Rules of the game

Reforming governments should take note of the theory of "mechanism design"—work on which recently won three economists the Nobel
November 25, 2007

In mid-October, the Nobel prize for economics was awarded to Leo Hurwicz, Eric Maskin and Roger Myerson for their work on mechanism design. Newspaper economics correspondents made what they could of this news, but obviously had only the vaguest idea of what mechanism design is.

Mechanism design is too important to get this kind of treatment. The philosophy behind the theory represents a genuine "third way" that ought to inform the reforming efforts of any government of whatever stripe.

The basic idea of mechanism design is simple. It is obvious that in any organisation, decisions should be made by whoever happens to have the necessary expertise. The people at the head of the tree will almost never have the command of detail or the local knowledge needed to make good decisions at the coalface. So an efficient organisation will decentralise authority to the most appropriate level.

The problem is that the guys lower down the pecking order are unlikely to have the same motives as the guys at the top. As David Hume put it: "In constraining any system of government… every man ought to be supposed a knave and to have no other end, in all his actions, than private interest.'' He didn't mean by this that everybody is corrupt; only that people will find excuses to look after themselves rather than the interests of their organisation, unless someone keeps an eye on them.

There are two traditional ways of dealing with this problem. The first is to rely on market incentives to generate efficient outcomes. Margaret Thatcher was dimly aiming at this solution when she introduced so-called internal markets into the NHS in 1990. But the idea that free markets are necessarily efficient is a myth. Perfectly competitive markets are efficient, and we should exploit this fact when appropriate. But very few markets are perfectly competitive. Among other requirements, a perfectly competitive market needs a very large number of buyers and sellers and a free flow of information about price and quality. So Tesco doesn't operate in a perfectly competitive market, let alone the Bristol Royal Infirmary. And even when markets are perfectly competitive, nothing says they need be fair—and who wants an NHS that isn't fair?

The second traditional way of running an organisation is as a command economy. Central planners make up rules intended to deal with every contingency, and hammer anyone they catch cheating. After the Soviet Union, central planning is now universally condemned, but the NHS ran quite successfully as a command economy when it was set up in 1948. However, there are two problems. Central planners seldom make good plans, and even when they do, it is only a matter of time before people in the organisation learn to tell themselves that it is fine to bend the rules if the guy at the next desk is getting away with it.

Mechanism design provides a third way, by combining the carrot of a market economy with the stick of a command economy. If this were all the new Nobel laureates were proposing, it wouldn't be very exciting; all mainstream politicians these days agree. However, the prize-winners offer a more scientific methodology.

In the jargon of the trade, the government or central planner is called the "principal." She has various "agents," whom she wishes to act jointly in the public interest. Where the principal can monitor the behaviour of her agents, she can set rules like an old-time central planner and apply the stick to any deviants. But what should she do when she can't observe their behaviour, or can't tell whether what they are doing is good or bad? She must apply the carrot by offering incentives based on what she can observe. For example, the manager of a factory may offer a bonus if output reaches a high enough level.

All this is common ground, but the next step is not. The fatal error that governments seem unable to avoid is to assume that people won't change their behaviour in response to new rules and incentives. NHS targets are a good example. Some aspect of the service is thought to be inadequate, and so a target is set without taking into account the fact that a manager will thereby have an incentive to move resources from other services in an effort to meet the target. The departing manager of the Maidstone and Tunbridge Wells Hospital Trust attributed the Clostridium difficile outbreak that resulted in 90 deaths to this cause.

The US congress made a similar mistake in 1990 when it passed an act that said drug manufacturers could sell drugs to Medicare at no more than 88 per cent of the average selling price. The problem was created by an extra provision which said that Medicare must also be offered at least as good a price as any retailer. This provision would only work as its framers intended if drug manufacturers could be relied upon to ignore the new incentives it created for them. But why would they sell a drug to a retailer at less than 88 per cent of the current average price if they would then have to sell the drug at the same price to a huge customer like Medicare? However, if no drugs are sold at less than 88 per cent of the current average, the average price will be forced up!

