distribution of these returns. Although all the econometric specifications had actual returns above the average random return, it is only the 5yr rolling OLS and the Kalman specifications that achieve returns consistent with rejecting the hypothesis of randomness at the 5% level.
Interpreting the Results of the Trading Strategy
Our results may indicate that increasing uncertainty increases investor focus on observable fundamentals. In addition to the precipitous drop in equities beginning in March 2000, volatility increased substantially and remained high through the end of 2002. In contrast, during a period of greater optimism about future dividend growth prospects, signaled by falling earnings and dividend yields, lagged or even contemporaneous fundamental data may be discounted in favor of a focus on more forward looking measures.
To investigate this hypothesis further, we conduct another series of cross-sectional regressions identical to those in Section 4. In this case, however, the sample is split between the period prior and the period following the March 2000 stock market peak. This follows the recommendation of Kan and Zhang (1999) to run split sample regressions to test the stability of risk premia. Instability in factor premiums may indicate the presence of useless factors. As presented in Table 9, the results of additional cross sectional regressions indicate a larger risk premium on the predicted returns in the second subset following March 2000 than in the initial sample from 1994 to 2000. The larger risk premium estimate is consistent with the conjecture that following the market peak, a greater focus was put on fundamental information. Lack of certainty about earnings and future economic prospects results in a higher level of surveillance of economic news by investors.
(Insert Table 9 here)
The far more striking result for the cross-sectional regressions in Table 9 is the instability of the risk premium on market betas. In Panels A and B, the CAPM beta risk premia are both large and significant for those regressions without the TVPFM predicted returns, consistent with the theory underlying the CAPM model. In the presence of the TVPFM predicted returns, the market risk premia are still positive and somewhat less significant. In Panels C and D the market risk