Corporate Prediction Markets

Many people are familiar with the predictive success of small, low-stakes “prediction markets” run by the University of Iowa (predicting political elections better than the pollsters) and firms such as Tradesports.com (predicting everything from whether Hamas will recognize Israel to whether Scooter Libby will be convicted) and the Chicago Board of Trade (predicting U.S. unemployment and other key economic indicators).  As colleagues Saul Levmore and Cass Sunstein have pointed out, these markets are routinely better at forecasting what is going to happen than any other available means.  The insight here is a Hayekian one: markets, when they work, are the best available mechanism for gathering, aggregating, and processing information.

Those familiar with this literature may also be familiar with the use of these markets by firms such as Hewlett-Packard and Google, which have made increasing use of prediction markets to help make business decisions.   Initial studies suggest that these markets provide relatively reliable predictions and are not easily manipulated by those who have a stake in decision making.  For example, HP’s market estimated sales of a particular product better than traditional forecasting methods.  Other results suggesting the usefulness of simple prediction markets have been seen at Google, Siemens, Intel, and many other firms. More promising still, David Pennock’s patent-pending dynamic pari-mutuel market and Robin Hanson’s market scoring rule make it possible to generate sound predictions even in very thin markets.

In a forthcoming paper, Michael Abramowicz, a law professor at George Washington, and I explore how these markets may help solve a number of corporate law problems.

Conditional prediction markets (e.g.,, what will Firm X’s stock price be in one year if the strategy proposed by a corporate raider is adopted) could allow corporations to estimate the effect of different major decisions, such as whether to acquire a target, whether to adopt a governance reform, or whether to dismiss a CEO, on stock price. If further experimentation shows the technology to be reliable, prediction markets could serve as an alternative or complement to shareholder voting as a means of disciplining corporate boards and managers.  Appropriate prediction market design can combat the “rational apathy” that plagues shareholder votes, while market participants would still have incentives to take into account the recommendations of corporate officials, to the extent that those recommendations may reflect information unavailable to market participants. Appropriately designed prediction markets may also serve as a weak substitute for insider trading, producing an alternative avenue for insiders to profit on and thus reveal inside information, without creating an unlevel playing field in the market for a firm’s securities.

To get an idea of where we are going, consider the CEO certification requirements for financial statements imposed by the Sarbanes-Oxley Act, which carry a penalty of up to 25 years in prison.  Suppose a CEO, when deciding to sign the document, believes that there is a possibility that certain executives have been shifting sales from one period into another in order to meet sales targets and Wall Street’s expectations about quarterly earnings. The CEO can rely on information to flow to her through the normal hierarchical channels—from line sales persons to sales managers to sales executives to finance and then finally to the COO or CFO—but she may have reason to doubt the veracity or completeness of the information she receives. For one, the individuals in this or any chain of command is subject to their own incentive mix, which may or may not be in accord with that of the CEO or the firm’s owners.  Moreover, the incentives have to be perfectly aligned for all links in the chain; any link may be able to distort the information in a way that the other links, even if acting in good faith, cannot know of or be in a position to change.  For example, a line sales person may distort the true figures and conceal the deception such that the lie gets reinforced instead of found out, or a senior vice president may be presented with truthful but bad information only to change it at the last minute before it gets to the top.  In addition, certain third parties, like the firm’s auditor or lawyers, may know about any misrepresentation but be unable or unwilling to convey it to the CEO because of restrictions on their access to the CEO or because their own (personal or intrafirm) incentives are not perfectly aligned with that of the CEO.  To take just one simple example, the partner on the account may not want to tell the CEO the truth told to him by a lower-level auditor because the partner is worried about alienating a key client and thus reducing his annual bonus. 

