** Bernd Ankenbrand - gexid ** Bernd’s presentation focused on the work he did with Nokia on prediction markets. Nokia used prediction markets (using the gexid software) to forecast KPI’s, and had about 120 participants. (KPI = Key Performance Indicator.) One note from Bernd’s presentation that I found interesting is that Nokia first defined and communicated what the different KPI’s meant on an internal site. Despite the importance to management, the average employee didn’t necessarily know which KPI’s were being measured or what they were. The project led to a unique learning experience for many of the participants. He did a lot of research on the participants, and I wanted to mention one specific correlation. The heaviest traders on the site were those that most valued anonymity and monetary incentives. While this may seem obvious, it is important to catalog. When it came to forecast accuracy, Bernd did say that it wasn’t terribly good. But what was really important is that compared to the existing forecast, the prediction markets were an improvement. This is something James Surowiecki has mentioned before: prediction markets aren’t a crystal ball, but they are almost always better than any existing forecasting method. Finally, congratulations to Bernd for finishing his PhD! ** John Delaney - InTrade ** The week before the conference, John e-mailed the Prediction Markets Google Group to ask what questions people at the conference would want him to answer. His presentation addressed both what was e-mailed to him on the group as well as via personal messages. I’ll address the important/interesting points here. It’s obvious that John wants InTrade to be much more successful than it has been, but he still clearly believes that InTrade can still be very successful. He mentioned that one key in their business is market makers; a few of their former market makers left, and that had a significant impact in the categories in which they participated. The new gambling laws in the US has had a significant impact on their business, but John is optimistic about the long-term potential of the industry in the States. Regarding InTrade and TradeSports, John was clear that the companies are separate legal entities, with different management and employees. That said, he said that like a divorce, there are still connections between the two that take some time to replace. They are also looking into ways that individuals can submit contracts to trade on the exchange, similar to what you can do at Inkling. Clearly this would still be fairly tightly controlled, and judgement of the contracts would be done outside of InTrade by an independent entity. John is disappointed that InTrade hasn’t grown more than they have. At the same time, he seems to be very optimistic about InTrade because their employees are still very motivated and morale continues to be strong. He also said he would fly to the US, but it would need to be for a good reason. Finally, John addressed the infamous North Korea missile market. He was asked, “Was it a mistake?” He said both No and Yes. It wasn’t a mistake in that the market was judged according to a strict interpretation of the rules. At the same time, it was a mistake in that they didn’t handle the PR issue particularly well. His lesson learned was that they simply need to be incredibly careful regarding their market definitions. ** Mat Fogarty - XPree (formerly EA) ** EA produces approximately 120 games per year. While they forecasted quality of their products internally, there was little accountability for the forecasts. This made for poor forecasts, and so executives didn’t really know which products truly stood out, and which didn’t. This caused them to regularly fill their supply chains with products that just wouldn’t sell in those volumes, a money-losing proposition. Mat managed to pitch the CEO, and received good support. From that, he managed to drive the project through. What he found was that he got support at the two extremes of the company; the executives liked it because they were being told better truth, and the lower-level employees liked having a forum where they could tell the truth unfiltered. Mat found the most resistance from the middle management in EA. EA’s first prediction markets were run on product quality, using Metacritic scores to cash out the contracts. (Metacritic produces well-regarded benchmark scores for the industry by aggregating and scoring reviews.) Prizes such as game consoles were given periodically, based on a lottery weighted by a users net worth, and each week the top trader received a stein. How well did they work? Well, the previous error rate on games was 6.6%, and prediction markets nearly halved it, to 3.5%. Mat also provided some interesting demographic information on traders, but my generalisation of it is that the more junior people were more likely to register, to trade, and to be profitable. Finally, Mat introduced his new prediction market company, Xpree. Their software is really focused on maximising participating in a prediction market by making it tremendously easy for users and administrators to participate. He showed some early screenshots of sliders and other new user interfaces which essentially remove the standard trading interface for ease of use. They have started using this software within EA to start predicting ship dates for products and product sales figures. Finally, they talked about incorporating a wiki into the process. A user could short-sell a contract, and then be directly to a wiki where they could explain why they traded the way they did. It will be interesting to see how this is implemented and utilised. My opinion is that if people have to sign their