Invest1 publisher2 min readPublished
Former Fed economist dates the fall in AI-exposed job postings to March 2022, eight months before ChatGPT
Job postings in AI-exposed occupations peaked as the Fed began hiking in March 2022, says a former Fed economist citing a study of 238 million postings. The same column then credits AI with keeping the slump going in jobs it can automate.
The Investor · Invest desk
What happened
- Postings for the most AI-exposed occupations peaked and began falling in March 2022, when the FOMC started raising rates, eight months before ChatGPT existed.
- Zanna Iscenko and Fabien Curto Millet analyzed 238 million job postings and tied the decline to the tightening cycle, noting that exposed jobs cluster in rate-sensitive sectors.
- Stanford's Digital Economy Lab reports that workers aged 22 to 25 in the most AI-exposed occupations are running roughly 19 percent behind peers in less-exposed fields.
- The Stanford researchers counter that the most exposed jobs are not generally the most rate-sensitive, and that the gap keeps widening as rates come down.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- contradiction If tightening drove the decline, the gap for young exposed workers should narrow as rates fall. Stanford reports it still widening, so the rate story can explain when the slump started but not why it has lasted.
- constraint A drop in postings in information, finance or professional services cannot count as evidence that AI is replacing workers until the rate cycle those sectors are exposed to has been stripped out.
- cost Firms that trim junior intake book a payroll saving now. On the columnist's account, they also shrink the pool of staff they would later promote into senior roles.
The strongest evidence for the rate story is the control group. The Economic Policy Institute found that young workers without college degrees, in occupations that score negative on AI exposure, saw their unemployment rise at a similar pace over the same period [4]. On that measure, AI has little to do with the work those people do [4]. If they lost ground as fast as the exposed group, the cause goes beyond AI. The Fed is the obvious candidate: postings in exposed occupations peaked the month it began tightening, and ChatGPT arrived around November 2022 [1][2].
The evidence allows three readings. Rates did it, and hiring in exposed occupations should recover against unexposed ones as policy eases. AI did it, and the March 2022 peak is coincidence. Or rates started the decline and AI is keeping it from reversing in the work AI can automate. The first reading struggles to stand alone against Stanford's widening gap [7], and the columnist, who spent a decade at the Fed [13], calls that pushback appropriate [7].
The reply the column offers Stanford is an AI explanation. "Firms are not firing their junior employees; they're hiring fewer of them," the columnist wrote, adding that the decline "concentrates where AI is automating work rather than where it complements it" [9]. So a piece that dates the slump to the Fed arrives at the third reading. I think the evidence it gathers supports that reading and nothing stronger. The Stanford authors are just as careful about their own side, calling their results "descriptive patterns, not causal estimates" and writing that they "do not see widespread, economy-wide job displacement associated with AI" [6].
The third reading would break if postings in automating occupations recovered as fast as postings in complementing ones once rates settle. The column does not say how far postings fell in either group, so the two effects cannot yet be sized against each other. The columnist concedes the wider limit, writing that "it is impossible to separate the impact of overlapping shocks in real time" [8].
Firms are making expensive decisions on hiring, education and regulation regardless [10], and the one the column documents is a cut to intake with no cut to headcount [9]. The columnist argues that firms hired juniors to turn them into senior professionals, with their output a byproduct [14], and calls the junior hire "a capital investment mislabeled" [11]. A firm that stops that spending books a payroll saving this year. On the column's account the cost comes later, because employers "still want experienced people, but many have simply stopped funding the process that produces experience" [12].
What to watch
- Whether hiring in AI-exposed occupations recovers against unexposed ones as rates come down; a relative recovery would favor the rate explanation.
- The next update from Stanford's Digital Economy Lab on whether the roughly 19 percent gap for workers aged 22 to 25 keeps widening.
- Posting data split between occupations where AI automates work and where it complements it, which is the test of the columnist's answer to Stanford.