Matching Markets and the Deferred-acceptance Algorithm

What this topic is

A matching market is one where both sides choose: a job needs to want you and you need to want the job. Price alone cannot clear it, because the seller cares who the buyer is. The deferred-acceptance algorithm, published by Gale and Shapley in 1962, produces a matching with a specific guarantee — no pair of participants both prefer each other to whoever they were matched with. That condition is called stability, and everything commercially valuable about the algorithm follows from it.

Why it matters

A pair who both prefer each other to their assignment is exactly a pair with a mutual reason to leave your platform and deal directly. Stability is therefore not a fairness property; it is a disintermediation-resistance property, and it is measurable in the transactions you keep. Markets without a stable clearing mechanism also tend to fall apart in a specific and well-documented way — offers move earlier, become exploding, and match quality collapses — which is why the algorithm was independently reinvented by an industry that had no economists in the room.

Who should learn it

Anyone building a marketplace where both sides select each other — talent, freelance work, mentoring, deal flow, internal mobility, shift assignment, capacity allocation across teams — and anyone who has watched matched users take the relationship off-platform.

What you will understand

• What stability means, why a blocking pair is a commercial problem rather than a mathematical one, and how to find one
• How deferred acceptance runs, and why it always terminates in a stable matching
• Why the proposing side gets systematically the better outcome, and what that makes the choice of proposer
• The impossibility results — no mechanism is strategy-proof for both sides, and the set of unmatched agents never changes

Prerequisites

• Signaling and the Separating Equilibrium
• Entry Deterrence and Strategic Capacity Commitment

Common misconception

"We can just score every pair and assign the best fits greedily." A greedy assignment optimises a number you invented and has no guarantee at all against blocking pairs. You will produce assignments where a firm and a candidate who were not matched to each other both wish they had been — and since both parties can see the market, they will find each other and transact without you. The second misconception is that stability is a fairness idea. Stable matchings can be wildly unequal: in a market of three firms and three candidates it is possible for one stable matching to give every firm its first choice and every candidate its last, and another to do precisely the reverse. Both are stable. The algorithm does not choose between them; you do, by deciding who proposes.