Large markets can apparently be seen as a predictive (or perhaps decision-making?) apparatus. The predictions made by markets are apparently better than the voted predictions of individuals and better than the predictions of the best-predicting individual.

An InformationDerivativeMarket is a market of derived securities (see FinancialDerivative?) whose payoffs are tied to outcome of real world events. If large numbers of people actively trade these securities, the prices that evolve can be used to predict the relative likelihood of the outcomes of the real world event to which the payoffs are tied.

For example, the Iowa electronic markets accurately predict U.S. Presidential elections. The Hollywood market predicts the revenue of movies better than the studios. I have heard that the futures market for orange juice predicts the weather for a county where most American oranges are grown better than the U.S. government’s forecasting service.

People at are working on adapting market techniques to work in small-scale situations where there are too few people to make a liquid market (e.g. a corporation trying to make sales predictions for next year). So far they have a system which gives very good results in the lab with about ten people when there is no public information. They are working on making it work when there is public information, and also working with Hewlett Packard to make actual predictions.

Some implementation issues include things like the moral hazard. For instance, how do you make it so that sales managers don’t try to sell less than they can in order to match their own predictions if they predicted too low accidentally?

I think this is a neat take on the idea of researching better ways for human groups to make decisions.


public exchanges:

real money

play money

both play money and real money

Other links

An academic discussion paper on using TradeSports? to analyze other financial markets (specifically, using the correlation between a Saddam security in TradeSports? and indicators in other markets to analyze the amount by which those indicates were affected by predictions about Saddam): Leigh, Wolfers Zitzewitz: What do Financial Markets Think of War in Iraq?

Specializations of the idea

note: this idea is sometimes called “information market” (example). I call it this because after reading a comment by SunirShah, I think that term is too broad (consider a market where people bought and sold information from each other).



Maybe it’s just the examples I’ve been looking at, but I’ve never been impressed by the futures markets; They always seem to predict accurately about a week before event X.

That is, I’ve frequently seen them be way off, until it was immanently obvious what the outcome would be, (to just about everybody.)

The event happens, and then the futures prediction market people go, “Would you look at that– we predicted it!” Wow! Look at the percent of times we accurately got it! WHOAH!

I’m thinking specificly of Howard Dean election outcomes, though I don’t have the link on hand.

Basically, I just see these markets responding to news that is available to everybody, and not some “secret Libertarian sauce” made of perfectly rational actors meticulously commiting to research.

But maybe I’m just looking at bad examples.

I think there are studies that show that some forms of information derivative markets predict better than the best individual in the market, and better than just taking a poll of all of the participants. But I don’t have time to read and cite the literature right now. If I did, though, I’d start with this paper, not because it’s the best but because it’s the first paper on this topic that I found after a quick search just now; I assume that by following its citations and finding who cites it, you could find the other relevant papers:

KY Chen, LR Fine, BA Huberman Predicting the Future

Another link. Researching.

My feeling: Depends on when you sample the market vs. poll the people.

Define external redirect: TradeSports FinancialDerivative CategoryMarketBased

EditNearLinks: SunirShah


The same page elsewhere: