Mercury's Blog

Chris Masse over at Midas Oracle has recently generated a fair number comments on his post “Are IBM Smarter Cities prediction markets too smart for people?” As the commenters rightly point out, what IBM was doing was NOT a prediction market, but instead a polling system.

This is something I’ve been talking about since May 2008. When you try to use a prediction market to forecast the “best” ideas, you get a “Keynesian beauty contest.” To quote (via wikipedia):

It is not a case of choosing those [faces] that, to the best of one’s judgment, are really the prettiest, nor even those that average opinion genuinely thinks the prettiest. We have reached the third degree where we devote our intelligences to anticipating what average opinion expects the average opinion to be. (Keynes, General Theory of Employment Interest and Money, 1936).

Starting from the wrong metaphor

Ideas in an organization are a long-term investment, where significant work and development is required before you can even get to the prototype stage of development. This is a very different concept than regular, known markets, such as the number of customers in the next quarter or “will me meet sales targets next year” which are asked in prediction markets. Prediction markets work well for the regular, known markets, because prediction markets involve liquid, easily-traded contracts.

Idea markets should be seen through the metaphor of venture capital. An individual sees lots of good ideas, but none of them are well developed. Just as a venture capitalist puts resources into ideas and those resources are locked-up and illiquid until the idea has proven its success or died, an individual in an idea market should see their votes as an illiquid, sunk cost.

The organization running the idea market then concentrates their resources where individuals concentrate their votes. But just like venture capital, this cannot be a one-time market. The organization needs to regularly run similar markets, say every quarter or every six months. If an idea gets resources but doesn’t make enough progress (or doesn’t look like it will be as effective as originally thought) the market will stop voting for it. As an idea makes good progress, it continues to receive resources for development. Finally, when an idea is implemented and made successful, all of the people that voted for it during its development will be rewarded. And just like venture capital, the earlier you contribute the more you are rewarded.

The key is that idea selection and development is long-term work, and thus when markets are used to forecast and help allocate resources, the market structure must match that long-term approach.

I’ve recently come across a couple of interesting notes in the prediction market industry.

#1 – Trading UK Housing Prices

CityOdds, based in London, has recently opened a prediction market (currently marketed as a competition) to predict a UK housing market index. While this is a play-money market, there is a £10,000 first-place prize.

CityOdds runs both fantasy markets, as well as real-money markets on currencies and other commodities. Mike Chadney, the founder, is a good guy and experienced “City” man. Check out the housing market here: http://www.cityodds.com/hpitrading.html

#2 – Predictify shuts down

Predictify made a bit of a splash when Scott Adams used them to forecast how many sales of his (then) new book were going to be sold.

It seemed a bit of an odd company; venture-funded, but with people who had no background in prediction markets at all. They offered a small number of markets where accurate forecasting would win cash, while most markets were just for leaderboard position. (My trading was perhaps typical: I’d log it and only scan the markets where I could win cash and ignore the rest.)

Well, they’ve died. According to an announcement on their website:

Due to the tough economic climate, we are planning to cease operations and shut down the company in the near future. If you have an account balance of $20 or more, please visit your account page and enter your withdrawal information by 11:59pm on August 31, 2009 to receive payment.

We sincerely enjoyed building and operating Predictify, and we’re glad that you could be a part of it.

The Predictify Team

My question is this: what does this mean for Crowdcast?

Crowdcast has an absolutely fantastic team who have great experience in prediction markets. But can they thrive as a venture-funded company? I’m hopeful, but perhaps am more skeptical that there are simply enough customers truly interested in their level of solution. Particularly as so many companies need hand-holding and thus a lot of expensive people-time through the early stages of implementation, are the profits sufficient to satisfy investors?

I don’t know, but I do believe they’ve at least got the people to give it a go. (Strangely enough, it was Crowdcast’s founder Mat Fogarty who originally told me about Predictify well before their launched.)

