The Problem with Prediction Markets

June 19, 2014

Features, General

Public prediction markets are speculative markets for making predictions on the outcomes of discrete events which accept trading from public participants.(1) A real-world example is the IEM, run by the University of Iowa, which allows users to speculate on events such as whether Republicans or Democrats will control the House of Representatives in the upcoming election. The structure of these markets is very simple. Users trade contracts through the exchange which pay out, in the case of the IEM, $1 to the person that has correctly predicted the event occurring. Because of this payoff, users price the contracts according to the probability that the user estimates for the events. Through the magic of markets, these estimates end up being highly accurate as compared with other forms of estimation widely used.

The IEM's market for 2014 Congressional control. DH/RH refers to party control of the House, and DS/RS refers to party control of the Senate. A price of 40 indicates a 40% probability of the event occurring.

Because they are highly accurate and seem to have many useful potential applications, there are a number of people out on the Internet who promote the idea of prediction markets,(2) hoping to bring them to bear on the social issues of the day. Unfortunately, efforts to make public prediction markets more common have been stymied by regulatory issues. In many cases, the regulatory problems are in good faith and due to genuine concerns. But some proposals suffer from a fundamental disconnect with the economics of markets.

What’s the idea behind these prediction markets?

Prediction markets work on the same fundamental principles that drive other markets for financial assets, such as stock markets or futures markets. The value of the assets is uncertain, so a large number of people (we’ll just call them traders), who believe they are well-informed, attempt to profit by trading based on their beliefs and information. Traders motivated by profit are incentivized to gather as much information as possible on events, to perform analysis on the information, and to invest in the market. As these traders trade with one another, the price begins to reflect all of the information and analysis performed by everyone, and it becomes very difficult to beat the market. The market becomes “informationally efficient.” When stock markets are efficient it’s hard to make returns in excess of the returns enjoyed by other investors, as we observe. When prediction markets are efficient it’s hard to make probability estimates more accurate than the probability implied by the market.

Applications of prediction markets

There are a few existing prediction markets operating. iPredict in New Zealand (operated by the Victoria University of Wellington) offers the broadest range of prediction markets, with markets on political events, climate change, scientific announcements, and financial assets. The IEM has held markets on political topics such as elections and Federal Reserve monetary policy announcements (both open now). NADEX offers prediction markets on asset prices and macroeconomic releases in the US. There are several prediction markets for the outcomes of sporting events, which, while interesting, are not that important economically. All of these markets, as far as I can tell, appeal primarily to gamblers or hobbyists.

In addition to the items above, other ideas for prediction markets that may be socially meaningful have included markets on public policy proposals,(3) the spread of diseases, weather events and corporate events. For many people, the idea of using markets to tell you whether events are likely or unlikely is aesthetically appealing. Many heated arguments could be quickly curtailed by pointing to the market and saying, “If you’re so sure of yourself, go make a million dollars.” Unfortunately, economic and political economy issues suggest that this dream may be a long way off.

Feedback between prediction markets and the underlying event

In a futures market, market participants try to estimate what the value of an asset or commodity (like wheat or gold) will be at the expiration of the contract. Because they are very large and liquid, futures markets in the United States are very efficient at estimating future prices. They are also efficient because of “arbitrage.” Arbitrage is a feedback effect which causes current prices for a commodity to rise when the futures price rises (if the prices aren’t consistent, traders can purchase commodities now and store them until the expiration of the futures contracts, earning risk-free profit). Thus futures prices aren’t a pure estimate; the futures price and the current price for a commodity form a simultaneous equation that gets solved by the market based on current and future expected supply/demand for the commodity.

This article has become much more interesting in hindsight.

