Learning, capital-embodied technology and aggregate fluctuations

Recent cyclical episodes in the U.S. and G-7 economies are asymmetric: recoveries and expansions tend to be long and gradual and busts tend to be short and sharp. A large body of work views the two recent cyclical U.S. episodes, namely, the “new economy” boom in the late 1990s, and the 2000s housing boom-bust as episodes where over-optimistic beliefs have played a significant role. These episodes have revived interest in expectations driven business cycles models. However, previous work in this area has not addressed the important asymmetry feature of business cycles. This paper takes a step towards addressing this limitation of expectations driven business cycle models. We propose a generalization of the Greenwood et al. (1988) model with vintage capital and learning about capital embodied productivity and show it can deliver fluctuations that are asymmetric as in the U.S. data. Learning, calibrated to match the procyclical forecast precision from the Survey of Professional Forecasters, is crucial for the modelʼs ability to generate asymmetries. Forecast errors generated by the model are shown to trigger recessions that mimic in magnitude, duration and depth the typical post WW II U.S. recession.

► A model with learning, calibrated to match data from the survey of professional forecasters.
► The model can generate the growth asymmetry of cycles as observed in U.S. data.
► Forecast errors produce upward and downward swings in investment, output, utilization and hours.
► Noise driven recessions are of similar magnitude, depth and duration to the average U.S. recession.