Covariance Estimation with Markov-Switching Generalized Autoregressive Conditional Heteroskedasticity Models with Applications to Portfolio Managemen

The objective of this paper is to implement and test the multivariate regime-switching GARCH model as a potential improvement on traditional methods for estimating the covariance matrix for multiple time series. I describe the characteristics and estimation of the primary model of interest, MS-GARCH, and some competitor models. I implement and backtest a portfolio management strategy based on risk minimization using MS-GARCH forecasts and evaluate performance relative to competitors. I find MS-GARCH to be an useful tool in portfolio construction, and to offer some significant advantages over more traditional models in terms of accuracy and interpretability when describing a process.