name (even user name) to wiki comments, it might be difficult to get them to tell the truth. ** Leighton Vaughn Williams - Nottingham Trent University **
Leighton gave a great presentation that explained the various boom and bust cycles of betting and taxation in the UK from World War II onward. He has done a lot of work in the past decade with the government to reduce the taxation on betting. In 2001, tax law on gambling was changed from a turnover tax to a gross profits tax. This encourages a low-price, high-turnover strategy, compared to the previous high-price, low-turnover strategy. For the betting public, tax was effectively axed. From a betting turnover of £7billion in 2001, this rapidly grew to £53billion in 2007. Though tax rates were cut, revenue stayed neutral and an entire industry exploded in growth. Finally, Leighton really stressed the Journal of Prediction Markets. If you aren’t already a subscriber, please sign up! An digital subscription is only £30 a year, and if you want a paper copy as well it’s only £49 a year. You can sign up for a free trial here: http://www.ingentaconnect.com/content/ubpl/jpm, or contact Christopher Woodhead for more details. If enough interest in the Journal cannot be sustained through subscriptions, the future of the publication may be in doubt. It’s academic, so they aren’t necessarily looking to turn a profit- just break even. Additionally, they are always looking for more articles, and they don’t necessarily have to be strictly “academic.” Interesting practical studies on prediction markets, including corporate prediction markets would also be very welcome. ** Will Speck - FT.com ** The FT.com partnered with InTrade to create a realistic market. One of the things they found with their market is that there needs to be a balance between both long term incentives and short term incentives. They had prizes for both, but different behaviours could be driven by a focus on one or the other. The participants on their site (FTPredict) saw the game as “trading,” and not “predicting.” This brings an important distinction when it comes to talking with users, and something marketers should be aware of. Will saw that prediction market data could be a good tool for the FT audience. FTPredict was seen as a compelling, content-based marketing tool for the Financial Times. There was quite a bit of interest in the project, and they will be doing more contests in the future. While it’s US-only right now because of the legal work required for the contest rules, that could potentially change in the future. ** Emile Servan-Schreiber - Newsfutures ** Emile’s lecture was focused on Newsfuture’s competitive forecasting software. It’s something that Newsfutures has found quite popular among certain corporate clients recently. While not a prediction market by definition, it is related and serves the needs of some of their particular clients better than a CDA model. He stepped through the Competitive Forecasting model, which lets users enter a high-low spread of where they believe a metric will occur. (ie, sales of widget X between 10000 and 11500). To win points, the forecast must capture the final value. The smaller your spread (more precision), the more points you can win. However, if your spread doesn’t contain the value, the users will lose more points with a smaller spread. Finally, it rewards early users by accumulating points constantly over time. Competitive forecasting has been popular for variables where there are few participants, are busy/high-level users, people are not familiar with trading, or have few training opportunities. Finally, Emile finished his presentation by talking about bet2give.com. It’s been pretty successful so far, but Emile mentioned that building a trading base has been a challenge, as their typical play-money traders aren’t interested. That said, they’ve been getting a completely new trader base interested in the site, and have seen some good growth since introducing the site. ** Henry Berg - Microsoft ** Henry Berg leads the Information Markets Group at Microsoft. It was started in 2006, and right now is about 4 people. Their focus is in prediction markets, estimate contests, and other types of information markets. The original prediction market effort at Microsoft was led by Todd Proebsting in Microsoft Research. This led to the development of the Information Forecasting Exchange, which was in place from 2003 through 2006. Todd is now an advisor to the Information Markets Group. Henry showed off the tool that Microsoft has built, and is calling PredictionPoint. Their goal was to lower barriers to participation. To do this, they moved to a betting paradigm, as most people are more comfortable with betting. Their user interface lets participants explore bets without typing and without finalising a trade. In addition to prediction markets, they have what Microsoft calls “Estimate contests.” These are similar to HP’s BRAIN algorithm, but based on a quadratic scoring rule. Microsoft is actively expanding their efforts in this space, and have run a few dozen markets in the past year or so. A little unusually, their markets are always a bundle of securities (typically 2-8 contracts) that are tied together. Instead of trading on the value of a metric, they trade on which band the metric will fall. (ie, less than 10k, 10-11k, 11-12k, 12-13k, more than 13k) [Chris Hibbert has talked about bands, scaled claims, etc. and this concept here.] The demo was very interesting, and I hope they eventually talk a little more about the results they’ve seen from their markets. ** Jesper Muller-Krogstrup & Oliver Pedersen - Nosco ** Nosco was started about 18 months ago, and I first met them at the Vienna Prediction Market Summit last year. Jesper and Oliver are clearly polished presenters and experienced in prediction markets. Nosco provides both a prediction market software package and consulting services around that package. They have built three specific products around prediction markets:
