Indirect and direct forecasting of volatility-timing portfolios

Abstract

Recent studies have challenged the usefulness of variance–covariance matrix forecasting for the purpose of minimum-variance portfolio construction, instead advocating for the direct forecasting of realized weights. This study examines the value of this direct approach when dimension reduction is handled in the portfolio construction problem via popular volatility timing strategies. Using empirical data from the 45 largest U.S. stocks, this paper reveals that the traditional indirect approach, which relies on volatility forecasts, consistently delivers higher out-of-sample portfolio Sharpe ratios. This finding is robust to random portfolio selection, forecasting horizons, and transaction costs. The results demonstrate the continued usefulness of volatility forecasting models in portfolio construction.