Not all recessions are the same

At the St. Louis Fed:

May’s unemployment rate dropped to 3.8%, the lowest in 18 years. Does this tell us something about the next recession?

with a link to the Economic Synopses of 2018-06-01: Recession Signals: The Yield Curve vs. Unemployment Rate Troughs by Kevin L. Kliesen.

The article looks separately at the yield curve and unemployment rate troughs. Shows both to be good indicators of recession. Compares the indications and finds they don't agree. Conclusion:

Overall, both indicators tend to be reliable signals of a coming recession. But as with all recession signals, the wise economic analysts should examine many indicators rather than betting the farm on one or two.

Fair enough.


I looked at the yield curve recently myself. And I like to use the change in employment, not terribly far removed from the unemployment rate, as a recession indicator. So I find this stuff worth reading.

But I find a couple odd things. They show a graph: Treasury Yield Curve, 1954 to 2018. And sure enough:

As Figure 1 shows, yield curve inversions have regularly occurred prior to periods of economic recessions since the 1960s.

But there is also a horizontal line on the graph, representing the "average expansion value" for the 1954-2018 period. As if that overall value was somehow relevant to every recession since 1954. Or since the 1960s, or what.

Maybe it is relevant. Maybe it isn't. Not all recessions are the same.

They show another graph: Unemployment Rate, 1954 to 2018. This graph shows a horizontal line at the 4% level. This line is not an average, but is offered to show that

This is only the third economic expansion in the past eight (current included) that has registered an unemployment rate below 4 percent.

Yeh. Recessions differ. Recoveries differ. So I don't really think that an "average expansion value" for the whole 1954-2018 period tells us very much.


A footnote to their comparison of the recession indicators says:

Since June 1954, the average monthly change in the unemployment rate is –0.003 percent, with a standard deviation of 0.18 percent.

They are looking at long-period average values, not only for the yield curve, but also for the unemployment rate. I'm not convinced that these averages are relevant.

Useful? Yeah, they are useful, if you use them to show that they are not relevant. Or if you manage to convince me that I'm wrong about the lack of relevance. Kliesen does neither. If he does, I missed it.

But he dangled that stuff in front of my face: the -0.003 average monthly change, and the 0.18 standard deviation. I had to see if I could duplicate his numbers. I got the -0.003 monthly change all right. But I got a standard deviation of 0.19, not 0.18.

It's probably me. I probably did something wrong: selected the wrong data, or selected the wrong STDEV function. No matter: I already think the number is not relevant.

But there I was, with Excel on my screen, and all that unemployment rate data, thinking that the long-term averages are not relevant numbers. So how could I resist? I figured the average by decade instead. And what the heck, standard deviation by decade too:

Graph #1

Blue: average monthly change in the unemployment rate, by decade. The numbers are large in the 2000s and later; otherwise small. The 1970s and 2000s show positive values, meaning unemployment on average increased; unemployment on average decreased in the other decades.

The three largest decreases in unemployment occur in the 1960s, the 1990s, and the current decade.

Red  Brown: standard deviation of the monthly change numbers, by decade.

The three lowest values occur in the 1960s, the 1990s, and the current decade.

I don't know what that means. But it's a hell of a lot more interesting than a couple overall average values.