Markets That Tip

Classical economics is built on diminishing returns. The second factory is less profitable than the first, the hundredth acre is worse land than the first, and this is what produces a stable equilibrium with several firms surviving. Expansion eventually stops paying.

Some things do not work that way, and where they do not, the economics is different in kind.

Start with network effects. A telephone with one user is worth nothing. A marketplace with no sellers attracts no buyers, and with no buyers attracts no sellers. The value of the product to each user depends on how many other users there are.

This produces a hard problem at launch (the chicken and egg) and an enormous advantage once solved, because a competitor must now offer not merely a better product but a sufficient reason to leave everyone you know behind. That is a far higher bar than being better.

Add zero marginal cost, which software has: the millionth copy costs nothing to make. Add learning curves, where unit costs fall predictably with cumulative production. Add data feedback, where more users produce better recommendations which attract more users.

Together these produce increasing returns, and Brian Arthur's insight was that increasing returns change the character of a market. With diminishing returns, an early lead erodes. With increasing returns, an early lead compounds, and the market tips toward a single winner.

The consequence is path dependence: the outcome depends on accidents of history rather than on which option was best, and once a standard is locked in, switching requires everyone to switch at once.

QWERTY is the usual illustration and the details are disputed: the story that it was designed to slow typists down is probably wrong, and its persistence despite the existence of faster layouts is real. Better examples are railway gauges, mains voltage standards, the shipping container's dimensions, and half the programming languages in production use. In each case the surviving standard is the one that got far enough first, which is not the same as the best one and is not obviously worse either, since a standard's value is mostly that it is standard.

Where increasing returns are extreme and physical, you get a natural monopoly: water pipes, electricity transmission, rail track. One network serving everybody is genuinely cheaper than two, so competition by duplication wastes resources on a large scale.

Breaking these up is not the answer, and there are three that are. Regulate the monopolist's prices, which requires the regulator to know costs the firm has every incentive to misrepresent. Own it publicly, which removes the profit motive and the information it carries. Or separate the network from the services running on it, regulate the network as a monopoly, and let competition happen on top, which is what has been done with electricity, telecoms and, patchily, rail.

The last complication is that many modern platforms are two-sided, and standard competition intuitions break on them.

A two-sided market serves two groups who value each other's presence. A newspaper serves readers and advertisers. A credit card serves cardholders and merchants. A nightclub serves men and women, and prices them differently. An app store serves developers and users.

The pricing structure is then not a simple markup over cost. The side that is harder to attract gets subsidised by the side that wants them: readers pay less than the paper costs to print, cardholders get rewards, developers pay a commission. Below-cost pricing to one group can be perfectly competitive rather than predatory, which is a conclusion that requires thinking about both sides at once, and which the US Supreme Court had to work through in the American Express case, deciding, controversially, that harm must be shown across the whole platform rather than on one side.

Which leaves the practical question that everyone has personally encountered. A messaging app is better than the incumbent and nobody moves.

The barrier is not price, ignorance or inertia. It is coordination. Switching pays only if the people you message switch too, so each person's best move is to stay while everyone stays: even in a world where everyone would prefer the alternative. That is a trap rather than a preference, and no amount of product improvement dissolves it.

What does dissolve it: a group switching together, a bridge that lets both be used at once, or regulation mandating interoperability. Mobile number portability is the cleanest example of policy dissolving exactly this lock-in, before it, changing network meant changing your number and telling everybody, and after it, the switching cost fell to nearly nothing and prices followed.