In a speech in Frankfurt in October, Peter Praet, a member of the executive board of the European Central Bank, told a conference of economists something curiously obvious. “Individual households are heterogeneous in many respects,” he said. “It is important to measure and analyze this heterogeneity because it can have important implications for aggregate figures.” People are different, he meant, and we need to understand how to understand the economy.
Praet had to state the obvious because until this year economists, in particular those who make forecasts, put their faith in models that ignored those differences. Those that the ECB and the International Monetary Fund used to predict the future relied on a “representative agent,” a single imaginary person who stands in for everyone.
The problem was that these models failed to predict the consequences of the austerity programs that several European countries adopted in 2010. It turned out that actual people didn’t behave like the imaginary proxy. Economists are learning that the poor and the wealthy respond differently to austerity and stimulus. This could present challenges to politicians. If people behave differently, then policy might have to treat them differently.
Attempts to divine—or calculate—how fiscal decisions affect the economy rest on the “marginal propensity to consume,” the likelihood that if you put a dollar in someone’s hand, he will spend rather than save it. Economic disagreements about stimulus are basically arguments over that question. The argument splits along the lines of an ancient tension in the profession—should an economist observe human behavior, or assume it?
The representative-agent models used by central banks and the IMF came out of a movement among economists in the 1970s and ’80s who assumed that people were rational and planned for the future. For example, Olivier Blanchard, now chief economist of the IMF, wrote in 1990 that when a government tightens its belt to reduce deficits, households might start spending, relieved that the problem is being handled and there won’t be an even bigger readjustment in the future.
If models could assume that everyone was the same rational person, there was no need for data on how people act. And conveniently, relying on a single person made the math behind the modeling easier.
Per Krusell, who now teaches at Stockholm University, was a brand-new Ph.D. in the early 1990s when he began work on a macroeconomic model that assumed more than one agent recognizing that people are different. He and Tony Smith, another new Ph.D., had a hunch that with enough computing power they could build a multiple-agent model. “If we had presented those ideas to our senior colleagues, they would have said, ‘No, go try something else,’ ” says Krusell. The resulting model won positions for Krusell at Princeton and Smith at Yale.
Still, their paper languished for more than a decade, basic research without a practical application. “We had a model that had something to say about propensities to save being different,” says Krusell. “Now the question [was], ‘How big are the differences?’ ”
There wasn’t much reason to find out. Representative-agent models had fairly accurate predictive power until the financial crisis and its aftermath. In January 2013, Blanchard and Daniel Leigh published a working paper for the IMF that was in essence a confession. When the fund forecast growth rates for various countries in 2010, it made mistakes, as did the European Commission. “Consumption may have depended more on current than on future income,” they wrote. The rational agent was supposed to plan for the future, confident that deficits were going down. Actual people did not. An IMF report in May looked back at the fund’s program in Greece, where real gross domestic product declined more than three times as much as predicted. The IMF’s mathematical assumptions about behavior had been wrong, the report conceded.
The challenges the world faced in 2010—low home values, credit hard to get, central banks unable or unlikely to lower interest rates further—“were nonexistent in representative-agent models,” says Christopher Carroll, who teaches at Johns Hopkins. At the ECB’s Frankfurt conference, Carroll presented a paper that bolstered Krusell and Smith’s model with microeconomic data. For Carroll, people differ in one crucial way. “The marginal propensity to consume,” according to the paper, “is substantially larger for low-wealth than for high-wealth households.” Rich people behave like the hyperrational agent. They plan for the future. They save during a stimulus, thinking about the taxes to come. And they can borrow during a fiscal contraction.
Poor people are what economists call “borrowing constrained.” They tend to have more needs than are being met, so when money arrives, they spend it. When the government stops spending and credit is hard to come by, the mythical everyman, like the rich person, continues to spend. But most real people don’t have access to credit, and they hunker down. Carroll’s findings have been confirmed by other academics in the last two years who looked at Italian and U.S. data.
Krusell reports a sudden interest in his 15-year-old model. He’s been asked to give a speech at a meeting next year of European central bankers looking for a better way to diagnose the continent’s problems. “You get a much bigger kick,” says Krusell, “when you realize it could be important for current policy.”