(Cross-post to IFTF's FutureNow blog.)
I had been co-administering the prediction market for the Open Ex program at IFTF and pulled together a summary for people interested in Prediction Markets. The IFTF market used InklingMarkets.com as a platform.
These ideas are pulled predominantly from the experience of running the market, rather than participating in the market.
(1) Don't buy-and-hold. Prediction markets are less enticing the further into the future the event might be. This is (obviously) because of staleness and the degree of uncertainty as a function of time, but also because a long-term market will hold someone's money trapped for a long time if they're "buy and hold." In other words, given no transaction costs for trading in and out, it's a poor strategy to buy and hold.
(2) Think ahead about how to prove something HASN'T happened. As a market administrator, you have to be very careful when constructing a market to clearly state winning conditions AND, in some cases, sources for establishing winning conditions. If you need to determine that an event has happened, since it's impossible to determine it *hasn't* happened I'd suggest requiring that the event needs to be reported in a limited number of specified, searchable venues. This might be a list of news sources, or even as specific as keyword search engine. Another route would be requiring that the players report on an event happening as proof that it occurred ( i.e. let them participate as news miner-researchers).
(3) Teach Game Strategy to Boost Confidence. There are two main flaws associated with trading. The first is given in this example: If I believe that Obama's going to win the election, but I think it's about 60% likely and the market's trading at 95%, then I should SELL to bring the percentage in closer to my prediction. However if I sell and the market closes with Obama winning - as I expected him to do - I would lose my trade. So the market force here actually pushes the market to an extreme price - 95%: the only people who buy at that price would think that 95% is an okay price. People who believe he'd win but aren't quite so certain, do not sell because they'd be betting against their actual prediction.
The game strategy that would fix this would be to SELL at 95, but make sure that you cover your sell before the market closes regardless of the price.
Alternatively if you were constructing a prediction market software application, you could enable bids to be put in directly for any prediction, but the odds would be different depending on the existing distribution of the predictions. (This would be more of a betting market than a prediction market.)
The second flaw is the assumption that people will be unmoved by mob assessment. Obviously the predictive value has been shown repeatedly in liquid markets with a lot of information and consensus building done outside the market (e.g. with elections, sporting events, etc.). However I would imagine in a small market without information, participants might second-guess themselves. This could be a good thing if there is also a way to engage conversation; it could be a bad thing if accurate voices belong to people who lack self-confidence, such as people who rely on their intuition but who function in a strongly analytical environment.
There is an argument to the contrary, that in order for the market to be accurate there needs to be arbitrage players who are acting in uninformed ways and for reasons like entertainment, etc.
Recommendations for Market-Makers
(1) Players MUST trade into and out of the stocks frequently in order to ensure the markets are valid, rather than tending to show only the extremists' views. This should be formally incentivized. One way would be that the winner is the person with the highest *weighted* score, where the weight is related to the number of trades placed. (Some serious thought would have to go into that.) Players must also be explicitly taught how to sell an overpriced stock and cover before market close.
(2) Beware insider trading. There should be conversations outside of the markets regarding the relevant news so that players can feel that they have some basic understanding of the issues. When I set up markets in the game I tried to consistently refer to at least one major news story so that players could get some context to begin with. If there were more time available, a daily email news feed on all open market topics would be a great help if the markets aren't limited to highly visible news (e.g. the French Presidential Election).
(3) New players probably don't understand how quickly scores can change. Because trades are settled for $0 or $100 (to reflect the 0 or 100% probabilistic outcomes) it's not like real-life stock trading; if you Buy for 51 and you're correct, you make 49 on each share. You can nearly double your balance in one successsful trade. (Or lose it.) The leaderboard should be emailed frequently, and top winnings on each *trade* should be publicized . "Fortunes" are made and lost quickly - this needs to be emphasized. "The game isn't over until it's over."
I'm very interested in how to best capture the meta-market.
(1) Options Markets. In financial markets, options are a way of pricing the expected future value of a security. An options market with a rule excluding people from engaging in both the options market and the underlying stock market simultaneously might enable a long term forecast in a short-term timeframe.
For example: Who will win the Nobel Prize in 2008? might have an associated options market: Will "other" in "Who will win the Nobel Prize in 2008?" be under 25% by midnight, Oct 31, 2007?
(2) Contingency Markets. I'm also interested in contingent markets for this purpose. A "contingent market" might be a pair of related markets: One market states that a thing happens, and you're paid the price of something if it does happen and refunded otherwise; one market states it doesn't happen and you're paid the price of the same thing if it doesn't happen and refunded otherwise. The difference in the prices is the markets' expectations of the thing happening on the price of the item.
An example: What will be the Net Revenues of a company if the CEO is fired? and What will be the Net Revenues of the same company if the CEO isn't fired? The difference in the pricing will show you the expectation of the CEO leaving on the Net Revenues.
In this way you could have a contingent market, stop it when the action is either taken or not (a definitive short-term action) and then return to benchmark at the end of the period of time -- let's say one year -- what has occurred. (Presumably everyone who has participated will have use of the funds in the interim and the settlement will really be an adjustment a year later.)