Individual homelessness: Entries, exits, and policy

Abstract

Homelessness is part of the lives of many people. But almost no one is homeless for all or most of his or her life. The median shelter homeless spell is well under a month, and even “chronic homelessness” officially entails spells of a year or so. I model homelessness as part of people’s lives in a dynamic stochastic framework in continuous time. I can explain many empirical regularities with a parsimonious model: for instance, why the last addresses of homeless people are concentrated in a few low-rent neighborhoods, why homeless entries are hard to predict, why recidivism is common and past homelessness is a good predictor of future homelessness, why some groups recidivate more often than others, why the hazard rate for shelter exit is single-peaked, why effective homelessness prevention programs do not alter the average length of homeless spells. I also examine policy. The optimal homelessness prevention program is Pigouvian and starkly simple. With an optimal prevention program in place, optimal shelter quality maximizes a simple and intuitive expression, and insurance programs always raise social welfare. Most of the previous economics literature about homelessness has been static, but most literature about homelessness outside economics has been dynamic. This paper tries to bring the two strands of literature closer together.