Vague Thoughts on Modeling Dynamic Scoring

NBER MONETARY ECONOMICS PROGRAM MEETING, NOVEMBER 5, 2004

Discussion of N. Gregory Mankiw and Matt Weinzerl, "Dynamic Scoring: A Back-of-the-Envelope Guide"

J. Bradford DeLong
U.C. Berkeley

Let me start by raising a doubt about one of the purposes of the paper--a doubt that I am not sure that I endorse, but that I do think is worth taking seriously. When I worked for the Treasury, I asked the career civil servants why they were so opposed to "dynamic scoring." Their answer was that every administration's bozo political appointees come in with their own model that says that their particular policies are absolutely flawless and totally great. The model errors that administrations bring with them, they said, are much bigger than you would believe: in fact, they hinted, my own model errors were bigger than they dared tell me. Confining analysis to "static scoring" was, they said, a way of minimizing the impact of model error over a sequence of administrations. It generated results that were wrong, yes, but the sum-of-squared-policy-errors was less than if each administration got to claim that its own particular model were true.

It seemed to me then that this might well be wrong: that the Congressional Budget Office, at least, had the expertise and the independence to do "dynamic scoring" in a sane and consistent way over time. I can't claim to have settled views on this, however. All I have are vague worries.

Let me say, next, that I definitely share another of the underlying motivations for this paper: I very much fear that our current system of capital leaves not just $20 bills on the sidewalk but $1000 bills, and not just one but millions. In quite a number of models the best way to tax capital is to impose a large capital levy at time 0, and immediately thereafter reduce capital taxation to the minimum amount necessary to meet the constraints imposed by one's distributional goals. (Dynamic consistency considerations modify this, of course, but I have never been able to figure out how in an appropriately general manner.) Our current system of taxing capital in no way fits any part of this pattern, and I have never heard a convincing argument that distributional goals cannot be met by taxing other forms of income.

My fear his risen substantially in the past few weeks as I have been trying to write my review of Peter Lindert's _Growing Public_. Peter narrates the growth of social insurance states across the OECD. One of his central questions is why, given the huge burden that raising the revenue to pay for the benefits of European social insurance states places on their economies, western European economies aren't poorer today. They are within striking distance of the U.S. as far as real output per capita is concerned. They are by and large even as far as real output per hour worked is concerned (but I do not believe that lower levels of hours per capita in western Europe are fully the result of differences in preferences). They accomplish this with much lower land endowments per capita than the U.S. does--and land is an important input in the U.S. production function. Why, if tax systems do have first-order effects on economic activity, isn't Europe much poorer?

Peter Lindert's answer is that western Europeans have crafted tax systems that lie lightly on the economy. Open economies keenly aware that capital is mobile and that their citizens by and large have not, they have faced incentives to reduce the possibility of business flight to a minimum and to make sure that their tax codes encourage domestic investment. The U.S., by contrast, has not. Lindert's judgment is that western Europe's tax system raises much more revenue without imposing a significantly larger burden on the private economy. He thus echoes Adam Smith's judgment in the fifth book of the _Wealth of Nations_, comparing late eighteenth-century Britain and France, that Britain's tax system raises three times as much revenue per capita as the French, and is felt to be only one-third as burdensome. And the French tax system was burdensome: it was, after all, the fiscal crisis of the late eighteenth-century French monarchythat was the trigger for the French Revolution.

These fears are magnified even further if you don't believe that the big prices in the economy accurately reflect social marginal valuations. If there are important labor rents earned by workers in capital-intensive industries, then the excess burden from capital taxation will be magnified and will fall on the fortunate rent-sharers in labor as well. If total factor productivity does not fall from the sky but is instead linked to investments in any of a number of ways, then any capital tax that reduces investment will reduce productivity growth as well. As far as American economic growth is concerned, the big thing in the past two decades has been the extraordinary reduction in the real price of computers and the concommitant incredible surge of capital deepening in information technology--a surge of capital deepening to which Steve Oliner and Dan Sichel attribute more than half of the surge in American labor productivity growth since the early 1990s.

Do we really believe that this technological progress in making computers was independent of investment in information technology--that Moore's Law would still have held had our businesses invested nothing at all in computers over the past two decades? No, we do not. Then by how much would more capital-friendly tax policies in the 1970s and 1980s that encouraged investment, including investment in information technology, have brought forward in time the high-tech productivity boom of the 1990s and 2000s? This is a first-order question to which I do not know the answer. But it is hard to think about this question seriously and not conclude that it is a powerful factor making it likely that this paper's estimates of growth effects and revenue offsets are more likely to be low than high.