I’ve been writing about ideas and innovation for about a year now. More and more evidence has, in my opinion, built up to show that the stock market metaphor is not appropriate for finding the best ideas from a prospective pool.

From Emile of NewsFutures, here are three links and a quote from each:

The imagination market, Information Systems Frontiers (July 2007)

Participants were able to trade shares of technology ideas over the course of 3 weeks, resulting in the market identifying the “best” idea as the highest priced security. Our findings suggest that information markets for idea generation result in more ideas and more participants than traditional idea generation techniques; however, using markets to rank ideas may be no better than other methods of idea ranking.

Examining Trader Behavior in Idea Markets: An Implementation of GE’s Imagination Markets , Journal of Prediction Markets (April 2009)

In this experiment, we examine the behavior of traders that have submitted the ideas on the market and their influence on the market’s outcome. An idea’s submitter is clearly motivated to have his idea valued highly by the market, both by the funding given to the top idea as well as smaller prizes given to the top three ideas. In general, founders tended to buy their suggested ideas at prices above the volume-weighted-average-price (VWAP) in significant volumes. We discuss the implications and mitigation strategies. A survey of market participants yielded mixed results regarding the market’s effectiveness at ranking ideas but very positive results regarding the quality of ideas proposed.

GE Global Research blog (link here)

From the comments,
“GE Healthcare IT attempted an imagination market a few months ago to bring forth some new ideas for the company’s future. It left me with very strong mixed impressions: on one hand, it’s wonderful that we’re leveraging the power of technology for mass collaboration and idea sharing. On the other hand, I felt that the tool obfuscates the very opinions it seeks to gather by due to the inherent complexities of market behavior.

My primary objection is the use of the stock market paradigm to evaluate these ideas. Simply, I find it too abstract to be useful in gathering feedback about the quality of an idea. Stock investment is done by trying to predict *the change in collectively perceived value of something over time*. However, when dealing with ideas, neither the collectively perceived value, nor the change in this value over time are valuable metrics; you want people evaluating ideas based on *their opinions*, not based on their attempts to predict changes in the investment decisions of others over the course of a few weeks. These are static ideas isolated from one another, not evolving companies that interact. I think a stock market is an unnecessarily abstract, and distracting way to retrieve simple information: what do people think of these ideas?”

I’d be interested in hearing from people that disagree with this…

I’ve been blogging here about prediction markets for two and a half years. In that time, a lot of discussion has taken place both here and elsewhere about specific prediction market contracts; what happened on those markets and when.

One of the problems with prediction markets is that the information from them is not just valuable when it’s “live” but also after the market has finished and been closed out. It would be useful to go back and observe changes in markets over time, and even more valuable to compare markets to each other. Unfortunately, the data is largely useless to the various vendors and typically gets deleted after a certain period of time. (Though the real-money markets like InTrade do tend to protect their data as a valuable asset.)

I’ve recently found a solution that could prove quite useful to the prediction market industry: Timetric.

The Future

Timetric is essentially YouTube for time-series data. Their current library of charts (nearly 100k time-series) includes everything from exchange rates to crime data to Twitter replies. These can be manipulated, compared, layered on top of each other, and more. Their standard charts can be easily dropped into blog posts, and the raw data easily exported for manipulation elsewhere. (Full disclosure: I’m friends with the founders, three very smart PhD’s from Cambridge University.)

In my ideal future Timetric would be the home for all prediction market data. Users could compare how InTrade, NewsFutures, and Inkling performed over time on the 2008 election. Those political charts could be compared to economic data or virtually anything else to draw further conclusions. In fact, Timetric is now being used in Guardian blogs (a UK newspaper) to enhance coverage; an example is here.

This is an example of an embedded chart. (It doesn’t display correctly in Google Reader, so please click to see the original post.) Be sure to tick the “Multiple Axes” box for the full effect.