The inventors of mechanism design coped with such problems by recognising that the system of rules and incentives created by the principal's attempts at reform create a deadly serious game for her agents to play. (The rules of this game are the "mechanism" of mechanism design.) John Nash—hero of the movie A Beautiful Mind—taught us how to predict how the agents will play such games once they have learned the ropes. The prediction is what we nowadays call a Nash equilibrium. This is a set of strategies (one for each agent) with the property that each agent's strategy is a best reply to the other agents' strategies. Agents will keep searching for a better strategy until there is none to be found. When this happens, they will find themselves at a Nash equilibrium.

The principal's task can now be seen as seeking a game for her agents to play in which the Nash equilibrium that results is as near to her goal as possible. But how is the principal to do this? The main achievement of our three prize-winners was to show how the best of all possible outcomes can sometimes be found.

One of Roger Myerson's results is particularly striking. A seller has a house to sell. Only she knows the least she will take for the house; everybody else believes her valuation is equally likely to be anything between £1m and £5m. There is only one possible buyer. Only he knows the most he will pay for the house; everybody else believes his valuation is equally likely to be anything between £1m and £5m. If the house is sold at a price between the valuations of the buyer and the seller, then both are better off. The economic surplus created by the transaction is the difference in their valuations.

A benign and all-powerful principal now appears. Her aim is to maximise economic surplus. To this end, she can force the buyer and the seller to bargain according to any rules she may invent. What is the best of all possible outcomes? It is remarkable that we can answer this question at all. It is even more remarkable that the answer is so unpalatable. The best possible outcome is that the house only gets sold when the buyer's valuation exceeds the seller's by £1m.

The efficient outcome would be for the house to be sold whenever the buyer's valuation exceeds the seller's, but this outcome is beyond the principal's reach, because she doesn't know the agents' valuations. If she asks them, they will answer strategically, because only dimwits put their final offer on the table straight away. The result is that the house often won't be sold even though everyone would have profited by its sale.

Such inefficiencies are usually unavoidable when information can't flow freely, which is usually the case in real life. Economists say that Myerson's result is second best, but the best that can be achieved given the principal's state of information, which makes the first best impossible. The lesson for governments with utopian aspirations couldn't be clearer. In the NHS, for example, more people will die if we set our sights too high than if we are realistic about what is possible in this imperfect world.

How can mechanism design be used to improve our lives? In 2000, the use of an innovative procedure to prevent collusion in the auction of 3G telecoms licences in Britain made £22.5bn for the taxpayer. After auctions, the most promising area is the regulation of imperfectly competitive industries. The story here is a kind of replay in miniature of the misconceived idea that we have to choose between a free market or a command economy. Current regulation law is based on the idea that perfect competition is good and monopoly bad. The result is a ramshackle edifice that largely ignores the advances in understanding imperfect competition that came with the advent of game theory.

Critics of regulation are right to complain that much regulation is so bad that we would be better off without it. That is to say, we would all be better off letting companies with lots of market power screw more money out of us than we need to pay, rather than tolerating the restraints to trade that current regulation creates. But regulation doesn't need to be bad. We understand some imperfectly competitive industries well enough that the current state of mechanism design is already adequate to get close to the best of all possible regulatory regimes, but the opposition from vested interests and the legal establishment is very powerful. Nothing is therefore going to happen without a prolonged campaign by enlightened journalists demanding that regulation be brought up to date.

What about the NHS? None of our theorems comes near encompassing the complexity of the problem, but the basic principles apply. If you are going to base incentives on targets, do some research on how managers are likely to respond. It will involve writing a new model for each institution you try to regulate, because one lesson of mechanism design is that "one size fits all" is a bad philosophy.

Such models need a lot of thought. Data needs to be gathered and analysed. All this will be expensive, but in the NHS, quick fixes cost lives. So does politically motivated micromanagement from Whitehall. We need to delegate responsibility to the appropriate professionals. As with the Bank of England or the BBC, there is a strong case for making the day-to-day management of the NHS independent of government, whose role would be restricted to determining funding and to ensuring that the NHS directorate honours its constitutional responsibilities.

It is optimistic of me to think that the award of the Nobel prize for mechanism design might provoke a public debate on these issues. But hope burns eternal even in the most rational of breasts.



Bookmark this page with:

Stumble Upon              Digg              Delicious              Reddit              Facebook