Regime-Switching Quantile Regression Forests

This paper extends quantile regression forests to analyze temporally heterogeneous data with potentially complex dependencies using hidden Markov models. The method assumes a latent homogeneous Markov chain to model the temporal evolution of unobserved heterogeneity, and a state-specific quantile regression forest to capture regime-switching dynamics. Estimation is carried out via the Expectation–Maximization algorithm using the asymmetric Laplace density as a working likelihood, in which posterior state probabilities are incorporated as weights in the bootstrap sampling procedure used to grow the trees. The performance of the proposed method is evaluated through an empirical application to financial data.