James D. Hamilton explores whether we’re in a recession by examining a variety of indicators. Courtesy of Econbrowser:.
Recession indicators
Many people may not care whether our current situation meets the formal definition of a recession, but as I’ve explained previously, you should. Here’s a summary of how I see the economy at the moment. I begin by discussing a new paper by UCLA Professor Ed Leamer, which has also been highlighted by Greg Mankiw, Frank Stephenson, Calculated Risk, and Brian Blackstone.
Leamer looks at four key series, asking what threshold each has to cross in order to be included in episodes that the National Bureau of Economic Research characterized as economic recessions. Leamer observes that it’s typically called a recession if the 6-month growth rate of nonfarm payroll employment falls below -0.5%.
 |
Leamer also says that in a recession, the 6-month change in the unemployment rate rises above +0.8%,
 |
the 6-month growth in industrial production falls below -3%,
 |
and we see a 6-month change below -0.4% in civilian employment as measured by the BLS household survey:
 |
Of these measures, the increase in the unemployment rate has crossed Leamer’s historical threshold, and with the latest data the drop in civilian employment is getting pretty close. But industrial production and NFP do not yet meet Leamer’s criteria, leading him to conclude that we are not yet in a recession. Things would have to get much worse, he writes, before he would designate the current slowdown as a recession.
The growth rate of nonfarm payrolls is also an indicator stressed by Calculated Risk, though CR favors the year-over-year change rather than Leamer’s 6-month change. Using a threshold of no change in year-over-year employment seems to satisfy Leamer’s sort of selection indicator, and on that basis the July NFP numbers would lead you to say we’re in a recession:
 |
One challenge with using any of these indicators is that the series can be revised over time. The patterns in the series as they’ve now been revised may be clear, but how good is the inference we can draw with the data actually in hand? Looking at real-time year-over-year NFP growth rates seems to be pretty robust by this criterion:
 |
And I must say that when you look at a graph of the level of the unemployment rate (rather than the changes plotted above), the recurrent mountains seem like a pretty prominent feature that could almost be taken as the defining feature of a recession.
 |
I argued in a paper published in 2005 that the recessionary pattern one’s eye detects in the unemployment rate is due to a nonlinearity in the time series, with the unemployment rate increasing more quickly than it decreases. In that paper, I modeled the unemployment rate as depending on an intercept and its previous two lagged values, with the intercept changing between one of three values– a low level for normal times, and two higher values for mild and severe recessions. A shift into recession causes the unemployment rate to rise quickly, while it drops more slowly in recoveries. The analyst doesn’t have data on whether the economy is in expansion, mild recession, or severe recession at any point in time, but we can draw an inference based on what we see happening to the unemployment rate. I was curious to see what that model says when you update it with the most recent unemployment numbers.
 |
Historically this seems to give you a pretty fair recession inference using the unemployment numbers alone, though it would have lengthened both the 1960 and the 1990 recessions into double-dipped downturns. The interesting thing to me is that the model also assesses the probability that another recession has now started as being 69%.
Once again, however, recognizing these patterns with all the historical data now on hand is an easier task than trying to call the events in real time. For example, here’s what the above graph looks like if you tried to base the inference on data available at the time, without knowing what the future is going to be:
 |
It’s obviously a lot harder call on this basis, though the signal from the latest unemployment jump is strong enough that you might still feel justified in making it.
A key challenge in trying to make these determinations is the continuing changes in demographics and the structure of the economy. For example, with a decreasing manufacturing share, the downturns in industrial production may be becoming less pronounced. One can always find some rule that seems to fit the historical data pretty well, but will it continue to work in the future? The more variables you’re free to choose from and the more variety in the rules you consider, the better you may seem to fit the available data, but the more problems you’re likely to encounter applying that to the new situations that are always developing. I do think there’s a pretty strong case, based on the employment and unemployment numbers, that we are currently in a recession. Nevertheless, we will be making the Official Econbrowser Declaration, for what it’s worth, using the GDP-based recession indicator index, which is deliberately very simple and limited, and hopefully therefore the most robust.
And with 1.9% second-quarter growth as currently reported, GDP growth doesn’t meet the criteria for a recession so far.
But the whole reason I’m interested in this question of whether our current difficulties should be classified as a recession, is that if we are in a true recession, the process is going to feed on itself, and more bad things are ahead of us. If it’s a real recession, it should be evident in the 2008:H2 GDP numbers.
And if you want an Official Econbrowser Declaration, you’re going to have to wait for those.
