Prediction Markets and the Election--Posner

The forthcoming presidential election has drawn attention to online predictions markets. The first, and one of the best known, is the Iowa Electronic Market (IEM), started in 1988 to bet on presidential elections. Participants can bet up to $500. The odds and hence the price of a contract are set by the bidders themselves, as in a stock market, rather than by the "house," as in casino gambling. A number of other prediction markets, some using virtual (i.e., play) rather than real money, have emerged, includingTradeSports.com, the Foresight Exchange Market, Newsfutures, Intrade, and the Hollywood Stock Exchange. IEM, on which I'll focus, has correctly predicted the outcome of every presidential election since 1988, and its predictions have been consistently more accurate than the polls. An interesting comparison between the Gallup Poll and the Iowa market in the 1996 presidential campaign (www.biz.uiowa.edu/iem/media/96Pres_VS.html) reveals that throughout the entire campaign the Iowa market’s predicted outcome was much closer (in margin of victory) to the actual outcome than the Gallup Poll was. Studies have found that prediction markets beat polls and other prediction tools even when a prediction market uses play rather than real money. The Pentagon planned to create a prediction market in which participants could bet on the likelihood of terrorist attacks, assassinations, and coups. The plan caused outrage and was abandoned. There was a serious objection to the plan: people planning terrorist attacks, assassinations, and coups have inside information which they could use to make a killing (pun intended) in the prediction market. The success of prediction markets is related to though distinct from the success of the "blogosphere" in ferreting out information that eludes the mass media. Both the blogosphere and prediction markets aggregate greater amounts of information than any centralized information gatherer can obtain. In the case of the blogosphere, it is easy to see why this is so. It is virtually costless (except in time) to become a blogger, and among the millions of people drawn to blogging are people with all sorts of pockets of specialized information, which the internet enables to be pooled rapidly. This pooling resembles the economic market, in which vast amounts of information, encapsulated in prices, are pooled (the basic insight of Friedrich Hayek, and the secret of capitalism’s superiority to socialism as a means of optimizing economic activity). Prediction markets provide an even closer analogy to the market, since they (or rather some of them, for others permit betting only with play money) provide financial rewards for correct information (as blogging rarely does), in this resembling ordinary commercial speculation. Someone who thinks he has superior insight into political processes will have an incentive to place a bet in IEM or some other political prediction market. This method of aggregating information--call it expert aggregation--is different from public opinion polling, which is based on randomness. The political pollsters quiz a random sample of likely voters for their likely vote; they do not ask them for an opinion of how other people will vote, a matter on which randomly selected respondents cannot be expected to have an expert opinion. The idea behind the prediction market is that the opportunity to make money or just the fun of betting on one's insights or hunches (the only reward that the virtual prediction markets offer participants) will elicit expert opinions--more so, certainly, than random polling, which anyway, as I have said, does not ask respondents for an opinion about anyone's voting except their own.. I don't think the success of prediction markets is due to a "wisdom of crowds" phenomenon--the idea that somehow large groups of seemingly nonexpert people are bound to "get it right." The "wisdom of crowds" is really just a matter of reducing sampling error. Suppose 100 people guess the weight of a person. Some will guess too low, some too high, but the average guess will be close to the true weight. If, however, just one person is asked to guess, the chances are great that his guess will be either too high or too low. One problem with prediction markets, a problem that occurred on the day of the 2004 presidential election, is that a market can swing on the basis of unreliable information until the information is corrected. (That happened last week when the price of stock in United Airlines plummeted on a mistaken report that the airline was about to declare bankruptcy.) Exit polls showed Kerry winning a disproportionate number of the votes cast early in the morning, and immediately the prediction markets predicted that he would win the election; and of course he lost. Another potential problem with the prediction-market model is that the limits of the bets that can be placed, illustrated by the Iowa market’s $500 limit, are so low. One understands why there are limits: otherwise there would be a danger of market manipulation. Expenditures on the current presidential election campaign will exceed a billion dollars. It must be that the prediction markets attract people who derive nonpecuniary satisfaction from successful bets and that among those people are likely to be a number who really do have insight into the issues bet on in the market, since their bets are more likely to be correct and therefore they are more likely to derive the satisfaction that comes from successful betting. Probably most people who bet on horse racing think they know something about horses, and probably most people who bet on the outcome of a political campaign know something about politics. It may seem odd, though, that a stranger would have a better sense of how people will vote than a random sample of people would know, each of them, how he or she will vote. But only about half of all eligible voters actually vote in a presidential election, many people refuse to talk to pollsters, some people do not make up their mind until the last minute (but may be hesitant to reveal their indecision to a pollster), some respondents will tell the pollster what they think he wants to (or will be impressed to) hear, and the number of persons sampled is never large enough to avoid a confident prediction of a point outcome, as distinct from a range (say a 95 percent probability that one candidate's vote percentage will be between 47 and 50 percent and the other's between 49 and 52 percent). There is an interesting question whether prediction markets should be thought of as "gambling” and perhaps prohibited. As a matter of policy, that would be a mistake, even if one thinks that gambling should be prohibited. The prediction markets are markets for speculation, rather than for game-playing or risk-taking. Slot machines, card-playing, roulette wheels, and other conventional forms of gambling do not generate socially valuable information. Speculation does. Commercial speculation serves to hedge commercial risks and bring prices into closer phase with value. Political, cultural, etc. prediction markets also yield socially valuable information. The outcome of elections is important to companies and even individuals for whom particular public policies are important; they may wish to make adjustments to avert or exploit looming political change. Politicians too need to have as sharp a sense as possible about the effects on the electorate of their and their opponents' strategies. Apparently they can get more accurate information from the prediction markets than from the public opinion pollsters.

