Tuesday, November 23

Information markets: how cool is this?

This Economist article on information exchanges astounds me. (You can only read it if you have a subscription, or if you buy access to it. I happened to read it in my paper copy of the magazine; if you know Fraxas personally, email me and I'll flip you the article in email.)

Essentially, the article describes (and prognosticates on) a relatively novel kind of exchange. Rather than selling securities in companies (aka stock) or derivative instruments, information exchanges sell futures on events yet to happen. That probably doesn't mean much to you unless you follow the markets, so here's a longer explanation cribbed from the article:

Take the presidential election: people could bet on George Bush by buying a contract that paid a dollar if he won and nothing if he lost. Anyone certain of a Bush victory should have been willing to pay up to a dollar for the contract. Anyone confident that Mr Bush would lose could have sold such a contract, expecting to pay nothing when the result came. With many participants buying and selling in this way, the market discovered a price for the contract—in effect, its best guess of the probability of a Bush win.

Think about that for a second. This is a mechanism that exploits people's (enlightened?) self-interest, in the form of their desire for money, to collect their opinion on a particular topic. It discourages the intentional and unintentional fraud inherent in phone surveys, because of the monetary stake. It also introduces a novel (for the opinion-gathering industry) and well-tested set of tools for opinion analysis: the discipline of financial economics.

That reminds me of the proof of Fermat's last theorem. (Simon Singh's account is a fantastic one, in case you care.) The proof relied partially on the realization that two very disparate branches of math were, in fact, solving the same problems from very different angles. Things difficult in one context are easy in the other, and vice versa. So perhaps thinking of opinion-modelling as a financial market will increase the accuracy of the opinion marketing. Additionally -- and this is a topic for a different essay -- could the same be true in reverse?

Another interesting aspect of information markets that isn't adressed in the Economist article is that these markets, if they become popular, allow people to monetize their knowlege in a relatively direct way. If I know more than the world in general does about a topic that's traded on an information exchange, I can exploit that knowledge for personal gain. Entrepreneurial current-events analyst could very well be a respected career choice.

Capitalism wins again!

1 comment:

JeremyHussell said...

The following is an edited transcript of Fraxas talking to Jeremy about this post.

Jeremy: The election polling angle still has a problem: people will buy and sell shares based on their opinion about who's going to win, not who they'll actually vote for. When voting time comes around, they may make their personal statement by voting for the candidate they like, rather than the one they think will win, in part because nobody will ever know who they voted for (unless they tell someone). If the public perception of who's going to win is way off, the market price will be just as wildly off as any telephone poll, if not more so.
Fraxas: that's addressed in the article.
Fraxas: and, as with all opinion research, you have to understand thoroughly what it is you're asking about.
Fraxas: but yes -- if perception is off then you'll have a market failure on election day.
Fraxas: but you can hedge against that by allowing derivatives on the contracts to be traded on the exchange.
Fraxas: By allowing trade of more than just the contract that pays on Bush being elected, you can circumvent that perception issue.
Fraxas: like, for example, you could float a Kerry contract as well
Fraxas: and a contract that pays in the case of an election being contested
Fraxas: and derivatives -- futures, contingent trades, options -- on each of those instruments.
Fraxas: s/instruments/contracts/
Jeremy: How does that help if 3/4 people think Bush is going to win, but plan to vote Kerry anyway?
Fraxas: I guess it doesn't.
Fraxas: but a lot of people answer "Do you think Bush will win?" and "who do you think will win?" differently.
Fraxas: anyway. you're absolutely right -- information exchanges measure opinion and not fact.
Jeremy: I suppose on the latter, they tend to respond with the name of the person they were going to vote for more often?
Fraxas: and "who do you think will win?" and "who will you vote for?" are two different questions.
Fraxas: (I was just making a general point. Most really good surveys have redundant questions in them, to catch variance like that.)
Jeremy: How do you make a market based on "Who will you vote for?"?
Fraxas: you can't.
Fraxas: well, in that specific instance you can't.
Fraxas: because it would require a non-secret ballot for payout.
Fraxas: ...and democracy really doesn't work without secret ballots ;)
Jeremy: Indeed.
Jeremy: Doesn't stop people from trying to figure it out anyway.
Fraxas: well, no.
Fraxas: but you know, those are all points to be noted in analysis.
Fraxas: and you're right -- markets like this elucidate the *consensus* opinion of their participants
Fraxas: so they're susceptible to GIGO.
Fraxas: I wonder if you'd get better results if you could establish a large enough pool of experts.
Jeremy: The trick, then, is telling apart the questions where the collective wisdom is correct, and the questions where you're better off asking an expert, and the questions that nobody can answer correctly.
Fraxas: or only using the tool when what you're interested in is what you can get.
Jeremy: BTW, page 270 of (Simon Singh's) "Fermat's Enigma" has a picture of someone who looks frighteningly like me.
Fraxas: yeah, I remember that
Fraxas: I was like "wtf, where does J find time to be an RA in london"
Jeremy: Kind of spooky, that.