New York Summit Recap, Part 2
Andersson mentioned many intriguing possibilities, including markets tied to merger & acquisition events, equity earnings, credit rating actions, credit defaults, bankruptcies and IPO announcements. There seems to be demand out there for such contracts despite the fact that they will be correlated to existing instruments. Hedgestreet also has an eye on economic data release derivatives. Lastly, Hedgestreet is working with external market makers in an attempt to make their markets more liquid, but doesn't have plans for automatic market-making.
Emile Servan-Schreiber, CEO of NewsFutures mentioned a series of corporate internal prediction markets, including Eli Lilly markets on health insurance costs, drug benefits, and access to physicians. Emile made a joint presentation with Bob O'Brien of Corning on a prize-based liquid crystal display market. The market aggregates information across the LCD supply chain, from component makers to TV manufacturers to retailers, and it will be interesting to see if its use reduces the volatility of LCD prices. Corning is also running a conditional market on price elasticity, i.e. LCD demand is projected given various price ranges. O'Brien expressed some concern that the markets might be exploited or contaminated by competitors, and said that suspect traders might be weeded-out over time, although specific orders were confidential. In addition to rewarding accuracy, the market's scoring rule takes into account certainty and timeliness. Only the top 3 traders in each "tournament" receive prizes.
Servan-Schreiber was eager to cite his research with David Pennock and others on the predictive accuracy of play money versus real money markets. In short, real money markets were not found to be significantly more accurate. I would add that in some cases, play money markets might actually be more accurate, as they are not subject to structural factors like hedging or interest rates. Of course, then the problems are what will compel participation and information discovery? Emile gave two suggestions on encouraging participation in play money markets. First, participants should value the information; it should be intrinsically relevant to them. Second, the interaction with other traders is important, especially the recognition that comes from successful trading. The second piece of advice might be somewhat problematic from a predictive standpoint, as traders motivated by reputational pay-offs will tend to be risk-loving, and this sort of behavior is among the litany of explanations for the favorite-longshot bias. The same can be said about discontinuous prize-based payoffs.
Citing another paper by David Pennock, Servan-Schreiber mentioned that simple averaging of predictions is not very different from averages weighted by prior trader performance. This is somewhat surprising, and perhaps discouraging for the neural-network-based Owise, which apparently had no representatives at the conference. Now, real money markets to a large extent do weight previous performance, as successful traders can redeploy their winnings.
Additionally, play money markets in corporations might have real money implications. As Bo Cowgill appreciates, this is another obstacle to running internal real money markets as they might in some cases be a reasonable proxy for a company's publicly traded equity.
David Pennock of Yahoo Research Labs reviewed his dynamic pari-mutuel markets, and described ways to maximize liquidity in combinatorial markets with generic bidding languages. Standard pari-mutuel auctions encourage late betting and do not allow for position liquidation. Therefore, it's not possible to "buy low and sell high" as/when information is incorporated into the market. Dynamic pari-mutuel markets overcome these limitations with a price function that reacts to incoming information, and a side auction where participants can hedge away their positions.
Markets that allow for combinatorial bets are problematic because with n events liquidity is potentially spread-out over 2^n outcomes or 2^2^n bets. Of course, the market will likely only need to trade a very small subset of these outcomes. A generic bidding language would allow for bets on any specific logical outcome, e.g. "I win $1 if (A and not B) or C", but then the auctioneer must decide how to match trades, and will likely choose to maximize trade by matching only those orders that leave the house with no risk.
James Surowiecki, author of The Wisdom of Crowds, interprets appeals to the "marginal trader" as a manifestation of reliance on experts and the desire for centralized decision-making. While he believes there are certainly experts in the world who are good at predicting, it is very difficult to know who they are ahead of time for specific questions, and so-called experts are notoriously over-confident in the accuracy of their predictions. (One factor may be that their expert-hood compels them towards further reputational pay-offs.) Surowiecki also described a variation on the jelly-beans-in-the-jar experiment, where after a certain number of rounds the best guessers were grouped together and played more rounds as a group — but these "expert" groups performed no better than groups assembled at random. This case is perhaps an example of retrospective overfitting, whereby some guessers just happened to do well on a previous trials. Surowiecki, like most prediction market enthusiasts, is acutely aware of what kinds of crowds tend to fail and went over the basics of what makes a good predictive group, stressing independence, diversity, and low to negative correlation in biases.
Thomas Malone of MIT argued that advances in communication technology are broadly responsible for the long-term evolution of government and the spread of democracy, and projects similar decentralization in the corporate world. However, communication technology can also abet centralized authority, and it seems that its impact must be assessed on a case-by-case basis. For example, would Charles Schwab have enjoyed so much success if existing brokers could have cheaply deployed retail-targeted trading interfaces via the internet?
Finally, it's been one week since the conference and legal issues might already be heating up in the US. During that week, Intrade's PredictionX market on whether the prediction market industry will suffer a "legality shock" in 2006 has crept from the high single digits into the teens, while the contract for "major political validation" in 2006 has fallen from around 50 to the 30s. Validation is more likely in the UK, although one conference participant claimed that "three letter agencies" in the US were very interested and open-minded about prediction markets.
Some operations seem to be rather illegal, but since the law is vague and archaic, entrepreneurs may feel that the cost of not doing business outweighs the probability and degree of legal punishment.