The Compensation Nobody Pays

There is a criterion used in almost every cost-benefit analysis performed by any government, and most people have never heard of it.

Kaldor-Hicks efficiency asks whether the gains from a change exceed the losses, whether the winners could, in principle, compensate the losers and still be better off. If yes, the change passes.

Note what it does not ask: whether any compensation happens. The criterion is satisfied by possibility alone, and once you know this you cannot read a cost-benefit analysis the same way again, because "this policy has net benefits of £2 billion" and "this policy makes people better off" are different claims and the first is routinely reported as the second.

The distinction matters most for trade, where the theory is explicit that there will be losers and the compensation is the entire reason the result feels acceptable.

Take the political economy first, because it explains why compensation systematically fails.

Mancur Olson's Logic of Collective Action, from 1965, makes the argument. A group's ability to organise depends not on how much is at stake in total but on how concentrated the stake is per member. A tariff on sugar costs each household a few pounds a year and delivers millions to a handful of producers. The producers will fund a lobbying operation. No household will spend an afternoon on it, and any household that did would be providing a benefit to everyone else for free.

So concentrated interests organise and diffuse ones do not, regardless of which is larger. Protection is easier to enact than to remove; subsidies outlive their justification; and compensation schemes for the losers of a change are legislated small, because the people who would receive them are dispersed and the people paying are not.

Now the assumption the standard case rests on, which the China shock literature tested and found wanting.

The gains from trade come from resources moving to more productive uses. That requires the resources to move. If a displaced worker finds new employment at a comparable wage, their output has been reallocated and the theory works. If they spend a decade out of the labour force, the output is not reallocated; it is lost, and the arithmetic changes.

Autor, Dorn and Hanson found that in the American regions most exposed to Chinese import competition, people substantially did not move and did not change industry. Employment and participation fell and stayed down for more than a decade. Disability claims rose. Marriage rates fell and mortality from drugs and alcohol rose in the same places, though attributing those to trade specifically is contested.

Mobility is expensive in exactly the way the model does not price. Selling a house in a declining town means selling into the decline. Schools, family, and knowing where things are do not transfer. And the cruelty of the calculation is that the people least able to afford the move are the ones for whom moving would help most.

So what works?

The evidence is reasonably clear that trade-specific programmes are the wrong design. Requiring proof that trade rather than automation caused a job loss is an administrative barrier that excludes most affected people, arrives late, and makes no economic sense. The displaced worker's situation is identical either way, and the distinction exists only to limit the programme's cost.

What has better support is support that does not require diagnosing the cause: wage insurance that tops up earnings when someone takes a lower-paid job, unemployment insurance with adequate duration, healthcare not tied to employment, and place-based investment that treats the region rather than advising people to leave it.

Place-based policy was unfashionable in economics for a long time, on the grounds that it is more efficient to help people than places. The China shock evidence has substantially changed that, because the finding is precisely that people do not move, so helping people without helping places means not helping people.

Which gives a set of questions worth asking whenever anybody reports that a policy has positive net benefits.

Who bears the losses, and how concentrated are they? Over what period: a decade of adjustment is not a transition, it is a generation. Is compensation actually funded and legislated, or is it in the discussion section of a report? And is the counterfactual the right one: not this policy versus nothing, but this policy versus the same resources deployed differently.

None of these questions undermines the finding of net benefit. They are the questions that determine whether the net benefit is a reason to do the thing.