Estimation of Markov regime-switching regression models with endogenous switching

Following Hamilton (1989), estimation of Markov regime-switching regressions nearly always relies on the assumption that the latent state variable controlling the regime change is exogenous. We incorporate endogenous switching into a Markov-switching regression and develop strategies for identification and estimation. Identification requires instruments, which can be found in observed exogenous variables that influence the transition probabilities of the regime-switching process, as in the so-called time-varying transition probability case. However, even with fixed transition probabilities, the lagged state variable can serve as an instrument provided it is exogenous and the state process is serially dependent. This is true even though the lagged state is unobserved. A straightforward test for endogeneity is also presented. Monte Carlo experiments confirm that the estimation procedures perform quite well in practice. We apply the endogenous switching model to the volatility feedback model of equity returns given in Turner, Startz and Nelson (1989).