In the James Room of the Renaissance Baltimore Harborplace Hotel, Erick Sager is demonstrating what happens when you admit that people die. Half-lit by PowerPoint, he explains that a seminal 1998 paper on the ideal level of government debt relies on an infinitely lived agent—it assumes that people are immortal.
This isn’t as crazy it sounds. A lot of macroeconomic predictions still rest on this assumption. It makes the math easier. Packed in the room with Sager are 19 Ph.D. economists from universities and government agencies, taking turns at the lectern. They’ve gathered for the annual conference of the National Tax Association. Infinitely lived agents certainly don’t sound crazy to them.
Sager, an economist for the Bureau of Labor Statistics, speaks in surprisingly plain English. He’s added to that 1998 paper, he says, “by relaxing the assumption that people are infinitely lived.” Economists call this a “life cycle” approach, and it produces a dramatically different result. Rather than hold debt, his life cycle analysis suggests governments should hold savings. For the U.S., that’s a swing on the order of $20 trillion.
William Gale from the Brookings Institution rises to respond. When you go home to your families for Thanksgiving, he says, you really ought to be able to talk with authority about the ideal level of sovereign debt. But, he laments, the research just isn’t there yet. “Better to talk about Trump,” says another economist. There’s a short bark of nervous laughter.
The PowerPoint slides continue, part of a session called “Structural Models of Fiscal Policy Effects.” A model, an economist’s basic kitchen knife, is a simulation of how people will respond to, say, an interest rate hike or a sudden flood of cheap labor. The people in the James Room build models to estimate what happens when the federal government changes the way it taxes or spends.
When Washington argues about fiscal policy, it’s really fighting over models. By the time the White House produces its budget, its Office of Management and Budget has already modeled what it hopes that budget will do. Majorities in Congress send their budget resolutions to their own preferred think tanks for modeling, too. Then, by statute, bills that come out of most committees must receive a “score”—a modeled result—from the Congressional Budget Office and, for revenue bills, the Joint Committee on Taxation. The CBO and the JCT have a reputation for straight-backed probity, but congressional staffers quietly haggle with both institutions over footnotes.
So Republican economists model against Democratic economists, with some referee economists in the middle. You say your tax cuts can be offset by economic growth. Oh, I ask? Well, are your agents life-cycle or infinitely lived? This is the knife fight in the kitchen, and it’s how the presumed mortality of imaginary people determines the size of your tax bill.
The discussion in Baltimore was planned months ago, but it took place last Friday, on Nov. 11. Structural models of fiscal policy effects have sudden relevance. A single party now controls the White House and both houses of Congress, which means a revenue bill is coming soon, with the first significant tax cuts since 2003. It’s likely to pass, and an appropriations bill will likely follow hard upon. Anyone who cares about taxes, spending, and the debt owed by the U.S. Treasury is going to have to start caring about the details of models.
Republicans have long argued that economic growth from tax cuts should be fed back into the model, year by year. They call this approach “dynamic scoring” or “macroeconomic analysis.” For the first time, macroeconomic analysis will likely prevail in next year’s official scores for major revenue bills from the JCT. Some Democrats, who’ve been suspicious of an approach that makes tax cuts look cheaper, are slowly warming to the same idea for appropriations bills. It could make infrastructure spending look cheaper, too.
Into this fussing over details strides Donald Trump. During the campaign, he proposed a tax cut that would cost, according to his own preferred estimate, $4.4 trillion. And to pay for it, his campaign proposed a new kind of analysis, an economic model radically more complex than what either academics or policymakers have tried in the past.
All aspects of Trump’s plan, including trade and regulatory rollbacks, would be part of the analysis. Together, the campaign argued, they would create enough growth, and therefore enough tax revenue, to offset all but about $200 billion of those tax cuts.
The real challenge of budgeting is to offer something, but at a discount. In 2017 dynamic scoring will let the Republican majority offer tax cuts without having to offset them entirely with spending cuts. It may even offer infrastructure spending—without having to renege on the promise of tax cuts. If the models are right, they’re right. If they’re wrong, the tax cuts will be a debt-driven Keynesian stimulus.
Dynamic scoring arrived on the Republican wave of 1994. In January 1995, as one of its first acts, the new GOP majority in Congress invited Alan Greenspan, among others, to a rare joint hearing of the budget committees. The representatives wanted to talk about macroeconomic models of budget changes. Greenspan, then the chairman of the Federal Reserve and thus in charge of the world’s best-known macroeconomic modeler, was skeptical.
Then as now, the CBO every year produces a 10-year projection of economic growth. This is the “baseline,” the fixed point from which everything else is calculated. Under “static analysis,” modelers in Washington make assumptions about human behavior. But as they project out into the future, they can’t change the CBO’s baseline gross domestic product. Under “dynamic analysis,” they can. Next year’s projected growth changes the baseline for the year after, and so on. If static analysis is arithmetic, dynamic analysis is calculus.
