The views expressed in this Testimony are those of the author and do not necessarily reflect the views of the Trustees and employees of The Urban Institute.
Mr. Chairman, Mr. Spratt and members of the Committee, thank you for the opportunity to testify.
Since leaving the Congressional Budget Office (CBO), I have had the opportunity to work on budgeting issues in a number of countries. It is remarkable how many different constitutional arrangements exist for dividing budgeting power between the executive and legislative branches of government. But few countries give their legislatures as much budgeting power as that enjoyed by the Congress of the United States.
Regardless of a legislature's constitutional power, its actual influence over budget decisions can be enhanced if it can draw on analyses done by an expert staff. That is true even in parliamentary systems where the executive branch has most constitutional power. But obviously, the analytic input from such a staff is most crucial where it is the legislature that is most important in making budget decisions.
There are major advantages in keeping the expert staff nonpartisan. It lends more stability as political power shifts and that allows the development of specialized skills in different areas of public policy. A nonpartisan staff often has more credibility with outsiders, and although there are exceptions, those analysts who try to combine rigorous policy analysis with political judgments typically do not do well with either. It is better to let analysts be analysts and to let elected politicians decide which of the analytic results can be sold to the voters.
I am, of course, biased, but I have little doubt that the existence of CBO has greatly increased the Congress's capacity to budget and enhanced its influence vis-à-vis the executive branch. CBO's forecasts give the Congress an alternative view of the economic and budgetary future; its cost estimates guard against the Congress unwittingly adopting programs whose costs are very different in the long run than in the immediate future; and its policy analysis helps the Congress decide what works and what doesn't work.
It is inevitable that some of CBO's output will be wrong and some of it will be annoying to one political party or the other, either because mistakes were made or good analysis was badly timed. But if one adds up the impressive volume of CBO cost estimates, analyses, and forecasts, a remarkably high portion is non-controversial and a remarkably low portion really makes someone angry.
I shall concentrate the rest of my testimony on a very few areas of CBO responsibility where I have strong views, but I would be happy to answer questions about other areas as well. I shall focus on CBO projections of budget aggregates that are used to formulate budget resolutions and on the issue of dynamic scoring of tax and expenditure policy changes.
Budget forecasts No one forecasts anything very well. That is true whether one looks at pundits forecasting the course of the war in Afghanistan, demographers forecasting worldwide birth rates, or pollsters forecasting the French presidential election. It is particularly difficult to forecast the budget balance, because one does not forecast it directly. One forecasts two much larger numbers revenues and outlays and takes the difference. Relatively small percentage errors in forecasting revenues and outlays thus imply very much larger percentage errors in forecasting surpluses or deficits. For example, in 2001, revenues totaled $2.0 trillion and the surplus $127 billion. Every one percent error in forecasting the former implied a 16 percent error in forecasting the latter.
I recently studied the history of errors and I would like to submit my results for the record. They are pretty discouraging. The average error made in the forecast of the budget balance used to formulate the budget resolution is over $100 billion for the first year covered by the resolution and over $400 billion five years out. These are errors made because of flaws in economic and technical assumptions and do not include the effect of policy changes. (They are also adjusted for the growth in the economy.) Ten-year projections were initiated only in 1997; so we cannot test them against reality. But the projection for the budget balance in 2007 changed over $800 billion between early 1997 and the summer of 2000 an amount equal to more than five times the value of the 2001 tax cut in 2007. If our view of the 2012 budget balance changes by a comparable amount over the next 3-1/2 years relative to GDP, it will be altered by a cool $1 trillion for that single year.
The importance of errors of this type depends on how a forecast is used. The 75-year forecast used by the Social Security trustees is bound to be off by huge amounts in dollar terms, but it is unlikely to be wrong about its basic qualitative conclusion that the economic burden of supporting the Social Security system will rise rapidly between 2010 and 2030. Similarly, flaws in economic forecasts are unlikely to obscure the qualitative nature of the budget effects of a tax cut or an entitlement increase. That is to say, if an entitlement increase is shown to cost very much more in year 7 than in year 4 by a good forecast, roughly the same pattern of costs is likely to be revealed by a bad forecast as well. Put yet another way, forecasts of changes in a baseline due to policy changes are likely to be more accurate than forecasts of the baseline itself.
But I believe that the Congress asks for too much when they ask CBO for a ten-year projection of the budget balance for the purpose of formulating a budget resolution. The projected budget balance is too erratic from year to year to be used for that purpose. Five years is about the outside limit for a budget resolution and even that is tenuous. (I realize that the House emphasized the first five years in this year's resolution.) There is nothing to prevent the Congress from requesting that CBO do an economic forecast for years 6 through 10 that would be hidden in an appendix somewhere and pulled out to estimate the effects of a particular tax or entitlement measure. That would allow the nature of phase-ins to be observed. But I would not compute a budget balance ten years hence and put it in a budget resolution, because that is essentially a useless exercise.
Although errors in forecasts are likely to be huge, there is one custom that tends to make forecasts seem even more volatile than they really are. CBO, the press and the public discuss the cumulative budget balance over five or ten years. That is likely to change by hundreds of billions from forecast to forecast and that seems like a lot of money. But adding the budget balance for one year out to that for five years out makes no sense, because the latter is so much less reliable than the former. It is truly adding apples and oranges. I wish CBO would expunge the columns from their tables that indicate cumulative totals, but the custom of using them has become so entrenched that I know that I am fighting a losing cause.
Improving budget forecasts Because forecasting is inherently difficult, there is not much that CBO, the Congress, or anyone else can do to greatly increase the accuracy of budget forecasts. However, there are actions that might result in minor improvements.
