« What I think I am nearly certain about | Main | What I've been reading »
The Candidate Doesn't Matter
Despite large swings in the market for presidential nominees over the past few days, on the Democratic side Clinton has dropped and Obama risen by almost 15 points and on the Republican side McCain has surged to take the lead (!), there has been very little movement in the winner take all market which is predicting a Democrat win. Thus, the markets seem to be saying that the candidate doesn't matter. In this election it's party and the Republicans are losing.
Winner Take All Market from IEM (click to expand).
Posted by Alex Tabarrok on January 4, 2008 at 10:59 AM in Economics | Permalink
Comments
Ok, Alex, the blue spike (to 95!) around mid 07...how does the explanation run?
Also, about a year earlier...it *was* a toss up whether the Dems or the Reps would win? Really? This tells me that IEM players are *not* fair and balanced (R/D) gamblers...and that current 40/60 split is so desperately modest.
Posted by: calmo at Jan 4, 2008 11:37:20 AM
Alex, do you have any thoughts on the meaning of sizeable arbitrage opportunities between the different election markets that Brian Weatherson pointed out on Crooked Timber a couple days ago? His upshot:
"I’d think most of the good arguments for taking betting markets seriously as a predictive device make assumptions that entail there aren’t arbitrage possibilities of this size."
Matthew Yglesias also recently blogged about why he thinks betting markets are silly, but his argument was pretty facile: he retrospectively calls the prices absurd and thinks he'd make a bunch of money if only he felt like playing. (I can't find the link right off.)
Here is a more sophisticated post summarizing an academic article challenging the betting markets as predictors of elections, but I cannot entirely wrap my head around it. I hope you guys or Robin Hanson will have something to say about these challenges.
Posted by: Lee at Jan 4, 2008 11:43:32 AM
As usual I tend to make less informational and more structural interpretations of these thin markets.
The comparable Intrade markets had a similar non-reaction. It could just be that the active trading capital is occupied with the nomination and primary markets and there is no reason to make significant allocations to the Dem/Rep markets right now, especially given the margining system.
Also, the markets were not surprised by the outcomes and predicted them more decisively than the polls, so if the markets were not surprised, we shouldn't expect a big shift in the Dem/Rep markets.
Posted by: Jason Ruspini at Jan 4, 2008 11:45:55 AM
The market in vote share seems to show a much more stable and narrower difference..
Posted by: Diversity at Jan 4, 2008 11:46:14 AM
Lee,
Brian was just confused about what it means to arbitrage, as he notes later in the comment thread.
The basic finding you should operate with is that prediction markets are good predictors of future events - this is a robust and *very* well established result using years of data from dozens of markets around the world for a great variety of events. Furthermore, prediction markets are but one species of financial and betting markets for which there are many thousands of studies showing similar results.
I haven't looked at the Erickson and Wlezien study closely but the way to interpret it is simple - they are telling you how to make money on the stock market. Do you believe them? If you do, how long do you think these profit opportunities will last? (i.e. before market prices adjust to reflect events even more accurately?)
Posted by: Alex Tabarrok at Jan 4, 2008 11:59:32 AM
Thanks for the response, Alex.
I have some more reading to do about these markets and the fundamentals of these sorts of Bayesian/betting ideas. It's an interesting topic, especially the appeal to my TANSTAFL-intuition that any problems found with them would be self-correcting.
I'd completely surrender to the idea, but my feel-it-in-my-gut skepticism keeps getting support from elite sneers on the blogs, and the fact that the msm mostly ignores it, excepting Slate.
Posted by: Lee at Jan 4, 2008 12:28:55 PM
I'd venture a guess that the spike was a first timer screwing up and putting in a bid way over market value. I know I've done it before. I bid that the Republicans had a 50% chance of winning New York in November. Biggest waste of $5.00 ever.
Posted by: Explain at Jan 4, 2008 1:12:00 PM
Alex, did I miss your mea culpa on debt? If you would like to increase my happiness, please write one.
Posted by: odograph at Jan 4, 2008 1:17:15 PM
I think the Republican party will have to be destroyed in order to save it. Most of them seem completely unwilling to concede how badly they've fouled up.
Since the Republicans seem doomed anyway, it's a shame the Democrats haven't gone for more principled but otherwise less "electable" types like Dodd or Gravel.
Posted by: TGGP at Jan 4, 2008 2:19:24 PM
Speaking as a "lapsed Republican" I think it's the hard right wingers (or the escapees to Ron Paul) who stay interested.
