Mercury’s Blog

If you're new here, you may want to get Mercury's Blog by Email or subscribe to my RSS feed. Thanks for visiting!

I’d like to bring your attention to a few things in the prediction market industry that you may or may not be aware of already.

—————–
First, in the run-up to the 2008 United States presidential election, Slate magazine has created a webpage where you can compare the current prices / probabilities of each candidate winning their party primary as well as the prices / probabilities for winning the general election.  Data comes from both the Iowa Electronic Markets and InTrade.  (I believe they expect to integrate Casual Observer data once those markets are set up.  Their website seems to be in a bit of a revision right now.)  While the page layout could be a little bit cleaner, it certainly does show some interesting data.  Perhaps the most intriguing issue is how InTrade and IEM deal with the candidates selected.

IEM only has contracts for Clinton, Obama, Edwards, and Field.  InTrade also lists contracts for several more candidates (such as Richardson, Gore, etc.).  When comparing them side-by-side, the results are interesting.  Clinton keeps the same level of support across the IEM and InTrade, (43.5 and 48, respectively), but Obama has significantly different levels of support (43.7 on IEM and 29.4 on InTrade).  Part of this can perhaps be explained by those extra candidates the InTrade allows trading on; Al Gore is currently priced just above 10.  I’ll let you draw whatever conclusions you might draw from this data.  My take is that in the more mature markets (IEM, which has been around for a lot longer and the limits on accounts mean less potential manipulation) Hillary and Barack are neck-and-neck for the Democratic nomination.

The same effect can be seen on the Republican nomination to a lesser degree.  Guilani has less support on IEM than InTrade (about 8 points), where Romney has about 2 points more support on IEM than InTrade.
——————-

Secondly, the Journal of Prediction Markets has published its first issue!  This is quite an exciting development for prediction markets in my opinion.  Hopefully it will spur additional research and visibility into these quite powerful tools.  The editor is Leighton Vaughn Williams from the Nottingham Business School, and the Journal is published by the University of Buckingham Press right here in the UK.

I’m also quite happy to announce that you will find an article that I wrote in this inaugural issue.  It examines small-scale prediction markets and discusses accuracy, behaviours, and other elements that are critical for success.  I believe there is a free trial period for the Journal if you click here.  Otherwise, please check out the website for the Journal of Prediction Markets here, and click here to see the rather esteemed Editorial Board.
——————-

Finally, the Second Workshop on Prediction Markets will be held in San Diego on June 12, 2007.  It’s part of the ACM Conference on Electronic Commerce (EC’07), which takes place from June 11-14th.  I currently plan on attending, and hope to see many of you out there.  It does have a more academic tilt than some other conferences, but should be very worthwhile.  You can check out their website for more details, and I certainly recommend attending if you can.
——————-

On a final cultural note, today was the unofficial UK National Workout Day.  As the first Monday after changing to Daylight Savings Time, today was the first day that you could come home after work where it was still light out.  Perhaps it’s because I live near the walking/cycling path on the Thames, but it was absolutely incredible how many people were out running/jogging/walking tonight!

There’s only one other time I’ve seen that many people out working out at once; last year right after the switchover.  After the first day, enthusiasm just isn’t as high, and you never see the same number of people out at once again.

When creating a new prediction market, or series of prediction markets, the creator must consider the structure of how forecasts are made.  The structure used can affect both the accuracy of the prediction, potential financial losses for the market institution, as well as how quickly the market will react to new information.

I will broadly generalise the existing structures into three categories: CDA’s, Algorithms, and (for lack of a better term) Others.  This categorisation is not based on the mathematics of the structures, but how they work when implemented.

(Origins of these thoughts were started at this post on Midas Oracle.)

CDA’s
These are the classic prediction markets, and are derived directly from how financial markets operate.  Orders are placed to purchase a futures contract (bid) and other orders are placed offering to sell futures contracts (asks).  When these match, and exchange is made.  Exchanges operating like this are the traditional CME/CBOT, the Iowa Electronic Markets and InTrade/TradeSports.

The reason these are used for real-money prediction markets is because they offer no risk to the market institution.  All trades are between individuals on the exchange; the exchange itself doesn’t offer liquidity as a market-maker.  The drawback to this system is that in order for the market to work, there must be sufficient liquidity.  If just a few people are trading, they simply may not come to agreement on a price and thus exchange contracts.  Without liquidity, the market will just languish and be unsuccessful.

There are two particularly great benefits to a CDA market, however.  One is that prices can make instantaneous jumps.  If a presidential candidate makes a huge gaffe, a contract that represents their election chances can dive from 60 to 20 in one trade, something that is difficult to achieve with an algorithm.  The second is that traders specify their opinions exactly, by placing bids or asks at specific prices.

