Guilherme Moura (UFSC)

Title: Regularized Wishart Stochastic Volatility Model

We extend Uhlig's (1994) Wishart Stochastic Volatility model by regularizing covariance predictions towards a prior reference matrix, ensuring stationarity and stabilizing eigenvalues. Our method provides closed-form sequential updating for filtering, prediction, and likelihood. We also enhance variance discounting with directional forgetting, allowing varying rates over time and across observation vector directions. An empirical portfolio selection application with up to 1000 assets shows that the method delivers lower-risk portfolios than benchmarks.