Go Menzie! The truth will out

A comment from Bob Flood at Econbrowser:

10 US recessions since 1950 last abt 3 quarters…so unconditional prob in any quarter is 30/276 = abt 10%. So prob no recession 90% . Pretending they are independent events, prob no recession thru Jan 2021 = .9^5 = 60%. But there all sorts of conditional things going on and instead of the censored regression first-pass regressions seen here, I’d like to see some quarterly GDP growth regressions with the residuals saved and then drawn out of sample to see – in maybe 1000 or so runs – the current-state conditional prob of recessions over various horizons. Nice undergrad macro topic.

I could follow the first part of that no problem: Ten recessions lasting on average maybe 3 quarters = 30 quarters of recession out of 276; about 10% are recession quarters, so about 90% are not. And then 90% raised to the fifth power for 5 consecutive quarters: 0.9*0.9*0.9*0.9*0.9 = I get 0.59, call it .60, 60 percent. Oh and 276 quarters = 69 years; 69 years since 1950 = 2019. Check.

That much I get, and I can check his numbers and they look good to me. And everything makes sense until I get to the part about regressions. Mental block. No matter: I read to see if one day somebody says something that makes it all make sense to me.

Menzie Chinn responded to Bob Flood:

So boostrapping to evaluate the growth elasticity bounds. I tried some plain OLS to get the point estimate; one problem is that growth is trending downward over the past 50 odd years, so how to deal with that in a good-fitting but not overparameterized way. But agree it’s an interesting topic to assign.

OLS is "ordinary least squares" which I think is the simplest form of regression but I never had it in school and I dunno. So when I read Menzie's remark, most of it fades into fuzziness in my mind and I'm left with this right here:

growth is trending downward over the past 50 odd years

That's what I get. Let me repeat it:

growth is trending downward over the past 50 odd years

People don't say that. You almost never hear it. What you hear is that since 1948 (or whenever) average growth is 3.3% (or whatever).

Nobody bothers to point out that on a graph the average is always a flat line; flat in this case at the 3.3% level and since 1948. Apparently we are supposed to think "remarkable consistency!" when we hear "3.3% since 1948". But it is mathematical tomfoolery!

Are we to believe that growth is NOT trending downward over the past fifty years? No one says this explicitly, of course. But it seems we are expected to focus on the reliable 3.3% number, and not worry our little heads about any downward trend. Now, though, Menzie Chinn has let that cat out of the bag.

Go Menzie!


PS, If the 50-year downward trend is serious enough to mess up economists' regressions, it is serious enough not to be ignored.

PPS, According to what Menzie says, you evidently don't even have to do a regression to know that growth is trending downward over the past 50 odd years.