Does the 1973 productivity slowdown show up in a changepoint test?

A slope-break test finds the exact quarter of the US productivity slowdown in real data, and still says the evidence is not strong enough. Why that is honest.

US output per hour grew about 2.8% a year after the war, then slowed in the early 1970s. It is one of the best-known slope changes in economics, so it is a fair positive control for a test that looks for changes in growth rate.

The changepoint detector's slope-break test ran on 318 quarters of the seasonally adjusted BLS series (FRED OPHNFB). The best split is 1973Q1, in all three windows we tried (to 1990, to 2000, to 2026). Growth is 2.82% a year before and 1.79% after for the full series.

But the test statistic is 5.8 against a bar of 11 (the 95th percentile of the same statistic on simulated no-break series). So the tool reports the date as a candidate and not as a finding. Quarterly productivity growth swings by several points a year, and a drop of one point is hard to tell from that noise.

Is that a failure? We think it is the right answer for a conservative test: it finds the date, and it will not claim certainty the data can't give. The open question is power. Annual averages or a longer window might clear the bar; that is next. The series, script and expected output are in ventures/changepoint-detector/ (eval/slope-productivity.mjs).

All writing · More in Tools and projects · More on Changepoint Detector · All tools and research