A Markov Regime Switching Model for Asset Allocation

This paper develops an innovative regime switching multi-factor model accounting for the different regime switching behaviors in the systematic and idiosyncratic components of asset returns. A Gibbs sampling approach for estimation is proposed to deal with the computational challenges that arise from a large number of assets and multiple Markov chains. Finally, a dynamic asset allocation problem is studied under this model. In the empirical analysis, the model is applied to study the sector ETFs. The idiosyncratic volatilities of different sectors’ returns exhibit a strong degree of covariation and state-dependent patterns, which are different from the dynamics of their systematic components. The out-of-sample dynamic asset allocation experiments show that the new regime switching model statistically significantly outperformed the linear multi-factor model and conventional regime switching models driven by a common Markov chain. These results suggest that it is not only important to account for regimes in portfolio decisions, but correct specification about the regimes’ structure and number is of equal importance