Kyle Fitzgibbons

​As you can see the employment ratio was about the same in 2007 as in 1990, and hence the aggregate data shows no evidence that either trade or automation reduced employment during the period studied by Autor, Card and Hanson, as well as Acemoglu and Restrepo.

Of course that doesn't mean these factors have not had a negative effect on overall employment, just that doing so would require a very sophisticated study. Unfortunately, the science of economics has not yet advanced to the point where that sort of study is feasible. And thus we are forced to admit that we simply don't know if there is any effect on overall employment.

​But I do think that we know that trade and automation raise real GDP.

How much of a role did trade play in the long-term decline in the manufacturing share of total employment, which fell from around a quarter of the work force in 1970 to 9 percent in 2015? The answer is, something, but not much.

Or to put it another way, absent the trade deficit manufacturing would be maybe a fifth bigger than it is – which is actually what EPI estimates too (Exhibit D in the linked paper). That wouldn’t make much difference to the long-run downward trend, but looms larger relative to the absolute decline since 2000.

Autor et al only estimate the effects of the China shock, which they suggest led to the loss of 985,000 manufacturing jobs between 1999 and 2011. That’s less than a fifth of the absolute loss of manufacturing jobs over that period, and a quite small share of the long-term manufacturing decline.

​I’m not saying that the effects were trivial: Autor and co-Autors show that the adverse effects on regional economies were large and long-lasting. But there’s no contradiction between that result and the general assertion that America’s shift away from manufacturing doesn’t have much to do with trade, and even less to do with trade policy.