It was February 2008, and the economic forecasters surveyed each quarter by the Federal Reserve Bank of Philadelphia were starting to worry. “The outlook for growth in the first half of 2008 looks much weaker now than it did three months ago,” the Fed wrote, summarizing the survey results. “However, the forecasters are not predicting a contraction.” Year-over-year GDP growth would be a modest 1.8% in 2008, they predicted, rising to 2.8% in 2009. Unemployment would average 5.1%.
The forecasters weren’t nearly worried enough. In reality, unemployment would hit 10% in October 2009, and GDP went negative, contracting -0.13% in 2008 and -2.5% in 2009.
Economists are not good at predicting recessions, although to be fair neither is virtually anyone else. But some forecasters are more accurate than others, and so Quartz decided to ask a group with a track record of accurately predicting economic and geopolitical events about the chances of a recession in 2020.
The Good Judgment Project was a research program created by Philip Tetlock and Barbara Mellers, professors of psychology at the University of Pennsylvania, to study forecasting and human judgment. Initially, they asked a team of volunteers to participate in a forecasting tournament sponsored by the US government, whose aim was to improve the accuracy of intelligence agencies’ forecasts. Good Judgment aggregated the forecasts of its participants and harnessed the wisdom of the crowd to win the tournament in its first year.
The researchers noticed that some of their participants were consistently outperforming their peers. (The forecasters were, at that point, amateurs and hailed from a variety of professions and backgrounds. What they had in common wasn’t an area of expertise but a particular style of thinking.) In the second year of the tournament, Good Judgment assigned those “superforecasters” to work together on teams and found that when they collaborated they performed even better. After Good Judgment won the tournament for the second year in a row, the US government shut it down and hired Good Judgment Inc., the project’s for-profit spinoff, to make predictions.
Quartz asked Good Judgment Inc. to survey a few dozen of its forecasters about the chances of a recession in 2020. (Technically, they were asked “What is the probability that the US economy posts two consecutive quarters of decline before 31 March 2021?” Because a recession is defined as two quarters of negative growth, a recession that begins in Q4 of 2020 wouldn’t be classified as a recession until the end of Q1 2021.)
On average, the Good Judgment forecasters said there was an 18% chance of a recession this year; the median forecast was 15%.
The forecasters at Good Judgment Inc. use a version of the Delphi method for rapid forecasts like the one they did for Quartz. With the Delphi method, every participant makes a forecast on their own and describes their reasoning. Then everyone sees everyone else’s comments and forecasts—anonymously, so as to judge them by their contents, not the forecaster’s reputation. Then each forecaster has a chance to update their original forecast. Those final, updated forecasts are shown in the chart above.
The Good Judgment forecasters put the chances of a recession slightly lower than does Moody’s Analytics, an economics research firm.
“Right now our various recession probability models put the risk of a recession in the next 12 months anywhere from 20-35%,” said Ryan Sweet, director of real-time economics at Moody’s Analytics. He said the biggest recession risk was an escalation of the trade war and US-China conflict more broadly.
A poll by the National Association for Business Economists found three out of four economists predicted a recession by the end of 2021. And Quartz’s Allison Schrager shared her own view of recession risks in December.
The Good Judgment forecasters left comments for each other describing their reasons for and against recession, and in the second round of the forecast they voted on the most useful comments. Below are the most upvoted comments on both sides, provided anonymously to Quartz:
“In the seven times since 1969 that the three month and 10-year yield curve on US treasuries has inverted for more than 10 days, a recession has followed within 18 months. The last inversion started in June 2019 and if this pattern holds, a recession should start before January 2021.”
“Large amounts of debt will slow consumer and business spending. We are already seeing this in autos with homes close behind. A decline would have happened sooner if not for pent up demand for housing due to an inadequate labor supply in construction. Large amounts of government debt also an issue”
“A variety of factors are building including the flip of the two year vs. 10 year bond rates, a stalling in the manufacturing sector, and a long run of economic growth that has to end at some point. China’s debt crisis is a very likely potential trigger. Once it hits, it’s likely to be big.”
“US interest rates are low, and consumer confidence is positive and steady. Global political events are not at the level of severity to cause much concern. Administration will work economic levers to help re-election.”
“Election year. Administration will want to keep the markets strong (economy might not cooperate). Economic factors will eventually overtake stock prices but not likely over the next 15 months.”
“Central banks printing money is keeping money cheap, keeping the surface tension of so many bubbles relatively strong. And while the US fiscal and monetary situations are problematic, the US (dollar) won’t cease to be an acceptable choice on the menu in the next 15 months.”
Recessions are typically caused by the combination of some sort of imbalance between supply and demand and a shock that makes that imbalance painfully apparent, says Sweet. The problem for forecasters is that, even if you can spot the imbalance, it’s hard to predict the type and timing of the shock. It was easier to point out that dot-com valuations were inflated in the late 1990s than to predict when investors would collectively wake up to that fact. Today, it’s easy to worry about high levels of corporate debt, but harder to tell whether it will become a crisis. Some imbalances resolve themselves, says Sweet, and these “garden variety corrections” don’t necessarily cause recessions.
Spikes in oil prices are a common cause of recessions, for example, but they’re exceedingly difficult to predict. That’s partly because markets already incorporate a lot of information about the future into the price of oil, and partly because the events that drive spikes in oil prices are rare and driven by random or idiosyncratic factors.
Likewise recessions can often be caused or exacerbated by bad monetary policy, but, at least in countries like the US, the central bank’s decision-making “generally reflects roughly the consensus of the economics profession,” writes economist Scott Sumner in a blog post. That means, almost by definition, that economists as a group will struggle to predict this sort of recession.
By the middle of 2008, the forecasters surveyed by the Philadelphia Fed were well aware housing prices were dropping and that economic conditions were worsening. Their forecasts had grown more pessimistic for six consecutive quarters, predicting ever-weaker growth and higher unemployment. But they did not foresee the shock that would send the global economy into a tailspin. In August, they predicted GDP growth of 0.7% in Q4, rising to 2.5% by Q3 of the following year. One month later, Lehman Brothers collapsed.