Market Efficiency: Tradesports vs. IEM

There is a great post over at Crooked Timber about the Iowa Electronic Markets. Daniel Davies, like many, finds serious inefficiencies in IEM and wonders why. I'll give him my standard answer: $500 deposit limit. Why does this matter? DD's post contrasts Robin Hanson markets and James Surowiecki crowds. While his contrast may have been incorrect (as Robin commented), its a useful metaphor for explaining why I expect less efficiency on IEM than Tradesports.

Let's contrast two types of opinion aggregation: the stock market and the Who Wants To Be A Millionaire crowd voting. While both are an improvement over one person's opinion, I believe the former generates better information.

In a market, the most skilled people get to wield the most money and have the most impact on prices. If I'm really good at picking stocks, I can get hired by a big hedge or pension fund, and use hundreds of millions of dollars to correct mispricings. In a sense, I'm getting thousands of times the "vote" as someone with only a hundred grand to invest. With a crowd vote, generally all votes are weighted equally. People who are superior decision-makers don't get extra influence.

Tradesports is more like a market. IEM, because of the deposit limit, is more like a crowd, where no one person has many more votes than another. So its not surprising that the folk on IEM are not betting very rationally. Someone who kinda wants to bet Kerry because its fun is not going to do so at an outrageous price, but he's not going to use the Black-Scholes equation either. And someone who wants to carefully calculate the right positions is not going to find IEM worth their while because of the low limit.

The difference between the systems is another manifestation of my idea about eliciting hope vs. belief. When you control a huge sum of other people's money, people who will fire you if you screw up, you force yourself to act more on belief. When you're controlling a little of your own money, a fair bit of hope gets mixed in. Not as much as when you are voicing costless opinions (like a vote), but still a fair bit.

Its also similar to why its easier to beat a poker game than the stock market. No matter how good a poker player you are, you still only get one seat in one game. Because the skills and experience to play poker well are rare, and there are many bad players, there are always opportunities for good players to make money. But in the stock market, its almost as if the best player can take as many seats at as many tables as he wants. As you might imagine, this makes the competition much, much tougher.

Aggregating opinions like IEM is still better than nothing. The audience does better than any one expert at WWTBAM. But if we let audience members bet $1 to $100 on their answers, and gave their votes weight proportionally, I'll wager the audience would do even better.

Share this

This sounds plausible. It

This sounds plausible. It would be great to have a careful statistical analysis comparing IEM to other no-limit political markets such as TradeSports, to see which one actually predicts better. Unfortunately IEM and TradeSports differ on other dimensions as well, such as where their traders are located. Perhaps some lab experiments comparing limited to no-limit markets are in order.

It seems to me that in the

It seems to me that in the medium run -- several years to a decade, say -- the appeal to other people's money isn't necessary unless the market doesn't bet on short-term things, or is already almost as smart as any greedy self-confident individual. (Well... or perhaps has weird statistical properties which make it impossible for a bettor with better knowledge to place safe bets, or is in a weird society where no one can scrape together a starting stake, or has some other pathology.)

If a smart bettor can bet on things which pay off after, say, a month (like sports results or next month's commodity prices), and he understands that the market price is consistently significantly wrong, then the growth in a smart bettor's worth (by something on the order of the ratio of AccuracyOfHisPrediction to AccuracyOfMarketPrediction) could be compounded every month. Then if a smart bettor can scrape together the resources to start with $50k or so in today's terms, either his advantage over the market must be small, or it won't take impossibly long before he can place such big bets that all by himself he can drive a reasonably-sized market to become more efficient: either the other bettors learn to bet not too much less accurately than he does, or they drop out of the market, or they go broke.

One thing- with online poker

One thing- with online poker these days, you can play many tables at once (as my roommate John does), like 4 or more depending on how big your monitor is, or in some cases how many monitors you have hooked up (limited, of course, by your mental capacity to handle 4+ tables worth of poker).

Brian - in the long run,

Brian - in the long run, that plus AI poker bots is going to be a big issue for online poker.

William - I think you're overestimating the size of the edge and how often it compounds. You don't get big short-term edges. If you look at a more reasonable model, say a good bettor can make 20-40% a year, then even with compounding it takes a long time to turn $50K into the billions of dollars that would love to earn that return.

But you're certainly right that eventually just being a better bettor is enough. How long depends on how much better you are.

"Brian - in the long run,

"Brian - in the long run, that plus AI poker bots is going to be a big issue for online poker."

I think poker bots will probably devastate the online game fairly soon, unless I'm missing some way to keep them out. I don't think I am.

Good post. Another factor -

Good post.

Another factor - often overlooked - is the actual predictability of the issue in question. Some things simply can't be predicted well, either by individuals or crowds.

Asking people to bet on a question which is very hard to actually predict would be a good way to distinguish between hope and belief. Maybe you could use such questions to evaluate the overall effectiveness of various prediction-market implementations. Something which is basically unknowable should sell pretty close to 50%.