Specification and the Technology-Hours Debate: What Can We learn from Bayesian VARSs?

Many recent papers, following Gali (1999), have found a negative response of employment to a positive technology shock identified as a permanent shock to labor productivity, contradicting the prediction of standard RBC models. In a recent paper, Christiano, Eichenbaum and Vigfusson (2003) get a positive response of employment measured by hours per capita when Gali’s assumption of a unit root in hours is relaxed. In this paper, we propose a new specification test to disentangle for the level/first difference models. We calculate posterior probabilities of various specifications of productivity-hours VARs using Bayes factors, measured by the Laplace approximation and by the posterior information criterion suggested by Phillips (1995, 1996) and Phillips and Ploberger (1994). Our results strongly support the results of CEV, namely, a specification of hours in levels rather than differences. However, resulting level VAR implies impulse responses , variance decompositions and conditional correlations with distributions so wide that meaningful inference of the role of technology shocks in business cycles is impossible