Cleansing effect of recessions

How can an industry grow more productive when no firm inside it improved?

An industry's measured productivity — output per hour worked — rises by four percent over two years. The natural reading is that firms got better at what they do: new machines, better methods, sharper management.

But suppose you audit every firm and find that not one of them improved. Each is producing exactly what it produced before, with the same efficiency. The aggregate still rose. That is not a paradox to be explained away; it is a clue that the aggregate number is measuring something other than what we assumed.

Start with what the industry figure actually is. It is not the average of firm productivities. It is total output divided by total hours — which makes it a weighted average, where each firm's weight is its share of the industry's labour. That word "weighted" is the whole answer, and it is worth being slow about it.

A weighted average can move in two entirely different ways. The values can change — firms get better. Or the weights can change — the same firms, with unchanged values, come to occupy different shares of the total. Only the first is improvement in the everyday sense. The second is selection: a change in composition, with no change in any member.

Test that with the smallest example that works. Two firms, each with a hundred workers. One produces 100 units per worker, the other 50. The industry average is 75. Now the weak firm fails and closes; its workers are hired by the strong one, which runs at its usual efficiency. Nothing inside either firm changed. The industry now produces 100 per worker. The four-percent puzzle is just this arithmetic at realistic scale.

So the next question is why a recession would move the weights in that direction rather than randomly. Here is the argument, made most sharply by Ricardo Caballero and Mohamad Hammour in 1994. Demand falls, so prices and margins compress. A firm survives only if it can cover its costs at the new, thinner margin. High-productivity firms have more slack to absorb that; low-productivity firms sit closest to the line and cross it first. Downturns are therefore a period when the exit rate rises and is unusually correlated with productivity. Between recessions, weak firms can persist for years on ordinary demand — the pressure that would sort them simply is not applied.

Two further channels reinforce it, and they are worth separating because they are different mechanisms. Reallocation: when a firm closes, its workers and its capital do not vanish, they move — and if they land in more productive firms, the weights shift again. And selective entry: when survival is hard, marginal new firms do not start, so entrants during a downturn are on average the better-prepared ones. Selection operates on exit, on entry, and on where the freed resources land.

Now I want to be careful, because the term "cleansing" carries a moral warmth the evidence does not fully support. Two objections have real force.

The first is the sullying argument, made by Gadi Barlevy in 2002. In a downturn, credit is scarce, and a firm's ability to raise money is not the same thing as its productivity. A well-run young firm with no collateral can fail while a poorly run old firm with assets to pledge survives. To that extent the recession is sorting on financing access, not on efficiency — selection on the wrong trait.

The second is empirical and specific. Lucia Foster, Cheryl Grim and John Haltiwanger examined US firm-level data and found the Great Recession was markedly less cleansing than earlier downturns: the usual link between low productivity and exit was weaker, and the shock's concentration in construction and housing meant many exits reflected where a firm happened to be rather than how well it ran. So the cleansing effect is a real mechanism whose strength depends on the recession, not a law that every downturn obeys.

And the arithmetic says nothing about the cost. The workers in the closed firm bear long-lived earnings losses, whatever the aggregate does. A rise in the industry figure and a fall in a thousand households' incomes are perfectly compatible, because they are not measuring the same thing.

Three refinements.

First, this does not mean recessions are good. It means one specific measured quantity can improve during them for reasons unrelated to improvement, and that using that quantity to score the downturn confuses selection with progress.

Second, it does not mean the exiting firms deserved to exit. Selection acts on survival, and survival correlates with productivity only as strongly as the market's pressures happen to correlate with it — which is exactly what the credit-constraint objection disputes. Whenever you see a selection story, the sharp question is: selection on which trait?

Third, the mechanism is not confined to recessions. Trade liberalisation, deregulation, and technological shifts all raise measured aggregate productivity substantially through reallocation between firms rather than improvement within them. Recessions are simply the case where the composition change is fast enough to be visible against the background.

• Decomposition methods that separate within-firm change from reallocation in real data.
• Why low-productivity firms persist for so long in ordinary times.
• The scarring literature on what happens to displaced workers over the following decade.

Every term the collection defines is gathered in the glossary.