“Full dynamic analysis of individual budget initiatives should be our goal,” Greenspan told Congress in 1995. “Unfortunately, the analytical tools required to achieve it are deficient.” He himself would prefer to reduce or even eliminate taxes on capital gains, he said, and he suspected it would lead to little or no loss in revenue over the long term. But he couldn’t prove it. “More to the point,” Greenspan added, “if we fail to achieve adequate reductions in outlays, budget scoring will not substitute for hard political choices.” An e-mail from his office on Nov. 16 read simply: “Though he wishes it were otherwise, Dr. Greenspan has not changed his views.”
Under Republicans, the models became increasingly embedded in the process. In 2003 the House required the JCT to perform macroeconomic analysis on tax legislation, and the CBO did the same for the president’s budget. In 2006, during a push for immigration reform, both the JCT and the CBO modeled what changes to the labor force would do for economic growth.
“Democrats were less enthusiastic,” says Doug Elmendorf. Now the dean of the Harvard Kennedy School, Elmendorf ran the CBO from the beginning of the Obama administration until last year. Until 2009, dynamic scoring had not been used on spending programs. It can be, but the economic growth takes longer to show up. Pay for universal pre-K, for example, and any economic growth driven by educated, well-adjusted citizens won’t appear for another 25 years. But under Elmendorf, the CBO produced a dynamic analysis of the American Recovery and Reinvestment Act of 2009, what’s commonly referred to as the Stimulus.
So far, the CBO and the JCT have produced analyses, not scores. The distinction is important. An analysis is just a piece of paper. A score is a statutory requirement and, depending on its outcome, it can trigger parliamentary procedure. Since last year, a House resolution has required the JCT to produce a dynamic score of all revenue bills. The Senate is likely to agree, which means that for the first time, rules of order will hang on macroeconomic analysis.
But the requirement is only for revenue bills—the ones that change the tax code. It’s not required for appropriations bills, the ones that spend money. Which means that a tax cut, say the $4.4 trillion one proposed by Trump, will look cheaper. But any check cut for infrastructure spending won’t.
Although they aren’t allowed to talk on the record, two JCT economists—the referees—are on the panel in Baltimore, too. After the standard caveat that their views don’t represent those of the Joint Committee, they present a paper. It shows how changes to the tax code can seem to have minor aggregate effects, but weigh heavily on specific groups of people—those who own houses, for example, or receive the earned income-tax credit. Those differences, in turn, affect the aggregate.
Here’s another thing aside from immortality that doesn’t seem absurd to economists. Until the last decade, most macroeconomic models assumed a single agent—that the entire economy is just a single person making choices. Again, this made the math easier. The JCT paper is part of a broader move among academics and central bankers toward “heterogeneous agent” models, in which different people respond differently. Thomas Barthold, the JCT’s chief of staff, confirmed that the paper is part of a “major” overhaul of one of its models for 2017.
Since the last time Washington passed a major tax cut in 2003, economic models have become more complex. Computing power is even cheaper. New access to massive data sets has opened up new ideas. And—this is a consistent theme at the Baltimore meeting—economists have been adjusting the assumptions in their models to account for a slow recovery that they don’t completely understand.
The International Monetary Fund, for example, moved toward heterogeneous agents—the people-are-different approach—after its models failed to predict Greeks’ reaction to their country’s austerity program. The IMF had assumed that, faced with budget certainty, people would be more willing to spend. In reality, rich people did. Poor people didn’t.
But even the smartest economists with the most powerful computers can’t capture the world’s every variable, and so modeling, particularly in a fast-moving legislative environment—the JCT analyzes 6,000 to 7,000 proposals each year—is the art of figuring out what can be ignored. The JCT’s models have embraced both heterogeneity and the life-cycle approach. That people are both different and mortal is too important to ignore. But buried in a presentation on its website is a curious admission.
One of its models requires “perfect foresight,” that people and businesses can predict what government will do in the future. “Perfect foresight means the model cannot solve if fiscal policy is unsustainable,” the presentation reads, “requiring counterfactual assumptions about how fiscal policy will be made sustainable.” In English: The math doesn’t work if Congress can’t be bothered to eventually balance the budget.
The world is full of people and institutions that don’t behave as they should. Before the financial crisis, models expected central banks to counteract aggressive fiscal stimulus, raising interest rates to avoid inflation. And yet, central banks all over the developed world are now keeping rates low and begging for aggressive fiscal stimulus.
In Baltimore, Nora Traum of North Carolina State University presents a paper titled Clearing Up the Fiscal Multiplier Morass. A multiplier is the expected extra effect of a fiscal policy: Spend $1, and you might get $1.60 of economic growth. “Morass,” because different assumptions create different multipliers. Recently, the JCT asked nine modelers, using three different kinds of models, to predict the effect on growth of three different tax reform proposals. For one reform, predictions on growth varied from –4.2 percent to 16.4 percent in the short run, and from 1.7 percent to 7.5 percent in the long run.
