Election Markets

by Brian on January 2, 2008

Since the U.S. Presidential primaries are about to start, it would be nice to be able to get a read on what the betting markets are saying in order to make some retrospective assessments of how well they predicted the result. Unfortunately, or perhaps fortunately for some of you, there’s no such thing as what the markets are uniquely saying. Indeed, there are some arbitrage possibilities (if you have access to each of the markets) the size of which you might find hard to believe.

Here are the bid and ask prices (bid first) for various Republican contenders on different markets. I’m leaving off the transaction fees the markets charge, but I don’t think these are large enough to stop these being actual arbitrage possibilities. (All prices are as of 15:45 GMT Jan 2. The first prices quoted are from Betfair, the second pair from Intrade and the last from IEM.)

Guiliani: 0.277/0.294; 0.285/0.289; 0.220/0.224.
Romney: 0.243/0.253; 0.230/0.245; 0.296/0.305.
Huckabee: 0.106/0.121; 0.094/0.095; 0.140/0.149.
McCain: 0.200/0.212; 0.227/0.232; 0.231/0.232.
Thompson: 0.012/0.029; 0.024/0.029; 0.038/0.043.
Paul*: 0.060/0.068; 0.071/0.075; 0.058/0.065.

Note that for IEM, the Paul quote is actually a rest of field quote, so you’d expect it to be a little above the other Paul quotes not a little below. And obviously there is only so much you can gain out of the Iowa markets, since they only allow you to play with $500. But I’m still struck by this divergence between the different markets. I’m reasonably confident that if we looked around at other betting markets we’d find an even greater divergence.

And perhaps this goes without saying, but 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. That’s to say, whatever the potential benefits of predictive markets, there is little reason to believe they are realised in the betting markets on the Republican primary.

{ 1 trackback }

Divergence between InTrade and BetFair political prices?? | Midas Oracle .ORG
01.03.08 at 8:34 am

{ 26 comments }

1

R. Stanton Scott 01.02.08 at 4:43 pm

I bought some RROF stock at IEM on a bet that Huckabee would do well and get his own stock, and the strategy worked. I also got some Edwards stock at IEM thinking that if he won Iowa his price would rise and I could make a bit. I made these choices based on my observation that the stocks tended to rise and fall more or less with poll results, and I figured that if I could successfully predict Iowa I could double my money, especially if polls fail to do so. We’ll see.

I did not think of the arbitrage possibilities, though I should have–this is the same as bookie shopping, in a way. That is, the spread on the Arkansas-Texas game will differ by several points from Little Rock to Dallas.

It strikes me that the predictive value of these markets depends on aggregating opinions, and only works if the aggregated opinion is based on good information and analysis. So a determination of the predictive power requires either a large n or detailed information about the participants, just as the determination of point spreads depends on the bookie’s particular market. Right?

2

R Johnston 01.02.08 at 5:27 pm

These markets have obvious problems with low volume of trading and nonrandom bias of some market actors. There’s also a less obvious problem with the inability to take a true short position. There’s, for example, a lot of money to be made shorting those Paul contracts–Paul has essentially no chance at the Republican nomination and the people currently owning and seeking Paul contracts are Paul partisans who aren’t looking to make money on those positions–but jumping into the markets to simulate doing that would require tying up a lot of money by purchasing unit bundles and then selling Paul (or RROF for iem). Rather than an essentially infinite return on the safest short investment you’ll ever find you’re looking at a 7% return on a moderate investment that ties up your funds until the convention, a considerably less attractive investment, especially once the limits on total return due to low volume are taken into account.

3

Neil 01.02.08 at 5:35 pm

Seems right, but these markets have been quite accurate in the past even with small numbers of participants (and they don’t seem particularly well informed either).

4

Slocum 01.02.08 at 5:46 pm

It strikes me that the predictive value of these markets depends on aggregating opinions, and only works if the aggregated opinion is based on good information and analysis.

Not really — that is, it works if the aggregated opinion includes not just ‘good information and analysis’ but also rumors, guesses and hunches. If these are based on nothing at all, they’re useless, but if they’re based on unrepresentative, local information, they still contribute to the overall quality of the prediction. If every participant bases his or her bet on who seems popular at the coffee shop, the individual predictions are not much good, but the aggregate is. Think of it as an alternate way of taking a poll.

5

digamma 01.02.08 at 5:48 pm

What’s the biggest arbitrage opportunity here? Buying Giuliani for 0.224 on IEM and selling him for 0.285 on Intrade? That 6.1 US cent profit won’t cover your transaction costs. Am I missing something here?

