Defense spending and fiscal multipliers: it's all in the variance

We provide estimates of U.S. government expenditure multipliers for defense and non-defense spending over 1939-2014, using a fairly standard DSGE model that includes anticipated military spending changes ("war news shocks"), and find the following. First, our model's war news shocks compare favorably to Ramey's (2011) narrative-based "defense news" shocks. Second, war news shocks have little effect on model variables regardless of the period under examination. Unanticipated military expenditure accounts for substantial movements in output, but only when observations from 1939 to 1954 are considered. Apart from that, movements in output are entirely driven by total factor productivity shocks. Third, our structural model can generate defense expenditure multipliers above unity under two conditions: (i) the multiplier is calculated using the peak of the impulse-response function and (ii) a large number of observations before and up to the Korean War are included. When multipliers are calculated according to Mountford and Uhlig's (2009) present-value definition, they never exceed unity, regardless of the sample under analysis.