Leonardo Melosi

Abstract: I develop and estimate a dynamic general equilibrium model with imperfectly informed firms in the sense of Woodford (2002). The model has two aggregate shocks: a monetary policy shock and a technology shock. Firms observe idiosyncratic noisy signals about these shocks and face strategic complementarities in price setting. In this environment, agents’ "forecasting the forecasts of others" can produce realistic dynamics of model variables, with associated highly persistent real effects of monetary shocks and delayed effects of such shocks on inflation. The paper provides a full Bayesian analysis of the model, revealing that it can capture the persistent propagation of monetary shocks only by predicting that firms acquire less information about monetary policy than about technology. To investigate the plausibility of this finding, I augment the model to allow firms to optimally choose how much information to acquire about the two shocks, subject to an information-processing constraint à la Sims (2003). This constraint sets the rate at which firms can substitute pieces of information about the two shocks. I find that, in the estimated model, firms’ marginal value of the information about monetary policy shocks is much higher than that about technology shocks. Hence, I argue that the estimated model predicts that firms acquire implausibly too little information about monetary policy.

Abstract: Marginal data densities are used to perform Bayesian model selection. They are hard objects to compute given their high dimension. We propose a new algorithm to evaluate the marginal data density for a broad class of macroeconometric models, using the outcome from a Gibbs sampler. Our method is suitable for settings where the parameter space can be partitioned so that the conditional predictive density can be analytically computed.