My colleague UCSD Professor Valerie Ramey has an interesting new paper looking at the effects of higher government spending on GDP.
Ramey (2012) approaches the question from a forecasting perspective. Suppose a certain event (examples of which are detailed below) causes you to revise your forecast of how high government spending is going to be over the next few years. How would this news cause you to change your forecast of how high private GDP (that is, all the components of GDP other than government spending) is going to be? If your prediction of private GDP goes up, that is evidence consistent with a fiscal multiplier greater than one– added government spending not only contributes directly to GDP from the accounting identity, but also helps boost private spending as well. If private GDP goes down, that suggests a multiplier less than one.
The forecasting models she looks at are vector autoregressions, in which one tries to predict a set of variables such as the log of real government spending per capita, log of real private GDP per capita, the marginal tax rate, and the 3-month T-bill rate based on what all of those 4 variables have been doing over the last year. In her simplest exercise, Valerie looked at how the forecasts of each of the 4 variables would change if real spending this quarter comes in higher than you would have expected according to the model. The top panels in the figure below are based on forecasting relations using data all the way back to 1939. The graph in the left panel shows how the model’s k-quarter-ahead forecast of real government spending would change in response to news of higher government spending at time 0, plotted as a function of k, how far into the future you’re looking. (It will help economist readers, but perhaps not anyone else, if I were to describe this as an impulse-response function based on a Cholesky factorization with government spending ordered first as in Blanchard and Perotti (2002)). Given the positive serial correlation in government spending, if you learn spending is about 0.4% higher this quarter, you’d expect further spending increases over the next year, with the graph normalized such that the news causes you to expect 1% higher real spending per person 4 quarters following the original information.
The right top panel shows how the model’s forecast of future real GDP per person excluding government spending would change in response to the news. The model predicts that private spending will be 0.7% lower after a year. This negative effect is statistically significant.
In part this inference is based on what happened during World War II. Between 1941 and 1944, real government spending increased by $75 billion (in 1937 dollars), but real GDP only rose by $60 billion. That’s consistent with the patterns above, in which private spending falls in response to higher government spending. One might argue that there were other special factors such as rationing that reduced private GDP at the time. The second row in the figure above leaves out the World War II data, and just bases the inference on what we saw over 1947-2008. The effects are similar to those found using the full sample. The last panel of the above figure leaves out both World War II and Korea. With less data, the inferences are less reliable, but the overall estimates remain quite similar to those for the full sample.
Valerie’s paper explores a number of other possible ways one could form the forecasting question. One concern is whether government spending rose in response to some other events that might have had direct effects on the economy, in which case the revision in forecasts might represent the consequences of those events instead of an effect of government spending itself. The next graph uses an alternative idea proposed by Perlotti (2011) of specifying the news variable in terms of defense spending alone rather than overall government spending (this is now a 5-variable VAR with defense spending ordered first). Results are essentially the same as for the first exercise.
In yet another approach, Valerie used a news series constructed in Ramey (2011) that is based on reading of Business Week and other historical sources to construct a series of changes in the expected present
discounted value of government spending caused by military events. Although the effects on private GDP are not measured as precisely using this indicator, the overall inference confirms the view that higher government spending raises GDP by less than the spending itself.
Ramey (2012) concludes:
Using a variety of
identification methods and samples, I find that in most cases private spending falls
significantly in response to an increase in government spending. These results
imply that the average GDP multiplier lies below unity.