The Wisdom of Profiteers: Prediction Markets and Democracy

intrade gambling or prediction market 2

[Editor’s Note: For more of Cherie Braden’s analysis of new financial instruments/developments, see Bitcoin Under the Liberal Microscope: How It Really Works and Why Does It Matter]

Despite the ever-present abundance of contrary opinions on what various markets will do (Should you short sell gold this week? Is bitcoin a bubble? Is there a bond bubble? Is the Euro safe? Is my financial fly down?), there’s one thing we know for sure: speculation itself never goes out of style, and any market inevitably consists of a collection of investors who are “in it to win it,” whether they aim for internal profit through a boom or collapse or external profit through market manipulation.

Where the assets on which stock markets trade are businesses, prediction markets (also called information markets) trade on an information asset: they query investors about the likelihood of an event and assign value to possible outcomes. Prediction markets have a proven potential to yield more accurate forecasts than individual experts for both short- and long-term queries (see “Accuracy and Forecast Standard Error of Prediction Markets” by Joyce Berg, Forrest Nelson and Thomas Rietzas), and a number of economists over the past decade have joined the bandwagon in wanting us to tap the predictions of these markets for guidance in making policy decisions. (For a detailed analysis of why prediction markets produce accurate results, see Puong Fei Yeh’s “Using Prediction Markets to Enhance US Intelligence Capabilities”.) A 2008 issue of Science included a sort of open letter to the US federal government requesting “modest reforms” to “facilitate the use of prediction markets [on U.S. soil] while still meeting the legitimate concerns of lawmakers and regulators,”  and the letter had some awfully big names in economics as signatories, Nobel Prize winners Kenneth J. Arrow, Vernon Smith, and Thomas Schelling among them (“The Promise of Prediction Markets”).

Companies have used internal prediction markets to crowdsource their employees in business decisions (Google, for example—getting around federal gambling restrictions by supplying employees with funds specifically for investment in said markets). DARPA’s short-lived (2001-2003) FutureMAP (Futures Markets Applied to Prediction) program sought to use prediction markets to improve strategic intelligence. (The program was canceled in response to an outcry concerning the questionable morality of betting on bad things happening.) In more recent news, a company called CrowdMed has launched a beta for a service that would aggregate the opinions of a board of “medical detectives” to diagnose patients/clients whose illnesses have eluded traditional in-office detection, using “patented prediction market technology.”

But feedback on the use of prediction markets isn’t fully positive at this point. Intrade, a for-profit experiment headquartered in Dublin and begun in 1999, has had more downs than ups lately (their home page alone suggests insolvency is likely unstoppable). Intrade officially stopped allowing US investors last November in response to a CFTC lawsuit, but no doubt many traders got around the ban using Tor. Aside from some obvious dangers in investing in an off-shore company—circumventing local governments’ regulations nullifies the protections those governments provide consumers, and Intrade was caught dipping into investor funds for personal use—analysts have increasingly wondered whether these open-to-the-public betting forums have sufficient liquidity (number of investors, amount of investment, and a market response fast enough to keep up with information changes) to generate accurate predictions. Writing for the New Yorker, John Cassidy has pointed to the volatility with which Intrade markets responded in the 2012 US presidential election, as well as to fears that they were being manipulated by political parties. Nonetheless, “numerous studies have suggested…that markets do lead to predictions that are more accurate than traditional forecasting techniques,” Puong Fei Yeh argues, “including those that rely on expert opinions.”  Puong gives an interesting case in point: “Orange juice futures prices have been shown to be better predictors of weather than the National Weather Service’s forecasts.” Whether this proves efficacy in predicting the weather through orange juice prices or ineptitude on the part of the National Weather Service might be hard to say, but there is a certain logic to the claim that price responds to information with extraordinary speed and is capable of applying competitive efficiency to aggregating the opinions of specialists (say, for instance, climatologists, meteorologists, and scientists with access to different weather forecasting models).

On a side note, aggregated poll data like Nate Silver uses might outperform prediction markets, even when we consider that prediction markets, as Silver points out, are “able to react to news events before their effects show up in the polls.” And one might even argue that implementing a Five-Thirty-Eight model in policy formation would be more democratic than technocratic—a sort of extrapolated referendum transcending the limitations of individual polls. But arguments for poll aggregation accuracy have not, to my knowledge, been widely accompanied by the suggestion that we should be using such aggregation in making policy decisions. It is this suggestion about prediction markets that I find potentially dangerous and wish to address.

