Wharton scrapped course bidding for a fake-money market that prices course bundles

Wharton's MBA elective allocation was a points-bidding auction, and by the research team's assessment the mechanisms schools actually used were inefficient, unfair, or both. Students guessed which bids would clear, hedging their true preferences; oversubscribed courses stranded classmates on bad schedules; the game rewarded strategists. Eric Budish's 2011 Journal of Political Economy paper laid out a theoretical fix — approximate competitive equilibrium from equal incomes (A-CEEI) — but flagged practical blockers: reporting complexity, computational complexity, and capacity violations.

Course Match, built by Budish with Gerard Cachon, Judd Kessler and Abraham Othman, closed those gaps. Students report preferences over schedules rather than per-course bids; the engine runs a massive parallel heuristic search solving billions of mixed-integer programs to land on an approximate competitive equilibrium in a fake-money economy for seats; and capacity constraints hold. In full-scale use it allocated courses for roughly 1,700 Wharton students facing up to 350 courses a semester, with summary statistics across two semesters showing the outcome both fair and efficient.

The adoption was de-risked with a novel lab experiment at Wharton: using an 'elicited preferences' method built from binary choices, Budish and Kessler found participants preferred the schedules A-CEEI gave them to the ones the incumbent bidding-points auction produced, with measurably less envy. They report the experiment helped persuade the Wharton School to adopt the mechanism and shaped how preference reporting works in the product. In 2017, Wharton spun the implementation out as a startup, Cognomos, with Budish as a founding director.

Clear whole schedules at equilibrium prices, not seats

Equal budgets make the allocation envy-reducing by construction: nobody is outbid because they started with fewer points, they are outbid because the seat was worth more to someone else.

Bidding over whole schedules eliminates the exposure problem — students no longer win course one while losing the course that made course one worth taking.

Equilibrium prices replace guesswork: prices rise for oversubscribed courses and students reallocate to bargains, so demand spreads without a midnight click race.

The computational trick — billions of small mixed-integer programs in a parallel search — is what turned a 2011 existence result into a system that clears 1,700 students against 350 courses.

The team tested the mechanism on real Wharton students before full rollout, so the school adopted it knowing participants could report complex preferences accurately enough.

Any scarce multi-item allocation falls apart when it clears item by item: people strategize, honest bidders lose. Clear all bundles at equilibrium prices and honesty turns nearly optimal.

The mechanism design and computational procedure are in the public domain, and Wharton owns the production implementation, having funded it; the disclosure in the NBER working paper notes the school spun that intellectual property out into Cognomos in 2017 with Budish as a founding director holding equity. The case has become the standard reference for taking a combinatorial-allocation mechanism from theory — where A-CEEI's fairness and efficiency properties were proven — to a working market clearing thousands of schedules a semester.