TCS Daily

The Wisdom of Tradesports

By William Wilson - November 27, 2006 12:00 AM

In the wake of the Democratic takeover of the House and Senate, much hay has been made over the failure of the prediction market Tradesports (or the Iowa Electronic Markets for the more legalistically squeamish). Going into November 7th, TS gave the Republicans a seventy percent chance of retaining the Senate. For much of the life of the contracts, the markets predicted both chambers staying in GOP hands. Were the markets wrong? The answer, unequivocally, is no.

The suggestion that Tradesports "missed" on the control of the Senate is based on a faulty understanding of probabilities. By Election Day, Senate control really came down to three close races, Virginia, Missouri, and Montana. A probability calculator shows that if the markets accurately gave each of those races about a 60% chance of going to the Democrat, then the odds that all three would go Democratic were only 22% (0.6 to the third power). Therefore, a rational person would choose the Democrat in the individual races, but would nonetheless choose the Republicans to retain control. After all, the Republicans needed only one of those three, while the Dems needed all three. Only if we posit that the Democrats had an 80% chance to win each race would Senate control have been even a 50-50 proposition.

The fact that the election went the way it did does not mean that the probabilities on individual races were inaccurate. We can't play out the election 10,000 times to find out the actual probabilities. If we could, John Tester would lose some of the time, Jim Talent would win some of the time, and George Allen might—might!—have made a smart campaign decision once in a while. In the end, the Republicans would probably keep the Senate more often than not—just like Tradesports indicated.

Tradesports' political markets should not be mistaken for perfect prognosticators—they are markets. Investors' financial incentives systematically root out personal bias and home team favoritism, but the prices do not (primarily) present glimpses into outcome probabilities. Many months ago, we bought Harold Ford to win Tennessee's Senate seat at 1-in-3 odds (that is, shares purchased at $3.30 would pay $10), thinking not only that his famed campaigning skills gave him a real shot at winning, but also that the national media would fixate on him at some point and generate irrational exuberance for his chances. This indeed occurred, and we profited by selling around $6.00. By incorporating a meta-level anticipation of future value unrelated to the actual chances of winning, Tradesports thus acts more like a commodity market than a crystal ball.

As a result, prices can be taken out of context easily. Scott Moore concluded an otherwise interesting article about Tradesports in the Wall Street Journal by consoling Republicans with the news that "the latest odds of Hillary Clinton winning the White House in 2008 have tumbled to 22%." This information presentation was deeply misleading. According to the Tradesports contract log, this tumble must have been from 23%, not exactly a catastrophic fall. And even her lowered total kept her neck-and-neck with John McCain, and far ahead of all other challengers.

More to the point, though, the prices do not represent the binary "win or lose" probabilities of a Clinton return to 1600 Pennsylvania. The clearing price for Clinton commodity speculation was $2.20; with two years until the election, as many people as not believe that at any future time the universe of TS bettors will pay more than $2.20 for her stock. Sure, some of these Clinton speculators may believe that her chances of winning the '08 election are better than 22%, but many of them may have also believed that near-term news cycles would be more favorable to her than not, or that the Democratic swell on November 7 would raise all Democrats' chances relative to Republican ones. And for every one of those traders, a paired trader has speculated the opposite to an inversely proportional degree (i.e., someone was also willing to pay $7.80 to make $10 for the proposition that Hillary would not become President in '08). The beauty of Tradesports—what differentiates it from a simple sportsbook—is that it allows such speculation; no one must hold a contract until expiration. The "odds-speak" shorthand conveys the right price, but shouldn't suggest an inappropriately probabilistic explanation of January 20, 2009.

Compounding that interpretive problem is the fact that Tradesports is an imperfect market. Prices do not move with the merry fluidity of the New York Stock Exchange, the market breaks down at the margins, and an alarming number of utterly irrational exchanges occur. Many of these problems are attributable to the constrained liquidity of the TS universe. The sheen of moral degeneracy that shrouds gambling keeps the Tradesports user base low relative to the generally interested population. And for the last month, at least, those few users have had a harder time funding their activities due in no small part to legal questions raised by the Unlawful Internet Gambling Enforcement Act of 2006 (UIGEA).

It should also be noted that TS is an information market and depends on good information. The fact that Tradesports showed Republicans retaining the House last June only reflects the information known at the time, not the failure of the market. All markets are only as good as the information people actually have, not what they will have in the future.

Similarly, traders must have a good understanding of the available information. On election night, as both Allen and Talent continued to lead the incoming returns on the news, the prices for Republican control of the Senate soared above 90. In retrospect, it seems clear that most traders were incorrectly interpreting incoming returns as precinct portions of homogeneous electoral wholes. Trades made based on the earliest reporting precinct returns were eventually harpooned by the real-time demographic heterogeneity of ballot tabulation; when the Jackson County numbers came in for Claire McCaskill, the SEN.GOP market crashed—big time.

Still, Tradesports is most valuable for its information barometrics. A remarkable accomplishment is that TS predicted the closeness of the respective Senate races in addition to their outcomes. The speculative capacity of the market opens the door for information evaluation, separating real news from spin. Karl Rove's late-October confidence momentarily bolstered the Republican markets, but Karl couldn't make it so, and the numbers quickly returned to Democratic headings. That is the truly amazing thing about Tradesports: it can distinguish between the non-event that was Giuliani's Presidential exploration announcement, which the market shook off as old news, and Barack Obama telling Tim Russert that he has "thought about the possibility," which doubled the price of "Obama as Democratic nominee in'08" within a week—and caused an "Obama to win in '08" contract to be created just two days later. Pundits could tell us there is a difference, but Tradesports registers the magnitude, urgency, and certainty of the information.

By the time we get to 2008, we might lack this valuable information evaluation. TS was right about the incoming Congress; the incoming Congress should do right by Tradesports and repeal UIGEA.


1 Comment

While I agree with many of points raised in Mr Wilson and Mr Wallach's article, especially with regard to the distinctions they made between the clearing prices and the actual odds of a given candidate winning, the authors make a rather basic error in their leading example of the market for the GOP retaining control of the Senate. The authors state that "if the markets gave each of [the three close senate races] about a 60% chance of going to the Democrat [sic], then the odds that all three would go Democratic were only 22% (0.6 to the third power)."

This statement, of course, would only be true in the truly exceptional case that all three races were completely uncorrelated with each other. It is far more likely (political scientists and stat heads can check or estimate this by tracking reliable poling data) that close elections show significant positive correlation. Certainly, one can imagine global events that would tip the odds of all three elections towards one party or another.

If the three elections were perfectly correlated the probability of the GOP losing control of the Senate would be 60%, not 22%. So, without estimating the correlation of these elections, we can only say that the odds of losing were = 22%.

TCS Daily Archives