Is Your Covariance Matrix Still Relevant? An Asset Allocation-Based Analysis of Dynamic Volatility Models

Ever since Harry Markowitz published his seminal paper on portfolio selection, investors have incorporated estimates of future volatilities and correlations into their asset allocation process. While portfolio construction methods continue to evolve, many investors continue to forecast volatility using traditional approaches that are ill-suited to the time-changing nature of volatility. In this paper, I analyze the performance of seven different multivariate-volatility models using a new, risk-parity based approach to determine each model's accuracy. I find that traditional, sample covariance methods perform poorly when trying to forecast short-term volatility, and that a more dynamic model often provides superior out-of-sample forecasts