In most prediction markets, by comparison, there is no anticipated feedback between the probability of an event occurring and its current price. Often, when such feedbacks are possible, they lead to public policy issues.(4) For example, in 2003 DARPA proposed a futures market that would include predictions of terrorist attacks in the Middle East. The proposal was cancelled shortly thereafter, mostly due to outrage-based political point-scoring, but if the market were successful there would be an incentive for traders to bet on terrorist events occurring, then funding terrorism campaigns to ensure that their predictions were successful. Notably, the incentive to fund such a terrorism campaign actually increases as the probability of a terror attack as gauged by the market declines. Similarly, if the market for a particular politician to win his election becomes large enough, one could foresee a large trader making donations to his campaign or funding smear campaigns to increase his chances of making a winning trade.(5)

As long as these feedback issues remain possible, selling politicians on the idea of cash prediction markets will remain a difficult task, even if they would be a huge epistemic win.

Prediction markets often have no natural hedgers

A second illuminating point of comparison between prediction markets and futures markets is in their relative attractiveness as hedging instruments. One of the successes of futures markets in the US has been to reap the benefits of vertical integration among firms in a supply chain while avoiding some of its downsides. A wheat farmer faces financial risks from the price of wheat, which could cause the harvest to be unprofitable, forcing the farm to shut down. A baker faces financial risk from a rise in wheat prices, a mirror of the farmers risks. They are symmetric natural hedgers, and using futures they can share the joint risk of the supply chain, while retaining the risk benefits of not being tied to any particular plot of land or purchaser. Meanwhile, the individual firms can focus on the area where they are knowledgeable and add economic value.(6) Because the futures prices are linear and continuous, they match the risks faced by hedging firms.

Binary events, the type which would be covered by prediction markets, often have an ambiguous impact on firm performance (unlike prices which directly feed into the bottom line). Many events which are currently traded on prediction markets or promoted as future possibilities for prediction markets, such as the outcome of a presidential election or a rise in global temperature, have only ambiguous impact on any given economic actor. While it’s hard to foresee what events would be traded if we had functioning prediction markets, intuition suggests that most binary events do not have transparent and direct impact on firm profitability. Trading off the risk of the event could simply mean you are trading into modeling risk in the outcomes. Some events have more or less clear outcomes on markets as a whole, in which case financial firms can use them as hedges which will ripple through the financial system to affect other asset prices, but they may not have any reason to do so if there are no uninformed traders participating in that market.

When prices are hedged, the natural hedging is symmetric: there is someone who wins as much as the other firm loses. Most binary events do not have this characteristic. An event like “a hurricane appears before the end of the year” may harm an insurance company that offers flood insurance, without there being some firm that benefits by the same amount who wants to take the other side of the risk. Some prediction markets actually do have linear payoffs, such as Cantor Fitzgerald’s proposed real-money Hollywood Stock Exchange, but these markets defy easy interpretation as probabilities.

The lack of natural hedgers isn’t just a minor point on the viability of public prediction markets. Gathering and analyzing information is costly, and someone has to bear that cost if a market is going to be informationally efficient. In traditional financial markets, the people who bear that cost are those who are using a market for some purpose other than to make money off their predictions. In the stock market, it’s people who need to move wealth from the present to the future, such as saving for retirement. In the futures market, it’s people who want to hedge their risk. In all markets, there is a segment of people who gain benefits simply from gambling. All of these people lose a little on average to informed market participants.(7) Their losses pay for the activities of the informed market participants and is essentially the cost of achieving their ultimate goal (hedging or investing). The only category of investors that would naturally be the losers to fund a prediction market are gamblers, which, as a matter of public policy, are discouraged in America. The alternative for prediction markets is to be subsidized by an entity which values the predictions, which means that some party (likely a government agency) has to decide that a prediction market is worth having from a cost-benefits perspective.(8)

This leads to the following prediction of my own: The only binary public futures markets which become large and liquid will be those that either appeal to gamblers or which are subsidized by a sponsoring organization.(9)

Most items of interest are not economically interesting

Many commenters who support prediction markets believe that one day there will be prediction markets for everything. They envision a world where every question can be easily solved by pointing to an existing prediction market (or creating a new one) and having their question answered to the full extent of human knowledge. Unfortunately, the number of assets that exist in a market category is inversely proportional to the liquidity of each of the assets. This is part of why we observe significantly greater liquidity in the public stock market than in the market for corporate bonds, which has a tremendous volume of different securities outstanding comparatively. This is something of a minor point given the other issues, because many people would be happy with even a few large prediction markets on pressing concerns, but it limits the scope of what prediction markets can hope to achieve even if they gain regulatory approval.

