Mechanism design (reverse game theory)

The rules are the game

Game theory asks: given the rules, what will people do?
Mechanism design asks the reverse: given what I want to happen, what rules should I write?

Three Nobel Prizes in thirteen years (Hurwicz/Maskin/Myerson in 2007, Roth/Shapley in 2012, Milgrom/Wilson in 2020) sit on this one idea: outcomes are shaped more by the rules than by the players. Below we're going to prove that twice. First with voting — four reasonable rules pick four different winners from the same 122 ballots. Then with auctions — a tiny twist in the rules makes "tell the truth" the best strategy for every bidder on earth.

Fact 1: Four rules. Four winners. Same ballots.

Below are 122 voters' preferences over four candidates A B C D. Nobody is lying. Nobody is changing their mind. Click a voting rule and watch the counting play out — then watch who wins.

The 122-voter ballot profile · pick a rule

Voters	Their ranking (favourite → least)
38	C▸A▸D▸B
28	B▸D▸A▸C
27	D▸B▸A▸C
26	A▸D▸B▸C
3	C▸D▸A▸B

The outcome was determined by the rule, not by the voters. C has the largest first-place base. B wins the runoff. D is everyone's acceptable second choice. A beats every single rival one-on-one. Each has a legitimate claim to be "the winner." The ballots don't choose — the rule does. Which means whoever writes the rule has already done much of the choosing, and "the people's will" isn't a well-defined object once there are more than two options.

Fact 2: Even majority rule can produce no winner

Three voters, three candidates A B C. Each voter is perfectly consistent. But look what happens when you ask the group to decide pairwise:

👤 Voter 1

A▸B▸C

👤 Voter 2

B▸C▸A

👤 Voter 3

C▸A▸B

Who wins depends entirely on what order you hold the votes in. Pick an agenda — the "chair" of the meeting gets to set it — and watch the winner change.

Pick an agenda · same voters · different winner

The Condorcet paradox, and it's not academic. In a cyclic majority, whoever controls the agenda controls the winner. The "killer amendment" in Congress and the ordering of amendments in committees are this theorem deployed deliberately. The lesson: procedural power is as decisive as voting power. Both Arrow's theorem (1951) and Gibbard–Satterthwaite tell you this isn't a quirk of specific rules — with three or more options, no "nice" ranked voting rule can be strategy-proof. All the trouble starts at three.

Fact 3: In the right auction, "tell the truth" is the best strategy

Flip over to auctions — the other arena of mechanism design. You want to buy something. Your true value is 70 (what it's actually worth to you). Three rivals are also bidding; their values are drawn uniformly from 0–100 and you don't know what they are. The question: how much should you bid?

The answer depends entirely on the auction format. Drag the bid slider and watch two profit curves — one for a first-price auction (highest bid pays their bid), one for a second-price (Vickrey) auction (highest bid pays the second-highest bid). Each curve is a live Monte Carlo of 2,000 auctions against random rivals.

Your true value is 70 · how much should you actually bid?

Your true value

what it's worth to you

💎 Second-price profit

—

pays 2nd-highest bid

💸 First-price profit

—

pays your bid

🎯 Your bid70

Two things you just discovered. First, the blue second-price curve peaks exactly at your true value of 70, and it's almost flat on top — bidding 65 or 75 instead of 70 costs you almost nothing. In Vickrey, truth is a dominant strategy and the mechanism is forgiving. Second, the orange first-price curve peaks around 50–55: you have to shade your bid below your value, and shade right. Shade too little → you win worthless victories. Shade too much → you lose good ones. First-price is a tax on unsophistication.

Fact 4: Change one assumption — and winning becomes bad news

Now change the setup. You're not bidding on something only worth 70 to you — you're bidding on an oil tract that has the same true value for everyone. Call it 100. Nobody knows that number exactly; each of you gets a noisy, unbiased estimate. (This is how oil leases, company acquisitions, and sports free-agent contracts actually look.)