... I don't know. As a more libertarian leaning California Republican I've been out of sync with the Evangelical connection. In, say, 1985 I didn't see it coming. I thought that a market friendly but less rigid party would emerge.
Evangelicals will be the last Republicans, right? And the party will be centered on their Litmus issues.
Posted by: odograph at Jan 4, 2008 2:31:47 PM
Alex,
SUE, e.g., was found and published several years ago, but still is a money maker.
Posted by: hubris at Jan 4, 2008 4:34:04 PM
Perhaps we can safely disregard the market here, as in retrospect many ought to have done with securitized mortgages... Item: betting markets got Virginia 2006 wrong until almost the last moment, even though the symmetrical polling trend in favor of Webb and against Allen was apparent months in advance, (in fact, nearly from the moment of Webb's announcement,) and even a dimwit like me called it six weeks prior. Item: the only candidate Hillary is likely to fear is McCain, because five years as a prisoner of war is a campaign bio item you really can't trump. So the market, once again, may not be thinking too clearly. But the Dems will keep Congress.
Posted by: Lee A. Arnold at Jan 4, 2008 8:07:38 PM
Evangelicals will be the last Republicans, right? And the party will be centered on their Litmus issues.
If so, then it would not be out of the question to expect a realignment. Huckabee is, for example, certainly not a "hard right winger" by any means; he would have been a fairly conventional religious left/conservative Democrat in many other eras, the kind that was mostly stamped out for a variety of reasons until the Democrats began nominating several in 2006 in order to retake Congress. (Remember, Carter won the Evangelical vote handily in '76 against Ford.) As one libertarian-leaning, I would not necessarily be happy with such a realignment, at least for quite a while, since most libertarian views don't command a majority of the population.
And of course, at this point in '92 Bush was a shoo-in for re-election, and Clinton was about to get 3% in the Iowa caucuses, etc., as everyone will remind people this week. Part of the Republican problem right now is that they have a lot of candidates who appeal to particular parts of the old coalition, but not enough to all
Posted by: John Thacker at Jan 4, 2008 10:02:27 PM
Item: betting markets got Virginia 2006 wrong until almost the last moment
So what? Anecdotal evidence of a time when a market got the result wrong is meaningless. Imagine there are 10 elections each one of which has a 60% chance of being won by the Democrat candidate, and each one of which is correctly priced by the market to be 60-40. You'd expect the Democrats to win about six of those races and hence the market would look like it got about four of them "wrong" - but did it? Obviously not. So pointing out a time when the market "failed" to pick the winner (not what the market is doing, by the way) is, by itself, evidence of nothing.
and even a dimwit like me called it six weeks prior
Well, then why didn't you make a killing at the hands of the obviously wrong market? And if it was so clearly wrong right up until the last minute, what was keeping everyone else who, like you, knew it was wrong from bidding the price to the correct level?
The point is not that the market is going to predict the future winner perfectly - that's impossible in a world of uncertainty, but that markets are probably the best estimate of the current overall likelihood of a particular result based on the information available currently.
Posted by: Brian Courts at Jan 4, 2008 11:09:30 PM
Ok, Alex, the blue spike (to 95!) around mid 07...how does the explanation run?
It's called noise and doesn't generally need an explanation. Any short-term change in price, in this or any highly traded market, is mostly noise. Taking a single, or even serveral, samples and trying to figure out which part is signal and which is noise is essentially impossible. Assuming we have a large number of samples, the noise can be filtered out and the signal extracted by looking at the data, but only over a a sufficient period of time.
Posted by: Brian Courts at Jan 4, 2008 11:22:10 PM
The 95 spike is surely an incidence of 'fat fingers':
http://bigpicture.typepad.com/comments/2007/03/beware_the_fatt.html
Posted by: Robert Wiblin at Jan 4, 2008 11:43:58 PM
Brian Courts, your thought experiment is less useful than anecdote. You are "assuming a can opener," as goes the old joke about economists. You would have to explain how you can PROVE that the market has "correctly" priced the chances in 10 elections -- in other words, you would have to specify the objective reason each candidate has a 60% chance of winning in each election, independently of the market price and thus validating the market price. For example, coin-flipping depends objectively upon chance in physics: are you tossing a lopsided coin? For another example, stock valuations are judged by calculations using underlying fundamentals. I suppose you could go from the percentage of voters in each party, but this is unreliable with regard to the White House, which has mostly been a personality contest. Failing an independent verification in your example, your judgment that the markets did NOT get four races wrong is nonsense.
