Predicting Job Loss?

Hardly a day goes by without a new prediction of job growth or destruction from AI and other new technologies. Predicting job growth is a growing industry. But how good are these predictions? For 80 years the US Bureau of Labor Statistics has forecasted job growth by occupation in its Occupational Outlook series. The forecasts were generally quite sophisticated albeit often not quantitative.

In 1974, for example, the BLS said one downward force for truck drivers was that “[T]he trend to large shopping centers rather than many small stores will reduce the number of deliveries required.” In 1963, however, they weren’t quite so accurate about about pilots writing “Over the longer run, the rate of airline employment growth is likely to slow down because the introduction of a supersonic transport plane will enable the airlines to fly more traffic without corresponding expansion in the number of airline planes and workers…”. Sad!

In a new paper, Maxim Massenkoff collects all this data and makes it quantifiable with LLM assistance. What he finds is that the Occupational Outlook performed reasonably well, occupations that were forecast to grow strongly did grow significantly more than those forecast to grow slowly or decline. But was there alpha? A little but not much.

…these predictions were not that much better than a naive forecast based only on growth over the previous decade. One implication is that, in general, jobs go away slowly: over decades rather than years. Historically, job seekers have been able to get a good sense of the future growth of a job by looking at what’s been growing in the past.

If past predictions were only marginally better than simple extrapolations it’s hard to believe that future predictions will perform much better. At least, that is my prediction.