Industry concentration

Why do some industries settle into three enormous firms while others support thousands of small ones?

Two industries, same country, same decade, same laws. One is large passenger jets, essentially two makers worldwide. The other is restaurants, hundreds of thousands of them, no firm holding a share worth measuring. It is tempting to reach for character — one industry consolidated, the other did not — but that is a restatement, not a reason. What is different about the production of the two things such that one number of firms is stable and the other is not?

Begin with the cost of making a unit, and ask how it changes with the size of the operation. Almost every process has fixed costs paid before the first unit — a plant, a design, a certification — spread thinner the more units you make. But the spreading does not go on forever. At some output the average cost stops falling appreciably: the plant is full, the design amortised, and doubling again buys little. Call that output the minimum efficient scale.

Now put that beside the size of the market. If minimum efficient scale is a hundredth of national demand, roughly a hundred firms can each operate at full efficiency and none can undercut the rest by growing. If it is a third of demand, three can, and a fourth entrant would be permanently more expensive and would not enter. The ratio of the two numbers gives a rough ceiling on how many viable firms the industry has room for — the argument traces to Bain in the 1950s and remains the honest first cut.

Test it on the fragmented cases, because it does well there. Ready-mixed concrete has modest plant scale but must be discharged within about ninety minutes of batching, so the market is not the nation at all — it is a radius of tens of miles, each supporting a firm or two. Restaurants have almost no scale advantage in the activity that matters, since the meal is produced and consumed on the spot, and customers actively want variety. Transport cost and heterogeneous taste both shrink the effective market relative to the efficient plant, and the ratio delivers thousands of firms.

Now try to break it. Soft drinks: bottling plants are not enormous relative to a national market, and many could run at efficient scale, so the ratio predicts a fragmented industry. It is one of the most concentrated industries there is, and has been for a century, in every large economy. The account is not merely imprecise here — it points the wrong way.

What is different? The fixed cost of a bottling plant is set by engineering, and once the market is large enough to support many such plants, it does. Advertising is not like that. A firm can always spend more, and if spending more raises what customers will pay, a larger market justifies a larger spend rather than admitting more entrants. Sutton's 1991 argument turns on this distinction: where sunk costs are exogenous — fixed by technology — concentration falls as the market grows, exactly as the ratio predicts; where they are endogenous, chosen by firms and effective at raising willingness to pay, concentration has a floor that does not fall however large the market becomes. Research spending behaves the same way, which is why pharmaceuticals and large aircraft look structurally like soft drinks despite sharing nothing else.

That is the load-bearing claim, and it is a claim rather than a fact: the number of firms is set by minimum efficient scale relative to market size only when the sunk costs are technologically fixed. If escalating spend did not reliably raise willingness to pay, the floor would not exist and growing markets would fragment everywhere.

One thing worth naming as wrong. Concentration is often blamed on lax competition enforcement, which certainly matters at the margin. But it cannot be the general answer, because the same industries come out concentrated across countries with very different antitrust traditions, and those countries simultaneously host the fragmented ones. A theory that varies by jurisdiction cannot explain a pattern that does not.

Two refinements. First, minimum efficient scale is about the whole firm, not a factory: distribution, brand, regulatory approval and software all carry fixed costs, and in many modern industries those dwarf the plant. Second, the relevant market is the one a customer can actually buy from, which perishability, transport cost and taste all shrink — which is why a globally fragmented industry can be locally near-monopolised.

The misconception worth correcting is that concentration is a stage industries mature toward. Plenty do not, and stay fragmented for centuries, because nothing in their cost structure rewards being large. Equally, three huge firms are not by themselves evidence of anything wrong; that may be what efficiency looks like when the efficient unit is enormous.

Two limits on all of the above. The ratio gives a ceiling on firm numbers, not a prediction of who fills it — history and first-mover advantage decide the identities, and nothing here makes the survivors the best ones. And where firms are fragmented, the binding constraint is frequently not scale economics but its opposite: coordination costs inside a large organisation that rise faster than the savings, an argument developed in this collection's explanation of the boundaries of the firm.

• Why concentration says little on its own about prices or profits, and what else must be true before it does.
• How digital goods, whose marginal cost is near zero, fit a framework built around plant scale.

Every term the collection defines is gathered in the glossary.