Multivariate Regime Switching Model Estimation and Asset Allocation

Markov regime switching (MRS) models successfully describe the cyclical behavior of time series by introducing hidden states and can better explain some stylised facts of asset returns. However, due to the complexity of the model, especially for multi-variate and multi-state cases, traditional maximum likelihood estimation (MLE) methods for MRS model suffers from strict assumptions and prone to converge to local optima. In this paper, we design a spectral clustering algorithm to predict hidden states of multi-variate MRS model by constructing feature vector and derive the parameter estimation. Monte-Carlo simulation results show that our algorithm is more robust than MLE. Meanwhile, we also give an application example of the algorithm by implementing a MRS asset allocation strategy in Chinese stock market.