Everyone bids their honest estimate. Watch what happens to the winner — the person who estimated highest — as the field grows.

Common-value auction · true value is 100, estimates are noisy

👥 Number of bidders5 🎲 Estimate noise σ20

True value

100

what the tract is really worth

🏆 Winner's bid

—

how high they estimated

💰 Winner's profit

—

bid − true value (first-price)

The counter-intuitive comparative static. Everyone's estimate is unbiased. Nobody is irrational. And yet bidding your honest estimate loses money, and loses more as the field grows. Why? Because winning is itself bad news: the winner is, by construction, whoever happened to be most over-optimistic. "Congratulations, you've paid more than anybody else in the world thought it was worth." In common-value settings, more rivals should make you bid less, not more — exactly backwards from the Vickrey slider above.

Capen, Clapp & Campbell (1971) discovered this at ARCO when they noticed oil companies were systematically overpaying for offshore leases. It's the standard account of acquisition premiums that destroy shareholder value, free-agent contracts that look insane in year three, and book advances that never earn out.

The four forces every rule-writer wrestles with

When you design any mechanism — an auction, a voting rule, a pay scheme, a matching market — four tensions keep showing up. The revelation principle says: if you can get good outcomes with any mechanism, you can get them with one where everyone just tells the truth. So the whole art is making honesty the best policy.

Design force	What it means	Shows up as
Truthfulness	Agents should want to report their actual values	Vickrey, deferred acceptance (medical match), honest tax returns
Efficiency	The highest-value user ends up with the thing	Pivot (VCG) mechanisms, kidney exchange cycles
Budget balance	Money in = money out, no outside subsidy needed	Market-clearing prices (and Myerson–Satterthwaite says you can't have this + truth + efficiency at once)
Robustness	Still works when people miscount, collude, or play dumb	Why Google uses GSP instead of VCG; why "one more bidder" beats a clever reserve

Bulow & Klemperer's blunt advice. A plain English auction with n + 1 bidders raises more revenue than the theoretically optimal mechanism with n. In plain English: stop optimizing the squeeze and go find another buyer. Effort spent on marketing the sale, lowering entry barriers, and making the process easy to join dominates effort spent on clever pricing rules. Klemperer's experience advising on real auctions is that failures were almost always about insufficient entry and collusion, never about choosing the wrong format from the textbook.

Where mechanism design has actually changed the world

Medical residencies. The National Resident Matching Program uses deferred acceptance (Gale–Shapley): applicants rank hospitals, hospitals rank applicants, and nobody has to strategize — honest ranking is a dominant strategy on the applicant side.
School choice. The pre-2005 Boston system punished honesty: naming your true favourite could cost you your second favourite. Al Roth's redesign (now used in Boston, NYC and most of the US) swapped it for deferred acceptance. The reform was a mechanism-design fix.
Kidney exchange. You can't buy or sell kidneys, so price can't clear the market. Instead: find cycles (my donor matches you, yours matches me) and chains starting from altruistic donors. Thousands of transplants that would never otherwise happen.
FCC spectrum auctions. Simultaneous ascending auctions designed by Milgrom, Wilson & McAfee raised billions — then bidders discovered they could coordinate by encoding market numbers in the trailing digits of their bids ("stay out of Dallas and I'll stay out of Phoenix"). The fix: click-box bidding, removing the usable digits. Every piece of info your mechanism reveals is a coordination device for bidders.
Deposit insurance. A mechanism-design fix to the bank-run threshold problem: change the payoffs so the run's threshold is never reached. The best cross-model illustration in the whole book.

Related models

♟️ Game theory — the forward direction: given rules, predict outcomes. Mechanism design is the reverse: given desired outcomes, write the rules.
🔥 Thresholds & feedback — bank runs, standing ovations and riots are threshold problems; deposit insurance is the mechanism-design patch.
🏔 NK rugged landscape — matching markets like kidney exchange are combinatorial searches on a landscape; mechanism design is choosing which landscape to climb.

← Game theory All models →