Justin Wolfers' case for prediction markets

from John Authers:

Prediction markets in sport get the results right – almost spookily so. In the 3,791 NFL football games played from 1984 to 2000, regression results showed that the spread in the spread betting line could explain 99.7 per cent of the winning margin. Over a big enough sample, in other words, spread betters’ collective judgment was virtually flawless. Moving to basketball, the spread available in Las Vegas spread explained 97.3 per cent of winning margins, in a sample of more than 9,000 games from 1994 to 2001.

They also work well for the Hollywood Stock Exchange: a market predicting movies’ opening weekend box office take is almost as uncannily accurate as the sports betters.

Economic prediction markets work better than consensus forecasts of economists. During the brief history of an economic prediction market run by Goldman Sachs and Deutsche Bank, the market came closer to predicting ISM business confidence surveys, initial unemployment claims, non-farm payrolls and retail sales than economic forecasters did.

Prediction markets can help track political risk. Intrade.com and the Iowa Electronic Markets have gained a lot of coverage during the protracted campaign for the Democrats’ nomination. They deserve it: in the 2004 presidential election they called the winner of the electoral college vote correctly in all 50 states.

More to the point, over history they perform better than the Gallup poll, which is much more expensive to compile. Their average error on presidential elections is 1.4 per cent: the equivalent number for Gallup is more than 2 per cent.

They have a much longer history than polls. There are records of election betting in Wall Street going back to 1884. The average betting turnover per election, in 2002 dollars, is $37m, equivalent to 54 per cent of campaign spending.

They appear to be hard to manipulate. When the long-shot presidential contender Pat Buchanan encouraged his supporters to buy Buchanan futures on the Iowa markets, they failed to move the markets for more than a few hours – once the price was obviously too high, sellers emerged and pushed it back down again.

Arbitrage opportunities are few. In 2003, when Arnold Schwarzenegger ran for the California governorship, two rival markets offered futures on the race. They were based in Ireland and the Caribbean. But arbitrage opportunities never emerged, despite a wild race with many twists and turns. The bid price in one market was never above the ask price in the other.

One surprising result is that it does not seem too matter (on the basis of some unscientific comparisons) whether the money invested in the market is real or just a toy currency. Economics would predict otherwise, but in an experiment covering a full season French soccer, a prediction market using toy money did just as well as a market where real money was at risk.

There are weaknesses: they are only as good as the information that is made available to them. The market in futures for the chances of discovering weapons of mass destruction in Iraq was as wide of the mark as everyone else.

And like other markets, they are subject to behavioural biases. One is representational – when asked to picture a great baseball player, most people imagine Babe Ruth, who was white. When Wolfers tested the theory that the market would persistently underrate teams with a large proportion of African-American players (who look nothing like Ruth), it was confirmed. The average return on teams with three white players or less is 97 cents for a $1 bet. For teams with seven white players or more, the return is little over 92 cents. Thus the market as a whole overestimated the chances of predominantly white teams.

In stock markets, Wolfers found a similar bias related to gender: analysts persistently underestimated the earnings of companies with female CEOs. (He also found that male analysts were particularly likely to underestimate the earnings potential of female-led companies).

But these failings are not unique to prediction markets. All markets are prone to the behavioural biases of their participants, and to lack of adequate information.