Federal reserve forecasts as fiscal news in an identified vector autoregression

Abstract

This investigation seeks to understand how government spending affects macroeconomic aggregates when policy changes are anticipated. This question is important since many policy actions are debated in congress and discussed in popular news outlets before being enacted. If agents are rational, they will incorporate news about future spending changes before they are enacted. The main contribution of this paper is to bring a new data set to bear on established methodologies. We collected Federal Reserve forecasts for the period 1965 – 2005 from online archives of FOMC meetings. We incorporate the forecasts as a measure of anticipated military spending to identify government spending shocks in a VAR. Two primary results emerge. When we use the raw forecasts, we find that GDP, hours, wages, and consumption all rise following a shock to the news variable. Output multipliers, when measured as the peak response to the shock, range from .5 to slightly larger than unity. When we instead incorporate forecast errors in the VAR we find just the opposite: hours increase while wages and consumption fall after a government spending shock, as is typical with the narrative approach to identifying government spending shock. Thus, the way in which the forecast data is incorporated into the VAR becomes crucially important to the results.