Traum simplified assumptions by creating two sets of conditions. In one, Congress is relaxed about the problem of long-term debt. In the other, the Federal Reserve is relaxed about long-term inflation. She’s trying to answer a question that’s been bothering macroeconomists for several years: Should you model what people and institutions should do, or what they actually seem to be doing?
“The recent recession,” says Traum, “brought out this need to evaluate our understanding of multipliers.” She prefers to give a range of estimates, rather than a single point, to avoid false precision. But she recognizes that politicians and the public aren’t academics and may need the clarity of a single point.
Elmendorf supports dynamic analysis, so long as it’s performed on all parts of the budget and not just tax cuts. He, too, prefers to give a range of estimates. “It’s good to be modest about what we know about the world,” he says. But not too modest. Make a tax cut or an infrastructure bill look more expensive than it is, and you cost jobs unnecessarily. “It’s not so obvious,” he says, “what the safe default way is to go.”
“Our central beef,” says Peter Navarro, “is that every think tank only has the inclination to focus on a small part of the problem.” In late September, he and Wilbur Ross released a white paper titled Scoring the Trump Economic Plan: Trade, Regulatory, & Energy Policy Impacts. Navarro is an economist at the Paul Merage School of Business at the University of California at Irvine, advising the Trump transition team. Ross, a billionaire investor, is being vetted for a cabinet position. (Navarro and Ross spoke to Bloomberg Businessweek before the election.) The paper lays out how, after a $4.4 trillion tax cut, Trump gets close to deficit neutral.
Over the summer, both Trump and the House Republicans sent their tax reform plans for analysis to the Tax Foundation, a conservative think tank. The foundation priced the GOP’s tax cut at a cost of $2.4 trillion over the next decade—static. Dynamic, it found a deficit of only $191 billion. Trump’s tax plan shook out at $4.4 trillion to $5.9 trillion static cost, $2.6 trillion to $3.9 trillion dynamic.
According to Kyle Pomerleau, who wrote both analyses, most of the economic growth that feeds back into the dynamic model comes not from reductions in the personal or corporate tax rate, but from allowing businesses to write off their investments immediately instead of over time. This will encourage the purchase of productive capital, which makes workers more efficient and creates growth. Feed that back into the model, repeat.
In Pomerleau’s modeling notes—always read the modeling notes—he lays out two crucial assumptions. First, like North Carolina State’s Traum, he doesn’t assume that politics or bond markets will force Washington to balance the budget over the long term. Second, the model “does not take into account the fiscal or economic effects of interest on debt.”
In almost all standard models, a chunk of new government debt raises interest rates and “crowds out” other investments. The Tax Foundation doesn’t agree with this assumption. “When we model tax plans, we take the global savings glut seriously,” says Pomerleau. Basically, the Tax Foundation believes that there’s so much money in the world that it will find a home no matter what the U.S. Treasury does.
This isn’t quackery. There is a global savings glut. But the assumption that there won’t be any crowding out of other investments is a significant departure from other models, the practice at the JCT and the CBO, and available research (though most of that research predates the recession). And it puts Republicans in an awkward position. Their plans score well if you don’t think government debt is bad.
For Trump’s plan, Ross and Navarro start with the Tax Foundation and keep going. “Once you get into dynamic scoring,” says Ross, “the question is to what do you apply the dynamic scoring?” They apply it broadly, finding $1.7 trillion in revenue from higher growth from a new trade policy, a bit less than $500 billion from rolling back regulations, and just under $150 billion from a new energy policy.
Cost-benefit analysis on regulation is even less certain than the macroeconomics of tax reform. The White House puts out an annual report on only a small portion of its regulatory code. Ross and Navarro believe that the key is hires, that federal agencies will be able to shift staff toward figuring out which regulations need to go. Once those recommendations are ready, the CBO would have to build it into a brand-new model.
“So what?” says Douglas Holtz-Eakin. Now with the American Action Forum, a right-leaning think tank, he ran the CBO for several years under George W. Bush. Like Elmendorf, Holtz-Eakin has trust in the strength and independence of the institution. Attempts to shift scores, he says, add up to “fine recreational yelling and screaming.”
What Holtz-Eakin and Elmendorf and even the people at the Renaissance Harborplace are counting on is the strength of budgeting institutions built over the last 40 years and a continuation of wonky fights over footnotes and process. You say crowd out, I say global savings glut. You say infinitely lived, I say life cycle. You say the Fed is an active agent, I say passive.
But models and even official scores don’t have any constitutional meaning. They serve only to chasten. It’s not clear what will happen when they encounter a president who can’t be chastened. “That’s going to be the interesting moment,” says Holtz-Eakin. “They’re going to put out a budget. They’re going to dynamically score it. And I can’t wait to see it.”