The Predictability of Systematic Risk Factors Correlations

Systematic risk factors can be used to create portfolios free of idiosyncratic risks but are not usually orthogonal to each other. Are correlations between pairs of factors predictable? Does accounting for the dependence structure among systematic factors improve traditional portfolio optimization techniques? To answer this question, I use asymmetric GARCH models combined with EVT-copula specifications to perform out of-sample forecasts and simulate one-day-ahead of five Fama-French five factors (FF5) returns. MinCVar ARMA-eGARCH-EVT-Copula portfolio specifications significantly improve the risk and cumulative return performance compared to the equally weighted portfolio. The results indicate that one can predict the FF5 dependence structure