Read on Twitter
Pleased to share a small extension of our work-in-progress on job training mismatch, which seems relevant as some industries gradually reopen and workers reshuffle. In this piece we think about training programs & occupational demand over the bus cycle.
https://libertystreeteconomics.newyorkfed.org/2020/05/job-training-mismatch-and-the-covid-19-recovery-a-cautionary-note-from-the-great-recession.html
https://libertystreeteconomics.newyorkfed.org/2020/05/job-training-mismatch-and-the-covid-19-recovery-a-cautionary-note-from-the-great-recession.html
There is debate re: the roles of agg. demand vs. skill gaps in the persistence of displaced worker earning losses ( @rothstein_jesse;
@Brendan_Duke; @JHWeissmann). COVID skill gap or not, we ask a different but related Q: how well do training programs target demand in recessions?
@Brendan_Duke; @JHWeissmann). COVID skill gap or not, we ask a different but related Q: how well do training programs target demand in recessions?
2 Findings:
- Displaced workers retraining through the Trade Act benefited if enrolled in normal times, but *not* the GR.
- Mismatch between the composition of occupations trained for, and new vacancies in those occs., grew during the GR.
Did workers train for the 'wrong' jobs?
- Displaced workers retraining through the Trade Act benefited if enrolled in normal times, but *not* the GR.
- Mismatch between the composition of occupations trained for, and new vacancies in those occs., grew during the GR.
Did workers train for the 'wrong' jobs?
Comparing recipients of both retraining vouchers and extended UI, to a control group that waived out of the training requirement for receiving UI, we use taker data to study SR earnings, and find positive wage replacement returns during normal times, but not the Great Recession.
Caveat: this data tends to understate recovery levels, but the qualitative bus. cycle result is robust to validation w/ microdata. Despite selection concerns, this picture nests both my JMP's positive training result (L) and an influential Mathematica study's negative result (R).
We then follow Sahin et al. (2014) and construct a reduced form training mismatch index capturing whether targeted occupations under- or over-trained for a given labor market’s needs, using online vacancies. A value of 0 is perfectly matched. The index ~doubles from 2006 to 2010.
This training mismatch was concentrated in a handful of states like Ohio, and driven by a few occupations like metal & plastic production workers. On average, all states *over*-trained for occupations that were no longer demanded, rather than under-training for new/emerging work.
Assuming mobility is limited w/in states, labor mrkts that never fully recalled its workers, may have also had a hard time predicting the permanence of new skills required. Training may thus be a less effective insurance vehicle during COVID should structural change be permanent.
This ongoing work (joint with Karen Ni, Fed RA and future HKS PhD student) is part of a larger project on training mismatch (beyond COVID). All input is very much welcomed.
