Index Models
Chapter 10
10-2
The Need for a Simpler Model
the input list (i.e., list of assets on the market) in the Markovitz portfolio selection model is very important, determining the accuracy of finding efficient portfolios _ however, it involves a lot of calculations _ for example, for n = 50 assets we need to
_
calculate (or, more precisely, estimate):
n = 50 expected returns _ n = 50 variances _ n(n 1) / 2 = 1225 covariances
_
in total, n(n + 3) / 2 = 1325 estimates!
_
10-3
The Need for a Simpler Model (cont.)
also, because we estimate returns, variances and covariances, small errors can have large effects
_
for example, if we estimate wrongly the covariance matrix (mutually inconsistent correlation coefficients), it is possible that the variance of the portfolio we construct is negative! _ hence, the need for a simpler model, that doesnt rely on calculations that many calculations
_
10-4
The Single-Factor Model
security returns tend to move together because of market risk _ suppose we can summarize all the common effects into one macroeconomic variable _ then we can write the return on stock i as ri = E(ri) + mi + ei where mi is the unanticipated effect of the common macroeconomic factors, and ei is the
_
unanticipated effect of firm-specific factors _ note that E(mi) = E(ei) = 0
10-5
The Single-Factor Model (cont.)
different firms have different sensitivities to macroeconomic events _ denote the sensitivity of firm i to the common set of factors (sensitivity coefficient) by i _ denote the variable that encompasses the
_
unanticipated effect of the common set of macroeconomic factors by F _ then mi = i F and the equation for the return on stock k becomes the single-factor model: ri = E(ri) + i F + ei
10-6
The Single-Index Model
now, we need a measure for F, the common macroeconomic factors _ since a market index corresponds to a welldiversified
_
portfolio, its return should respond only to the common macroeconomic factors _ hence, we can use a market index (say, S&P 500) to approximate our macroeconomic variable _ the single-index model _ investors are more interested in risk premiums rather that returns
10-7
The Single-Index Model (cont.)
then we can write the return on stock k as ri rf = i
_
+ i [rm rf] + ei or: Ri = i + i Rm + ei where Ri, Rm are excess returns _ we can decompose the excess return on a security into three components:
_
i
= return if the excess return on the market portfolio is zero
_
i
[rm rf] = return due to market movements _ ei = return due to unexpected firmspecific factors
10-8
Why Beta?
covariance between returns on stock k and market portfolio (index): _ hence,
_
2
0),(0 ),(),(),( ),( ),( ),(),(),(
mi mmi mimmimi mimii mi fmfimfimi
R R Cov R e Cov R R Cov R Cov R e R Cov R R Cov r r r r Cov r r r Cov r r Cov
= ++ = ++ = ++ = = = =
2
),(
m mi i
r r Cov
10-9
Return Variance and Covariances
note that the variance of returns is _ hence, there are two sources of risk: _market (idiosyncratic) risk: i
_
2
m
2
_
2
firm-specific risk: ei
covariance between returns on stocks i and j: 2 2 2 2 ) ( ei m i i m i i i e R Var
_
+ = + + = ),( ),(),(
mji jmjjimii jiji
e R e R Cov R R Cov r r Cov
++ + + = =
10-10
Why Is It a Simpler Model?
the input list for n = 50 assets consists of:
_
n = 50 estimates of expected returns _ n = 50 estimates of sensitivity coefficients ( i ) _ 1 (one) estimate of the variance of the market portfolio (index) _ n = 50 estimates of firm-specific risks (
_
ei
2)
in total, 3n + 1 = 151 estimates, as compared to 1325!
_
10-11
Advantages and Disadvantages
_ Advantages:
easier to generate the Markovitz frontier _ allows the specialization of security analysts by industry
_
_ Disadvantages:
overly simplistic decomposition of risk macro
_
vs. micro (ignores, for example, industry specific events) _ resulting portfolios might be inefficient
10-12
Estimating the Index Model
the single index model equation has the form of a regression equation: Rit = i + i Rmt + eit _ this defines a line with intercept i and slope
_
, with eit being the deviations from the line for the individual returns _ this line is called the Security Characteristic Line (SCL) _ it can be estimated using standard estimation techniques
10-13
Estimating the Index Model (cont.)
_
_
to do that, follow the steps:
gather historical data on stock prices (usually closing price), market index and risk-free asset
(T-bills) _ construct one-period returns (for a onemonth or one-week holding periods) for the stock, the market index and the risk-free asset _ this yields the variance of the return on the market index _ construct excess returns for the stock and the market index _ estimate the index model equation and obtain estimates of i , i , and ei
2
10-14
Security Characteristic Line
Ri Rm SCL
10-15
Portfolio Risk
the single index model equation for a portfolio has the same form: Rp = p + p Rm + ep where p, p, and ep are weighted returns of the individual stock counterparts _ the variance of the firm-specific term ep
_
decreases as the number of stocks included in the portfolio increases _ this is another example of the effects of diversification on risk
10-16
The Effect of Diversification on Risk
Diversifiable risk Market risk number of assets
p
2
2 2
p m
10-17
Problems with the CAPM
remember that the CAPM holds that E(ri) = rf + i [E(rm) rf] _ implication of the CAPM:
_
the market portfolio is efficient _ relationship between risk and expected returns
_
in practice, the CAPM is not directly testable, because it makes prediction about ex ante returns, while we only observe ex post returns
_
10-18
Testing the CAPM using the Index Model
remember that the beta coefficient in the index model is the same as the beta in the CAPM _ we can write the index model equation as ri rf = i + i [rm rf] + ei _ take expectations of both sides: E(ri) rf = i + i [E(rm) rf] _ according to CAPM, a stocks should be equal to zero, on average
_
hence, we should find that our estimates of s are centered around zero (Jensen, 1968)
_
10-19
Estimated Alphas
10-20
More Practical Insights
the beta coefficient, variances of return on market index and of firm-specific deviations can be estimated from historical data _ a source of such information is Merrill Lynchs
_
Security Risk Evaluation book (beta book) _ differences from index model:
uses returns rather than excess returns _ ignores dividends
_
10-21
Adjusted Beta
estimated beta coefficients tend to move toward one over time _ reasons:
_
average beta for all stocks is 1 (market beta) _ firms become more diversified over time _ they eliminate more of firm-specific risk
_
Merrill Lynch calculates an adjusted beta to compensate for this tendency: where a and e are adjusted and estimated betas, respectively 3 1 3 2 + = ea
_
10-22
Tracking Portfolios
suppose an investor identifies an underpriced
_
portfolio P ( p > 0) and wants to invest in it _ still, if the market as a whole declines, she would still end up losing money _ to avoid that, she can construct a tracking portfolio T, with the following structure:
a proportion p in the market index _ a proportion (1 p) in the risk-free asset
_
since T is constructed from the market index and the risk-free asset, its alpha coefficient is zero
_
10-23
Tracking Portfolios (cont.)
next, she can buy P and short-sell T at the same time _ eliminates the market risk _ still, the investment will yield a return (because of the portfolio Ps positive alpha) _ note: the portfolio is not risk-free it still has the firm-specific risk _ this strategy is what many hedge funds do
_
