Bayesian Filtering Estimation of Statistical Dynamic Factor Mode

Statistical factor analysis has been widely used in many areas of investment science such as risk management, portfolio selection, trading strategies, etc. This dissertation mainly investigates the estimation of dynamic factor model in the Bayesian framework, using the techniques of particle filter with online parameter learning such as marginalized particle filter and particle learning. We also compare our results with the offline conventional method such as Kalman filter combined with EM algorithm in root mean squared error criterion.;In the real data analysis, regime switching or structure break in the factor structure will make the estimation of static model difficult and lead to the problem of model misspecification. For solving this issue, we construct a regime switching dynamic factor model and compare its performance with conventional method using EM algorithm in the context of statistical arbitrage trading strategy.