VAR-ifying Predictions: Inferring Implicit Models from Published Forecasts

Surveys are widely used to measure economic agents' expectations, but forecasters rarely share information on how these are generated. In this paper, I introduce a novel identification strategy to recover forecasters' belief generation model. I show how my strategy can be applied even when there is no information on the forecasts for some variables. The method relies on a mapping between local projections and VARs under linear data generating processes. In an empirical application, I focus on central banks' interest rate forecasts, which are often kept private while forecasts for other variables are disclosed. Using Federal Reserve's Tealbook data, I show that the method can reasonably recover the central bank's private model and private interest rate forecasts by relying solely on published forecasts.