First run as the Iowa Political Stock Market for the 1988 presidential election by Robert Forsythe, Forrest Nelson, George R. Neumann, and Jack Wright (Anatomy of an Experimental Political Stock Market, 1992), proposed as idea futures by Robin Hanson (1990), and surveyed by Justin Wolfers and Eric Zitzewitz (Prediction Markets, 2004), after Wall Street betting on presidential elections from 1868 to 1940.
Lets people trade contracts that pay out if a defined event happens, so the trading price becomes a running estimate of the event's probability.
An internal market on whether the new product ships by March, read as a running probability
Each contract pays $1 if the product ships by March and nothing if it slips, so its price reads as the market's chance it ships. The other traders react to three pieces of news; you trade after each one. Buy when you think the price is too low and sell when you think it is too high: every trade moves the price, and the more you trade the more each contract costs. The thin black line is the price without your trades. Illustrative.
With your trades: price in cents, read as percentWithout your trades
An example
A maker of industrial sensors ran an internal market for its new controller: one contract paying a play-money dollar if the product shipped to customers by March 31, open to the 140 people in engineering, operations, and sales. The program office's status report said on track.
The market opened at 50 cents. When the beta slipped two weeks, traders sold it down to 31. A test engineer who knew the slip came from packaging, not firmware, bought twenty contracts and pushed it back to 40. A fix from the chip supplier lifted it to the mid-50s, and a clean final test to about 67.
Two-in-three was well short of the status report's green. The vice president of engineering asked the market's most active sellers, anonymously through the market's comment board, what they were seeing. Three pointed to the same thing: the certification lab's queue, which no status report tracked. She booked a backup lab that week, and the launch event moved from mid-March to early April as a hedge.
Why it works
The people closest to a project know things that never reach the status report: a vendor's tone on a call, a test that keeps failing intermittently, a queue at an outside lab. Asking them in a meeting gets the official answer, because nobody wants to be the one who said it would slip. A market lets them say it with a trade, privately, and pays them for being right, so the price gathers what many people know into one number.
Because a contract pays $1 if the event happens, a price of 67 cents is the market's estimate of a 67 percent chance. The price also moves as news arrives, so it tracks the estimate over time, and its gap with the official forecast is itself a signal worth chasing. The method needs enough traders, a question with an unambiguous answer, and stakes that matter to them.
The thinking behind it
Robert E. Forsythe Robert E. Forsythe, a University of Iowa economist, wrote with Forrest Nelson, George R. Neumann, and Jack Wright Anatomy of an Experimental Political Stock Market (1992), in the American Economic Review. Their market, the Iowa Political Stock Market, let traders buy contracts on the 1988 presidential vote with real money, for teaching and research, and it forecast the result better than the opinion polls. It became the Iowa Electronic Markets.
Robin Hanson Robin Hanson, an economist at George Mason University, proposed idea futures in Could Gambling Save Science? Encouraging an Honest Consensus (1990), a paper for the Eighth International Conference on Risk and Gambling in London: markets in which people bet on disputed claims, so that prices gather what they know. He later designed the logarithmic market scoring rule, described in Combinatorial Information Market Design (2003), in Information Systems Frontiers, which many prediction markets use to set their prices.
Justin Wolfers Justin Wolfers, an economist, wrote with Eric Zitzewitz the standard survey of the field, Prediction Markets (2004), in the Journal of Economic Perspectives. It explains how the prices of contracts that pay off on future events gather dispersed information into forecasts, and it sets out when those prices can be read as the crowd's probability that the event will happen, the reading this page's figure relies on.
How to run it
- Write a question with one answer. Name the event, the date, and the source that will settle it, so no trader can argue about the payout.
- Set the contract. Pay one unit if the event happens and nothing if it does not, so the price reads as a probability.
- Recruit the people who know. Open it to everyone with a view of the work, across functions and levels, enough of them to keep it trading.
- Protect the traders. Keep positions anonymous and make clear that no one is judged on what they bet.
- Make the stakes real. Use money, play money that buys prizes, or a public leaderboard, so being right is worth something.
- Watch the gap. Compare the price with the official forecast and ask what the traders are seeing when the two diverge.
- Settle and keep score. Pay out on the stated source, and track how well the market's prices called the outcomes over time.
Where it goes wrong
The market is too thin. With a dozen traders, one person's mood sets the price; spot it when the price jumps on single trades with no news. Widen the invitation, or add an automated market maker.
The question is vague. A contract on whether the launch is a success gets settled by argument, and traders stop trusting it. Rewrite every question so a stranger could settle it from a named source.
The price threatens someone. A manager who sees a market betting against her project may lean on the traders, or the market may become a way to embarrass a team. Agree in advance what the prices are for, and keep them away from performance reviews.
When to use something else
- Delphi Forecasts and Predictions. When too few people know enough to keep a market trading, and a structured panel of experts will do.
- Wisdom of Crowds Estimate. When one independent guess from each person is enough, with no need for a price that moves over time.
- Superforecasting. When a few trained forecasters will give and update their own probabilities.
Where it sits in SOLVE eX
Go to the source
- Anatomy of an Experimental Political Stock Market, American Economic Review 82(5); the market ran for the 1988 election, 1992
- Could Gambling Save Science? Encouraging an Honest Consensus, Eighth International Conference on Risk and Gambling, London, 1990
- Prediction Markets, Journal of Economic Perspectives 18(2), 2004
- Historical Presidential Betting Markets, Journal of Economic Perspectives 18(2), by Paul W. Rhode and Koleman S. Strumpf, 2004