Summary | Rationale | Powerpoint | JPG Slide | JPG Image | Credits | Comments | All Nuggets
Computational Properties of Prediction Markets
How does the crowd get its wisdom?
Summary
Prediction markets use bets on the outcomes of events and have been quite successful in harnessing the "Wisdom of Crowds," giving mostly accurate predictions both in popular events (Oscars, sporting events elections, sporting events) as well as many corporations using them internally to predict sales and the implications of new features. These tools have become so valuable that the Commodity Futures Trading Commission has recently asked for public comments for regulations of these event contracts.
How do these markets aggregate information so well? We can create computational models of these markets to determine what kinds of knowledge these markets can help us understand. Research in this direction will help us develop contracts that will maximize the accuracy and usefulness of the results while limiting problems of manipulability and limited volume.
Rationale
A typical market, such as found on intrade.com, will have a security such as 2016.OLYMPICS.NTH.AMERICA that will pay off $100 if 2016 Summer Olympics are in North America and $0 otherwise. One can show that a risk-neutral user who believes the probability of this even is say 33% would buy this security if the price were below 33 and sell (possibly short) this security if the price is above 33. One could thus use the price as an aggregate predection of the probability that the event will occur. Considerable analysis of past markets have show the surprisingly accurate value of these predictions. If you rounded off the predictions, Intrade accurate predicted the outcome of every state in the 2004 electoral college and the 2008 senate races.
There markets aggregate disparate information over its many users to compute these probabilities similar to parallel and distributed computation mechanisms. Thus theoretical computer scientists can help develop accurate models of these processes that would lead to the development of securities that will improve both the quality and quantity of information received, help limit the effects of market manipulability and give accurate results even if there is low liquidity in the markets.
Contributors and Credits
Lance Fortnow
Image Ideas
or if you need one that I have rights to:
Comments
- to give feedback on this nugget, just add another bullet to this list.
- I expanded "CFTC" and embedded the link into the phrase "public comments". - Salil
- Would be nice to mention TCS in the summary, and if possible, make a stronger case for the role of TCS in this topic (eg by some recent successes, or connections to other things we've done). - Salil
- The tagline seems a bit off-topic to me. Research in prediction markets is more about how to gather the information of the crowds, rather than the philosophical question of why crowds should be good at predicting things. How about changing it to "How can we harness the wisdom of the crowds"? The pictures are really great! - Shuchi
- Explain better the role of TCS. It would also be nice to have a specific example of what one would hope to achieve here (what is the grand goal?) - Valentine
- I disagree with Suschi. Understanding how the crowds predict is a computational question not a philosophical one. I made some minor changes and added a sentence about computation in the summary. I feel funny talking about my own work in the area (where do give a model of simple prediction market) and don't think we should be too technical either. - Lance
- The role of TCS definitely comes out more clearly now. Perhaps the second par of the rationale could go a bit further in this direction by starting along the lines "The process by which prediction markets compute these probabilities seems to share many features with previously studied computational models, such as XXX [listing features and/or relevant computational models]. Thus, theoretical computer scientists can help..." - Salil
- Minor suggestion: "models of these markets based on computational models" -> "computational models of these markets" - Salil
- Note from designer Elaine Park: I was surprised to find a dearth of images of dollars in fists, which I thought would best represent betting. I did find an image of a racetrack, and one of a silhouette of crowd, and I thought it might work if I went in and drew the dollars there, but I'm afraid it just might get messy. As a proposed alternative, here is a prediction chart that should be instantly recognizable to everyone.
- I edited along the lines of Salil's last two suggestions. I like Elaine's B-Ball chart. - Lance
- The basketball tournament chart might be too much of a boy thing. Unfortunately, I can't think of anything better. -- Richard