\ Prediction Markets | TechnoTaste

Prediction Markets

The latest issue of the journal Episteme has all the buzzwords of current topics. It's about Web 2.0, Wikipedia, prediction markets. And epistemology. Big time.

But don't bother. This journal is quite pretentious, and the articles are not particularly well written. Worse, there doesn't seem to be much in the way of new thought there. Rather it seems like a handful of scholars who have just caught on to what's been going on and are talking about it as thought it's the latest and greatest. It reminds me of going to a session at the American Anthropological Association meetings in 2003 and listening to a sincere anthropologist give a 6 point talk on how to use Powerpoint.

The only gem seems to be a long-ish article by Larry Sanger (Wikipedia co-founder) on expertise on Wikipedia:

Sanger, Lawrence M. 2009. "The Fate of Expertise after WIKIPEDIA." Episteme 6:52-73.

I'm not sure I agree with most of what he has to say – I need to stew on it a bit more. But regardless I think it's an interesting piece of opinion and history from someone who was there at the beginning.

You can imagine that, in an election that's as enthusiastically watched as this one, there's a lot of attention to predicting the outcome. Sites like fivethirtyeight.com use sophisticated statistical models to make predictions based on polling data. I've written about this previously.

Another mechanism getting a lot of attention, however, is the prediction market. Prediction markets are just like stock markets, except that instead of investing in shares of a public company you invest in a future outcome, such as Barack Obama winning the election. When you buy a stock on the NYSE or NASDAQ, what you're really doing is making a prediction that a particular company will do well in the future, or at least that it's stock price will do well. Prediction markets take that basic mechanism and turn it into a way to predict. In a prediction market, the price of an outcome going up and down reflects the collective intelligence about how likely that outcome is.

At least that's the idea. The New York Times has an interesting article today about recent problems with prediction markets around the upcoming presidential elections. Apparently, among three top prediction markets, InTrade, BetFair, and the Iowa Electronic Markets there's been a great deal of variation in predictions at any given time. In other words, the predictions don't agree. This is not supposed to happen.

Prediction markets work on the principle that if you give enough rational, self-interested, maximizing people the chance for tangible benefit from a future outcome, all those people acting in their own interest will give you a very good idea of what's going to happen. It's in their interest to predict as best they can, and when lots of people do that, the invisible hand works to set the right price / prediction. Meanwhile, I can bet on a prediction at any one of the prediction markets. Theoretically, I'll put my money in the one that gives me the best chance for a gain. If InTrade is predicting a John McCain win, I'm buying Barack Obama as quick as I can if I'm a Democrat. And vice versa. The effect should be to reduce the variation in price across the markets.

But apparently that isn't happening. The NY Times article notes that InTrade saw a giant swing towards John McCain recently that wasn't matched in any of the other markets. Apparently, there was a single 'institutional investor' who was betting big on InTrade for 'political reasons' and skewing the results. At least that's the story they tell. But that's not supposed to happen either. In a big market like InTrade, outliers should not be able to push the price around that much. The bigger the market, the more buying or selling it takes to artificially change the result.

It begs the question – how big does the market have to get in order to be insulated against outliers. Or, can it ever really be insulated at all? We're starting to put a lot of stock in prediction markets. For sure, some of it is justified – the Iowa Electronic Markets have successfully predicted the winner of the presidential election many, many times.

But I don't think we know enough about how incentives work in prediction markets yet, especially on issues like presidential elections. I'm not going to fight 100s of years of economists who have relied on the assumption that people seek profit for themselves. No doubt that's a powerful motive. But it's not the only one. Knowing that people are paying attention to prediction markets, that those markets have the power to sway perceptions and votes, maybe my incentive to invest in John McCain isn't only based on wanting to walk away with a new flat screen TV, it's also based on my Republican ideology. I'm willing to bet my $ on the chance that swaying the market could sway the election.

Is that so far fetched? I don't think so. NY Times notes that big-time political sites like RealClearPolitics list the prediction market prices right there next to the polls. I'm not claiming that the prediction markets are broken. I'm just claiming we don't know. The assumption that profit is the only incentive at work is unfounded in my view. In high-stakes situations like this, where other powerful incentives exist, we really need a deeper understanding of what drives people when they bet.