We present a feasible estimation method for maximum likelihood estimation of factor models in which the common factors are subject to ARCH-type effects. Our approach consists of an EM step to estimate the factor loadings and the idiosyncratic variances, followed by standard estimation of the conditional variance parameters. Our proposed procedure yields significance speed and robustness result in two applications to UK stock returns, one using 26 sectoral indices, the other 266 individual companies. We also discuss restrictions in the coefficients, unequal number of observations, seasonal factors and models in which the factor variances affect the mean. Consistent estimates for Engle's (1987) factor GARCH model are explicitly considered.
Download is not available
