A Continuous‐time Markov Chain Approach for Modeling of Poverty Dynamics: Application to Mozambique

Abstract This paper explores the use of continuous‐time Markov chain theory to describe poverty dynamics. It is shown how poverty measures can be derived beyond the commonly reported headcounts and transition probabilities. The added measures include the stationary situation, the mean sojourn time in a given poverty state and an index for mobility. Probit regression is employed to identify the most influential factors on the transition probabilities. Moreover, sensitivity analysis shows that the results are robust against perturbations of the transition matrix. We illustrate the approach with pseudo‐panel data constructed from a repeated cross‐section survey in Mozambique, using a pairwise matching method to connect households in the 2003 sample to similar households in 2009. Results reflect high and persistent poverty levels with considerable movements into and out of poverty. An estimated 57 percent of the poor in the first wave remained poor in the second wave and 43 percent moved out. Likewise, 64 percent remained non‐poor and 36 percent moved in. The corresponding stationary poverty headcount is 45 percent with respective mean sojourn time of 6.9 years in poverty and 8.4 years out of poverty. Conditioning the Markov chain on covariates identified by probit regressions indicates that poverty dynamics are responsive to household characteristics and livelihoods.