We use the control function approach to identify the average treatment effect and the effect of treatment on the treated in models with a continuous endogenous regressor whose impact is heterogeneous. We assume a stochastic polynomial restriction on the form of the heterogeneity, but unlike alternative nonparametric control function approaches, our approach does not require large support assumptions.

A general equilibrium search model makes layoff costs affect the aggregate unemployment rate in ways that depend on equilibrium proportions of frictional and structural unemployment that in turn depend on the generosity of government unemployment benefits and skill losses among newly displaced workers. The model explains how, before the 1970s, lower flows into unemployment gave Europe lower unemployment rates than the United States and also how, after 1980, higher durations have kept unemployment rates in Europe persistently higher than in the United States. These outcomes arise from the way Europe's higher firing costs and more generous unemployment compensation make its unemployment rate respond to bigger skill losses among newly displaced workers. Those bigger skill losses also explain why U.S. workers have experienced more earnings volatility since 1980 and why, especially among older workers, hazard rates of gaining employment in Europe now fall sharply with increases in the duration of unemployment.

Consider two agents who learn the value of an unknown parameter by observing a sequence of private signals. The signals are independent and identically distributed across time but not necessarily across agents. We show that when each agent's signal space is finite, the agents will commonly learn the value of the parameter, that is, that the true value of the parameter will become approximate common knowledge. The essential step in this argument is to express the expectation of one agent's signals, conditional on those of the other agent, in terms of a Markov chain. This allows us to invoke a contraction mapping principle ensuring that if one agent's signals are close to those expected under a particular value of the parameter, then that agent expects the other agent's signals to be even closer to those expected under the parameter value. In contrast, if the agents' observations come from a countably infinite signal space, then this contraction mapping property fails. We show by example that common learning can fail in this case.