A1565
Title: Regularized Wishart autoregressive stochastic volatility
Authors: Roman Liesenfeld - University of Cologne (Germany) [presenting]
Guilherme Moura - Universidade Federal de Santa Catarina (Brazil)
Abstract: An existing stochastic volatility (WSV) approach to model high-dimensional financial return series is extended with time-varying volatility and correlations. The WSV is a state-space model with a multiplicative evolution of the precision matrix driven by a multivariate beta variate. Predictions of the covariance matrix are weighted moving averages with exponential forgetting that discounts information in past returns uniformly over time and across the space of the return vector. The extension of the WSV regularizes this forgetting scheme by shrinking the covariance matrix predictions toward a pre-specified target matrix. This restricts the variation of the latent covariance matrix, thereby ensuring stationarity of returns and stabilizing eigenvalues of the covariance matrix predictions. Furthermore, this regularized forgetting is combined with time-varying and directional forgetting to achieve robustness to sudden changes in the correlation structure triggered by transitions from one market regime to another. The regularized WSV maintains closed-form sequential updating formulas for filtering, prediction and likelihood evaluation, facilitating practical implementation. To evaluate its performance, a historical dataset of returns of US stocks is used, and the ability is considered to predict the covariance matrix and the weights of the global minimum variance portfolio. The results show that the regularized WSV approach performs well compared to several alternative models.