6

R 01.02.08 at 6:16 pm

A question re (2): In the cases where the actual outcomes are different from the polling predictions, do the markets tend to reflect the polling, or the actual outcome better?

7

Brian 01.02.08 at 6:16 pm

I’d be pretty stunned if the transaction costs were more than the profits here. The limit at IEM is $500, so let’s say we buy $500 of Guiliani there, for a return (if he wins) of $2232. We then sell $1500 at Intrade, and have to pay out $427 if he wins.

So before transaction costs, we’re up $1000 if Guiliani loses, and $300 if he wins. Could transaction costs eat up really all of that? Those would be some killer transaction costs. (I haven’t read enough of the fine print of either site to be sure, but I think IEM is very low transaction costs, and Intrade only charges commissions on wins.)

Slocum, it might be anther kind of poll, but that doesn’t mean it has any value. The Kos site polls always have Edwards winning by a mile. That’s a poll, but not one with much value. I’d like some assumptions about well-informedness of the opinions being polled before I trusted this, and if the betters aren’t even well informed enough to know when there’s free money on the table, I don’t think they pass that test. (This isn’t meant to be a complaint about prediction markets in general, just a concern that such markets need to meet some minimum standards to have epistemic value, and these particular markets don’t seem to meet them.)

8

Marc 01.02.08 at 6:21 pm

Two relevant points are the degree to which these small markets are not liquid, and the degree to which partisans can choose to spend money to promote their favorite candidate. The latter can be particularly difficult to root out. It’s entirely rational to believe that a few dollars invested in raising your candidate’s “price” on the futures market will have a larger impact on their fortunes than a more substantial sum given to the campaign for TV ads.

9

Stuart 01.02.08 at 6:42 pm

Surely bets on a political topic like this don’t need to be informed at all to be accurate – if the sample of people betting is broadly representative of all likely voters then the informed part is they know their own plans to vote (and of course they are likely to have strong clues about a number of other peoples voting choices). This assumes that people tend to bet more heavily on their own favourite than on other contenders, but I think this is quite likely.

10

Brian 01.02.08 at 6:56 pm

Stuart, it also assumes that the amount of money they have to bet is each roughly equal. But we know that in almost every election going that isn’t the case. There’s a small exception for American presidential elections, because although within most geographic areas the rich vote Republican, the rich areas tend to be Democratic. So there isn’t *as much* of a bias in the results. But in most elections, the majority of money will be significantly to the right of where the majority of votes will be.

11

R. Stanton Scott 01.02.08 at 7:47 pm

I don’t think bettors will buy more stock in their favorite candidate unless they think their favorite actually has a chance to win. I buy stock in a particular candidate because I expect the ratio of return in case she wins to be more than my calculation of his odds of winning. That is, my Romney stock at IEM pays about 3-1, and I bought it because I think he has a better than 3-1 chance of winning the nomination.

The participants almost certainly do not represent the voting public in the statistical sense. Even if they did, and bettors could buy into the market only by spending one single unit on their own preferred candidate, what you would have left is nothing more than a poll.

So I can’t think of any way these markets can accurately predict outcomes (at least no better than polls) unless they are based on stock purchases by people who have accurate information about the underlying dynamics of the election which they analyze effectively. If they use their analysis to place bets, the market might have some predictive power.

12

Cranky Observer 01.02.08 at 8:20 pm

> The participants almost certainly do not represent
> the voting public in the statistical sense. Even
> if they did, and bettors could buy into the market
> only by spending one single unit on their own
> preferred candidate, what you would have left is
> nothing more than a poll.

Now, if the Powerball Lottery Group were allowed to offer a game using selection of candidates with a $2 ticket and a variety of prizes including one of $10 million…

Cranky

13

Hedgie 01.02.08 at 9:06 pm

Brian, not sure I agree with your math. Also, your example isn’t really arbitrage (riskless profit), it’s simply a long and a short position on two exchanges with substantial risk. If I buy $500 of Guiliani at $0.224 that’s 2232 shares. If I sell $1500 of Guiliani at $0.285 that’s 5263 shares. If Guiliani loses you are correct, you win $1000 (short $1500 minus long $500). However if Guiliani wins you must cover your shorts at $1. You’re paying out net $0.715 to cover for a total of $3763, then net out your $1732 gain on the long position. Therefore if Guiliani wins you lose $2031, which intuitively makes sense since you took a short position 3x your long position. Of course that assumes you hold the shorts til Guiliani wins.

14

GreatZamfir 01.02.08 at 9:08 pm

@ Brian: have you actually made that bet you describe? If not, why not?

15

Brian 01.02.08 at 9:12 pm

I don’t set up accounts on betting sites because I’m not confident I would use them wisely. Not having accounts is one of the best tying myself to the mast strategies I know of. It is sorely tempting to und the knots when the siren song involves free $$.

