O Papaspiliopoulos and G Sermaidis
Monotonicity Properties of the Monte Carlo EM Algorithm and Connections with Simulated Likelihood
Abstract: In this note we show that the Monte Carlo EM algorithm, appropriately constructed with importance re-weighting, monotonically increases a corresponding simulated likelihood. This is result is formally proved but also intuitively explained by a formulation of the problem using auxiliary variables.
Keywords: Importance sampling, fixed random seeds, incomplete data, latent stochastic processes.