Mechanism design

Mechanism design is the branch of economics that asks the reverse of the usual game-theory question. Game theory takes the rules as given and predicts how rational players will behave. Mechanism design fixes the outcome you want – the goods go to whoever values them most, the public good gets funded in proportion to how many people want it, the treasury spends on what the members actually prefer – and then asks what rules would make self-interested players produce it without being asked to be honest, public-spirited, or even well-informed about each other. It is sometimes called inverse game theory for that reason.

The field was formalised by Leonid Hurwicz, who introduced the idea of a mechanism as a message-and-outcome system that could be compared on informational as well as allocative grounds, and extended by Eric Maskin on implementation and Roger Myerson on optimal design under private information. The three shared the 2007 Nobel Memorial Prize, with the official motivation “for having laid the foundations of mechanism design theory”. Their Nobel lectures – Hurwicz, Maskin and Myerson – remain the most accessible primary statements of what the field is for.

This matters to a DAO because a DAO is nothing but a mechanism. It has no manager who can override a bad outcome, no court that will read intent instead of text, and no way to punish a member for voting in their own interest. Whatever the rules reward is what the organisation will do. Most of the recurring failures catalogued under how DAOs fail are not failures of virtue; they are mechanisms behaving exactly as written.

The vocabulary: what a mechanism has to satisfy

A mechanism is judged against a short list of properties, and almost every design argument in DAO governance is really an argument about which of them to give up.

• Incentive compatibility (IC). Reporting your true preference is at least as good for you as any lie. The strong form, dominant-strategy incentive compatibility, holds no matter what anyone else does; the weaker Bayesian form holds only in expectation, given beliefs about others.
• Individual rationality (IR). Nobody is worse off for taking part than for staying out. A voting system that lets a majority expropriate a minority is not individually rational for the minority, which is why exit rights keep reappearing as a design patch.
• Budget balance. The mechanism does not need an outside subsidy to run. Quadratic funding is famously not budget balanced – it requires a matching pool – which is a design choice, not an oversight.
• Efficiency. The allocation maximises total value. This is the property most often sacrificed, because the impossibility results below say you frequently cannot have it alongside the others.

Matthew O. Jackson's survey Mechanism Theory (2003) is organised around the difficulty of getting incentive compatibility, efficiency, voluntary participation and budget balance at once, and the case of dominant strategies shows its shape. When preferences can take any form, the only mechanisms that pick the efficient outcome and make truthful reporting a dominant strategy are the Groves schemes, named after Theodore Groves, who found them; Jerry Green and Jean-Jacques Laffont proved that nothing else qualifies. The VCG mechanism described below is the best-known Groves scheme. A Groves scheme cannot in general balance its budget, and it can leave a participant worse off than if they had stayed out. Balance can be restored by setting one participant aside to absorb the surplus, at some cost in efficiency, and Jackson shows that the loss per person shrinks towards zero as the group grows.

The single most useful theorem in the field compresses the search space enormously. The revelation principle, set out in Myerson's Nobel lecture, says that any outcome achievable by any mechanism – however baroque its messages, rounds, or side-payments – is also achievable by a direct mechanism in which every player simply reports their private information truthfully. So a designer never has to invent clever message formats: it is enough to ask what truthful-reporting rules are feasible, and if none produces the outcome you want, no elaborate procedure will either.

The principle has a limit, which the prize committee's scientific background to the 2007 award sets out. A mechanism with a truthful equilibrium may also have other equilibria that produce worse outcomes, and players may end up in one of them. Eric Maskin's implementation theory, begun in 1977, asks for mechanisms in which every equilibrium produces the intended outcome.

What the impossibility results rule out

Mechanism design is as valuable for what it forecloses as for what it builds. Three results bound almost every governance argument.

Gibbard–Satterthwaite (Allan Gibbard 1973, Mark Satterthwaite 1975): for any voting rule over three or more outcomes that is deterministic, non-dictatorial, and can produce every outcome, there exist preferences under which some voter does better by misreporting. Strategic voting is not a defect of a particular ballot design – it is unavoidable. Any claim that a new voting scheme “eliminates strategic voting” is therefore false as stated; the honest version is that it changes which misreports pay, and how hard they are to coordinate. This is the theorem that governs the comparison between token-weighted voting, conviction voting and the rest – and, one level down, between the ballot methods that count whatever weight those rules assign.

