Thursday, July 31, 2008

I just don't get no respect -- about prediction markets

You've heard of prediction markets like InTrade, right? Basically, like gambling, but on real-world non-sporting events, so as to aggregate the market's knowledge about the future, and reward those who know and "share" their knowledge. GMU professor Robin Hanson did a lot of the work in formulating them and encouraging their development.

Now, if only he could recognize the insights of others.

Back in January, his blog posted Peter McCluskey's idea to use prediction markets to unveil another kind of information: how a presidential candidate impacts prices in financial markets, such as oil futures, government bond yields, etc. And how does he propose to do it? The prediction market would host a bet on a measure of their correlation. Which measure? The (modified) ratio of how much the candidate's contract price (i.e. market's estimate of chances of winning) changes to how much the financial security's price changes ... on election day. ( a so-called "shock response future")

Great idea, I thought -- but wrong measure. Election day by itself is unreliable. After all, more than just the election will influence the financial security's price that day. Plus, the market has already largely incorporated the influence of whoever's expected to win except in very close races. Worse, it's way too easy to manipulate: want to "prove" Democrats make interest rates low? Eat a loss by buying treasuries at an absurd price at a critical time, just like that gentleman who ate a loss on oil just to be the first to buy oil at $100/barrel.

So, I told them exactly that and suggested a better measure: don't just look at election day: measure the correlation all the way up through election season. Find how often e.g. oil's price goes up as a Democrat's chances of winning go up, and look at how closely they track each other, each day or week. That's far more robust against manipulation, and extracts much more relevant information. Yet arguing the point with McCluskey and Hanson was like talking to a wall: they were responding to distortions of my idea that seemed to have only a partial understanding of it. For example, Hanson argued that no, no, no, silly: we need a metric available before the election ... which mine is. No no no, Hanson really meant something completely different from what he actually said.

Well, McCluskey went ahead and launched his inferior futures market on the InTrade site. And, just two days ago, Hal Finney launched a celebration of those futures markets' "success" with a blog post that developed the implications of the prices those contracts traded at. Major posters gave McCluskey a good pat on the back.

(Re-)Enter Silas.

I remarked that I had suggested a better metric on the earlier thread, and asked if the people there would be more interested in a futures market for my idea. A few were, and one major poster reluctantly admitted that I had a better metric and deserved acknowledgement.

A major implication of my criticism, let's not forget, was that too many other forces impact a financial security's price on any given day (including election day) and the market has already incorporated most of the impact of whose expected to win (or, of course, Senate/House elections could impact as well...). This means we should expect the market McCluskey made to be rather useless -- traders will view the correlation on just election day, as effectively random -- a 50/50 chance either way. And what do we see? Yep: "49.9-50.1" -- or an implied 50/50 chance.

As Finney sheepishly avers: "These values are so close to the 50% mark that it appears that the markets do not expect any significant movement in oil prices or interest rates on election day, that can be attributed to developing information about which party will win. As critics have noted, this could be because they don't see much effect of political parties on these values, or else because they expect that the election day results will be a foregone conclusion and there will be no surprises in that regard."

Or, reading between the lines, despite all the kudos we're giving McCluskey, he created a market that provides completely useless information, even though I told him long before how to make it useful.

Hanson tried to save face by replying again, so I had to gently correct his misguided attempt to trivialize my insights. Okay, not so gentle -- but when you're so wrong, and for such wrong reasons, why do you expect a huge amount of respect right back? Especially when you're the one always complaining about how stupid it is that you have to gain social status just to get people to listent to your good ideas!

What's especially interesting is when Hanson alleges that, duh, of course he had considered my idea. It's just an obvious variation! And yeah, in a way it kind of is. But judge his attempts at response for yourself -- are those the remarks of someone who has considered the idea and rejected it? Would he constantly respond to misunderstandings of the idea? Make so many misstatements about it? Not notice his true complaints applying to his own idea?

Furthermore, even if he does think the election day correlation is so much more important, why didn't he suggest the further "obvious" variation of simply increasing election day's weighting in the correlation, which would still retain the measure's robustness against noise and manipulation?

His latest response to my damaging criticisms? Silence. A wise, wise choice.

Will they go ahead and belatedly implement my idea before I go ahead myself with it? It will be very, very funny when they do.


Robin Hanson said...

Maybe you should call me on the phone and have a regular conversation, instead of leaping to all these conclusions.

Silas Barta said...

!!! The last time a computer-savvy expert that I ticked off invited me to talk on the phone, it was a ruse to gather enough information to sic his friends on me. Once bitten, twice shy.

I'd be glad to vidchat, skypechat, or IM chat later; maybe you've already emailed me with your info for this.

Anyway, I don't think it's fair to say I was "leaping" to conclusions -- you gave very strong evidence in that January thread that you hadn't thought of the idea or at least not considered it thoroughly. Plus, if it was so obvious, it should have also been obvious to further modify my suggestion to give more weight to election day.

Silas Barta said...

Yikes! I hadn't even authorized my email to be shown in my profile. Oh well, you should have it from O/B.

Robin Hanson said...

My phone info is on my home page.

jck said...

"Or, reading between the lines, despite all the kudos we're giving McCluskey, he created a market that provides completely useless information..."
Agreed, most of the trading was in the contracts was due to the exploitation of subsidy due to initial miscalibration.
Here is an old comment on Peter's blog.
“A large fraction of the trades in NONDEM.PRES-GOVT.DEBT and NONDEM.PRES-TROOPS.IRAQ fit a pattern of driving the price down.”
You should not be too surprised, you have provided a free bet actually you have paid somebody to bet on a Republican President funded by shorting NONDEM.PRES-GOVT.DEBT or NONDEM.PRES-TROOPS.IRAQ and they will win regardless of the outcome.