Factor estimation using MCMC-based Kalman filter methods

Abstract

An exact MCMC-based solution for the Kalman filter with Markov switching and GARCH components is proposed. To motivate the solution, an international equity market model incorporating common Markovian regimes and GARCH residuals in a persistent factor environment is considered. Given the intractable and approximate nature of the model's likelihood function, a Metropolis-in-Gibbs sampler with Bayesian features is constructed for estimation purposes. To accelerate the drawing procedure, approximations to the conditional density of the common component are also considered. The model is applied to equity data for 18 developed markets to derive global, European, and country-specific equity market..