ABSTRACT This paper compares the statistical and economic performance of state‐of‐the‐art high‐frequency (HF) based multivariate volatility models with a simpler, widely used alternative, the Exponentially Weighted Moving Average (EWMA) filter. Using over two decades of 100 U.S. stock returns (2002–2023), we assess model performance through a Global Minimum Variance portfolio optimization exercise, with and without short‐selling restrictions across multiple forecast horizons. We find that the EWMA model cannot consistently be outperformed by more complex HF‐based volatility models at the daily and weekly forecast horizons, even delivering significant utility gains when including transaction costs due to a favorable balance between turnover and ex‐post portfolio volatility. At the monthly horizon, the EWMA remains competitive against most of its competitors. Our findings hold across alternative specifications, including different estimation window lengths, portfolio sizes and smoothing parameter values, emphasizing the continued relevance of parsimonious volatility specifications, such as the EWMA model, in realistic investment settings.
Romero et al. (Fri,) studied this question.