Estimating, Filtering and Forecasting Realized Betas

A strategy for estimating, ?filtering and forecasting time-varying factor betas is proposed. The approach is based on the multivariate realized regression principle, an omnibus noise ?filter and an adaptive long memory forecasting model. While the multivariate realized regression approach allows for an accurate estimation of the betas also when more than a (non-orthogonal) risk factor affects stock returns, the omnibus noise ?filter and adaptive long memory forecasting model, by accounting for the time series properties of factor betas, allow for accurate estimation and forecasting.