Modeling local trends with regime shifting models with time-varying probabilities

In this paper we show that persistence and switching of trends are phenomena that appear in most long-lived stock return series. We model stock returns using a family of models based on hidden Markov models with duration-dependent transition probabilities. Trends are correlated so that aggregates such as indexes exhibit the same persistence and switching behavior as single stocks themselves. Hidden Markov models can thus explain medium-term momentum.