Forecasting High-Dimensional Portfolios

Abstract In this paper, we investigate the usefulness of forecasting in a high-dimensional framework where the number of assets is larger than the temporal observations. The benefit of forecasting lies in the concept of timing , which means anticipating future market conditions. We find that when high-dimensional econometric approaches are used, forecasting either the mean or the covariance is better than predicting both and then approaches based on static estimates. Moreover, we find that timing portfolios also perform better than the naive strategy. Considering the portfolio returns over time, we find that a possible explanation for the better performance of volatility-timing portfolios is that they better manage risk during periods of high uncertainty.