MIXMCM: Stata module to estimate finite mixtures of non-stationary Markov chain models by maximum likelihood (ML) and the Expectation-Maximization (EM) algorithm
Legrand Dunold Fils Saint-Cyr, Laurent Piet · RePEc: Research Papers in Economics · 2018
mixmcm fits finite mixture of Markov chain models using conditional mlogit via the EM algorithm. The command estimates the parameters of the transition probabilities of agents under the assumption of a finite mixture of homogeneous types in the population, with each type following its own first-order Markovian process. That is, typically, agents are observed at several dates (or time periods) t={1...T} as located into a finite number of states (the modalities of depvar) j={1...K} (with K>=2). Each agent belongs to a specific homogeneous type g={1...G} (with G>=1). Agents' transitions from state j={1...K} to state k={1...K} are thus observed but the type that agents belong to are not.