News, noise, and fluctuations

the long version

We explore empirically models of aggregate fluctuations with two basic ingredients: agents form anticipations about the future based on noisy sources of information; these anticipations affect spending and output in the short run. our objective is to separate fluctuations due to actual changes in fundamentals (news) from those due to temporary errors in the private sector's estimates of these fundamentals (noise). Using a simple model where the consumption random walk hypothesis holds exactly, we address some basic methodological issues and take a first pass at the data. First, we show that if the econometrician has no informational advantage over the agents in the model, structural VARs cannot be used to identify news and noise shocks. Next, we develop a structural Maximum Likelihood approach which allows us to identify the model's parameters and to evaluate the role of news and noise shocks. Applied to postwar U.S. data, this approach suggests that noise shocks play an important role in short-run fluctuations. Keywords: Aggregate shocks, business cycles, vector autoregression, invertibility. JEL Classifications: E32, C32, D83.