True and Non-Obvious

Paul Samuelson was asked by the mathematician Stanislaw Ulam to name a proposition in the social sciences that was both true and non-obvious. He took some years to answer, and the answer was comparative advantage.

Here is why it is non-obvious. Suppose one country is better at producing everything than another: better at cars, better at wheat, better at software. Intuition says the second country has nothing to offer and trade can only hurt it.

Ricardo's argument is that this is wrong, and the reason is that the better country still has to choose. Its workers and factories can make cars or wheat, not both, so producing wheat costs it the cars it could have made instead. If that opportunity cost is higher than the other country's, both gain by the second country producing wheat and trading for cars.

What matters is not absolute skill but relative cost. A surgeon who types faster than their secretary should still not do their own typing, because an hour spent typing costs an hour of surgery.

The theorem is real and the gains are real, and the accompanying result is inside the same theory and gets left out of the summary.

In 1941 Wolfgang Stolper and Paul Samuelson proved that opening to trade raises the return to a country's abundant factor and lowers the return to its scarce one. In a rich country abundant in capital and skill and scarce in low-skilled labour, trade with a labour-abundant country raises returns to capital and skill and lowers wages for low-skilled workers.

That is not a caveat appended by critics. It is a theorem in the standard model, proved by one of the model's architects, and it says that trade produces losers who can be identified in advance by what they own and what they do.

The standard reply is that the winners gain more than the losers lose, so the winners could compensate the losers and everyone would be ahead. This is true, and it is a statement about possibility.

Then came the evidence about what actually happens, and it is the most important development in trade economics in thirty years.

David Autor, David Dorn and Gordon Hanson examined American local labour markets by their exposure to Chinese import competition after China joined the WTO. Their finding (the China shock) is that exposed regions suffered large falls in manufacturing employment, and that the effects did not dissipate. Wages fell, labour force participation fell, disability claims rose, and the local economies did not generate replacement employment. More than a decade later, the gaps were still there.

The model assumes displaced resources reallocate. In these regions they largely did not. People did not move, because moving is expensive when your house is worth little precisely because the town is declining, and because families, schools and networks are not portable. New industries did not appear, because the agglomeration that made the place a manufacturing centre had been the reason to be there.

Two things need saying about this carefully, because the result is politically useful to people who want the wrong conclusion.

The first is that the aggregate gains were real. Consumers across the country paid less for goods, and the sum of those small savings is large. The China shock is a finding about distribution and adjustment, not a demonstration that the trade was net negative.

The second is that trade is not the main cause of manufacturing job losses in rich countries. Most careful estimates attribute the larger share to automation: output rose while employment fell, which is what productivity growth looks like. Trade's contribution is real, concentrated, and smaller than its political salience. Conflating them matters, because a tariff does nothing about a robot, and a region losing jobs to automation gets no help from trade policy.

So why did the compensation never arrive? The American Trade Adjustment Assistance programme is the case study. It spent a small fraction of the estimated gains from trade. It required workers to prove that trade specifically had caused their job loss, which is difficult and excludes anyone displaced by the automation happening at the same time. And it offered retraining: to workers in their fifties, in regions where no employer was hiring the newly acquired skills.

The political economy is the explanation, and Mancur Olson wrote it down in 1965. A household saving fifty pounds a year will not organise; a town losing its largest employer will. Concentrated interests organise and diffuse ones do not, whatever the relative totals, which is why protection is far easier to enact than to remove and why compensation schemes are legislated small and administered narrowly.

Which is also why the comparison that matters is not tariff versus nothing. Estimates for steel, sugar and tyre tariffs consistently put the cost per job saved in the hundreds of thousands, paid by consumers and by downstream industries that use the input. The right comparison is a tariff against the same money spent directly (on wage insurance, relocation, income replacement or investment in the affected place) which almost always reaches more people for less.