Predictable Signals in Excess Returns: Evidence from Non-Gaussian State Space Models

The present work investigates predictable components in size-based and value-weighted market portfolios excess returns from NYSE, AMEX, and NASDAQ stocks over US Treasury bills using various Gaussian and non-Gaussian versions of state space or unobserved components models. Our state space or unobserved components model improves on Conrad and Kaul (1988) by taking into account fat tails that are widely documented in the returns series. Statistical hypotheses tests show existence of predictable components in excess returns for most size-based portfolios (Cap-1 through Cap-9) even at percent level of significance. However, for value-weighted market and largest size-based portfolio (Cap-10) the hypothesis tests fail to reveal existence of any predictable component. The results for most size-based portfolios are in conformance with Conrad and Kaul (1988) except the value-weighted market excess returns as well as the largest size-based portfolio (Cap-10). Conrad and Kaul (1988) isolated time-varying expected returns in weekly size-based excess returns using the same methodology but in a Gaussian setting. However, our results on value-weighted market excess returns are in line with Bidarkota and McCulloch (2004) who investigated value-weighted market excess returns in CRSP data.