Volcker, Greenspan, Bernanke, Yellen … and Hal?
Machine learning may soon help central bankers decide policy “The capability is here. The biggest hurdle is … cultural”
Artificial intelligence (AI), which is already steering cars, could help steer the world’s biggest economies in the next half-decade. Britain’s central bank has been developing computer algorithms for forecasting economic conditions and helping determine interest rate policy. Other monetary authorities are close behind. “The capability is here,” says Andrew Lo, director of MIT’s Laboratory for Financial Engineering. “The biggest hurdle is the cultural barrier. You’ve got a lot of central bankers who are not as open to technology.”
The Bank of England, under the direction of Chief Economist Andy Haldane, has quietly become a pacesetter in exploring the possibilities of AI. Paul Robinson, who heads the bank’s two-year-old Advanced Analytics unit, says the goal is to assist rather than replace humans. He says “many” central banks are at roughly the same stage of research and predicts that AI will make a meaningful
contribution to monetary policymaking “certainly within five years.”
Improvements would be welcome. Economists are, by their own admission, notoriously bad at making predictions. Consider the forecasts for 2015’s U.S. gross domestic product issued by Federal Reserve policymakers at the end of 2014. All 17 overestimated the eventual rate of growth, the closest by 0.2 percentage point, the furthest by 1.3 percentage points. The actual number was 1.9 percent.
Machine learning allows a computer to acquire a skill for which it hasn’t been explicitly programmed. Google’s self-driving car learns to drive by detecting patterns in vast amounts of driving data. Hedge funds, such as Two Sigma Investments and Renaissance Technologies, are already using AI to help make investment choices.
At central banks the principal task is to set an interest rate on short-term borrowing that guides the economy to a sweet spot between unemployment and inflation. Because rates work with a lag, doing so depends on forecasting economic conditions 6 to 12 months down the road.
One thing that makes monetary policy tricky is that a rate change will alter the conditions you’re trying to predict. And long-established connections among economic variables can change. For example, the inverse relationship between unemployment and inflation—what economists call the Phillips curve—seems to have disappeared in recent years. Robinson, of the Bank of England, concedes that AI works well when the structure of the economy is “invariant,” or stable, but is “less useful when it does undergo shifts.”
The U.S. Federal Reserve is moving gradually on AI. It uses computer models, in particular one called FRB/US (pronounced “ferbus”), to help with forecasts. FRB/US is a “selfcontained set of equations, data, programs, and documentation,” according to the Fed’s website. It’s useful for generating answers to specific what-if questions: What will happen to unemployment if 10-year Treasury yields rise by 2 percentage points?
Unlike a machine-learning system, FRB/US doesn’t learn on its own. “For the foreseeable future, the best approach will involve a combination of empirical rigor captured in models, together with human judgment,” says David Wilcox, director of the Fed’s division of research and statistics.
Some AI champions, such as Google Chief Economist Hal Varian, are also skeptical about AI’s ability to make economic forecasts, but for different reasons. As he sees it, the technology is ready, but the data—the copious supply of raw numbers that AI programs sift through to reach conclusions—are wanting. “The data sets are so small. GDP is released quarterly, so 50 years of data is only 200 observations and only seven recessions,” he wrote in an e-mail.
AI will soon have a lot more data to chew on. Web scrapers such as MIT’s Billion Prices Project are already combing the internet for real-time price points relevant to inflation.
Economists and computer scientists agree there will always be a role for human beings in central banking. As Michael Feroli, chief U.S. economist at JPMorgan Chase, puts it: “I don’t see why, in principle, you couldn’t have a computer set monetary policy. Having it testify before Congress is another matter.”
The bottom line The machine-learning branch of artificial intelligence could help central bankers set interest rates within five years.