Scott Sumner warns that predicting the next recession is a mug’s game. The dismal record on forecasting certainly offers support. “Whenever I hear that someone has accurately predicted a recession, my evaluation of that person declines,” writes the economist at the Mercatus Center at George Mason University. “Economists should not be trying to predict recessions; the point is to prevent them.” Success with the latter may be even more challenging than the former, but that’s an issue for another day. Meantime, Sumner notes that “recession predictions have a corrosive effect on economics as a science.” Why? Because “these predictions lead non-economists to believe that economists should predict recessions, and give undeserved reputational points to lucky permabears.”
For some context on Sumner’s reasoning, he offers the following analogy:
If I’m standing next to a statistician at the roulette wheel, and he predicts the ball will not land on the green numbers (36 out of 38 odds) for each of 220 spins, then I will assume he’s a good statistician. If he occasionally predicts the ball landing on the green (2 out of 38 odds), I will think less of his statistical skills, even if it does land on the green.
The point, of course, is that the economy has a bias for growth while recessions are the exception. The quality and degree of growth is another matter. But in a binary framework it’s clear that your powers of prognostication will shine if you simply predict growth from here on out and ignore the rare interruptions known as recessions.
Consider that US economy has been in recession roughly 30% of the time since 1857, according to NBER’s business-cycle records. A “dumb” prediction that the economy will expand, in other words, turns out to be a “smart” forecast most of the time.
Despite the low probability that the economy will contract as a general proposition, there are huge incentives for anticipating recessions. But as Sumner correctly points out, predicting downturns is difficult, perhaps impossible. It may appear otherwise at times, but it’s not obvious that random predictions are any worse (or better) than formal forecasts.
But while it’s rarely productive to divine the future for the business cycle, looking backward is another story. The case for nowcasting the macro trend, based on the numbers available in real time, has a pretty good record–assuming the analytics are prudent. The main caveat is that a high-confidence signal that a recession has started is only available after the tipping point has arrived. The opportunity to add value with the associated analytics is all about reducing the lag time between the start of a recession and the point in real time when it becomes obvious—and compelling—that the economy is contracting.
The gold standard on this front is the Chicago Fed National Activity Index—the three-month moving average (CFNAI-MA3) in particular. The index’s vintage data shows that CFNAI-MA3 issued a recession warning on Mar. 24, 2008—roughly three months after the peak in economic activity.
It’s true that there was no shortage of analysts predicting a recession prior to Mar. 24, 2008. But there were many advising that the economy would continue to grow. The problem is that most of the recession and growth predictions were more or less ad hoc and so separating the wheat from the chaff was a hopeless game of trying to pick winners and losers. A credible methodology for analyzing the business cycle requires a clear set of rules. That alone invalidates most of the recession forecasting efforts.
Looking into the future isn’t totally worthless, but it’s best to stick with a systematic effort and limit the time horizon to a few months. On that score allow me to make a shameless plug for the monthly updates at CapitalSpectator.com that aggregate a broad set of economic and financial indicators to anticipate the near-term outlook for the macro trend–here’s last month’s report. Note that the mandate for this research is limited to a binary framework—growth or recession? The record over the last several years has been encouraging, but readers are invited to judge for themselves: you can review the analytics here, clicking through previous monthly updates that are listed at the bottom of each report.
But let’s be clear: the degree of confidence for nowcasting is substantially higher relative to forecasting. That’s no surprise, even if the usual suspects would have us believe otherwise.