November 1, 2004
From the desk of Jane Galt:What do the markets say?
Lots of people are parsing the polls. But that's tiresome, plus I don't know anything about polling, so people won't listen to me. (Waaaaaaaa!) On the other hand, there's another way of tracking the race that I do know something about: electronic betting markets. These markets (the most famous are Tradesports and the Iowa Electronic Markets) let you bet on who you think will win the White House tomorrow (or next month, if it comes to that).
The markets have two advantages. They let us pool our collective wisdom, which as James Surowiecki showed in his excellent new book, The Wisdom of Crowds, is a better method for predicting the future than asking even the smartest of experts. (This is particularly true of questions about matters of popular opinion, such as election outcomes.) And they force people to put forward their very, very best guess by making them put money on the line.
So what do the markets say? Iowa gives a 55% probability that Bush wins. Tradesports says something similar. Tradesports has another great feature: it has people betting on outcome by state. So I did a little number crunching over my lunch hour, looking at the state data two ways. Assuming that the prices, which are based on $100 contracts, represent a probability of winning in each state (i.e. a price of 96.7 translates into a probability of 96.7% that Bush carries the state), I simply divided the states by their price: anything with a price above 50 went into the Bush column, while anything with a price below 50 went into the Kerry column. Using that method, we get an electoral vote count of 286 Bush, 252 Kerry, with Bush carrying Ohio, Iowa, Florida, New Mexico and Colorado, but losing Wisconsin, Minnesota, New Hampshire, Michigan and Pennsylvania.
But there's a more sophisticated way to look at the data, which takes into account the fact that the probability of Bush's carrying Ohio is not the same as the probability of his carrying Alabama: we can use an Expected Value calculation.
(nooooooo!!!! I hear you cry. Not math! Please, I'll do anything!)
Really, it's very simple, and very useful. To get the expected value, you just multiply the value at stake, by the probability that it will happen.
Let's look at two states with 10 electoral votes: Maryland and Arizona. Tradesports has priced the Bush/Maryland contract at 3, giving us a probability of 3% that Bush will get Maryland's 10 electoral votes. Maryland's expected value for Bush is therefore
0.03 X 10 = .3 Electoral votes
In Arizona, on the other hand, the Bush contract is priced at 92. So we calculate Arizona's expected value as
0.92 X 10 = 9.2 Electoral votes
If the election were being held in just those two states (with just those two candidates), we'd add Bush's two expected values together. If he passed 10 (50% of the 20 total electoral votes available), we'd expect him to win; if he didn't get past 10, we'd expect Kerry to win. In this case, his total expected value would be
9.2 + .3 = 9.5 electoral votes = Bush loses
Luckily for Bush, there are more states out there. Running the whole country through an expected value calculation gives us an expected value for Bush of 273.138, and 264.862 for Kerry -- in other words, a real squeaker.
Interestingly, I designed the spreadsheet I used to look at this months ago, and outside of about 5 states, nothing has really changed, except that the numbers have gotten more extreme -- except for the few battleground states, the probability of Bush winning in a given state is generally either 95+% or less than 5%.
I also used it to test the thesis of The Emerging Democratic Majority by looking at growth in Bush states v. Kerry states. At least by the Tradesports data, the theory doesn't hold up: in the last 20 years, the solid Kerry states have lost 9 electoral votes, while the solid Bush states have gained 8. You can get the whole spreadsheet here.
My prediction? Well, I'm not so good at predictions -- I thought Bush was toast three months ago. But I will say that at the current market price of roughly 45, I find the John Kerry contract to be a pretty attractive value proposition.
But of course, all this speculation will be meaningless once the election happens. Tomorrow night, I'll update the spreadsheet to show how the markets did at predicting.
Update Great piece on the markets from Linda Seebach at the Rocky Mountain News.
Posted by Jane Galt at November 1, 2004 4:06 PM | TrackBack | Technorati inbound linksMegan - take a look at this:
http://fairmodel.econ.yale.edu/vote2004/index2.htm
Then take a look at his August 15th interview on the Ny Times:
http://www.nytimes.com/2004/08/15/magazine/15QUESTIONS.html
Nice, Jane. I feel a little silly that the expectation approach didn't occur to me.
I've use similar methods, but I dropped the "expected value" approach because a price of, say, 95 on a Bush win in Alabama is tantamount to certainty that Bush will win there, while a price of, say, 5 on a Bush win in Massachusetts is tantamount to certainty that Bush will lose there. The outcome in each state is really binary (with the minor exceptions of Maine and Nebraska): win or lose. Carrying on with the example of Alabama and Massachusetts, the expected value method (using my made-up prices) gives Bush 9.15 electoral votes and Kerry 11.85 electoral votes, when the real result will be 9 for Bush and 12 for Kerry. In practice, the bias might run the other way, but there is a bias in the expected-value method, so I wouldn't take any solace from the result you obtained.
