Defining prediction market contracts

A lack of thought in the definition of prediction market contracts can cause problems when the time comes to settle the market.

In the early hours of Wednesday July 5 2006 the Democratic People’s Republic of Korea (that’s the one that isn’t democratic, or a republic, although it is in Korea) fired several medium and long-range missiles which all landed in the Sea of Japan.

The prediction market site, TradeSports, was running a market on whether North Korea would conduct a missile test with the missile landing outside its territory by July 31 2006. However, even after the government of North Korea had announced the test and the New York Times had reported it, TradeSports ruled that the conditions for paying out contracts had not been met. The issue was that the contract conditions specified that the United States Department of Defense had to confirm that the launch had taken place. In practice, because the U.S. considers monitoring foreign missile tests to be the remit of its intelligence agencies, it was the Whitehouse that acknowledged the test and the Defense Department never made a statement. People holding contracts were annoyed but TradeSports stuck to its strict interpretation of the conditions.

TradeSports was not a conventional bookmaker but a platform for people to trade contracts with each other, so this wasn’t a case of TradeSports wanting to avoid having to pay out. Indeed, had they ruled the other way, there would have been a group of equally annoyed market participants who had sold contracts and were now liable for the pay out. In hindsight, the controversy would have been avoided if the contract conditions had specified confirmation by the U.S. Government, without prescribing which bit of the government.

On Hivemind’s Agora platform it is straightforward to create and configure new prediction markets. The danger is that this ease-of-use might detract attention from the task of designing the market and carefully defining the events or variable that will be used to settle it.

When people challenge each other to bet on something the next step is usually to pin down the terms of the bet. Similarly, designing a prediction market forces you to clarify what you really mean. For example, some organisations have used prediction markets to predict whether a project will be completed on schedule. When creating such a market they have to define what “completed” means. What if the project delivers on time but only after de-scoping? Addressing these issues can be a valuable exercise in itself, independent of the information that the market may generate.

The appraisal of possible outcomes should be as exhaustive as possible because unanticipated developments can create ambiguity. In 1994 the Iowa Electronic Markets (IEM), a political prediction market run by the University of Iowa, was running a market on that year’s midterm elections for the U.S. Senate. One of its markets had three contracts that would pay out $1 times the share of Senate seats held by Democrats, Republicans or “Other”. On the eve of the election the Democrat and Republican contracts were trading at 0.466 and 0.509 respectively.

Immediately after the election on November 8 the Republicans had won 52 seats to the Democrats’ 48. The following day, before vote counting was complete, Richard Shelby, the Senator for Alabama, switched from the Democratic Party to the Republican Party. Shelby’s seat hadn’t even been up for contention in the election (Senate terms are six years with one third of seats being contested every two years). Shelby’s defection increased the share of seats held by Republicans from 0.52 to 0.53. Given the size of the change and the fact that participants in the IEM are limited to $500 stakes there wasn’t much point in people lawyering up, but had there been more at stake it could have ended in court.

Hivemind’s Agora platform can be used by organisations to run their own markets. There are also blockchain based prediction market platforms, such as Augur, on which anyone can define a market. Inevitably this has led to a proliferation of badly specified contracts. One example asked, “Will the weather be good for the Bastille day military parade in Paris tomorrow?” Settling a contract on “good weather” is obviously fraught with difficulties. A spectator at the parade might regard hot and sunny as good but a soldier participating in full ceremonial dress might equate good with several degrees cooler. If you think the Bastille Day weather market was ill-defined you probably won’t be participating in the market for “Does god exist?”

As well as allowing anyone to create a prediction market, Augur also differs in the way markets are settled. Market creators specify a reporter to determine the tentative truth but participants can dispute this truth by staking a platform specific crypto-token, REP. Crowdsourcing “truth” this way risks eroding the distinction between fact and opinion and encourages epistemological relativism. A well-designed prediction market should have a well-defined specification of the underlying event. This specification will often rely on a relevant authority and the selection of this authority should constitute part of the contract definition. If you don’t trust the specified authority to accurately call the event then don’t participate in the market. Prediction markets built on blockchains make use of “oracles” which are external data feeds that a smart contract can check to see if the conditions of the contract have been met. Of course, if errors occur in this data feed contracts might be settled incorrectly.

Another issue might occur if there is a change in the relevant authority. For example, the U.K. Retail Price Index (RPI) was first produced in 1947 by the Ministry of Labour which became the Dept. for Employment and Productivity in 1968, then the Dept. of Employment in 1970. Responsibility for RPI was passed to the Central Statistics Office in 1989 but this became the Office of National Statistics in 1996. The ONS still publish RPI although it hasn’t been classed as a “national statistic” since 2013 when it lost this designation due to concerns about an inherent bias in the way that it is calculated. Ideally prediction market contracts, particularly those relating to longer time horizons, should have provisions for contingencies like this but situations not covered by the original terms might still arise. To deal with these cases it is worth stipulating who will make the final decision when interpretation of a contract is not straightforward.

An authority’s version of truth may evolve over time. The first published estimate of official GDP is based on partial data. This number may be revised as more data becomes available. The potential for revisions even exists in traditional sports betting where winners have been disqualified and also-rans elevated to the podium weeks, months and even years after competition. Sports bookies avoid this problem by settling bets based on standings at market close because allowing for revisions after long delays and trying to reclaim money from some bettors would be a logistical struggle. As with the choice of authority, this arrangement constitutes part of the deal that participants are signing-up for. Similarly, when running a prediction on a published but revisable statistic, such as GDP or unemployment, a date should be specified. The market should be settled based on the published value of the statistic as of that date.

To help navigate the issues involved in designing prediction markets Hivemind will offer advice to organisations who want to use Hivemind Agora to run their own markets. We want to maximise the chance that these markets will address the question the organisation wants addressed and obtain the information they require.

Related posts:

Ma ei ole kohanud inimest, kes ütleks, et ta viitsib ja tahab ja suudab alati ja alati minna jooksma või mõnda muusse treeningusse. Alati on hetki, kui tuleb peale see paha “sõber” EI VIITSI. Mis…

Nothing highlights how far I exist from traditional masculine qualities quite like people’s surprise when I’m able to express emotions comfortably. As a straight, 6ft 5in tall man, I can only guess…

Beginner-friendly Writeup of the room Introductory Researching from TryHackMe with answers. “Introductory Researching TryHackme” is published by LeArN@lEaRn.