Forecasting Regional Income Inequalities Based on Markov Models

Abstract

It is known that the simple Markov model overpredicts the long run horizon mobility of the income distribution process. Dissolving the homogeneity assumption of the Markov model we can have better forecasts. One generalization of the Markov model, the Mover–Stayer model assumes heterogenous population: some units are moving according to a common Markov chain but there are some (unknown) units whose are not moving at all. They are called stayers.

Based on Frydman, 1984 methodology we compute both the Markov and Mover–Stayer models for Hungarian micro-regions income data and find that the Mover–Stayer model fits better the regional relative income data than the simple Markov model. Using likelihood ratio test statistics we show that the difference is highly significant.