- A news exchange
- An idea exchange
- An information exchange (traditional prediction market) The News Exchange system has been used by TV2, a television station in Denmark. As stories are written, contracts are immediately created. In this case, the predictions are attached to the stories and pushed out as content alongside the story. They reported that though it was a bit of a struggle to interact with writers at first, it did improve. The system has 21,000 users and over 2,000 markets. The Idea Exchange system is being used by Danske Bank, and kicked off quite recently. The bank’s goal is to catalog employee innovation, with the goal of eventually accumulating 1 million+ ideas. While currently in trial stages, they have already exceeded initial targets. Finally, they also have their Information Exchange, which is Nosco’s standard prediction market. In their experience, they found three main drivers in the public markets:
- competition structure (it’s more fun to compete with friends)
- simplicity
- update & feedback (short time to market)
They also found that it’s easier to attract people with one BIG prize, instead of a series of smaller prizes, as a BIG prize simply generates more publicity. When talking with people in Denmark, Nosco specifically avoids discussing “prediction markets.” Terminology, particularly in a country such as theirs, is very important. A “prediction market” isn’t a buzzword, and people often form mistaken impressions of what they’re doing because of the two words alone. It’s also clear that Nosco has a very talented graphic designer and web designer on their team. Their screenshots, their graphics, and their presentation were all very polished and tied together nicely. It was a very informative presentation!
** Jed Christiansen - Mercury Research & Consulting (me) **
My talk was essentially the same as my talk at ConsensusPoint’s New York conference in September. I explained a bit more about my research project, which was published in the first issue of the Journal of Prediction Markets. I ran a series of prediction markets (using the Inkling platform) in the summer of 2006.
The problem with many business prediction markets is that only a small number of people may know enough to trade on a market. The question I wanted to answer was: how small could a prediction market be and still provide calibrated results?
What I found was that in markets with 16 or more traders, the results were calibrated. (Estimated probability = Observed probability) Additionally, events with a small probability of occurring were well calibrated. I also examined the behaviours that traders exhibited, and effects of manipulation.
These results were all written up in the article; please click here to contact me if you have any questions.
As I mentioned, I first ran these markets in the summer of 2006. I also ran these markets this past summer, and hope to publish the results soon.
** Peter Sorensen - University of Copenhagen **
Peter has been studying the academic aspects of prediction markets for some time, now. He has written papers with Marco Ottaviani, such as “Aggregation of Information and Beliefs in Prediction Markets.” (Marco was at the London Business School, but has since moved to Northwestern University in Chicago.)
Peter’s talk centered around the theory of prediction markets. He gave a good background of the general theory, and talked specifically about some aspects of the Favorite-Longshot bias. It was a good talk, and Peter clearly has done some interesting work in prediction markets.
** Gunther Fadler & Peter Gollowitsch - Pro:kons **
Gunther and Peter’s work in prediction markets stretches back to 1999! (Well before the “Wisdom of Crowds.”) Their original company ran from 1999 to 2002, and ran over 20 prediction markets. In 2002, the company was re-organised/re-structured as BDF-net. The new company’s focus was on prediction markets, consensus finding, and new media projects.
Their talk centered around a huge prediction market project they have been doing with SRG, the Swiss public broadcasting organisation. (Essentially the Swiss version of the BBC.) This was a large and complicated project that is currently on-going, revolving around news coverage of the 2007 Swiss elections.
The project is particularly complex, as they the prediction markets are served on seven different websites (for the different television/radio station website), and perhaps more importantly, there are four official languages in Switzerland! Additionally, Switzerland has a very unique political system, and the markets had to take those complexities into account.
Their solution is a master database, with “skins” for each individual website. So far they have 2000+ users, with 26,000 visitors per month. They are planning on integrating wikipedia, YouTube, GoogleMaps and other services into the solution, too.
Finally, congratulations to Pro:kons for winning the Cisco Austria Web 2.0 aware for social software! This shows that there is some very exciting prediction markets work being done that simply hasn’t made it to the forefront of the english-speaking community. It was great to hear about their work, and I believe we’ll be seeing more of them soon.
————-
Well, that’s it! It was a great conference, with a wide range of speakers from across the prediction market spectrum. As always, contact me with any questions or clarifications. - Jed Christiansen | 
