The information in aggregate data from Markov chains

Abstract

SUMMARY Estimation in Markov chains from aggregate data is discussed. The information about different aspects of a chain, such as individual transition probabilities, the equilibrium distribution and linear combinations of probabilities related to prediction, is investigated. The loss of information due to aggregation is examined by comparing estimation from aggregate and panel data. We find that although aggregate are generally much less informative than panel data, in some instances they give good estimation of equilibrium distributions and mean occupancy numbers for states in a chain.