A major frustration facing revenue forecasters is that it takes a very long time to get information on recent tax receipts. CBO and OMB will have little information on the causes of the recent surprising shortfall in revenues by the time that they have to provide budget updates next summer. Different causes can have very different long-term implications, detailed information on 2001 tax returns will not be available until October or November, and even that data will not be perfectly accurate.
Changes in reporting could help a lot. For example, corporations do not immediately divide their tax payments between payroll and profit taxes. If they were asked to report HI tax collections a proportional tax revenue estimators would immediately have valuable information on total earnings in the corporate sector. Further valuable information would come from reporting stock options on W-2's for individuals or in the aggregate for a corporation. Of course, any increase in reporting comes with a compliance cost imposed on business, but I believe that these suggestions would not be very costly. It is also possible that a small infusion of money into the IRS could expedite the processing of returns, so that revenue estimators would not have to wait so long for basic information.
In my view, our statistical agencies do not have the budgets necessary to produce high quality statistics. Canada does better. Fundamentally, the Bureau of Economic Analysis (BEA) and the Bureau of Labor Statistics (BLS) should have more resources for basic research, so that their data collection techniques could keep up better with the rapidly changing structure of our economy. Of more immediate interest, the income side of our GDP accounts that is vital to revenue estimators is not given the same attention as the product side that is of more interest to business economists and other observers of the economy. Although the two sides should be equal in theory, there have been major statistical discrepancies in recent years. It is very difficult to make a decent forecast, if we have bad information on past history.
One could go on and on about deficiencies in official statistics deficiencies that could be ameliorated with minor infusions of money. The Administration requests a healthy increase in the BEA budget this year. I hope that the Appropriations Committees find a way to fund the Administration's request.
Again, I want to emphasize that better and more timely historical data will not enormously improve the accuracy of forecasts. It won't help us predict another terrorist attack or a Mideast oil embargo. But it may occasionally save us from a big mistake and that would be worthwhile.
Estimating revenue and expenditure feedbacks For many years, the Congress has been frustrated by the inability of the Joint Committee on Taxation or the CBO to provide a complete accounting of the revenue and outlay effects of behavioral responses to policy changes. It is commonly believed that no behavioral responses are considered. That is not true. Micro responses play a role in making estimates. For example, revenue estimators would take account of the effects of a change in gasoline taxes on the demand for gasoline when they make the revenue estimate that appears in the report on the legislation. They would not go further and estimate the impact on GDP or on the CPI or on other macro variables. Thus they miss the impact on other revenues because of the effects on GDP growth or tax indexing changes, and the impact on outlays because of changes in unemployment compensation or because of changes in COLA effects on indexed programs like Social Security or food stamps.
There is nothing to prevent CBO from doing studies to inform the Congress of the findings of academics and others as to the complete dynamic effects of specific policy changes. In fact, CBO has done such studies on capital gains tax rate changes and other things. The Congress will probably be disappointed by the wide range of uncertainty on such matters, but it is no wider than CBO has to deal with when forecasting the economy more generally.
The real practical problems come if CBO is asked to do dynamic scoring for a complex tax or reconciliation bill. Such a bill usually contains numerous provisions some pro-growth and others anti-growth. Dozens of technicians often work on different provisions of the bill simultaneously at JCT and CBO. If he is doing dynamic scoring, analyst A may decide that his provision increases the GDP growth rate next year by 0.1 percent. That should force every other analyst to re-estimate the effects of their provision whether or not they think their provision has any effect on growth. Two hours later Analyst B may decide that her provision reduces growth 0.2 percent. Again, everyone, including Mr. A should redo their estimates. Moreover, every change in the assumed GDP or the CPI or the unemployment rate will affect almost every other type of tax revenue and entitlement outlay, whether or not it is affected by the legislation. The budget baseline would have to be recomputed with every significant piece of legislation. The implied management problem is made even more difficult by the fact that the Congress often makes important changes in the language of bills at the last minute and much of the CBO scoring effort takes place very late at night and can last until dawn. Careful dynamic scoring would only be possible if Congress allowed several days for scoring instead of several hours, and even then it would be extremely difficult, if not impossible.
Another problem is that Congress would find it difficult to deal with an ever changing baseline. Before Gramm-Rudman the Congress used to change its baseline with the summer budget update provided by CBO. That would change the estimates attached to all pieces of legislation then being considered. But it was decided that this was too disruptive to bargaining over the details of bills. The Congress decided to let the earlier baseline be used throughout the year.
The Congress could do dynamic scoring of individual bills without changing the baseline, but this would often lead to illogical and inaccurate results. The effect of one program change on the cost of other programs can often be substantial. For example, anything that changes the CPI has a relatively large impact on outlays for indexed entitlement programs and personal income tax revenues.
Apparently, there are discussions about adding statements to the text of cost and revenue estimates where there might be an important effect on macro variables. These would be separate from official numerical estimates. Probably it will be practically necessary to confine the discussion most of the time to qualitative rather than quantitative statements. Nevertheless, such statements may be helpful to the Congress. Although I bemoan the recent ineffectiveness of the Budget Enforcement Act, it must be admitted that pay-as-you-go rules created a tyranny of numbers that did not allow the Congress to apply much judgment in assessing the value of tax and entitlement measures. Now the Congress has more room to decide whether a provision is better or worse than the partially static, numerical estimates imply.
If CBO and JCT start making judgments about macro variables that would supplement official cost and revenue estimates, they will have one more activity that will make people angry. They will have to make some very unpopular statements. For example, good analysis will counter-intuitively show that there are some tax cuts that decrease growth, and some tax increases that increase growth. I am happy that I will not be answering phone calls from Members after such judgments are made.