Comments

[Michael F. Martin]

Nice summary of prediction markets.

Could you explain what you mean by the following: "bring prices into closer phase with value"? (Perhaps in another entry?) Do you have a theory of how the time required for processing and communication of (even public) information to markets creates temporary shortages and surpluses in market prices. Doesn't this violate the efficient market hypothesis?

[neilehat]

Excellent! Now we have the "prediction pools" to contend with. In addition to voter fraud, hanging chads pollsters, Radio/TV Talk Show Hosts and the like. Does this really all mean that Dewey defeated Truman? Something in the back of my mind is screaming, "GAMBLER'S FALLACY"!

[Michael Hochster]

Do the studies which show prediction markets beat polls show that they do so per unit cost of collecting data? It is not so impressive if an expensive prediction market beats a cheap poll which could be made as accurate by spending an equivalent amount on a larger sample size.

[Blake]

The prediction markets should be fairly cheap, you are playing bookie, and bookies make a profit. Since you want the information and probably have money you most likely would like to subsidize it in the same way a bookie taxes the odds. Also they aren't talking about accuracy like a larger sample size fixes, they are talking about accuracy of results, so it IS impressive if they can get better results even if it costs more money. Cheap wrong info

[Matthew]

Firstly, I want to say that I am a big proponent of election markets as hedging instruments, but I think their predictive power gets overstated by simple statements like "they have predicted X," since a contract can "claim" different things throughout its long life. I don't follow IEM, but so far in the contract life of the presidential binary on Intrade, both Obama and McCain have traded at >50%, and Hillary has traded at 49% while holding a commanding lead over the field. So, any of these three could win the election, and Intrade in turn can claim to have predicted the results correctly via the free market.

Where we can see that the markets are pricing odds of winning incorrectly is in the arbitrage bounds of the markets, and in these, you can tell there are many failings in these political markets. The most simple one to observe is that there can be at most one winner of the election, so owning all candidates should be worth 100% less the cost of holding that position. However, I have frequently seen the sum of all bids exceed 100% by as much as 10%. Taking a short position in any subset of candidates such that the sum is greater than 100% is arbitrage. (Note: I just pulled up the historical graph of IEM and observed the same phenomenon with their prices on Dems vs Reps rather than bids on people.)

This, to me, indicates that the predictive powers of Intrade are not as strong as statements like "IEM [...] has correctly predicted the outcome of every presidential election since 1988" would indicate. And 10% north of arbitrage indicates the markets could be extremely far off.

[Berend de Boer]

Mr. Posner, you wrote: One problem with prediction markets, a problem that occurred on the day of the 2004 presidential election, is that a market can swing on the basis of unreliable information until the information is corrected.

Although true, you have to take into account the volume. That wasn't much volume. I and a couple of investors dived in when Bush dropped, but he never got below $33 or so, and there was never a case that we could scoop up as much as we wanted to. And it didn't last that long. But it was the icing on the cake.

[Berend de Boer]

On my last post: that was on intrade.

[Jack]

Michael: Not sure if this is the answer you're seeking.........

"Could you explain what you mean by the following: "bring prices into closer phase with value"?"

......... but consider that a product, say a home, does not sell until the value (perceived or actual) is higher than the asking price.

That's how we KNOW, as a study divulged last week, that the oil prices are due to a "market" gone awry, ie. being manipulated. Absent significant shortages the price of a bbl of oil should about the cost of replacing the bbl sold plus a reasonable profit, under the system that has served us for many years.

I guess that brings up a question as to whether we've priced oil correctly in the past by selling it only for production costs plus a small profit. Since no more is being created should oil be sold cheaply enough to burn? or should our government set a higher, intrinsic, price meant to conserve diminishing supplies to be used more sparingly in applications where there are no substitutes?

Pricing oil at higher prices would have the effect of subsidizing all of the alternatives evenly (rather than having pets such as US produced corn ethanol) and draw out the most efficient substitutes to "burn" thus saving our oil to lubricate our windmills and make resins or for other higher value uses.

[robert]

Judge:
Interesting comment on the "wisdom of crowds" phenomenon. Might it have some relevance to the actions of juries?

[Jim]

Too bad the economists at Bear Stearns and Lehman Brothers and Merrill and AIG didn't check the prediction markets. Even the pollsters would have gotten it right had they asked, "Do you think that greed and stupidity will have a negative effect in the long run?"

