... the people behind the Urban Institute research. In traditional interview format, our experts talk about the nature of their work, offer insights on what they've learned, and describe the personal goals that keep them going.
“When I first came to Washington we had almost no reliable information about the costs of government programs, or about who would gain or lose from proposed policy changes. In the years since the Urban Institute was founded, such information has become standard, in no small measure because of the central role the Institute has played in developing microsimulation.”
–Senator Daniel P. Moynihan, 1998
During the 2008 presidential campaign, the Tax Policy Center (TPC) used its state-of-the-art microsimulation model to produce the only comprehensive and objective analysis of the candidates’ tax plans. The study revealed each plan’s effects on economic growth, the national debt, and families’ tax burdens. TPC’s conclusions were cited hundreds of times in print and broadcast news, by numerous research and advocacy organizations, and by the candidates themselves. Jeffrey Rohaly, TPC’s director of tax modeling, deciphers the tax model underpinning this and other TPC analyses.
Five Questions Archives
December 15, 2008
1. TPC became the go-to source on tax policy for the media and other groups during the campaign. Why?
TPC’s top researchers have earned respect across the political spectrum. Len Burman, Bill Gale, and Gene Steuerle (who just left Urban to become vice president of the Peter G. Peterson Foundation) are all highly respected by their peers and by policymakers. Our tax model was built by experts who had worked at the Congressional Budget Office (CBO) and at the Treasury. No one really quibbles with our numbers because they’re consistent with the numbers put out by government agencies—the CBO, the Treasury, and the Joint Committee on Taxation—which means our estimates are on the right track.
One of the first issues we worked on, the alternative minimum tax, helped to build our reputation. The alternative minimum tax (AMT) wasn’t dealt with when the Bush tax cuts were passed in 2001. We showed that, unless something was done about the AMT’s growing reach, it would hit a third of all taxpayers by 2010. Our estimates matched those of the government agencies, only they weren’t out there talking about the effects like we were.
We’re credible also because we’re nonpartisan. People may disagree with the implications others draw from our numbers, but we don’t put a spin on our findings. The left and the right both use our numbers. In the presidential campaign, you heard Sen. Obama say many times that his tax plan would give a cut to 95 percent of working families—that’s a number we produced. You heard Sen. McCain say that 40 percent of the population doesn’t pay income tax—that’s from our analysis. And any time either of the campaigns would talk about taxes, we’d get dozens of phone calls from reporters asking us if what was said was true.
So I think the sheer volume of information we put out, combined with the credibility we have earned over the years, have combined to make the TPC the place to go to get information about taxes.
2. For many readers, the tax model may be something of a black box. What is a microsimulation model? And what data go into it?
The “micro” part refers to the micro dataset we use. We don’t use aggregate data on the economy as a whole. Instead, we use a dataset of individual records the IRS provides to the public. The model relies on this “public-use file,” which is a representative sample of roughly 150,000 households, with information on these households’ income, number of children, and allowable tax deductions and credits. The IRS removes personal details from these records, so we don’t know certain demographic characteristics, such as household members’ ages or where they live. We also don’t get information about income not reported on tax returns, like welfare. So we use imputation techniques and statistical matching techniques to fill in the missing pieces. Government agencies get all the details, so the fact that our numbers are consistent with the government numbers means we’re doing a pretty good job filling in the holes. So that’s our database—at the individual level, it’s more or less everything you’d need to know to calculate a person’s income tax.
The “simulation” part is exactly what it sounds like: we simulate the tax system. I think of the tax model as a giant version of TurboTax. Each household in our sample is run through a tax calculator that tells us what that household owes under current tax laws. We can test new tax proposals by changing tax rates, brackets, or other parameters and then run the model again to see how each household’s tax burden would change. We can add up the changes and see how overall tax revenue would be affected. Or we can narrow our findings to see how certain groups or income classes would fare.
We can also look ahead, estimating for the future by aging individual records, basing wage growth and other factors on CBO forecasts. Every January, when CBO releases its budget and economic forecast, we update our aging routine. We update the underlying data whenever the IRS releases a new public use file.
3. If TPC’s tax model resembles government agency models, what makes our work unique?
For one thing, we release a lot more information. CBO releases information but not usually on individual proposals. The Joint Committee on Taxation has issued a few distribution tables in the past seven or eight years; the Treasury hasn’t put out a comprehensive distribution table in a while. We’ve put out more than a thousand distribution and revenue tables over the past seven years, and we analyze any bill that gets publicity or reaches the Ways and Means Committee or Senate Finance Committee.
Also, we get requests for analyses from people on the Hill, from other research organizations, and from the media. And, for the most part, we don’t turn people down, unless our resources are stretched to the limit. We’re able to respond quickly too, though not at the expense of getting our analysis right.
A particularly quick turnaround was our analysis of President Bush’s proposed tax cuts in January 2003. At the time, it was clear that he was going to propose this second round of cuts. To be ready to analyze his plan the day it came out, we pulled information from news reports about what would likely be in the proposal. We were lucky because the plan more or less matched what we were expecting, so we were able to release distribution tables just a few hours after the president announced his proposal, and our numbers were reported in that evening’s news. That analysis and our follow-up work through May of that year—when the 2003 tax act finally passed—further established our reputation as the go-to source for distribution analysis.
With the presidential candidates’ tax plans, we worked behind the scenes with the campaigns for a long time, trying to nail down exactly what their plans entailed. Publicly, the candidates talked in broad strokes, but we needed more detail to run the plans through our tax model.
4. How accurate is the tax model? And what are its limitations?
The primary goal of modeling is to get as close to reality as possible. But when our estimates are off, or when official government estimates are off, it’s usually not because there’s something wrong with the methodology or the underlying database; it’s because predicting the future state of the economy is so hard. Capital gains, for example, are extremely sensitive to economic fluctuations. In our tax model, we rely on publicly available government forecasts of the state of the macroeconomy. I’m comfortable with our estimates since I know we use the best available forecasts and employ reasonable and widely accepted methodologies in producing our estimates.
The tax model is best for looking at incremental changes to the tax system, such as the presidential candidates’ tax plans. While it may sound like both candidates were proposing sweeping reforms, they’re really working within the context of the current tax system. Changing tax rates, modeling new credits, expanding current credits, and so on is not that hard to model. But dealing with fundamental changes in the tax system—such as instituting a value-added tax—definitely is. We can estimate how behavior will change when people face a tax on consumption instead of income, but we don’t have past experiences to build on. In general, accurately capturing behavioral responses to policy changes is the most difficult part of modeling.
5. How has the tax model evolved?
The tax model has evolved in the scope of taxes it includes. At first, the tax model could simulate only individual income taxes and payroll taxes. Over the next few years, we added a bunch of additional capabilities. We built a model of the federal estate tax, the tax on wealth you pay when you die. That took us a year to build, but now we have a capability that very few models have. Then we added the capability to include the corporate tax in our distribution tables. Soon we’ll be able to include federal excise taxes.
We’ve also made a lot of improvements. We can now measure the effects of retirement savings initiatives. On the education front, we can estimate changes to the Hope Credit, the Lifetime Learning Credit and model more complicated reforms of education tax incentives. And our distribution tables have become a lot more sophisticated. With distribution tables, you’re grouping people by their ability to pay taxes, so you need an accurate measure of income. When we first started, we used adjusted gross income, but that leaves out sources of untaxed income. We’ve since developed new and better measures of income that capture those variables and better represent a taxpayer’s ability to pay.
