Ja. Christopherson et al., Performance evaluation using conditional alphas and betas - A better job at predicting returns., J PORTFOLIO, 26(1), 1999, pp. 59
Unconditional measures can incorrectly measure alpha and beta when portfoli
o managers engage in dynamic trading strategies or change their alphas and
betas in response to publicly available information about the economy. The
authors advocate conditional performance evaluation (CPE) to measure dynami
c alphas and betas. Comparing a sample of 261 manager portfolios over 1980-
1996 to the Russell 3000,they find that a portfolio of the toy quintile of
CAPM alphas outperforms the bottom quintile by 1.45% annualized, while the
spread for quintiles of CPE alphas is 4.00%. When style indexes are used to
compute alphas, the spread between top and bottom quintiles of CAPM alphas
becomes a perverse -2.41% annualized. On the other hand, the top quintile
of CPE alphas outperforms the bottom by 2.61%. The top quintile also outper
forms the average manager by 1.70% and the Russell 3000 by 2.42% annualized
. While higher CPE alphas do not guarantee superior returns, they are more
likely to successfully forecast: returns than CAPM alphas.