Have you heard the one about the two economists?

Economist 1: Look, there’s $20 on the ground!
Economist 2: No there isn’t. If there were, someone would have picked it up already.

Prediction markets are completely legal in several different countries, but in none of them have prediction markets become integrated into the financial system in a meaningful way. No hedge fund uses prediction markets to hedge their positions, and no company uses them to de-risk their event exposure. Clearly this is only suggestive, because prediction markets are a new technology, and new technologies take time to be adopted (although adoption in the world of finance is often fast). A possible counterargument is that institutions have alternatives to prediction markets, as they can create over-the-counter contracts like credit-default swaps and weather insurance which relate to their business. These over-the-counter contracts can be viewed as creating a private prediction market. But if prediction markets are as valuable as they would seem to be, the cost reduction of a single, liquid market should outweigh the benefits that come from over-the-counter customized contracts (which tend to be expensive).

Prediction markets may reduce transparency

One underappreciated outcome of creating a prediction market, if it is successful, is in increasing the value of information related to that market. A property of valuable information is that to retain its value it must remain scarce (hence copyright laws, information wants to be free, etc.). If there is an entity producing information that is relevant to a large prediction market and the entity currently releases the information publicly, the release of that valuable information creates an uncompensated benefit to the public, a “positive externality.”

Realistically, with an existent prediction market the value of this information can lead to three possibilities: the entity bears the cost of compiling the information in secret and releasing to the public simultaneously (the Fed model), the company compiles the information without paying any particular attention to confidentiality and releases it to the public simultaneously (the Open model), or the entity compiles the information in secret and sells it to the highest bidder (the Thompson model).(10)

With the Fed model, the organization bears the cost of keeping the information confidential, but everyone has equal access to it after it’s reported. This is an unstable configuration, because those confidentiality costs aren’t core to the organization’s mission, so they will seek to recoup them unless they are morally or legally obligated not to do so.

With the Open model, the company doesn’t bear any additional costs and the public still benefits from the information when it is released, but confidence in the market is undermined, to some extent, because some of the information is inevitably leaked and so certain traders are trading with an advantage. This reduces the liquidity of the market because it reduces the profitability of dealing operations (dealers are uninformed traders, generally speaking), and it reduces desire to trade in the market from public investors who feel it is unfair.

The Thompson model,(11) in which the information is simply kept secret, has the downside of some traders trading with an advantage. It also has the downside of the information not being released to the market, which means that the assumptions underlying market prices are less transparent and thus less interpretable.

In the Thompson and the Open cases, prices are informative as early as possible, as the information is reflected in the markets as soon as it is leaked or sold. In the Fed and Open models, the information is released to the public eventually, which helps with transparency. But only the Open and Thompson models are stable without regulation, and the regulatory burden of enforcing the Fed model may cause some organizations to simply stop compiling the data in a releasable format. The model that holds for a particular organization depends on the regulatory environment, costs of confidentiality, and value of the information.

This alone is not to say that prediction markets would be bad for information, of course. In many cases (probably most cases) more information would be gathered than under our current regime, because such information would have more value. But the details of that outcome depend critically on the particulars of the situation. It does create a little bit of a political conundrum though, as politicians don’t want to make voters feel that they are on an unequal playing field, and they also don’t want businesses tangentially related to prediction markets to face burdensome regulation.

Personal Opinions

Despite the seeming public policy and economic challenges of making prediction markets more prevalent, I am in favor of them, I just don’t have high hopes. I originally started researching this post with a plan of writing a promotional piece on prediction markets and their potential benefits, but the deeper I dug the more stark the challenges seemed to be.(12) I foresee prediction markets as being more successful in corporate strategy than in public event prediction in the short-term (maybe one day I’ll write a post on the incentives of adopting internal prediction markets, but I have a feeling it won’t be pretty). I genuinely appreciate any comments, suggestions or arguments (or tweets, or shares for that matter), which can be left in the comments below, or sent to me at, or on twitter @ThatBJTerry.


