In this paper, we have two goals. First, we try to identify the stock market states and outline their statistical properties by using Multi-states Duration-Dependence Markov-switching models. Results show that the three-state model outperforms other models. An application to Tunisian stock market reveals that there exists three different states and each state represents different features of Tunisian stock market. Second, we construct a turning index based on the smoothed probabilities in order to identify the different Tunisian market cycle phases. The relevance of our index was documented from the synchronization between the values of the turning index and the values of TUNINDEX index return. It is well-adapted in order to account for extreme events.
