Adam highlights the inevitability of prediction markets’ integration with CRM/ERP systems. Bo leads discussion to the right direction. I’d like to speculate further on Bo’s first point.
The integration with business intelligence software, corporate data warehouses and document indexes will definitely facilitate the task of information collection by traders. I claim that such an integration will also enable the introduction of artificial intelligence agents, namely virtual actors that would trade on the relevant information and knowledge extracted by the CRM/ERP software.
This introduction of a hybrid prediction market, a prediction market populated by both human and artificial agents (term just coined), could solve some problems, such as low trading volume and would by default incorporate extracted explicit knowledge, while also creating some new, like for example the need to fine tune its algorithms. But, after all, I believe that it could trigger a new fascinating research field (machine learning powered, data and textual actors featuring in a prediction market) and empower new hot business applications.
A lot of questions arise. To name a few:
- Could or will such hybrid markets perform better than traditional pure ones?
- Do virtual agents negate the typical virtues of a prediction market (e.g. tabula rasa opening, harnessing ‘just’ the collective intelligence, etc.)?
- Does a market-based meta-algorithm perform better than traditional data mining meta-algorithms (e.g. pure artificial markets vs adaboost)?
- Will such markets be included in the declaration of establishment of The Institute of Prediction Markets or in a forthcoming literature review?
The journey will probably be long but fertile; the ideas exchange is now open.
George,
A few years ago when I worked for Accenture in their Technology Labs, much of our research was around a grand vision we called “Reality Online” which basically meant every physical object had a virtual double: humans, machines, widgets, etc. “Dead” objects would have a bunch of sensors reporting out about their status, etc. in order to track their health, repair them before they completely break down, be aware of their environment and react accordingly, etc. And humans would have virtual doubles representing them in all sorts of actions, programmed with rules on how to behave in various situations. A crude example: want to schedule a meeting with 6 other people? Just tell your virtual doubles to work it out.
Thus to your point, I could see your virtual double trading in a prediction marketplace after you’ve “programmed” it to behave a certain way and/or it has learned your trading behavior after monitoring you and learning over a period of time. And of course I don’t just mean “buy shares when the price reaches $35″.
George Tziralis’ vision is not Sci-Fi. I’m all for the day dreams of advanced applied researchers. Some of these day dreams may be interesting and become reality, one day.
Tom Malone at MIT is interested in this idea too. I think he is exploring it with an outside corporate research partner.
Also, note that what you’re describing is exactly what WeatherBill does. Weatherbill markets itself as a form of insurance, but it is basically a prediction market on the weather. As you describe, it has an algorithmic trader in the market that looks at vast quantities of current historical weather data.
Using this data, the automated trader formulates a prediction and acts as a market maker. It not only provides liquidity for the traders, but does so at a price that won’t rip off the company.
It might be interesting to see how predictive WeatherBill’s prices are. However, it would be hard to generalize on the basis of the results. Other human + machine “hybrid” prediction markets would have different algorithms and might not behave the same.
The range of available, relevant data is more important than whether the algorithms can be described as “artificial intelligence”, etc. Ultimately the methods are going to be buying or selling because of new highs, selling or buying because of new lows, absolute and relative (that is, prices and spreads). So “buy shares when the price reaches $35″ may be more correlated to more sophisticated-sounding methods than expected.