Are Government Spending Multipliers Greater During Periods of Slack? Evidence from 20th Century Historical Data
Michael T. Owyang, Valerie A. Ramey, Sarah Zubairy
A key question that has arisen during recent debates is whether government spending multipliers are larger during times when resources are idle. This paper seeks to shed light on this question by analyzing new quarterly historical data covering multiple large wars and depressions in the U.S. and Canada. Using an extension of Ramey’s (2011) military news series and Jordà’s (2005) method for estimating impulse responses, we find no evidence that multipliers are greater during periods of high unemployment in the U.S. In every case, the estimated multipliers are below unity. We do find some evidence of higher multipliers during periods of slack in Canada, with some multipliers above unity.
Review by Sebastian Fleitas
For a very long time the size of the expenditure multipliers has been one of the most vivid economic debates. For instance as recently as 2009, when the Obama administration proposed a fiscal stimulus package, there was a heated discussion regarding the relative size of the expenditure and tax multipliers. The reason fuelling this narrative is perhaps clear: ascertaining the potential impact of a particular proposed measure is key when designing the fiscal policy.
The paper by Owyang, Ramey and Zubairy, which was distributed by NEP-HIS on 2013-02-08 tries to answer this question: Are government spending multipliers greater during periods of slack for the US and Canada when we look at the historical data? The argument behind it is to consider that the expenditure multipliers will be greater in times of crisis, that is, during periods without full employment of labor and capital in the economy. This argument follows the idea that to wake up animal instincts, you need to have something in the forest when guys go out to hunt.
The answer that the authors offer is counterintuitive, which makes the paper very interesting. They find that the expenditure multipliers were higher in periods with high unemployment in Canada but they were the same for both periods in the US. To arrive to this conclusion the authors first construct high frequency (quarterly) historical data for the US and Canada. The procedure they follow to build the database is documented in an online available annex of the paper (here). After this process they have data on GPD, GDP deflator, government spending and the unemployment rate for the period 1890q1 to 2010q4 for the US and from 1921q1 to 2011q4 for Canada. The other key variable is the “news” variable, which reflects the changes in expected present value of government spending in response to military events as in Ramey (2011), which in turns directs to Ramey (2009).
Regarding the econometric approach, the authors use Jorda’s (2005) local projection technique to calculate impulse responses. The idea in Jorda (2005) is that, in contrast to VAR approaches which linearly approximate the data generating process to produce optimal one period forecasts, when we are looking at impulse response analysis we should care about the estimation of longer horizons. In this context, it is a better approach to estimate the impulse responses consistently by a sequence of projections of the endogenous variables shifted forward in time onto their lags using ordinary least squares (OLS) with standard errors addressing heterogeneity and serial correlation. The authors estimate a set of OLS regressions of different number of leads of the log of per capita government expenditure and GDP, over their lags and the variable news for periods with high and low unemployment and a quadratic trend. The coefficient for the variable “news” is the impulse response at that certain number of lags.
Finally, the paper made me think of three comments. First of all, the paper shows a very interesting and creative way to proceed when the data needed for the study is actually not available for that historical period. Besides combining sources of information, the authors constructed quarterly series of the variables. Since the paper was prepared for the American Economic Review Paper and Proceedings, it is a very short paper and the procedure to construct the variables is explained not in the paper but in the Annex. Given the lack of data, assumptions about the data generating process must be made. However, and besides the obvious limitation of space, the reader could miss an explanation about the assumptions that are made in the methods used and, also, what implications these assumptions have for the results, in particular about what is the source of variation that allows the identification of the coefficients. Maybe a section in the paper or in the appendix discussing these issues can shed light about what are the potential problems of different assumptions.
The last two comments are related to what is exactly the interpretation of the results. The first one directly follows from the last sentence of the paper. The authors state that they do not adjust for the fact that taxes often rise at the same time as government spending, which turns these multipliers not equal to pure deficit financed multipliers. However, it seems plausible that the effect of the multiplier on the GPD depends on whether this increase in the government was financed by taxes or by debt. If that is the case, and if the episodes when the former and the latter happen are mixed in a non-random way between the periods of high and low unemployment, then it is possible that the value of the coefficients can reflect not only the effect of the exogenous shock but also the effect of different ways to finance it.
The last comment relates to the consistent estimation of the parameters of the model. In the paper the “news” about military expenditure is taken as the only source of exogenous shock in this economy during the period of two years, four years and the time of the peak of each response. This “news” variable reflects exogenous innovations to the expenditure from a military source. However, it would be relevant for the paper to discuss the existence of other (non-military) sources of exogenous shocks to the expenditure. The relevance of this issue is because, given that the estimation of the parameters of interest is done by OLS, the consistency of the estimates requires zero covariance between the ¨news” and the error term of the equation, and this assumption can be violated if there exist this kind of non-military shocks and they are correlated to military “news”.
Overall I think this is a very interesting paper because of the results they find and also because of the construction of historical data. I found the results very puzzling and therefore a big motivation to continue trying to understand the relationship between GDP and public expenditure.