
GM M GM GM GM
+ + = | o
33
The sensitivity of GM to the market, measured by |
GM
, is
the slope of the regression line.

The intercept of the regression line is o
GM
, representing
the average firm-specific return when the markets excess
return is zero.

Deviations of particular observations from the regression
line in any period are denoted e
GM
, and called residuals
(i.e. each of these residuals is the difference between the
actual security return and the return that would be
predicted from the regression equation describing the
usual relationship between the security and the market).
Thus, residuals measure the impact of firm-specific events.
34
Estimating the regression equation of the single-index
model gives us the security characteristic line (SCL).

The SCL is a plot of the typical excess return on a
security as a function of the excess return on the market.

Compute o
GM
and |
GM
:

Let y
t
: excess return on GM in month t
x
t
: excess return on the market (S&P 500) in month t
n: the total number of months.

35
The estimate of beta coefficient (i.e. the slope of the
regression line SCL):





The intercept of the regression line:

=
= =
= = =
n
t
n
t
t t
n
t
t
n
t
n
t
t t t
GM
x x n
y x y x n
1 1
2 2
1 1 1
) ( ) (
) )( ( ) (

|
n
x y
n
t
n
t
t GM t
GM

=
= = 1 1
) (

|
o
36
Month y
t
x
t
x
t
y
t
(x
t
)
2
January 5.41 7.24 39.1684 52.4176
February -3.44 0.93 -3.1992 0.8649
March -8.79 -0.38 3.3402 0.1444
April -8.08 -1.01 8.1608 1.0201
May 7.10 4.92 34.9320 24.2064
June -0.03 1.18 -0.0354 1.3924
July -2.36 -0.83 1.9588 0.6889
August -3.55 -0.91 3.2305 0.8281
September -1.16 -4.18 4.8488 17.4724
October -1.02 3.97 -4.0494 15.7609
November 6.32 6.25 39.5000 39.0625
December 2.43 3.90 9.4770 15.2100
Sum -7.17 21.08 137.3325 169.0686
37
135 . 1
) 08 . 21 ( ) 0686 . 169 ( 12
) 17 . 7 )( 08 . 21 ( ) 3325 . 137 ( 12
2
=

=
= =
= = =
n
t
n
t
t t
n
t
t
n
t
n
t
t t t
GM
x x n
y x y x n
1 1
2 2
1 1 1
) ( ) (
) )( ( ) (

|
month per % 59 . 2
12
) 08 . 21 )( 135 . 1 ( 17 . 7
=

=
n
x y
n
t
n
t
t GM t
GM

=
= = 1 1
) (

|
o
38
Compute residuals:

For each month t, our estimate of the residual is the
deviation of GMs excess return from the prediction
of the SCL:
Deviation = Actual Predicted Return




These residuals are estimates of the monthly
unexpected firm-specific component of the rate of
return on GM stock.
)] (

[ ) ( ) ( t R t R t e
M GM GM GM GM
| o + =
)] ( 135 . 1 59 . 2 [ ) ( t R t R
M GM
+ =
39
Month y
t
(actual) x
t
(actual) Predicted y
t
Residual (Residual)
2
January 5.41 7.24 5.63 -0.22 0.05
February -3.44 0.93 -1.53 -1.91 3.63
March -8.79 -0.38 -3.02 -5.77 33.28
April -8.08 -1.01 -3.74 -4.34 18.87
May 7.10 4.92 2.99 4.11 16.86
June -0.03 1.18 -1.25 1.22 1.49
July -2.36 -0.83 -3.53 1.17 1.37
August -3.55 -0.91 -3.62 0.07 0.01
September -1.16 -4.18 -7.33 6.17 38.12
October -1.02 3.97 1.92 -2.94 8.62
November 6.32 6.25 4.50 1.82 3.30
December 2.43 3.90 1.84 0.59 0.35
Sum -7.17 21.08 126
40
Hence, we can estimate the firm-specific variance:




The standard deviation of the firm-specific component
of GMs return:


which is equal to the standard deviation of the
regression residual.

=
=
12
1
2 2
6 . 12
2 12
126
)] ( [
2
1
) (
t
GM
t e
n
e o
month per e
GM
% 55 . 3 6 . 12 ) ( = = o
41
Practitioners often use a modified
index model that uses total rather than
excess returns (deviations from T-bill
rates) in the regressions:



instead of




* e br a r
M
The Industry Version of the Index Model
e r r r r
f M f
+ + = ) ( | o
42
To see the impact of this departure:






If r
f
is constant over the sample period, both
equations have the same independent variable r
M

and residual e.

Thus, the slope coefficient will be the same in the
two equations.

e r r r r
f M f
+ + + = | | o
e r r
M f
+ + + = | | o ) 1 (
43
However, the intercept is really an estimate of



The apparent justification for this procedure is
that, on a monthly basis, r
f
(1 - |) is small.

But, note that for 1, the regression intercept will
not equal the index model alpha.
). 1 ( | o +
f
r
44
Betas estimated form past data may not be the
best estimates of future betas.

This suggests that we might want a forecasting
model for beta.

One simple approach would be to collect data on
beta in different periods and then estimate a
regression equation:

Current beta = a + b (Past beta)

Given estimates of a and b, we would then forecast
future betas using the rule:
Forecast beta = a + b (Current beta)

Predicting Betas
45
However, there is no reason to limit ourselves to
such simple forecasting rules.

Why not also investigate the predictive power of
other financial variables in forecasting beta?

Rosenberg and Guy find the following variables
help predict betas:

Variance of earnings.
Variance of cash flow.
Growth in earnings per share.
Market capitalization (firm size).
Dividend yield.
Debt-to-asset ratio.
46
Rosenberg and Guy also find that even after
controlling for a firms financial
characteristics, industry group helps to
predict beta.