[Matt]

Would there really be a danger of market manipulation if there were no maximum? It seems unlikely that buying up Iowa prediction shares to change the Iowa price would be the most productive use of campaign funds. Even if there were attempts at market manipulation, there would be plenty of people ready to buy on the other side as soon as the price didn't match the true value of the shares. Any sum of money spent on political campaigns will be insignificant compared to amount people are willing to spend to make an easy profit.

[a shop]

Hello.
Please come this site.

[Robin Hanson]

There is an academic literature on price manipulation in prediction markets that, so far, suggests it isn't much of a problem.

[Thomason]

With all due respect - this is what you write about in the midst of an economic meltsdown?

[Michael Caracappa]

Judge Posner writes: The Iowa Electronic Market "has correctly predicted the outcome of every presidential election since 1988."

That assertion carries little significance without also stipulating how far in advance the prediction was made. Did the IEM predict the correct outcome a week ahead? A day ahead? An hour ahead? My own impression of the IEM, having followed it for some years, is that IEM predictions border on worthless until the exit polls start coming in on election day. Then, by Golly, it's almost always right on!

That aside, the IEM doesn't really predict. It sets a probability of victory. Today (Sept 17), the IEM gives Obama a 57% probability of winning. So in 100 election "trials," you'd expect Obama to win 57 times and McCain to win 43 times. Therefore, if McCain, with that probability of victory, were to win a particular "trial," IEM wouldn't be wrong. Neither would IEM be right if Obama were to win.

[Jim]

Each week there is an electronic poll on AOL which is state specific and has an n of about 500,000. While I do not know the exact demographics of AOL users, the site itself is fairly liberal and with a sample that large, there must be some degree of both randomness and central tendency.

The poll has for the last two months shown an overwhelming majority for McCain and no one is wagering. Either it is a total fraud or very accurate. We will soon see which.

[Jim]

Oops! Here are the demographics for AOL:

AOL Demographics

35 million members worldwide
2.7 million peak simultaneous users
400 million e-mails sent and received daily
13.4 billion Web URLs served daily
70 minutes online per member daily
More than 2.1 billion instant messages sent daily across the AOL network

Gender
Male: 43%
Female: 57%

Marital Status
Single: 38%
Married: 62%

Education
College Grad: + 51%

Age Composition
18-24: 15%
25-34: 18%
35-44: 25%
45-54: 24%
55-64: 12%
65+: 7%

Household Income
$50k+: 63%
$75k+: 40%
$100k+: 23%

Occupation
Professional/Managerial: 30%
Self-employed: 11%

Source:
http://www.aolyellowpages.com/advertisers/1/H_why_demog_aol.htm

[neilehat]

When it comes to predictions, especially elections I think the best approach can be borrowed from old Leo Deroucher, "it ain't over till it's over". That is, unless of course, the Supreme Court decides too step in and elect Presidents.

[Jack]

The demographics should explain the McCain bias on the AOL polls. HH Income begins above median so fully half of the nations HH's are left out which would include a large number of young and minorities.

Also, I'd take a small position on predicting that those defining themselves as "conservatives" today would be more prone to "vote" in a voluntary poll.

[robert]

The Supreme Court did not "elect" anyone. By a 5-4 vote it stopped the recount. But more importantly, by a 7-2 vote the Court found a violation of the Equal Protection clause. You read that correctly: two justices from the minority on the recount joined the majority as to the equal protection violations. Don't believe me? Just Google 'Bush v. Gore'.
And finally, the NY Times (that's right, the NY TIMES!) found, after its own exhaustive investigation, that George W. Bush beat Al Gore in Florida.
Now, neilehat, care to change your name
to "tinfoilehat"?

[Handmaiden to the Translational Biomedical Sciences]

I think that you are wrong to say that the wisdom of crowds is really just a matter of reducing sampling error, because I think that Galton-type experiments typically show that the average of a large number of individual guesses is better even than the best individual guess. (This was the case with his own original observation of villagers guessing the weight of an ox at a fair, and is also the typical experience of psychology or economics professors recreating the situation by having their students guess the number of marbles in a jar or some such.) Thus, by averaging the guesses, one is not just reducing the chance of landing on a widely-off guess.

[neilehat]

robert, "Tinfoil" hats huh? Sounds like something out of Tom Terrific. No wonder the World no longer takes us seriously. Just remember, on election day, to get out and vote and vote often. This is true Democracy at work and it ain't over till it's over.

[Jack]

Florida and "Equal Protection clause?" Ha! Are they still leaving the methods of voting up to each voting district? And using good vote counting machines in the "nice" areas? What's the latest on K. Harris and the very extensive purge list of "felons" who have served their time ........... and those of similar names? Aaah, yes a vote recount stopped indeed! ...under a decision not to be "used as precedent".

[neilehat]

Jack, K.Harris huh? I heard she wanted to be a Senator, but the payoff didn't materialise after the act. I wonder why?

[Chenglin liu]

Dear Judge Posner,
I use your book Economic Analysis of Law (7th ed.) in my law&Econ class at St. Mary's University School of Law. It's a great book!
I have a question on Figure 7.1. regarding war on drugs. It seems that qs and qd should be ps and pd. I would appreciate it very much if you could send me an email at cliu@stmarytx.edu
Sincerely,
Chenglin Liu