Returning to reality, I think the reason why most people didn't see the Virginia election coming, was because they didn't observe the two men in the debates, nor match it to the mean polling trend which had been steady since mid April 2006 and was inexorable by early September. The bettors were going with conventional wisdom about the conventional wisdom.
Why didn't I bet? Very simple: I don't gamble. I believe that it is spiritually debilitating.
It's perhaps a saving grace, though, that you stuck "uncertainty," "probably," "estimate," and "likelihood" into your last sentence, because metaphysics releases you, also, into pure belief.
Posted by: Lee A. Arnold at Jan 5, 2008 1:52:56 AM
I'm not sure betting markets are the best possible predictor, but even assuming for the moment that they are ... there is no reason to believe they will always be right.
In the area of prediction, the "best" may still be fallible.
Posted by: odograph at Jan 5, 2008 9:56:12 AM
Isn't it a straightfoward matter to compare the predictive power of these markets with that of traditional polls?
Posted by: wonderer at Jan 5, 2008 8:08:36 PM
I was very surprised that Hillary's intrade contract for the Democratic nomination fell 20 points after Iowa. What surprised me was
1) Losing wasn't that unlikely -- polls were pointing to the possibility for weeks.
2) The implied conditional probability of winning the Democratic nomination based on winning Iowa seems crazy. This is underdetermined by the betting market price, but if you figure Hillary as a 50-50 shot in Iowa based on the polls, and use the betting market prices of 70 before Iowa and 50 after, the conditional probabilities work out to
45% wins Iowa, wins nomination
25% loses Iowa, wins nomination
5% wins Iowa, loses nomination
25% loses Iowa, loses nomination
With this set of conditional probabilities, if she wins Iowa, she has a 90% chance of winning the nomination, even before New Hampshire has voted!
Posted by: Zach at Jan 6, 2008 4:26:24 AM
Wonderer, I'm not sure how prediction markets are tested. They are continuous predictions (running over months) for a discrete event (a single election).
Are they "right" if they center in on the right answer in the final days? Are they "wrong" because they didn't have that same answer months earlier?
heh. I guess this can be filtered through one's beliefs. If they were "wrong" months earlier that might be (as I believe) just the limits of human prediction. On the other hand, a true market believer might say they were "right" and that conditions just changed later.
Posted by: odograph at Jan 6, 2008 8:02:33 AM
Zach,
I trade on InTrade, not with big money but I've been doing well, increasing the size of my investment from $150 to $650 over the course of four years.
The numbers you quote for Hillary Clinton are not unrealistic. She was the national frontrunner with both a huge lead in national polls, and a smaller one in New Hampshire polls. The other candidates were absolutely depending on stopping her in Iowa.
When trading these contracts, my "rule of thumb" was that with a Hillary victory in Iowa she would have close to 100% chance of winning the nomination. However, this does not apply to the other candidates, only to the national frontrunner. I was expecing that with an Obama win in Iowa, the probabilities would be 50/50 between him and Hillary, and with an Edwards win it would look something like 2/3 for Hillary, 1/6 for each Obama and Edwards.
Your chances of winning the nomination after Iowa depends on how you're doing in national polls. Hillary was unique because all her opponents were depending on an Iowa bounce
Posted by: Anders at Jan 6, 2008 8:31:57 AM
Thanks, Anders. The Hillary contract does seem consistent with your view of the probabilities. I'm amazed that the expected value of the Hillary contract could vary by 40 points by virtue of the Iowa caucuses.
To back up your theory of other candidates needing a bounce from Iowa, the Hillary contract to win New Hampshire appears to have fallen from 80 to 50 immediately after Iowa, and is now near 20.
Posted by: Zach at Jan 6, 2008 12:43:12 PM
Odograph, it should be fairly straightforward to test the relative predictive power of polls and markets. Take a given day, N days before the election, on which a poll has been conducted. See how close the market prices that day and that poll come to the actual election result. Repeat for as many polls (N1, N2, etc. days before the election) and as many different elections as necessary to gather enough data points.
Posted by: wonderer at Jan 7, 2008 7:32:30 PM
提供钢管无缝钢管数据恢复心理咨询心理辅导与心理治疗bjxlzx.cn等服务
Posted by: baom at Feb 9, 2008 1:47:28 AM