Algorithms
For the purposes of this post, I classify both Robin Hanson’s MSR (Market Scoring Rules) and David Pennock’s DPM (Dynamic Pari-Mutuel Market) in the same category.  In both structures, traders purchase shares in an outcome, and with each share purchased the price of the contract changes.  (The magnitude of the price change depends on the number of contracts already purchased.)

There are drawbacks to algorithms, however, and they mirror the advantages of CDA’s.  Firstly, it takes them more time and trading to respond to instantaneous price jumps.  In thick markets, this largely doesn’t matter.  But in thinner markets with less volume (such as many corporate markets), this could certainly become an issue.  A second drawback is that to make the algorithm work it may take some financial subsidisation.  In a play-money market this really shouldn’t matter, and specifically applies only to the MSR structure, which is currently a bit more common than the DPM structure.  Finally, while algorithms deal nicely with contracts that range from 0-100, they can be difficult to use in other ranges because of how the prices are set to move with each share purchase.

The huge advantage to Algorithms is that they offer infinite liquidity to traders.  This can be a very important factor for new exchanges, particularly as new users want to get up and running immediately.  Trading immediately, and seeing the effect of a price change, can be a very valuable tool in order to get prediction markets adopted in an organisation.

Others
NewsFutures has innovated other types of structures that can be considered prediction markets, but are structured quite differently.  The most prominent example of these can be seen in their Global Risks Prediction Market, run for the World Economic Forum.  This is their Competitive Forecasting structure.  Traders earn points based on how early they make their predictions, the spread of their predictions and if the bounds of their prediction contained the final answer.  This kind of solution works for companies that have specific concerns about the standard trading mechanisms, such as the time (from work) that users take to trade on the market, the desire for more data on trader’s opinions, etc.  More can be read on the site regarding the specifics of how it works.

Summary
This is just a summary of the variety of structures that a prediction market can utilise.  Each solution has it’s advantages and drawbacks, and should be considered within the context of how it will be used within the organisation, and the organisational needs.  I will be posting more in-depth descriptions and analysis of each of these in the future, and look forward to your comments.

When trying to build support for a series of prediction markets in a business, one fact must be understood: different divisions have different goals. This matters because each function within a business should have prediction markets pitched to them individually, with a focus on how their needs can be met.

For example, I have posted before about the three types of markets that are ideal for prediction markets in the corporate sector. They include project management, quantifying an industry, and quantifying risk.

Let’s think about a company that really needs predictions markets, one that requires massive research and development work before a product is released, and think about what divisions of that company requires.

One set of stakeholders that need to promote prediction markets are the research and development directors, project managers, or whatever title is appropriate. What they need to know is progress against targets for their project, ie, the project management prediction markets. This could be a delicate situation; the project manager may be surprised by a result showing their project months behind! In the long run, having this forecast is far better for the company, as decisions can be made much earlier to control costs and damage from slipped projects. If a company is banking their Q4 results on a product in the pipeline, when would they prefer to know that the product is going to slip to Q1? (Before or after they’ve put marketing cash behind it and discussed it with industry analysts?) Project managers and company leadership will be very interested in the results from project management markets.

Another important set of stakeholders is the sales/marketing divisions in a company. They are interested in sales, so the prediction markets regarding sales predictions are clearly ideal for this audience. These might be sales for next quarter, sales across the next year, or other time periods. The same kind of markets can be used to help determine which products to put marketing muscle behind, or even fund in the first place. (This was the focus of a New York Times article on prediction markets in a company called Rite-Solutions.) Prediction markets that inform key marketing decisions should be the focus of any efforts to pitch a project to the sales/marketing divisions.

Finally, any number of divisions in a company could be interested in a pitch that quantifies risk. A metric on product quality (as measured by known software bugs, service outages, or any other hard data) would impact both development personnel like engineers, the marketing division, and potentially company leadership at the top. These data points are important measures on the entire company’s performance, and therefore touch many functions throughout the organisation.

So just as different types of markets answer different kinds of questions, the benefits of the specific markets should be pitched differently within a company. Each function deals with different problems, and those should be identified and benefits from the appropriate type of prediction market promoted separately. This support will help ensure the widespread adoption and eventual success of your organisation’s prediction markets project.
—————-

Reminder: If you want to receive these blog posts directly in your inbox, just click here to sign up for a simple e-mail subscription. As always, click here to subscribe to Mercury’s Blog with your favourite RSS reader, such as Google Reader.