Predicting the state of the stock market using switching factors

In this thesis models for the identification and prediction of the state of the equity market are compared. In these models factors that follow Markov switching processes are extracted from several macro economic and financial variables. The market regime is determined by the state of the Markov process. The models are estimated over a period that includes the credit crisis that started in 2008 and a sub period that does not include this crisis. This shows that the models with normal error distributions are less robust than the models with Student t error distributions. Parameter estimates, the identification of the states and the correlation between the excess returns and the dynamic factor change dramatically for the first when the crisis is included in the estimation period, while remaining almost the same for the second. The incorporation of more states in the Markov chain slightly helps as the crises behavior of the dynamic factor can be partly captured by the additional states. The out-of-sample prediction of the dynamic factors and the states in VAR models is not of good quality. Evaluating the out-of-sample performance of the models with investment strategies, only the VAR models based on the four state models deliver significant improvements