The “price signal” gives individual investors an incentive to devote cognitive power to predicting outcomes with accuracy (instead of through wishful or pessimistic thinking, desire to conform to an ideological view, or other motivations). (For a rundown on prediction market payout structure, the now-defunct Intrade provides a good primer.) Economists like Justin Wolfers and Eric Zitzewitz (see “Prediction Markets” in The Journal of Economic Perspectives), John McGinnis (who devotes a chapter to promoting the use of prediction markets in his 2012 book Accelerating Democracy: Transforming Governance Through Technology), and the signatories to the above-mentioned open letter in Science see this demonstrated accuracy as a reason to implement prediction markets in decision-making. “An information market, like a deliberative group, ‘aggregate[s] infromation…about future events.’ But what distinguishes the information market is the availability of a price signal” (from Markets as Games of Chance by Ryan P. McCarthy). The idea McGinnis and others promote is that we can “harness prediction markets…to address important policy issues” (Hilary Anyaso in a review of Accelerating Democracy).

But transforming prediction markets into decision-making mechanisms externalizes incentives and thus tampers with the price signal—a result I’d like to hypothesize by example here—and consequently I think we should have serious concerns about extending the reach of the prediction marketplace to democratic policy implementation, even while we acknowledge the forecasting accuracy these markets have amply demonstrated. I suggest that prediction markets by themselves are insufficient to ensure effective policy choices. I don’t deny that they have the potential to be usefully employed for decision-making under specific circumstances (see later in this article), but I am not convinced we can externalize prediction market outcomes to decision-making without exposing them to manipulation that would compromise the price signal, resulting in either reduced accuracy or, if accuracy is retained, a failure to represent democratic interests.

Aside from obvious (and warranted) objections that voter power can’t be financially staked without unfairly disenfranchising/excluding the poor, I question whether this plutocratic decision-making model would even be technically viable. I personally am not in favor of increasing the presence of technocratic mechanisms in policy-making, but attacking them on moral grounds seems to meet an objection of “But they’re so accurate, they work so well. Don’t you want better decisions to be made?” But the accuracy evidenced in markets so far might only be reliable precisely because these markets aren’t part of a democratic implementation scheme. Their results are internal to a market or to a market and a non-democratic organization. There’s not the motivation to manipulate results for external (policy result) reasons.

Take the example of predicting the outcome of implementing a new environmental policy—aimed to decrease a community’s carbon footprint—that would minimize busing (assigning children to schools nearer their homes) and encourage parents to allow their children to walk to school. Asking a prediction market whether effecting this policy would in fact decrease carbon emissions would probably lead to an answer “yes” (it’s safe to say less driving = less exhaust). Asking the same market whether such a change would have a significant impact on overall carbon emissions could also provide a clear answer (numbers can be crunched to determine how much mileage a community racks up in transporting children to schools, the results could be compared with overall carbon emissions, and a large number of market investors could give considered input on whether the dent the policy would make in overall emissions is statistically significant).

But what about weighing the value of this result, and the value of the policy, against other consequences of the policy? The market could predict that the policy will significantly reduce carbon emissions, but still not address a long list of other concerns. Does the policy force children in poor neighborhoods to go to failing, under-funded local schools, perpetuating the socioeconomic divide? Does it endanger children? Does it punish families with parents unable (because of work, disabilities, or other reasons) to accompany their children on a potentially dangerous walk between home and school? What I’m saying is that the questions answered in prediction markets are necessarily limited. All the above questions and more could be put to the test in a prediction market, but I still don’t see how the market could be relied on to internally generate an all-things-considered value based on any number of limited-option questions.

But here might be a rebuttal to my complaint: If a single question upon which a policy decision hinges can somehow be determined, then put before the jury of the prediction market, its outcome is implementable as a value-answer. Using the earlier example of the school policy, if it were already somehow determined that the impact of the policy on working/disabled/etc. parents would be negligible, the impact on safety would be negligible, the impact on access to quality education would be negligible—essentially that all consequences besides the carbon emission consequence were either negligible or significantly less important than the carbon emission consequence—the prediction market’s answer might be said to reflect its investors’ considered opinion on whether implementing the policy would ultimately be a good idea. Still, narrowing the question down to an equally implementable set of all-things-considered options would entail considerable agenda setting in advance: a prediction market doesn’t make proposals based on a necessarily optimal set of solutions—the solutions are offered to it in multiple-choice format, and could exist within any range of optimality.