Myerson–Satterthwaite (1983): when a buyer and a seller each hold private valuations and either may refuse to trade, no mechanism is simultaneously efficient, budget-balanced and individually rational. Some gains from trade are always lost. This is the formal reason a treasury cannot simply “buy out” a dissenting member at a fair price by asking both sides what the stake is worth, and it is the theorem lurking behind every buyback, ragequit valuation and Harberger tax proposal.

Arrow's impossibility theorem (1951) sits underneath both: no rank-order rule for three or more options can satisfy unrestricted domain, unanimity, independence of irrelevant alternatives and non-dictatorship at once. Aggregating preferences is not a solved engineering problem waiting for better software.

The practical upshot for a DAO is modest but real. You cannot design your way to a mechanism with no strategic play and no losses. You can choose which strategic play is cheap and which is expensive, and you can make the expensive kind visible.

Auctions: the worked example the field was built on

Auctions are where mechanism design is most fully solved, and where the theory earned its second Nobel: Paul Milgrom and Robert Wilson took the 2020 prize “for improvements to auction theory and inventions of new auction formats”.

The canonical result is William Vickrey's: in a sealed-bid auction where the winner pays the second-highest bid, bidding your true value is a dominant strategy. You cannot improve your outcome by shading, because your bid determines only whether you win, not what you pay. Generalised to many goods, this becomes the Vickrey–Clarke–Groves (VCG) mechanism: each winner pays the externality they impose on everyone else. VCG is efficient and dominant-strategy incentive-compatible, and it is also fragile in exactly the ways that matter on a blockchain – it can generate very low revenue, and it is vulnerable to a bidder splitting into several identities, which is the general form of the Sybil problem.

That fragility is the bridge to crypto. A permissionless system cannot assume one person equals one bidder, cannot assume bidders are not the same party, and cannot assume the auctioneer is honest – the auctioneer is a validator with its own position. Every clean auction-theoretic result has to be re-derived under those weaker assumptions, which is most of what blockchain mechanism-design research has been doing since 2019.

Paul Klemperer argues in What Really Matters in Auction Design (2002) that real auctions fail mostly through collusion and through bidders deciding not to enter, rather than over the refinements most auction theory studies. His opening example is the six European auctions of third-generation mobile licences in 2000: Germany and the United Kingdom raised over 600 euros per person, while Austria, the Netherlands, Italy and Switzerland raised 100, 170, 240 and 20. An ascending auction lets bidders signal how to divide the lots.

In Germany's 1999 spectrum auction Mannesmann's opening bids signalled to T-Mobil how to split the ten blocks on sale, and the auction ended after two rounds with each firm taking half at a low price. The same format deters weaker bidders who expect the strongest to win. Klemperer's remedies are sealed bids, or an Anglo-Dutch hybrid that runs an ascending auction until two bidders remain and then takes one final sealed bid from each, and he concludes that much of good auction design is standard antitrust economics applied to auctions. Designing working auctions and matching markets of this kind is the subject of market design.

Where blockchains actually do mechanism design

Using cryptography and economic incentives together to design a decentralised protocol is usually called cryptoeconomics, and much of it is mechanism design applied to blockchains. Three deployed examples show the field applied at scale, and all three are readable in primary sources.

Transaction fee markets. EIP-1559 (Buterin, Conner, Dudley, Slipper, Norden and Bakhta, created 13 April 2019, now Final) replaced a first-price gas auction with a protocol-set base fee that is burned and adjusts with demand, plus a priority tip the block producer keeps. Tim Roughgarden's economic analysis (arXiv:2012.00854, December 2020) is the reference treatment: it formalises what a transaction fee mechanism must satisfy for both users and block producers, and shows the base-fee burn is what makes the producer side incentive-compatible – a producer who kept the base fee would have a reason to manufacture demand and inflate it. The EIP itself states the point plainly, that burning the base fee “removes miner incentive to manipulate the fee”.

