Factor models of stock returns: GARCH errors versus time-varying betas

This paper investigates the implications of time-varying betas in factor models for stock returns. It is shown that a single-factor model (SFMT) with autoregressive betas and homoscedastic errors (SFMT-AR) is capable of reproducing the most important stylized facts of stock returns. An empirical study on the major US stock market sectors shows that SFMT-AR outperforms, in terms of in-sample and out-of-sample performance, SFMT with constant betas and conditionally heteroscedastic (GARCH) errors, as well as two multivariate GARCH-type models.