Instead of relying on command-and-control mechanisms or computerized “internal controls,” which are all expensive, unreliable, and subject to abuse, the CEO could create a prediction market in which firm employees, auditors, and anyone else that the CEO wishes could trade on, say, whether the firm will be required to restate its earnings over some period of time.  This internal prediction market, even a thin, private, fake-money market, might give the CEO a better prediction about the veracity of the firm’s earnings than any command-and-control model, either firm created or imposed by regulation.  For one, it will allow any link in the chain, either at the firm or at its auditor, to convey its objective estimate of the probability of earnings restatement to the CEO through trading behavior.  And, as the theory suggests and practice supports, any attempts to manipulate the market, say by the audit partner or the SVP trying to cover up the true picture of the firm’s earnings, would only increase trading profit opportunities for those with accurate, objective trading information.  Especially with a market providing more robust financial incentives, the richness and simplicity conveyed to the CEO through the market price, reflecting the trading behavior of all market participants, cannot be equaled through any non-market means.  Just as the price of a pencil conveys simply all of the various costs of every element of its design, construction, and delivery—from tree and graphite mine to box on the store shelf—so too here does the price of a prediction market contract tell the CEO more than she could learn from any other sources.

For the foreseeable future, we imagine that corporations will probably create prediction markets solely for informational purposes. But if prediction markets prove sufficiently successful, one might imagine corporations committing to bind themselves to the decisions of particular prediction markets. Suppose, for example, that a corporation is considering building a new plant. It might launch two heavily subsidized conditional markets predicting the corporation’s stock price several months in the future. One conditional market would predict the stock price if the corporation committed to building the plant, and the other, the stock price if the corporation committed to not building the plant. After about a month (the exact time should be randomized to prevent last-minute manipulation), the corporation would check which prediction was higher, and make its decision accordingly. Because such markets should look after shareholders’ interests, implementation should reduce agency costs. Perhaps someday, corporations will allow shareholders to influence corporate policy by sponsoring prediction markets instead of by voting. This approach might be used not only to resolve specific policy issues (perhaps more than shareholders now are permitted to resolve in votes), but also to make personnel changes in management or the Board of Directors.

Comments

[Robin Hanson]

The idea of a decision-contingent prediction market is powerful and has wide application, including in corporate governance. FYI, over ten years ago I tried to publicize the general applicability of this approach, and illustrated it with corporate governance examples. For example, see http://hanson.gmu.edu/dumpceo.html . Over the years I didn't have much success getting academic journal referees to publish my suggestions. I presume and hope a Chicago law school prof will have better odds.

[Todd Henderson]

Thanks for commenting (and reading), Robin. Indeed, your work has been a tremendous inspiration, and we use it as a foundation for many of our ideas. As for the failure to get these ideas published, I think it has more to do with the march of progress than anything else. The work of yours and others in the intervening years has made the case for these markets, both theoretically and practically, much stronger. While some of the suggestions we consider are more radical and future-looking, many are implementable today by firms, as our interviews with market participants suggest. If you would be wiling, I'll send you a copy of the paper for your comments.

[Robin Hanson]

I'd love to comment on your paper, and hope that it can help entice some firms to try this stuff on non-trivial decisions.

[Chris. F. Masse .COM]

Hello Chicago,

Do you have marketing data suggesting that play-money or real-money traders would be interested in these topics?

No liquidity, no trader satisfaction, and no predictions.

The predictive power of the prediction markets is an offspring of their liquidity.

Speaking of the alternatives to CDA, still you need trader interest.

As for your paper, I'd happy to list it.

Ciao,

[Jason Ruspini]

Chris, Your question seems to assume that it would be beneficial to open the markets to the public. This again sounds like the exchange industry perspective, and doesn't speak to prediction markets in the strict sense. Conceivably (though there will most likely be a de facto blending), the former may always be larger and more profitable, just as the housecleaning industry is today.

If we, for the sake of argument, reduce "liquidity" to the number of motivated traders, it's not clear where the marginal informational gain levels-off -- the number of traders such that you have adequately captured the full distribution of opinion. In a real-money market the distribution of budgets may actually degrade the informational interpretation of price, as certain biases can have a disproportionate effect on the market (although this would be somewhat canceled by the fact that a large budget might correspond to past success in the market). With equal starting budgets, this number is in most cases probably more than say 30, but insofar as the information is valuable, the company might prefer to not give it away by immediately making the prices public, which would be the way to maximize liquidity. The optimal number of traders from the perspectives of companies may well be in the hundreds or low thousands, and might be limited to employees and contractors who will be encouraged to not leak the markets' information. Rewarding the employees with real-money will be simple if done through bonuses, and there would be sufficient motivation.

