Mechanism Design

Thinking Tools4.3 Crafting Solutions for the Future StateAlso known as Reverse game theory

Developed by Leonid Hurwicz, Eric S. Maskin, and Roger B. Myerson (1960 to 1981), who shared the 2007 economics Nobel for the work.

Starts from the outcome wanted and works back to rules, payoffs, and information flows under which people pursuing their own interest produce it.

Three rules for splitting a $3 million budget, and whether each makes the truth pay

Three plant managers share a $3.0 million capital pool; each truly needs $1.0 million. Pick a rule, then pick what one manager reports: the line says how a self-interested manager comes out, assuming the other two do whatever pays best. The rule is good when honest reporting is the best move. Illustrative.

The rule

What this manager reports

  • Pick a rule and a report.Honest while the other two pad: this manager's share falls to $0.83M against a $1.0M need. The rule punishes the truth.Padding is the best reply: with every manager asking $1.3M, each still gets $1.0M, and the requests no longer say who needs what.The report changes nothing: the formula gives $1.0M by headcount. Nobody gains by lying, and nobody's real need is used either.Padding gains nothing, and neither does honesty. The money follows headcount, and a manager with a real surge in need cannot say so.Honest, and best: the plant gets the $1.0M it asked for and pays nothing, because its request pushes no other project out.Padding costs more than it gains: the extra $0.3M carries a $0.4M charge, the value of the projects it pushes out of the other plants' budgets.
  • Honest while the other two pad: this manager's share falls to $0.83M against a $1.0M need. The rule punishes the truth. (rule: Split by requests, report: True need, $1.0M)
  • Padding is the best reply: with every manager asking $1.3M, each still gets $1.0M, and the requests no longer say who needs what. (rule: Split by requests, report: Padded to $1.3M)
  • The report changes nothing: the formula gives $1.0M by headcount. Nobody gains by lying, and nobody's real need is used either. (rule: Fixed formula, report: True need, $1.0M)
  • Padding gains nothing, and neither does honesty. The money follows headcount, and a manager with a real surge in need cannot say so. (rule: Fixed formula, report: Padded to $1.3M)
  • Honest, and best: the plant gets the $1.0M it asked for and pays nothing, because its request pushes no other project out. (rule: Pay for displacement, report: True need, $1.0M)
  • Padding costs more than it gains: the extra $0.3M carries a $0.4M charge, the value of the projects it pushes out of the other plants' budgets. (rule: Pay for displacement, report: Padded to $1.3M)

An example

Every year the plant managers of a mid-size manufacturer asked for about 30 percent more capital than they spent, and every year the chief financial officer cut every request by a quarter. The managers knew it and padded accordingly. The requests had stopped telling anyone where the money was needed.

The CFO stopped trying to catch padding and changed the rule instead. He started from the outcome he wanted: requests close to real need. Then he asked what rule would make that the best move for a manager looking after their own plant. Splitting the pool in proportion to requests rewarded padding. A fixed formula by headcount removed the reason to pad, and also any way to signal a real surge in need. The rule he adopted funded each request in full, but charged a plant for any amount above its prior-year spend at the return of the best project that money pushed out of another plant's budget. In the first year under the rule, requests came within six percent of what was later spent.

Why it works

Most management responses to gaming are detection: audit the requests, cut them across the board, punish the worst offenders. Each response teaches people to game more carefully. Mechanism design reverses the question. It takes the outcome as fixed and treats the rules as the unknown, then asks what rules would make the wanted behavior each person's best move, given what they know and what they want.

The central test is whether the truth pays. Under a rule where honest reporting is the best response whatever the others do, nobody needs to be trusted or caught, and the information people report can be used. The usual way to get there is to make each person bear the cost their choice imposes on others, so that the private calculation and the shared one point the same way.

The thinking behind it

Leonid Hurwicz Leonid Hurwicz founded mechanism design, which treats the rules of an economic institution as the unknown to be designed for a desired outcome. In Optimality and Informational Efficiency in Resource Allocation Processes (1960) he framed institutions as systems for communicating and processing information. He introduced incentive compatibility, the requirement that participants do best by acting as the rules intend. He shared the 2007 Nobel Memorial Prize in economics with Eric Maskin and Roger Myerson.

Eric S. Maskin Eric S. Maskin, an economic theorist, asked which social goals can be implemented at all: for which goals there is a mechanism whose equilibria give exactly the desired outcomes. His answer, set out in Nash Equilibrium and Welfare Optimality (circulated in 1977, published in 1999), is the condition now called Maskin monotonicity. It tells a designer in advance when no rule, however clever, can produce the outcome wanted.

Roger B. Myerson Roger B. Myerson, a game theorist at Northwestern and later the University of Chicago, laid foundations of mechanism design that won him a share of the 2007 prize. His Optimal Auction Design (1981) derived the auction that maximizes a seller's expected revenue, and is a founding result of auction theory. An auction is the plainest case of the designer's problem: the seller writes the rules, and the buyers decide how to bid under them.

How to run it

  1. State the outcome you want. Write it in terms of behavior, such as requests that match real need.
  2. Name who holds the information. List what each participant knows that you do not, and what each wants for themselves.
  3. Write the current rule's best reply. Work out what a self-interested participant does under today's rule, and how far that is from the outcome.
  4. Draft rules that change the payoff. Try two or three rules, such as a formula, a charge, or a reward for accuracy.
  5. Test each rule against honesty. For each rule, ask whether telling the truth is the best move whatever the others do.
  6. Make choices carry their cost. Prefer rules where each person pays for what their choice takes from others.
  7. Pilot and watch the reports. Run the rule for one cycle and compare what people reported with what happened.

Where it goes wrong

The first failure is designing for honest people. A rule that works only if participants report truthfully is a request, not a mechanism. Test every draft by asking what a smart, self-interested participant would do.

The second is a rule nobody understands. If managers cannot work out their best move, they fall back on habit, and the incentive does nothing. Explain the rule with a worked example before it starts.

The third is ignoring the side effects. A charge that stops padding can also stop managers from asking for money they really need. Watch for requests that fall below later spending, and adjust the charge.

When to use something else

  • Auction Design. When the problem is selling or allocating one scarce thing to whoever values it most.
  • Agency Theory. When one party acts for another and the question is how to align a single agent's pay with the owner's interest.
  • Game Theory. When the rules are fixed and the job is predicting how the players will act under them.

Where it sits in SOLVE eX

Go to the source

  • Optimality and Informational Efficiency in Resource Allocation Processes, Mathematical Methods in the Social Sciences, 1959 (Arrow, Karlin, and Suppes, eds., Stanford University Press), 1960
  • Nash Equilibrium and Welfare Optimality, Review of Economic Studies 66(1); first circulated as an MIT working paper, 1999
  • Optimal Auction Design, Mathematics of Operations Research 6(1), 1981

Back to the catalog