Understanding the Lesser Depression
*J. Bradford DeLong, U.C. Berkeley, September 2011
(2) The Great Moderation
2.2 The Great Moderation:
The Employment-to-Population Ratio
Here in Figure 1 is a time-series chart of the single key quantity you need to study to understand business cycles: the civilian adult employment-to-population ratio.
Every month the Bureau of Labor Statistics (BLS) of the U.S. Department of Labor sends its interviewers around the country to conduct the Current Population Survey (CPS). They ask a random sample of civilian American adults questions. ￼One of the questions is: “last week, did you do any work for pay or profit?”1 The proportion of American adults who answer “yes” to our first CPS question—“did you do any work for pay or profit?”—is the civilian adult employment-to-population ratio: the fraction of American adults who say that they have jobs.
The first thing that catches my eye in the chart is the high-frequency annual wiggle pattern. Almost every single year the share of American adults with jobs reaches a peak in the early summer, declines in the late-summer vacation season, climbs to a lesser peak in the fall before Christmas, and then collapses in the winter to rise again in the spring and the early summer. More people want to work in the (early) summer and fall. Some businesses—construction in the northeast and midwest comes to mind—cannot effectively function in winter. Other businesses see sharp increases in the demands for what they make and sell as the Christmas rush approaches.
Once you seasonally smooth the data to eliminate the spring and fall bulges and the winter slump in unemployment, the course of the employment-to-population ratio over the past fifty years shows:
- The surge in employment from the coming of feminism (and other relatively slow-moving changes in Americans' desires and opportunities to work.
- The post-WWII business cycle.
- ￼The largest such post-WWII business cycle: the Lesser Depression in which we are currently enmeshed.
We would like a view of the data that filters out (1) (or as much as possible of (1)) and that let's us look at (2) or (3) by themselves. And we would like to construct such a filter in a relatively automatic way, rather than asserting what the long run trend is and so baking our conclusions into our assumptions. We want to allow the data to speak. We do not want to torture the data until we confess.
￼A natural first way to cut the data (to me at least) is to detrend the seasonally-adjusted adult employment-to-population ratio data by a ten-year trailing moving average. Why trailing? So we can look at the present through our filter. Why ten? Because we have ten fingers? Why a moving average? Because we have seen enough moving averages over our lifetimes that we have a sense of how they behave--a sense we do not have for other trendier, fancier statistical detrending filters.
Filtering the employment-to-population ratio through this ten-year trailing moving average filter produces Figure 4. And let us process the data one more step. Johann Carl Friedrich Gauss said that when you are analyzing variability, you should measure it by squaring deviations from the average and then averaging those squared deviations. That produces Figure 5.
￼Figure 5 tells us that when you measure trends with a trailing ten-year moving average and look at the employment-to-population ratio, four features over the past third of a century jump out at you:
- The steep economic downturn of the Volcker disinflation around 1980.
- The rapid rise in the employment-to-population ratio as American feminism took full hold in the late 1980s.
- The "Great Moderation" that ended just a few short years ago (and an analogous period of extremely low variability in the 1970s, 1960s, and if we look further back in the 1950s as well).
- Our current "Lesser Depression".
And note that we have not quite succeeded in our task of filtering out movements of type (1). In Figure 5 the late 1980s look like a volatile time. It is certainly true that the employment-to-population ratio in the late 1980s was indeed far away from its ten-year moving average. But this was not because of any depression or recession. And this was not because employment was in any sense artificially elevated because of a high-pressure inflationary economy. Instead, the employment-to-population ratio in the late 1980s was far above its ten-year trailing moving average because (a) the moving average was still depressed below normal levels by the downturn of the early 1980s, and (b) the late 1980s were the high-water mark of the feminist revolution as female labor force participation increased at its fastest pace.
This shows that our automatic statistical procedure has not succeeded in completely purging our data series of movements of type (1): long-run changes in Americans desires and opportunities to work. But we would not expect any automatic, mechanical procedure to do a perfect job of filtering out such phenomena, would we?