But that there are other, I think more powerful, factors making it likely in my estimation at least that the estimates are more likely to be high than low. It's not that there is anything wrong with their modeling. It's that what they call a "tax cut" is not quite what my friends in the reality-based community think that a "tax cut" really is.

In this paper, a "tax cut" is a permanent and credible reduction in taxes accompanied instantaneously by an equally permanent and credible reduction in the permanent level of government spending. But that's not what a "tax cut" is here in America. Consider that the Congressional Budget Office tells us that in fiscal 1981 federal spending gross of offsetting receipts was 23.4% of GDP. Then came the Reagan tax cut. And revenues--including offsetting receipts--did fall from 20.8% of GDP in 1981 to 18.5% of GDP in 1983. But spending did not: 23.3% of GDP in 1984, and still nearly unchanged at 22.5% of GDP in 1993. Similarly for our latest round: revenues that peaked at 21.9% of GDP in 2000 are projected to be about 17.2% of GDP in the current year, fiscal 2005--with roughly one-third of the share decline the result of the collapse of the dot-com bubble, one-half the result of the Bush tax cuts, and the rest due to cyclical and other factors. But spending? 19.2% of GDP in 2000, and projected to be 21.4% of GDP in the current fiscal 2005.

Perhaps if we had a return to the caps on discretionary spending and the Budget Enforcement Act rules for revenues and entitlements that were in force during Clinton's term--during which federal spending fell from 22.5% of GDP to 19.2% of GDP--a "tax cut" in the real world would correspond to a "tax cut" in the models of Mankiw and , for then passing the tax cut would require that the Congress pass and the CBO score offsetting spending cuts. But that's not the world we live in. In the world in which we live in, a "tax cut" is a change in policy that leaves the question of how the government budget constraint is going to be enforced hanging wide open.

For example, the Clinton administration's fiscal policy was a direct result of Reagan's 1981 tax cut, and during the Clinton administration federal taxes as a share of GDP were higher than at any time since the Korean War. A temporary tax cut followed by an increase of taxes to a higher level is a supply side loss, not a win, even if you are a believer in Ricardian equivalence and convince yourself that the fall in national savings rates in the 1980s was unrelated to Reagan fiscal policy.

If we were going to get concrete, and, say, take the Bush tax cuts of 2001 and 2003 and try to assess their dynamic supply-side effects within a model in which the government budget constraint is satisfied as it will be satisfied, what would we have to do? I believe that we would have to model the satisfaction of the government budget constraint as a probabilistic combination of three different possibilities:

1. Recognize that there is a chance that the tax cut will be reversed. Perhaps once again, as in the early 1990s, an unwillingness to cut spending combined with mounting debt and debt servicing costs changes the complexion of politics. I would have said that the chances of this are high given the 7% of GDP long-run fiscal deficit that America appears to have, and the unwillingness of any politician to propose cuts in the growth rates of Medicare and Social Security spending. But the chances of this have dropped since Kerry reached his peak bubble value of 80% on the Iowa Electronic Markets Tuesday afternoon.
2. Recognize that there is a chance that there will be a long period of rising debt and debt burdens accompanied by cutbacks in spending shares as various institutional mechanisms force the legislature to face and try to meet the government budget constraint. We know what George W. Bush thinks of the pay-as-you-go mechanisms that restrained Congressional action so effectively in the 1990s: he thinks they are a joke: "You know what pay-go means? It means you pay--and [Kerry] goes and spends!" I would say that the chances of this have also dropped since Tuesday afternoon.
3. Last, there is the remaining possibility: that the government budget constraint itself will take its own non-policy steps to make sure that it is met, and generate an Argentina-style meltdown. By the principle that probabilities sum to one, I conclude that the chances of this have risen since Tuesday afternoon.

Now it seems to me that the Ramsey model is likely to be far off in its assessment for any of these three scenarios, even the second.

In the absence of *effective* institutional constraints on Congressional action along the lines of the 1990s Budget Enforcement Act, and given the current complexion of American politics, it seems to me that the paper's estimates of offsets are much more likely to be high than low. Indeed, I fear that the current Congress will ignore the government budget constraint until asset prices bring it to its attention in a striking way--which means that they will ignore it until the markets conclude that option 3 has a significant probability. If true, I would think that with the current Congress the offset would be negative--that the static revenue cost of the tax cuts understates the expected value of the burden imposed by reduced government spending and other effects.

The paper's results are conditional on the existence of some effective institutional framework, like the Budget Enforcement Act, to make sure that the government budget constraint is satisfied the way we would like it to be satisfied. The paper simply doesn't work if the government budget constraint is satisfied in a way that it would be unsuitable for the ears of the sensitive and high-strung (for example, the characters of "The Importance of Being Earnest" who are told to skip the chapter on the fall of the rupee) to hear.