Block-space allocation. Proposer–builder separation turns block construction into a repeated auction between specialised builders and the validator who holds the slot. It is a mechanism-design response to MEV: rather than pretend ordering has no value, PBS makes the value explicit and auctions it, on the theory that a competitive auction returns it to validators instead of to whoever runs the fastest bot. Whether that competition materialises is an empirical question, and the concentration numbers on the MEV page are the honest answer to it.

Public-goods funding. Quadratic funding, from Buterin, Hitzig and Weyl's A Flexible Design for Funding Public Goods (arXiv:1809.06421, September 2018), allocates a matching pool by the square of the sum of the square roots of contributions – a rule chosen precisely because, under its assumptions, it is the one that funds public goods at the socially optimal level. Its assumptions are also its weak point: it presumes distinct, non-colluding contributors, which is why every deployment pairs it with identity infrastructure. Retroactive public goods funding is the alternative bet – that judging results after the fact is an easier problem than eliciting preferences beforehand.

Why DAO governance is the hard case

Classical mechanism design assumes players cannot make binding side deals. Break that assumption and most of the guarantees evaporate, because a mechanism that is incentive-compatible for an individual is rarely incentive-compatible for a coalition that can pay its members to deviate together. Vitalik Buterin's On Collusion (April 2019) is the standard statement of the problem in a blockchain setting, and its conclusion is uncomfortable: the properties that make on-chain governance attractive – transparent, verifiable, tradeable votes – are the same properties that make collusion cheap to arrange and easy to enforce.

Two DAO-specific pressures follow directly.

• Votes are for sale because they are transferable. If a governance token can be bought, borrowed or delegated, then influence can be rented for exactly one decision. That is the whole subject of vote markets and governance bribery, and it is why vote-escrow lockups exist – not to reward loyalty, but to raise the cost of renting a position for a single vote.
• Identity is not free. Every mechanism that treats participants as distinct people rather than distinct balances – quadratic voting, quadratic funding, one-person-one-vote – imports an identity assumption that the base layer does not supply. In practice the mechanism is only as strong as the Sybil resistance bolted on beside it.

Buterin's Moving beyond coin voting governance (August 2021) sets out the resulting design space, and DAOs are not corporations (September 2022) makes the case that decentralisation is worth its cost in exactly the decisions where an adversary would otherwise capture a single point of control. Both are useful correctives to the assumption that a DAO should converge on corporate governance with extra steps.

Game theory has a branch built for coalitions, and it is as old as the field: John von Neumann and Oskar Morgenstern's Theory of Games and Economic Behavior (1944), the book that founded game theory, treated games in which players form coalitions alongside games between individuals. Mechanism design grew mostly from the non-cooperative side. Alex Evans's crash course in mechanism design for cryptoeconomic applications (BlockChannel, 2017) points to the mismatch: mining power and token holdings are concentrated, and widely studied mechanisms such as the Groves schemes break down when participants collude.

Cooperative game theory asks which coalitions form and how each divides its joint payoff; a division that no group of players could improve on by breaking away is said to be in the core. Yoad Lewenberg and four co-authors applied it to Bitcoin mining pools, which combine their members' computing power and share the rewards, in a 2015 paper. For some network parameters, especially under heavy transaction load, they found the core empty: no way of sharing a pool's rewards is stable, and some miners always gain by switching pools.

How Caper approaches this

Caper's answer is to keep the mechanism small enough to reason about, and to price the two things that usually go unpriced: liquidity and exit.

Every caper token trades against a bonding curve rather than an order book, so the price of any trade is a deterministic function of circulation, computable before you sign. There is no discretionary market maker to model and no liquidity that can be withdrawn, which removes a whole class of strategic behaviour rather than trying to disincentivise it.

The sharper move is where a caper puts the voting rule. A proposal carries up to four executable options plus a standing “Do nothing”; holders rank all of them on one ballot, and a Borda fold passes the leader only if its share of the weight actually cast clears 1.5 / option_count. That is a ranked rule and Gibbard–Satterthwaite applies to it in full – a caper does not sidestep the theorem.

What it adds is a second gate the theorem says nothing about: a legislative pass executes nothing by itself, it earns a trigger, which records the caper's trailing average token price as a baseline and opens the market window, and the action goes through only if the time-weighted price over that window closes at or above the baseline. Strategic ranking is still possible; it simply no longer settles the question on its own. The families the first phase sits among – token-weighted, quadratic, conviction, ranked-choice – are surveyed in voting mechanisms.

