The St. Louis Fed last week pondered the question: “Is the U.S. Due for a Recession?” In a blog post the bank advised that after a long expansion “there is a concern that, even though the economy looks good right now, the next recession may be lurking just around the corner.” On the short list for possible smoking guns, the post continued, is the low unemployment rate, which is currently at 4.1%, which is near a two-decade low.
Looking for a recession when the unemployment rate is low appears to be a classic case of counterfactual thinking, but Hyman Minsky famously pointed out in his financial instability hypothesis that stability is destabilizing and in a capitalist system every expansion creates the seeds of its own destruction. By that logic, the low jobless rate may be a warning sign after all.
The St. Louis Fed entertains the notion, explaining:
In U.S. economic history, it seems the longer an economic expansion continues, the lower the unemployment rate becomes. Thus, if a long economic expansion increases the likelihood of a recession, as the idea of positive duration dependence suggests, then a low unemployment rate may indeed suggest the increased likelihood of recession.
To be fair, the bank recognizes that estimating recession risk is best pursued in a multi-factor framework. And for good reason: no one indicator is flawless in the art/science of looking for reliable and timely recession signals.
It may appear otherwise if you’re in a Minsky state of mind and looking at the jobless rate these days. After all, every recession in the postwar era has started with a low unemployment rate, and so one can argue, naively, that a low jobless rate alone equates with a high recession risk. But this is misleading because low unemployment usually signals continued growth. Looking to those few times when low unemployment marks the start of a new recession is only valuable with hindsight.
As an example, consider the jobless rate as it appeared in the summer of 2016. The rate had dipped under 5% for the first time since the previous recession ended. One could have argued that the decline, after a relatively long run, had finally hit bottom and a new downturn was near.
Of course, we now know that unemployment continued to fall. A surprise? Not really. In late-August 2016, I reported that a broad review of the economic data strongly suggested that recession risk was low and that a rebound after a soft patch of growth was kicking in. With the benefit of hindsight, that turned out to be a good call. Dumb luck? Hardly.
Monitoring and modeling a large set of indicators, for the financial markets and the real economy, is a tough act to beat for estimating recession risk in real time that comes closest to nirvana in this niche, namely: relatively high reliability and timeliness. History, after all, indicates no less. on analyzing the business cycle, a backtest of vintage data for a broad set of indicators looks encouraging over the last two NBER-defined recessions. Similarly, the three-month average of the Chicago Fed’s National Activity Index dispensed relatively early and accurate warnings of recession risk in the last two downturns.
The one constraint is that it’s virtually impossible to reliably predict recessions. The next best thing is modeling the data for a reliable real-time warning that tells us that a new downturn has started in the recent past. A one-to-two-month lag is about as good as it gets, assuming you want to keep false signals to a minimum.
As for Minsky, his financial instability hypothesis deserves respect as a big-picture review of the cyclical nature of capitalism, but it was never designed for use as a real-time framework for estimating recession risk. Even worse is trying to extract trustworthy business-cycle signals from lagging indicators, one at a time.
Fortunately, there are productive solutions for minimizing uncertainty about the biggest known risk factor. You won’t find them, however, by cherry-picking the data in a bid to find a short cut to genuine insight for estimating recession risk in real time.