Market share paradox and heterogeneous chains (Q1203757)

The paper studies and resolves a paradox in an application of a Markov chain model to market share estimation. The problem arises when a common transition matrix \((p_{ij})\) is used for all individuals (customers), when in fact the population is heterogeneous with individuals of different types having different transition matrices. The main result of the paper implies that heterogeneity can have a very substantial impact on the computed stationary distribution.