Jane -
Geekmedia has had a page tabulating the Tradesports results by state, including an EV-distorted map (which I suggested to the site operator). Downloading the page and opening it in Excel made it real easy to do your EV-EV result. At 15:00 EST on Nov 1, the result is a little better for Bush - 273.78 EV, and at 18:00 EST, 274.17 EVEV.
Please suggest doing an "expected value" result to the site operator - it's probably too late for this year, but it would be nifty to have for 2008
" I find the John Kerry contract to be a pretty attractive value proposition."
If 45 looked good enough for enough people, the price would go up. That's the whole point of the market based indicator.
Or are you missing something?
You might want to check out IEM, because they showed today Kerry overtaking Bush in the Winner-Take-All market. BTW, I'm unconvinced about their predictive power. Once again, they appear to be good aggregators of conventional wisdom (i.e. it's very close) but say little to nothing about what will happen in the future.
Stimulated by this post, I've run a numerical analysis on the tradesports' state-by-state data which gives the following probabilities:
Bush Win: 57.5%
Tie: 1.5%
Kerry Win: 41%
Bush also wins with a tie since the House casts the deciding vote, so it's 59% that Bush wins, 41% Kerry.
So shorting Bush in the States and buying nationally is theoretically an arbitrage strategy, but it's actually arbitraged better than the spread already, so no new information or money making opportunities available.
The expected value calculation, while interesting, isn't completely valid. The problem is that Bush can either win AZ or not - he can't possibly get 9.2 EV from it.
But that is the point of EV calc. Of course he can't get 9.2 EV, but he has a .92 chance of getting 10. Just like you can't get 0.5 heads flips of a fair coin, it's still the E(heads) for a fair coin flipped one time.
Jane,
Contracts at Tradesports are worth $10. Though, for example, the convention practiced on the website is to display a bid of $9.57 as 95.7, so I understand your assumption. Of course, this doesn't change your analysis.
-CJ
OK, I'll retract the last part of my comment about it not being completely valid.
They let us pool our collective wisdom, which as James Surowiecki showed in his excellent new book, The Wisdom of Crowds, is a better method for predicting the future than asking even the smartest of experts
I agree, but that sure didn't help DARPA's Terrorism Futures Market, did it?
"... but that sure didn't help DARPA's Terrorism Futures Market, did it?"
Well, DARPA's problem was that they didn't anticipate the public reaction correctly. Perhaps the first step for DARPA should have been a "The public will accept a Terrorism Futures Market" futures market :)
I'm surprised Jane.
The Economist's final poll has Kerry winning.
Shouldn't that be enough for you?
;)
There's something funny going on at the Iowa electronic markets...
DEM04_L52 closed yesterday at 0.305. But sometime today it traded at 0.474.
REP04_G52 closed yesterday at 0.230. But sometime today it traded at 0.070.
Obviously somebody is trying to manipulate the market.
There haven't been any significant changes in the polls since yesterday, so that can't explain shifts like that. Plus, these big movements are in a small Kerry win and a large Bush win. But the market barely moved for a small Bush win, even though that is the intermediate result.
I saw it too. I think the explanation is simpler (see Occam). People are realizing that G52 is a loser on either side, and are ditching it. This is creating dislocations in the sums of the Dem and Rep number, as DemL52 approaches 0.5, and so does RepL52. Why would anyone manipulate this market? No one cares about it except poll obsessives (present company included!).
Best Poll for undecided voters;
Since 1956, Weekly Reader students in grades 1-12 have correctly picked the president
http://www.weeklyreader.com/election_vote.asp
Weekly Reader kids select Bush in Presidential Poll
The students who read Weekly Reader’s magazines have made their preference for President known: they want to send President Bush back to the White House.
The results of this year’s Weekly Reader poll have just been announced, and the winner is President Bush. Hundreds of thousands of students participated, giving the Republican President more than 60% of the votes cast and making him a decisive choice over Democratic Senator John Kerry.
Since 1956, Weekly Reader students in grades 1-12 have correctly picked the president, making the Weekly Reader poll one of the most accurate predictors of presidential outcomes in history.
Charles Manski would call your assumptions naive.
(link in Name)
GT, nope, someone has manipulating these markets for some time. Donald Luskin analyzed the data. Thankfully they're self-correcting. Bush is now at 52.
handy tool:
http://www.geekmedia.org/tradesports
Manipulation - Donald Luskin makes a pretty compelling case:
http://www.nationalreview.com/nrof_luskin/luskin200410181132.asp
DARPA - The original concept was envisioned as an "information futures" market "based on economic, civil and military assumptions" according to Robin Hanson, one of the project's architects. Terrorist events were used strictly as a demonstration scenario that resulted a premature fundamental misunderstanding and the end of consideration:
http://www.fenews.com/fen34/one_time_articles/geopolitical_futures/geopolitical_risk.html
I notoiced that tradesports correctly picked the presidential winner in each state if you go back a day or two. Far better than the polls. I wonder how the senate race predictions came out. I don't have the back data.
Comments are Closed.