A0880
Title: Flexible combination strategies for multivariate volatility forecasting
Authors: Christoph Frey - Lancaster University (Germany) [presenting]
David Happersberger - Invesco (Germany)
Abstract: Accurately forecasting multivariate volatility is crucial for risk management and portfolio allocation, yet model uncertainty and the curse of dimensionality often hinder the identification of a single optimal forecasting model. Additionally, it is well-documented in the literature that model combinations, across various domains, can significantly improve the accuracy of out-of-sample forecasts. The aim is to propose novel combination strategies that extend existing approaches by integrating advanced weighting schemes and dynamic model selection to improve the predictive accuracy of multivariate volatility forecasts. The methodology leverages multivariate conditional volatility models, realized covariance matrices, and dynamic-factor models, combining them with model weights that adjust based on recent individual forecast performance or portfolio utility. Comprehensive empirical analysis demonstrates that the proposed combination methods consistently outperform individual models and traditional equal-weighted combinations in terms of statistical accuracy and economic relevance, while also reducing parameter dependence.
