This article is the first of a series describing the variety of prediction market institutions in use. I’ll start with the simplest form of prediction markets: a double auction (buyers bidding prices up, sellers bidding down) for a simple asset that represents the outcome of interest. The asset will pay $1 if some outcome comes to pass, and will be worthless if not. This article describes three variations on this basic model.
The calculation of value for the buyers is simple: if the asset’s expected value is higher than the best price asked by a seller, then they should buy. Buyers who are not worried that new information will change the value before they can withdraw their offers should be willing to leave standing orders that prospective sellers can accept.
The calculation for the sellers is more subtle. Those who believe the asset’s expected value is lower than the best current price offered by a buyer should sell short, accepting the current bid price in exchange for a liability that they may have to pay back later. Fortunately, the liability is limited; the most they will have to repay is $1 for each share. The market operator will usually have a mechanism for managing their cash to ensure they will be able to pay back the liability if the outcome goes against them. Until the maturity date of the asset, they have the proceeds to invest as they think prudent.
That’s the starting point for Prediction Markets. Let’s call it the “short selling model”, since that’s the most salient feature. It’s sufficient to model markets that provide predictions (by publishing the market consensus of the estimated probability that an event will occur), encourage research, reward insight, and allow insurance and hedging of exposure to various risks.
The first variation defines two complementary assets in place of the original one. The second asset pays $1 when the first is judged at 0, and pays nothing when the first is decided to be worth $1. This variation was first proposed by Robin Hanson in 1988. This version makes the process symmetric between the seller and the buyer. They both spend money to buy assets that may pay off at some date in the future.
The short selling required in the first model can be a psychological barrier to some traders, who may not see that it differs from short selling in the stock market. In the stock market, the assets have unlimited growth potential (consider Google), which means that the liability can also grow without bound. Prediction market assets have a known maximum payout, which means the maximum liability is limited as well. While the risks and reward are the same in both versions, people’s unconscious aversion to short selling may make them more hesitant to sell than to buy.
The variant (call it “buying complementary assets”) has both sides buying assets, so it sidesteps the psychological barrier. Another variation (called “buying a basket of assets” or just “buying a basket”) also uses complementary assets to eliminate short selling, but selling doesn’t look the same as buying. In this version, the market is always willing to buy complementary pairs from traders (or sell them) for $1. Since one or the other of the assets will pay $1, this is an equal trade. The difference in this model is that a trader who want to bet against a position does so by buying a pair of assets, and selling the unwanted position. Buying the pair and selling one asset has the same effect on a trader’s holdings as buying the complementary asset. Since the difference can also be hidden in the user interface, the only advantage I know of for the “basket” model is that in a play money market, users can start out with balanced baskets of assets to encourage trading.
After a trade, the buyers in the three variations all have the same portfolios: if the price was p, their cash is reduced by p, and they have one additional positive asset. The sellers in the complementary asset and basket models are in a similar position: their cash is reduced by (1 – p), and they have an additional positive asset (though of the complementary position). Sellers in the short selling model is the only one that’s different: their cash is increased by (1 – p), and they are “short” one unit of the asset. If the asset ends up in the money, they owe that money to the market. Of course, after taking any of these positions, traders can sell or buy the same asset in order to close their position. They can do this to lock in gains, or to cut losses, but either way, they’re contributing to the price-setting activity of the market.
TradeSports (aka InTrade, TEN) uses the short selling model. Since they come to Prediction Markets from the betting world, they use the short selling as a way to keep more money in their customers’ accounts. When you sell short, your balance increases. The software reserves margin funds as assurance that traders will be able to cover their short positions, but it gives a generous allowance for claims that won’t settle for a while, and for offsetting positions in each trader’s account. Since traders are often more willing to bet in favor of a proposition than against, this added incentive to sell is probably a good trade-off for TradeSports.
The Iowa Electronic Markets (IEM) and NewsFutures follow the “buy a basket” model. On IEM, there is an explicit step (and a separate screen) for buying baskets of assets. Traders who don’t master this step can only buy assets, but can still bet in either direction, since there are separate markets in the different assets. I’ll talk later about why separating the markets is a mistake.
On NewsFutures, the explicit step of buying the basket has been hidden in the common case where there are only two outcomes; it’s only visible in multi-position claims. NewsFutures doesn’t currently have any multi-position claims, so I can’t verify this. (I’ll talk about markets with multiple outcomes in a later posting.) The Foresight Exchange follows the simpler “buying complementary assets” model.
Market operators who depend on trading for their profits want to encourage more trading. One of the obstacles is that traders are relatively unwilling to leave orders on the books (some estimates are that 80 percent of traders seldom or never leave standing orders.) If the market operator can encourage traders to leave more standing orders, there will be narrower spreads, and more trading opportunities. IEM is non-profit and NewsFutures and Foresight Exchange use only play-money, so they don’t take extra steps to encourage book orders. TradeSports uses its fee structure to encourage book orders. Only the person who accepts an offer out of the trading book pays a transaction fee. The trader who left the standing order trades for free.
From these basic market foundations, variations have been developed that support asking complex and interdependent questions, that support questions the basic mechanism can’t answer, and that predict well even when there are few traders. I plan to talk about all the variations in later posts.