Political prediction markets have steadily gain popularity. However, it was not until this year when the prediction markets entered the mainstream. The market prices on the outcomes such as the results from the primaries and Brexit vote were treated equally, if not more valid than the opinion polls. This went so far even to imply that there is something wrong with the polls that would suggest a different outcome. After all, the markets must know something that we don’t, right?
This question got its answer as the votes were tallied. The political prediction markets have failed spectacularly. Not only that the markets missed the outcomes, but they were completely misleading. What was obvious to the pollsters, was not clear to a sophisticated pool of market traders. How come? The experience in using prediction markets for risk assessment of project outcomes may provide some clues.
Prediction markets are about disclosing private information in the form of a trade. This is important to remember. So, if one structures the outcome “stock” to be in a different domain then the available information, then such information becomes irrelevant. For example, in project prediction markets the outcomes must match the “traders’ information domain; project managers do not place bets on the outcomes of work packages, neither do work crew staff place bets on the outcomes of the bidding process. In other words, unique insights of the traders and the definition of the outcome i.e. “stock” must match. So, are the traders in political prediction markets statisticians that can interpret changes in polling numbers? Experts in sociology that can analyze social trends? Probably not. They are likely a biased sample of a segment of the population; a pool of upper and middle class overeducated individuals. In fact, while they were wrong in predicting the outcome of the overall vote, they were correct in predicting the outcomes within their demographics. Again, the insights must match the definition of the outcome. If there is not relevant private information, then traders start relying on the market price data as the source of information - creating typical herd-type behavior and feedback loop as described in a recent Slate article.
However, the problems do not end there. One of the most significant issues in application of prediction markets in project risk analysis is the managerial control problem. The traders do not only consider the likelihood of an outcome, but also the capacity/willingness of the management team to address it proactively. In other words, the betting odds were driven by the confidence that the "manager" would prevent the unfavorable outcome. For example, if the outcome is schedule overrun, then a significant delay in some of the activities on the critical path may prompt project management team to accelerate subsequent activities in order to make the deadline. Something similar was going in political prediction area; ‘the manager” was perceived to be in control. That signal has overshadowed other more relevant information coming from the polls. To be even more ironic, “the manager” took the prediction market prices as a signal that no further actions are required, creating a perfect reinforcing loop in which the reality fails to be acknowledged.
In spite of these and similar difficulties such as trading volumes, the number of traders, the total available funds and others, the markets could provide a great tool for forecasting and risk assessment. In fact, in project management applications many of these methodological issues have been already addressed. Hence, there is no reason why this should not be the case with political prediction markets.