Lucas (1987) argues that the gain from eliminating aggregate fluctuations is trivial. Following Lucas, a number of researchers have altered assumptions on preferences and found that the gain from eliminating business cycles are potentially very large. However, in these exercises little discipline is placed on preference parameters. This paper estimates the welfare cost of business cycles, allowing for potential time-non-separabilities in preferences, where discipline is placed on the choice of preference parameters by requiring that the preferences be consistent with observed fluctuations in a model of business cycles. That is, a theoretical real business cycle world is constructed and the representative agent is then placed in this world. The agent responds optimally to exogenous shocks, given the frictions in the economy. The agent's preference parameters, along with other structural parameters, are estimated using a Bayesian procedure involving Markov Chain Monte Carlo methods. Two main results emerge from the paper. First, the form for the time-non-separability estimated in this paper is very different than the forms suggested and used elsewhere in the literature. Second, the welfare cost of business cycles is close to Lucas's estimate.