What the second phase buys is that misreporting costs what it claims. Once a proposal is triggered the default is that it passes, so the burden falls on whoever objects: they cannot simply say the action is bad, they have to sell into the curve at the price their own selling creates, and anyone who wants it to pass after that has to buy at theirs. Preference is revealed by a trade rather than asserted in a tally – the same instinct behind futarchy, without a separate prediction market to bootstrap. The mirror-image attack is the obvious one and worth naming: an actor with enough capital can hold the price above the baseline for the whole market window, so this substitutes a price-manipulation problem for a preference-misreport problem rather than escaping strategy altogether.

Exit is priced separately, and it is the part that answers the individual-rationality problem. A member leaving hands in governance tokens together with the soulbound vote tokens their own ballots minted – none are minted for trading – and redeems a fair share of the treasury that grows with both their holdings and the vote tokens they have earned. The governance tokens handed in are sold on the curve in the same transaction, so the member leaves with the curve's price for them as well as the treasury share.

A member on the losing side of a decision is therefore never trapped, and a majority that wants to expropriate a minority has to buy them out. The individual-rationality guarantee is conditional in exactly one way: the second factor is earned at the ballot box, so a holder who has never voted has nothing to redeem against, and can leave only by selling on the curve. For the mechanics themselves, see what is a caper and rage-quit and exit rights.

References and further reading

• The Sveriges Riksbank Prize in Economic Sciences 2007 – Hurwicz, Maskin and Myerson, “for having laid the foundations of mechanism design theory”
• Mechanism Design Theory – scientific background on the 2007 prize, compiled by the Prize Committee of the Royal Swedish Academy of Sciences, 15 October 2007
• Perspectives on Mechanism Design in Economic Theory – Roger Myerson's Nobel lecture, the clearest primary account of the revelation principle and of what mechanism design is for
• But Who Will Guard the Guardians? – Leonid Hurwicz's Nobel lecture on enforcement and the limits of self-implementing rules
• Mechanism Design: How to Implement Social Goals – Eric Maskin's Nobel lecture on implementation theory
• Mechanism Theory – Matthew O. Jackson, survey written for the Encyclopedia of Life Support Systems (EOLSS Publishers, 2003); working-paper version revised December 2003
• Theory of Games and Economic Behavior – John von Neumann and Oskar Morgenstern, Princeton University Press, 1944; 60th-anniversary edition 2007
• The Sveriges Riksbank Prize in Economic Sciences 2020 – Milgrom and Wilson, “for improvements to auction theory and inventions of new auction formats”
• What Really Matters in Auction Design – Paul Klemperer, Journal of Economic Perspectives 16(1): 169–189, Winter 2002
• Transaction Fee Mechanism Design for the Ethereum Blockchain: An Economic Analysis of EIP-1559 – Tim Roughgarden, arXiv:2012.00854, 1 December 2020
• EIP-1559: Fee market change for ETH 1.0 chain – created 13 April 2019, status Final
• A Flexible Design for Funding Public Goods – Buterin, Hitzig and Weyl, arXiv:1809.06421, 17 September 2018, the quadratic-funding paper
• On Collusion – Vitalik Buterin, April 2019, on why coalition-proofness is the binding constraint on-chain
• Moving beyond coin voting governance – August 2021; and DAOs are not corporations – September 2022
• Bitcoin Mining Pools: A Cooperative Game Theoretic Analysis – Yoad Lewenberg, Yoram Bachrach, Yonatan Sompolinsky, Aviv Zohar and Jeffrey S. Rosenschein, Proceedings of AAMAS 2015, pp. 919–927
• A Crash Course in Mechanism Design for Cryptoeconomic Applications – Alex Evans, BlockChannel, 17 October 2017; an introduction to the vocabulary above for protocol designers, with Namecoin's allocation of names as a worked example
• CS364A: Algorithmic Game Theory – Tim Roughgarden's Stanford course, autumn 2013: twenty lectures with notes and videos, from Myerson's lemma and VCG to spectrum auctions and kidney exchange; published as Twenty Lectures on Algorithmic Game Theory (Cambridge University Press, 2016)