Advances in Investment Analysis and Portfolio Management, Volume 9

if the benchmark model is the true asset pricing model that is able to correctly

price the cross-section of asset returns.

5

However the Jensen measure can be

viewed as the abnormal returns of the strategy compared to an alternative

passive strategy that invests in the risk-free asset and the market index that have

the same risk characteristics as the strategy (see Elton & Gruber, 1995). We

refer to performance from Eq. (7) as the Jensen measure.

We estimate the Ferson and Schadt (1996) performance measure by the

following regression:

r

it

= ␣

i

+ ␤

i

r

mt

+

冘

L

l=1

␦

il

r

mt

z

lt⫺ 1

+

␧

it

(8)

z

lt⫺ 1

is the de-meaned lth information variable (l = 1,...,L) realization at time

t ⫺ 1 and ␤

i

is the average conditional beta of strategy i with respect to each of

the benchmark portfolios. The ␦

ilk

coefficients are the estimated coefficients of

strategy i in the conditional beta function with respect to each of the L

information variables. The intercept ␣

i

is the Ferson and Schadt (1996) measure

which equals zero under the null hypothesis of the strategy exhibiting no

abnormal performance. The conditional performance measure allows the

strategy betas and factor risk premia to vary through time. Ferson and Schadt

(1996) use the approximation that the portfolio beta in period t is a linear

function of common information variables that are known at that time. The

additional terms, r

k

t

z

lt⫺ 1

(l=1,...,L), captures the covariance between the

conditional betas and the market risk premium. We refer to performance from

Eq. (8) as the FS performance measure.

We correct the test statistics of the performance measures for the effects of

heteroskedasticity using White (1980).

DATA

Investment Universe

We explore the impact of the different models on domestic U.K. asset

allocation strategies. We follow Fletcher (1997) and Grauer (2000) and use

industry portfolios as our investment universe. An alternative approach is to use

large portfolios of individual securities as in Chou, Li and Zhou (2001).

However, the focus on this study is on asset allocation strategies rather than

stock selection strategies. We leave the issue of stock selection strategies to

future research.

We use 10 U.K. industry portfolios and the domestic risk-free asset as the

investment universe. We collect the data from Datastream unless otherwise

257Characteristics Versus Covariances