LIMITING GOVERNMENT: THE FAILURE OF "STARVE THE BEAST"

For nearly 30 years, many Republicans have argued that the most effective way to control federal government spending is to "starve the beast" by reducing federal tax revenues. Moreover, two Nobel laureate economists, Milton Friedman and Gary Becker, have endorsed this argument. Friedman (2003) summarized

How can we ever cut government down to size? I believe there is one and only one way: the way parents control spendthrift children, cutting their allowance. For governments, this means cutting taxes. Resulting deficits will be an effective-I would go so far as to say, the only effective-restraint on the spending propensities of the executive branch and the legislature. The public reaction will make that restraint effective.

Becker and his colleagues Ed Lazear and Kevin Murphy (2003) described this effect as "the double benefit of tax cuts." (Lazear is the recently appointed chairman of the Council of Economic Advisers.) This argument has been effective in unifying the Republican Party in favor of reducing federal taxes, but at the cost of undermining the more traditional Republican concern about fiscal responsibility.

Problems with Starve the Beast

There are three major problems with the starve-the-beast argument: (1) it is not a plausible economic theory; (2) it is inconsistent with the facts; and (3) it has diverted attention away from the political reforms needed to limit government growth.

The Implausibility Issue

It is most implausible that reducing the current tax burden of federal spending would reduce the amount of federal services that voters demand. Orthodox price theory, of which Friedman and Becker are among the leading exponents, is unambiguous in concluding that reducing the price of a good or service increases the amount demanded. Reducing the current tax burden of federal spending has much the same effect as a price control, increasing the amount demanded relative to that supplied from current revenues, an effect that Friedman and Becker have consistently and correctly opposed in private markets.

The Facts

Second and more important, the starve-the-beast hypothesis is not consistent with the facts, at least since the beginning of the Reagan administration. Figure 1 shows current federal spending and receipts as a percent of gross domestic product by calendar year from 1981 through 2005. As this figure illustrates, most of the changes in the relative level of federal spending were coincident with changes in the relative federal tax burden in the opposite direction.

What is going on? One condition that contributes to this pattern is the variation in the unemployment rate, since an increase in the unemployment rate increases federal spending and reduces federal tax receipts. So any estimate of the relation between changes in federal spending and the level of current receipts should control for changes in the unemployment rate. Second, one should also control for the change in net interest payments as a percentage of GDP because interest payments are a fixed expenditure that is independent of the conditions that affect other spending.1

The relation between current federal spending and receipts from 1981 through 2005 is best summarized by the following least-squares regression:

D(S) = 2.761 - 0.145 R + 0.598 D (U) + 0.514 D (I) + u, (.848) (.045) (.055) (.197)

Adjusted R^sup 2^ = .848 S.E.R. = .223 D.W. = 2.386

where

D (S) is the change in current federal spending as a percent of GDP,

R is the level of current federal receipts as a percent of GDP,

D (U) is the change in the unemployment rate,

D (I) is the change in net interest payments as a percent of GDP,

and the figures in parentheses are the standard errors of the estimated coefficients.

The most important finding from this regression is that the level of current federal receipts as a percent of GDP has a significant negative effect on the change in current federal spending as a percent of GDP. A 1 percentage point increase in current federal receipts as a share of GDP apparently reduces the change in current federal spending as a share of GDP by about one-seventh of 1 percent a year indefinitely. As expected, the change in the unemployment rate has a strong positive effect on the change in current federal spending, and the change in net interest payments also has a significant positive effect.2 Using the sample 1949 through 1980 produced an estimated coefficient on the federal tax share of GDP that is also negative but not statistically significant.

Using both sample periods (1949-80 and 1981-2005), I found no significant relation between the change in the current federal spending share of GDP and the lagged federal deficit as a percent of GDP. In sum, there is no significant evidence that a recent high deficit ever had an effect similar to that of reducing a child's allowance; the difference is that the federal government has a credit card with no effective debt limit. Federal spending is better described as buying government services at a discount equal to the deficit, the costs of which will be borne by someone sometime in the future. For no extended period did these estimates reveal a significant positive relation between the change in federal spending as a percent of GDP and the level of federal receipts as a percent of GDP, the necessary condition for the starve-the-beast hypothesis to be confirmed. Starve the beast just does not work.

Another value of the above regression is that it provides a basis for estimating the current federal receipts share of GDP at which there would be no change in the current federal spending share. Assuming no change in the unemployment rate or in the interest payments share of GDP, federal receipts of about 19 percent of GDP would be necessary to stabilize current federal spending as a percent of GDP. Since federal receipts were 17.8 percent of GDP in 2005, a tax increase of about 1.2 percent of GDP would be necessary to prevent a continued increase in the federal spending share of GDP.

Given that total federal spending was 20.4 percent of GDP in 2005, however, even current receipts of 19 percent of GDP would not be sufficient to balance the budget, unless expenditures for defense and net interest payments are reduced by about 1.4 percent of GDP. Assuming no change in the defense and net interest spending shares of GDP, current receipts of 19.9 percent of GDP-a tax increase of about 2.1 percent of GDP relative to 2005-would be necessary to balance the budget within five years. A reduction of defense spending on completion of the U.S. military role in Iraq may be the only way to balance the budget without increasing current federal receipts above 19 percent of GDP. This reduction is well within the range of recent experience: spending for defense and net interest payments declined by 1.2 percent of GDP from 1992 to 1994 and by 3.3 percent of GDP from 1992 to 2000.

The Problem of Fiscal Discipline

The third problem, of course, is that the starve-the-beast perspective has led too many conservatives and libertarians to be casual about the sustained political discipline necessary to control federal spending directly, succumbing to the fantasy that tax cuts would solve this problem. President George W. Bush, for example, has proposed and won the approval of most congressional Republicans for large increases in federal spending for agriculture, defense, education, energy, homeland security, medical care, and transportation, and he has yet to veto a single spending bill. As a consequence, real per capita federal spending during the Bush administration has increased at the highest rate since the Johnson administration.

What to Do?

Why not balance the budget without a tax increase? From my perspective, that would be desirable but most unlikely in the near term. As long as the tax burden of federal government is lower than about 19 percent of GDP, the foregoing estimates suggest that our political representatives would resist a net reduction in federal spending as a percent of GDP. If our political system is biased in favor of larger government spending than a majority of the voters prefer, as is surely the case, we need to identify and correct these biases. My favorite rule to reduce these biases would be a constitutional amendment requiring that total federal spending in any fiscal year not exceed 110 percent of total federal receipts in the second prior fiscal year without the approval of a supermajority, say 60 percent, of the total members of each house of Congress or in any year in which a declaration of war is in effect. The 110 percent rule would lead to a small annual surplus in strong economic years, a small annual deficit in recession years, and a roughly balanced budget over time.

A 10 percent growth of federal spending over two years would maintain the federal spending share of GDP about constant, consistent with an annual increase of real GDP of 3 to 3.5 percent and an inflation rate of 1.3 to 1.8 percent, well within the range of recent experience. The 110 percent rule would require that tax cuts be matched with spending reductions within two years, and there would be strong incentives to increase economic growth and to maintain a low inflation rate.

Conclusion

If our political system then leads to decisions that roughly reflect voter preferences, the longer-term challenge for those of us who favor limited constitutional government is to try to convince voters to reduce their demand for the services financed by federal spending. Until that time, some increase in federal taxes appears to be a necessary part of a fiscal policy to balance the budget.