[Michael Abramowicz]

Chris,

I'm not sure why you think that heavily subsidized markets using Hanson's market scoring rule would not solve the problem. A sufficiently large subsidy should create trader interest, and anyway Hanson's approach should still give honest forecasting incentives even if only a very small number of people are participating.

And we will certainly send you the paper once we have a draft. Thanks for all your great work on compiling information about prediction markets (invaluable for me in compiling information for a book project that I am currently completing).

[Chris. F. Masse .COM]

A prediction exchange (a.k.a. betting exchange) serves two purposes: 1. Primarily, aggregating the maximum possible of speculators (including those willing to make bets in the mainstream people's opposite directions). 2. Secondarily, generating probabilistic predictions that could be of interest to commentators and, possibly (that's what Robin Hanson brought to civilization), decision makers.

Now, let's say you have an internal corporate prediction market on a Sarbannes-Oxley topic, where traders are subsidized. You say that a "sufficiently large subsidy should create trader interest". I would like to question this statement.

a) Are you sure that the traders are interested both in making money thru speculation and in the Sarbannes-Oxley topic? Or would they lie to you about the latter because they'd be aim at the former only. It won't work if the trader interest is artificial.

b) To paraphrase what NewsFutures' Emile Servan-Schreiber said the other day (about play money speculators), in your scheme, the traders "are not risking one cent from their own pocket". If traders risk their own money, then there's a self-selection process. That way, the clueless people decide to stay out of the trading.

c) Of course, a) and b) can be rebutted by saying that what counts is to have at least, say, 15% of informed traders (taking over the uninformed traders). Then I ask the tough question: What marketing data do you have that convinced you that knowledgeable executives and managers would spend work time at the office playing this trading game?

Time is the most precious element that these people have. I know that for a fact, since most of my subscribers read my site feed from the office, on weekdays. And when they interact with me, they ask me the hard questions about the usefulness of prediction markets.

To come back to my two introductory points, most scholars put an emphasis on #2, and overlook #1. There's no other way that the marketing approach in order to research trader interest. Asking people; watching their behaviors; testing. That's an area of research snubbed by the scholars. Minding #2 without minding #1 is like rowing with one arm only; you perform circles but you go nowhere.

Good luck with your paper. I hope that many readers will make harsh comments, because that's an incentive for progress.

[Michael Abramowicz]

Hi again Chris,

I agree with what you say for a scheme in which the traders are not risking a cent. Our intent was simply to say that perhaps even without subsidy, prediction markets can be effective, but we do not make any strong claims about that. Our broader focus is on hypothetical subsidized markets. With enough subsidy, people who know about the topic should be drawn to participate, and even "artificial" interest should be enough to generate meaningful predictions.

I do agree that analysis of actual markets and user behavior is an important area of research (though one in which I have not been much involved). At the same time, one should be cautious about inferring much from small-stakes markets about what would happen if we had markets with subsidies in the millions of dollars.

[alex kirtland]

Hi. This is a great discussion. I would love to see a draft of your paper, too, when it's available. I'm particularly interested to hear what prediction market ideas companies think can be implementable today.

Also, Chris is asking the right questions. Adoption (or generating appropriate levels of liquidity, if you prefer), is a difficult problem for any new corporate application to overcome. Currently, corporate wiki's and blogs are facing the same hurdle. Everyone can imagine how useful they would be (just look how successful they are outside of companies), but getting companies to use and adopt them for internal use is easier said than done.

[Cara Fletcher]

The success of those "prediction markets" run by the university of Iowa is really good and they know what are they doing.

[Grants]

The Government Grant provision federal agencies could well be applied to by using this electronically mediated agency. The different options available could well be identified and utilized. In contrast the Catalog of Federal Domestic Assistance could well be accessed to identify and utilize the diverse grants and assistance available for Government Grants.