Explaining U.S. federal deficits: 1889-1998

Economic Inquiry, July, 2002 by Brian L. Goff, Robert D. Tollison

The graphs of the 95 percent confidence intervals for these out-of-sample forecasts appear in Figures 3 and 4 along with the actual percentage changes in Debt/P for 1971-98. As Figure 3 shows, the upper limit of the forecast from the model without the political variables is either at or below the size of actual deficits during the 1980s; that is, it undepredicts deficits over several years in the 1980s and for at least one year in the 1990s. In contrast, Figure 4 shows that the model with the political variables over-predicts deficit growth through several years in the 1980s, with actual deficits approaching or exceeding the upper limit of the forecast only in the 1990s. These results indicate that the deficits of the 1980s might in fact have been smaller than expected. Of special note is the fact that the nonpolitical version of the model include the variables on which the deficit growth of the 1980s is usually blamed, that is, the temporary reductions in income due to recession in the early 1980s and the te mporary increases in government defense spending.

Tax Smoothing and Symmetry of Shocks

One limitation of the estimates presented in Table 1 is that the linear coefficients estimated for temporary income and government spending provided only partial information about the tax smoothing hypothesis. A linear coefficient imposes an implicit restriction of symmetry on the effects of increases and decreases in the temporary series; asymmetrical effects between increases and decreases in either temporary series are ruled out. For example, if governments respond to short-lived events requiring increased spending more so than for short-lived events requiring lower spending, a linear coefficient will not detect this effect.

To estimate a version of the model that allows for asymmetric effects of the temporary series, we first computed two new variables from the existing series. One series contains only positive values with the years of negative values equal to 0; that is, positive temporary spending and positive temporary income. The other series contains only negative values with the years of positive values equal to 0; that is, negative temporary spending and negative temporary income. We then reestimated the model (the second column) from Table 1 with these four temporary measures. The coefficients and standard errors for the temporary income and spending variables are listed in Table 2. (The coefficients for the other variables and the diagnostic measures are listed in note 21 and are nearly identical to their prior values). (21) Also listed in Table 2 are F-statistics testing the null hypotheses of equality of the positive and negative measures for each temporary series both independently and jointly. The coefficients for t he temporary income measures are negative and very similar in magnitude, whereas the coefficients for the temporary spending series are both positive, but the positive spending series has a coefficient about 1.35 times as large as the negative spending coefficient. However, none of the F-tests rejects the null of coefficient equality, indicating a high degree of symmetry between temporary increases and decreases in spending and income.