Not Your Average Prediction Market

Dr. Ajit Kambil, Global Research Director of Deloitte, authored an interesting piece on the use of prediction markets by CFOs. He presents a nice summary of how prediction markets work and their benefits — that they are a great way to aggregate dispersed insights and capture information that changes over time.

Some implementations of prediction markets compute probabilities of outcomes. To use Dr. Kambil’s example, one could ask, “Will the DJIA end above 10,000 before the end of the year?” Market participants who are bullish about the Dow buy shares. Those who are bearish sell. As trades come in, the market mechanism adjusts the probability of the event actually taking place.

While this is pretty interesting, getting to a collective forecast that gives you a probability distribution of possible outcomes, rather than just a point forecast, of the DJIA itself is awkward. To do it, you’d have to launch different markets around buckets of outcomes — as in, DJIA in 8,000-9,000, DJIA in 9,000-10,000, DJIA in 10,000-11,000, and so on — and then reason about the probabilities of each.

Crowdcast builds probability distributions automatically. This has two important implications. First, it enables a simple user interface and finer grained expression of beliefs. Rather than asking people to choose predefined buckets, they can select a precise range, as wide or narrow as they wish. And second, it supports some great applications for business. For instance, you can get alerts when the likelihood of hitting your target ship date falls below 50%.

Dr. Kambil also discusses the information dissemination “feature” of prediction markets. Prediction markets, like the public stock market for instance, not only collect information, but they also distribute it. While this is great for some applications — surely you’d want to know the price (and other indicators) of Apple stock before you bought some — it’s a problem for others.

In the enterprise, it is often the case that the more valuable and important a metric, the more secret it is. Revenues or earnings per share are but two examples. We’ve struggled with this reality for some time. Initially, our take was that we should simply concentrate our efforts on problems characterized by “public” metrics. Then we implemented access control lists, which enabled our customers to publish questions about sensitive metrics to a subset of the participants.

While this worked for some applications, the overall approach just didn’t sit well. There is knowledge in the enterprise about revenues and plenty of other sensitive metrics. And we’re all about harnessing knowledge. Won’t let the cat out of the bag yet, but we’ve cracked it! Details very soon.