Measuring Fiscal Shocks in Structural VARs Using Narrative Data

We describe a methodology that integrates the narrative approach to the identification of macroeconomic shocks into existing structural VAR settings while allowing for measurement error in the narrative measures of the shocks of interest. We apply our methodology to the fiscal VAR setup of Blanchard and Perotti (2002) and make use of the tax shocks constructed by Romer and Romer (2009). We find that the macroeconomic effects of unanticipated changes in taxes in the United States are significantly larger than in the typical application of the Blanchard and Perotti (2002) identification strategy. A cut in total federal tax revenues of one percent of GDP raises GDP by 2 percent in the first quarter and has a maximum output effect of 3 percent after one to two years. In contrast, the Blanchard and Perotti (2002) identification restrictions result in an impact multiplier of 0.4 percent and a maximum multiplier of 1.3 percent. We also use newly constructed data that allows identification of the distinct effects of personal income and corporate income tax changes. An unanticipated tax reform decreasing personal income taxes, but leaving cyclically