  1. For clarity, this post concerns only public prediction markets, and not internal prediction markets which operate under different assumptions and constraints. Also, I focus mostly on markets for discrete events, although the category of prediction markets includes continuous events as well. Back
  2. One notable figure is Robin Hanson, economist at George Mason University. I don’t know his specific views on many of the issues discussed herein, so I definitely don’t speak for or against him. Back
  3. Futarchy is a proposal to use prediction markets more or less directly to replace some of the apparatus of governance. Back
  4. This is counterbalanced by the fact that the larger a market is, the harder it is to manipulate, and traders with opposing bets and opposing contributions could tend to cancel each other out. Unfortunately, there is no economic “law of gravity” which denies this possibility, especially in the case where the vast majority of actors are honest so manipulation of the underlying event is relatively cheap. Back
  5. An interesting proposal in which this is the intended effect is Scott Sumner’s approach to nominal GDP level-targeting. He proposes creating a futures market for nominal GDP, which the Fed would trade in to keep the level in line with its target, in the same way it currently trades Treasury bills to target its policy rate. Back
  6. This example courtesy of Larry Harris’ Trading and Exchanges, 2003, a truly excellent book on market microstructure. Back
  7. There are several other classes that I elided for brevity. According to Harris’ breakdown of “utilitarian traders” they are investors and borrowers, asset exchangers, hedgers, gamblers, fledglings, cross-subsidizers, and tax-avoiders. Back
  8. Internal prediction markets use some combination of subsidy and non-monetary incentives, and seem to work well. Internal prediction markets are, in my view, heavily underinvested in. Back
  9. If there are natural hedgers that would hedge binary propositions, they could support a prediction market, but none are extant. Back
  10. I made up all these names, I actually don’t know if this has been explored elsewhere. A pointer would be awesome if so. Back
  11. Named for Thompson Information Services. Back
  12. One of the most interesting documents I encountered is by, of all people, the MPAA. It is their comments to Cantor Fitzgerald’s abortive application with the Commodity Futures Trading Commission to turn the Hollywood Stock Exchange into a real-money futures exchange. Back

By BJ Terry

B.J. is a former investment banker and Berkeley EECS graduate. B.J. enjoys elaborate cooking, effective altruism, and powerlifting. He can be found meditating in between coding sesh's.

View all posts by BJ Terry

Get the Statricks Digest!

Want to know when we add cool new features or crank out an interesting new blog post? Subscribe here - we'll keep you posted, not spammed!

  • RobinHanson

    The 2003 DARPA markets were not actually about terrorist attacks.

    While it can sometimes be worthwhile to win market bets by changing the world, that is in practice a rare situation, especially when little money is at stake in the market.

    I’ve focused on having the customer be someone willing to pay to get the info that a market would create. For example, firms wanting to forecast their sales or project completion dates. Yes if there is no such customer, there may well be no market, even when they are legal.

    In my experience firms usually accept the claim that market estimates would be cheap and more accurate than current estimates, but still don’t want them because they would be internally politically disruptive.

    • BJ Terry

      I wouldn’t be surprised if the market wasn’t really “about” terrorist attacks, although the articles linked from Wikipedia seemed pretty decisive about it being included, so I felt it was fair game at least for an example.

      In practice so far no prediction market has gotten big enough to make it worth manipulating the world, but we see plenty instances of manipulations in normal markets. If you are using a prediction market as just a cheap way of aggregating the most readily available information, so it’s not a deep market, then it would probably never be worth it (although even in that case, there are manipulations in the penny stock world, which sort of sits near the lower bound of markets liquid enough to provide efficient prices). With subsidized markets you can calibrate your subsidy to never be profitable enough to justify actions like that (which means you can never acquire information more costly than the actions required to ensure the event, which makes sense).

      Not surprising at all that political disruption is the main issue with internal prediction markets.

      • RobinHanson

        For more info on the DARPA markets see: Thin markets that couldn’t hedge much risk can still be very effective at aggregating info. There are lots of useful questions one can ask that have very low risk of inducing much harmful sabotage in the world.