The Determinants of the Separation Hazard in a Model with Learning and Time-varying Match Quality

Abstract. People and organizations enter relationships, learn about them, adapt to them, and sometimes decide to leave them. This paper develops a learning model of relationship dissolution. The model also allows relationship quality to vary over time, following an AR(1) process. The model nests two possible assumptions about the causes of relationship dissolution. One is that relationships dissolve when agents find out that they are in fact bad (pure learning model). The other is that relationships dissolve because they change over time (pure shocks model). This paper analyzes the effect of parameters on agents’ separation decision and the resulting separation hazard. It also examines how one can empirically distinguish between the pure learning model, the pure shocks model, and a mixed model. Observing an increasing and then decreasing separation hazard is not sufficient evidence for the pure learning model. The shape of the variance of the transition function is enough to distinguish between the three models. One can also use the impact of a bad observation on the separation hazard to make such a distinction.