On Portfolio Optimization: Forecasting Covariances and Choosing the Risk Model

Abstract

We evaluate the performance of different models for the covariance structure of stock returns, focusing on their use for optimal portfolio selection. Comparisons are based on forecasts of future covariances as well as the out-of-sample volatility of optimized portfolios from each model. A few factors capture the general covariance structure but adding more factors does not improve forecast power. Portfolio optimization helps for risk control, but the different covariance models yield similar results. Using a tracking error volatility criterion, larger differences appear, with particularly favorable results for a heuristic approach based on matching the benchmark's attributes.