Highlights

All bets are off at the Pentagon (FT features)

For many people, the first time they heard of the Pentagon’s plan to accept bets on terrorist activities was when the bizarre-sounding idea was abandoned.

The Policy Analysis Market, championed by the Pentagon’s research laboratory, the Defence Advanced Research Projects Agency (Darpa), would have traded futures contracts that paid out if particular events, including terrorist attacks, took place. It was widely attacked as both ghoulish and nonsensical.

Tom Daschle, the Senate Democratic leader, complained that it “could provide an incentive actually to commit acts of terrorism”.

But although the Pentagon may have abandoned its proposal, the controversy will not kill the essential idea of using betting to make predictions. Not only do many economists believe this method beats other forecasting techniques but they have also been demonstrating it for years. Precursors of the Policy Analysis Market have been used successfully to predict election results and even to make corporate sales forecasts.

Net Exchange, the company that developed the technology Darpa proposed to employ, hopes that it will be widely used both in the public and the private sector to replace more bureaucratic methods of forecasting. “We were talking to pharmaceuticals companies before all this and have been talking to them since,” says Charles Polk, Net Exchange’s president.

Darpa was hoping to exploit a simple idea: markets collect information by offering money for it. For instance, opinion polls predicting election results are notoriously unreliable because people often lie to pollsters. But offer the wrong odds on George W. Bush’s re-election and punters have every incentive to take the bet. In doing so, the theory goes, they reveal information more truthfully than they would to an opinion pollster. Economists and entrepreneurs are now developing information aggregation markets solely designed to gather information.

These markets create an asset to answer a specific question – for instance, a security that pays out $100 if, and only if, Mr Bush is re-elected. If his chances are 50-50, the asset should be worth about $50 (leaving aside complications).

The theoretical impetus behind this is the “efficient markets hypothesis”, which states that the market price of an asset accurately reflects all available information. If the “Bush to win” asset were priced at $30, well informed investors could take advantage by buying it and, in doing so, would quickly correct the price.

Following the collapse of the technology bubble, the efficient markets hypothesis looks a bit shaky. But proponents of information markets argue that they do not need to be perfectly efficient to be useful; they merely need to do better than other forecasting methods.

Robin Hanson, associate professor at George Mason University in the US and an adviser to Net Exchange, explains: “You have to compare these markets to the alternative. People point to market overvaluation during the bubble but the press and the analysts did not do a better job than the market of calling stock prices. You might point to individuals who did better than the market; but without hindsight, which advice would you follow?”

Information aggregation markets are particularly attractive when different people know different things but a bureaucracy or hierarchy is obstructing the free flow of information. The US intelligence services have been criticised for failing to co-ordinate information between their various agencies. Corporations often face similar difficulties.

Hewlett-Packard, the US computer company, has already appreciated this point. Kay-Yut Chen, a company researcher, teamed up with one of the field’s pioneers, Charles Plott of the California Institute of Technology, to use information markets to make better internal sales forecasts. These markets were minuscule compared with typical financial markets; they each operated at evenings and lunch times for a week, with fewer than two dozen active participants trading assets that paid out just $1 if future sales or revenues fell within a specified range.

Although the markets were thin and operated for low stakes, they performed better than Hewlett-Packard’s official sales forecasts, perhaps because the market traders came from different divisions in the company and had different information to bring to the market.

The trouble with such illiquid markets is that few trades are made and it is easy for one participant to blunder in and move the price a long way. Thin markets are also unlikely to attract the attention of informed speculators, because there is little money to be made.

While the experience of Hewlett-Packard showed that even thin information markets can beat other forecasting methods, it is better for an information aggregation market to have lots of trading activity. Prof Plott comments: “Thin markets can work; but I’d rather have a thick one.”

To get round the liquidity problem, Hewlett-Packard has looked for alternative economic mechanisms. Mr Chen explains: “When you run the market inside a corporation, you need knowledgeable people. But these people are busy and it’s hard to get them to play often enough.”

He and his colleagues developed a new approach, which is now being field-tested: instead of running a market, they run a one-off survey asking people to assign probabilities to different outcomes. They are later rewarded for their accuracy. The survey results are aggregated using a mathematical formula rather than a trading floor.

Prof Hanson’s design for Net Exchange and Darpa took a related approach to get round the problem of thin markets, creating a hybrid of an information aggregation market, which elicits a little information from each participant, with a method akin to Mr Chen’s, which is designed to obtain more. Prof Hanson claims that by pooling related markets in this way, his hybrid design increases liquidity, a necessary precursor to analysing the kind of political events that would have interested the Pentagon.

Prof Plott prefers to improve market liquidity more directly, by attracting more participants with information markets that are faster, easier and more fun to use. He is guarded about the hybrid methods, which have yet to establish a record.

Even if Prof Hanson’s design had worked perfectly, most people view the idea of betting on terrorist attacks with distaste. However, according to Mr Polk of Net Exchange, the so-called “terrorism futures” project was never designed to predict such attacks but to serve as proxy for general events such as a breakdown of the Middle East peace process.

“You could never try to predict or prevent a particular attack using the Policy Analysis Market,” he say. “It would be absurd.”

Prof Hanson also recognises the objections to such a market. “There are two concerns. To put it simply, one is that bad people might give up their information and make money. The other is that bad people might be willing to lose money to spoil the information in the market.”

Since the sums involved are small, neither Prof Hanson nor Mr Polk was worried about terrorists making money out of Darpa. Both suggest that it would also have been hard to rig the market. “It is very hard for manipulators to remove all of the information from the market,”, claims Prof Hanson. Mr Polk argues: “If you wanted to move the price in the wrong direction, you would have to make a lot of trades. That increase in volume would be noticed and is important information in itself.”

Although the Darpa project was cancelled, Mr Polk and Prof Hanson are hopeful that other firms will use their technology. “The biggest benefit is when you have a really important question to ask, which is why pharmaceuticals companies might be interested in using the technology to choose between different research projects,” suggests Prof Hanson.

But Prof Plott, who has tested information aggregation markets with several companies, sees serious political obstacles.

“These markets will eventually prove their worth but you have to be very careful taking the idea outside the laboratory.

“The press coverage is blatantly ignorant. Most companies treat their experiments with information aggregation markets as proprietary, because managers are afraid of being publicly portrayed as idiots.”

Prof Plott suggests that, in trying to overcome this conservatism, some proponents of information markets have claimed too much: “Net Exchange probably proposed more than it should have proposed to Darpa. But the problem is, in order to get a client to bite, people are tempted to promise more than they can deliver.”