A year and a half ago at the London Prediction Markets conference a number of people talked about what the prediction market industry should do; one of the big ideas was a warehouse for old prediction market data. Now that warehouse doesn’t need to be built; Timetric has done it. All we need to do is fill it with data.

The Challenge for Prediction Market software vendors

My challenge to software vendors is to create feeds that can be easily imported into Timetric. Don’t throw away your data… give it away to someone who values it!

While it’s not explicity stated on their site, they can easily take RSS feeds (specifically Atom) of data. This means that as people trade on contracts the Timetric will be able to update data for the contract nearly automatically. Doing so will make prediction market data permanent, and widely available to academics and the public. It will enable individuals to do their own experimentation and potentially be a great tool for prediction market enthusiasts to debate the merits of methods and approaches for different markets.

Software vendors can also provide a simple CSV or XLS file with datetime information in the first column and values in the second column. While it won’t update like Atom feeds, it does provide the same data for easy import into Timetric.

So what prediction market software companies are open to creating an Atom feed of contract data?

What companies will provide data for a permanent archive?

“In God We Trust; all others must bring data.”

– W. Edwards Deming

There has been more discussion recently on Midas Oracle, here and other blogs about the value of enterprise prediction markets. Part of this is because we don’t have a good measurement of value, and partly because a lack of information.

When it comes to value, there’s one great way to determine if a prediction market is valued by a company: they keep using it! No matter what people “think” the value of an enterprise prediction market may be, if the actual customer is willing to pay for it, the tool is valued.

But another big part of why we’re even having this discussion is because there’s no clear perspective on what’s going on with the industry. Why? Think of the parable of the blind men and the elephant. (Quote from Wikipedia).

A group of blind men (or men in the dark) touch an elephant to learn what it is like. Each one touches a different part, but only one part, such as the side or the tusk. They then compare notes on what they felt, and learn they are in complete disagreement. The story is used to indicate that reality may be viewed differently depending upon one’s perspective, suggesting that what seems an absolute truth may be relative due to the deceptive nature of half-truths.

So the problem is, we’re all talking our of our collective a**es.

There’s no way we can talk intelligently as a community unless we have a shared understanding of what’s actually going on in the industry.

The measurement

I propose that we start by looking at two measurements: retention rate and customer growth rate. Retention rate measures how long a customer stays active. This should be a relatively good indication of the value a prediction market provides to a client; the longer they pay for it the more valuable it is to them!

Customer growth rate is exactly that, how quickly the industry is growing and finding new clients.

While rates won’t provide a total magnitude on the size of the industry, it should provide a good proxy of value to customers and growth of the customer base. It’s not the sum-total of what can be measured right now, but I think it’s a solid first step.

A proposal

I hereby make a public proposal. In order that the entire industry can talk about a common set of data, I volunteer to act as the point of contact for data aggregation. Enterprise prediction market vendors would only need to provide limited data, and would get in return a comparison of their own company’s statistics to the industry at large. The public would get the aggregated statistics (only, no company-specific details) on the retention rate and growth rate of the industry.

The information that would be required of the Enterprise Prediction Markets software vendors is:

A list of clients (who don’t have to be named, code names/numbers okay) and for each of those:

• Date (by quarter) when the client was first invoiced
• Date (by quarter) when the client was last invoiced

I believe that invoicing is a good proxy for measuring the start and end of a prediction market, but if a software vendor would like to use other measures that’s fine, as long as they fairly represent the start and end of a paid prediction marketplace.

Thoughts?

I’d be interested in your thoughts on this. So far I’ve gotten some solid interest from some of the software vendors, but am posting this here to gather some additional interest. (Yes, a little public pressure.)

I think that if we’re confident enough in the prediction market industry, we shouldn’t be afraid of the data!

PS- Because of a recent spate of comment spammers, I’m moderating all comments so you see a bit of a delay before your comment shows up here. But once I approve it, you won’t need to wait if/when you comment again.