Hidden Markov models: Theory, applications and new perspectives

We are working on discrete latent variable models and dealing with model and variable selection algorithms to analyze multiple time-series and panel data with many categorical and continuous variables, including missing values. We are interested in extending the R package LMest created to estimate the hidden (latent) Markov models and related to the models proposed in the book titled “Latent Markov Models for Longitudinal Data”. The focus is also on more parsimonious parametrizations of the Markov chain transition matrices, on the causal formulations of these models, and on a tempered expectation-maximization algorithm to cope with the problem of local maxima arising when the parameters are estimated with the maximum likelihood method.