That said, looking into the motivations of investors is a whole other ballgame. As Bennett Haselton points out on Slashdot, “[E]ven markets with a cap on betting limits could be manipulated by a single trader willing to spend a lot of money to distort the marketplace odds.” If one corporation knew a particular market question’s outcome were going to be tapped for external use in policy-making, it could slam the market with bets on its desired outcome (by buying multiple shares or by incentivizing individuals to buy shares on its behalf)—not only would that corporation be able to dominate the odds by itself, but remaining investors who bet on a different outcome would be encouraged to hedge their bets by cutting their losses or switching to the prevailing answer. Part of ensuring the efficacy (not just the prediction accuracy) of using a prediction market in policy-making might be ensuring that its participants stand to gain or lose only within the context of the market.  Otherwise, what’s to prevent an investor from risking an immediate loss in the market to bet on a more significant gain connected with pushing a non-favored (and perhaps unfavorable) idea into the winning circle?

Where prediction markets minimize strategic voting, they encourage strategic betting. I don’t mean to argue that a prediction market is inherently more manipulable than decision polls, but I think we should recognize that it would not be less susceptible to manipulation. Slashdot’s Samzenpus says that prediction markets have the potential to minimize partisanship in decision-making. Investors are less likely to predict doom and gloom as the result of a candidate’s being elected if they stand to lose financially from making partisan claims that aren’t actually based in fact—so if these markets could be effectively protected against manipulative forces, there’d be a definite value to the way their predictions are seated in a presumed causal nexus instead of the perceived preventative force of promoting hyperbolic fantasies of an unlikely future.

But as Dan McArdle, my brother-in-law, pointed out to me, preventing manipulation would mean making “the smallest bet size larger than the value of the outcome to interested parties; then people would still be incented to vote purely… But that seems both ridiculous and impossible.” Impossible is right: the “value of outcome” would vary across a range of investors, and would be virtually immeasurable for each (investors would have to be vetted for a prediction market much like members are chosen for a court’s jury!) According to legal scholar Cass Sunstein, “Existing evidence does suggest that the risk [of prediction market manipulation] may be more hypothetical than real. Several efforts to manipulate election markets have been made, and they have not succeeded: In a short time, canny investors see that prices are inflated or deflated, and the price rapidly returns to normal. More experience is required to know whether manipulation will work in other contexts.” These markets (a famous example would be the Iowa Electronic Markets), though they seek to prophesize event outcomes, are not a source for policy decisions. We don’t decide an election based on a prediction market.

Companies like Google have been able to use prediction markets to their benefit in a situation where many outcomes are more foreseeable and direct than they are for a nation: corporate success tends to boil down to financial success on the collective level in a way that national success cannot. Employees of a publicly traded company are contingent members of a profit-driven organization; citizens of a nation claim a membership of less (though not nonexistent) contingency that often only indirectly corresponds to national profit, as well as being in correspondence with other ideals. So asking a prediction market “What’s good for the company?” (any consideration of which would be essentially financial, even if union-motivated, because the relationship between employee and employer is financially contingent) might be easier than asking it “What’s good for the nation?” (as plenty of citizens put other ideological concerns way ahead of money, and citizenship itself, conceptually, isn’t financially contingent, even if in practice it often is). In prediction markets utilized by companies, the overall merit of a policy can readily be reduced, at the discretion of the company’s governance, to one question before submitting that to the market—the “value” of the question or its outcome in determining policy has been predetermined by an executive body—no need for a democratic assessment of that value.

I think perhaps two main factors (aside from the what was mentioned earlier—price signal response speed to information and the power to aggregate the compartmentalized views of various specialists) contribute to present prediction markets being so “eerily accurate”: 1) Many, like Betfair, exist apart from policy implementation—their investors don’t have a hand in manipulating policies through their outcome predictions, and so investments are internal to the market (investors gain or lose based on correct prediction of an outcome, not based on the outcome itself) 2) Prediction markets created by companies, wherein outcome affects investors both as investors and as employees, are not essentially democratic, so any outcome on the docket will have already been determined by the company to be potentially beneficial—no outcome can lead to an individual investors’ interest winning out over the company’s interest, so, unlike in the market manipulations discussed earlier, there’s not the external disincentive to giving a genuine considered opinion.

Again, Nobel-Prize-winning economists’ views of prediction markets’ potential have been much more favorable than mine, and I am neither an economist nor a financial expert. And despite the gloom Intrade has recently cast over for-profit (on the broker end) prediction market ventures, I think we can expect corporate use of prediction markets to steadily increase. With their proliferation in the corporate world, we’ll probably be faced with mounting evidence of accuracy and more pressures on government to explore their use in policy creation. It’s worth considering whether the conditions that have produced this accuracy will be equally met in a public, democratic scheme.