16

stuart 01.02.08 at 9:59 pm

Brian, aren’t we talking about Primaries here? The effect you postulate might affect a competition between a Democratic and Republican candidate, but surely would be fairly minor even if it existed in this case, I would think.

17

Robin Green 01.02.08 at 10:23 pm

Stuart, what are you saying, primaries aren’t important enough for anyone to spend any money on? Er…

Or perhaps you’re saying that betting markets aren’t important enough for anyone to spend any money on to try to influence?

18

dsquared 01.02.08 at 11:03 pm

13: yes, Brian, I think you’ve misunderstood the pricing conventions or something. You buy and sell the same amount at two different prices.

19

stuart 01.03.08 at 12:11 am

Robin, I have no idea whether people try to influence betting markets in that way, and I didn’t think I was responding to any post about that. I wouldn’t have thought that the results of these markets are generally widely enough publicised to be worth the effort: this may be changing as online betting becomes more commonplace, or maybe it has and I am behind the times.

I was commenting on the point by Brian that suggested there is a bias towards the right wing in betting markets once you account for size of bets. This might well be true in general, but I was just thinking this effect would be much smaller, if it exists, in the US primaries because you have mostly left wing candidates competing with each other, and the same for the right wing candidates (obviously there is probably some overlap). Are there significantly right wing Democratic candidates, and left wing Republicans in each race that this effect would happen to any noticable degree? Or are there going to be Republican bettors putting lots of money on the Democratic candidates in some significant pattern (but which candidate(s) is likely to attract Republican bets?)

20

jim 01.03.08 at 12:56 am

There isn’t enough volume on IEM to sustain arbitrage. I originally bought some RROF. When Thompson split off, I sold the Thompson. When Huckabee split off, I went to sell the RROF. I found that on each uptick, I could only unload 5 or 10 at a time at the current bid price. So the notion that you’d buy $500 worth of Giuliani at .224 is unrealistic. There’s maybe $5 or $10 worth available at that price. Anyone who came in with the idea of buying $500 worth would bid up the price massively.

21

ato chiffre 01.03.08 at 10:28 am

There is simply not enough liquidity on these websites to make arbitrage worth while; moreover, I would check whether you have actually got the Betfair percentages correct.

22

dsquared 01.03.08 at 1:22 pm

20, 22: if there isn’t enough liquidity to make arbitrage worthwhile, then that establishes Brian’s larger point a fortiori; if there isn’t enough liquidity to make arbitrage worthwhile, there certainly isn’t enough liquidity for the market prices to be treated even as reliable indicators of the supply and demand in that particular market, let alone anything else in the real world.

Digging up my old spreadsheet, I notice that these exchanges are still apparently operating on a pricing convention which implies a negative time value of money, btw. (for example, on IEM, to get $1 with certainty in eight months’ time would cost you $1.018 now).

23

caveat bettor 01.03.08 at 3:43 pm

A 1% difference (either for a contract probability priced under 10% or over 90%) is not much of an arbitrage opportunity. Besides the liquidity issue which has been mentioned several times, there are a couple of other significant factors:

1) cost of carry (which reduces liquidity), and
2) credit risk, namely the increasing difficulty of making withdrawals, given the increasing regulatory constraints

24

Brian 01.03.08 at 3:51 pm

Sorry about getting the math wrong in the earlier cases. There was an arbitrage possibility, but not the size of one I suggested. My apologies for the mistake.

25

Brian 01.03.08 at 3:56 pm

By the way, some of the biggest arbitrage possibilities are gone now. Guiliani is now down to 0.220/0.224 on Intrade. Josh Marshall thinks that this is because of Guiliani’s late tanking in the Iowa polls. But those polls were all out 24 hours ago, and his numbers were much higher. I think it is because of some active CT readers!

This is one more reason why I don’t take these arbitrage chances. It requires acting quickly and getting the math right. Last time I tried to do that, I wrote post 7.

26

alexis 01.04.08 at 10:41 am

@23, *market* prices are always reliable indicators of supply and demand in a given *market*. The markets show what supply and demand is, rather than being indicators of some noumenal quantity.

It would be more useful to say “the market is not liquid enough for price discovery to be reliable. I.e. pace considerations in @24, a trader cannot usefully enter into transactions at observable prices that (ceteris paribus) generate a better return than ‘risk free’ investments, for the same risk. They can smell lunch, and it appears to be free, but it isn’t.

Secondly, of course you will see a ‘negative time value of money’ when there are no arbitrage opportunities. Observing a tractable arb opportunity is equivalent to observing a ‘risk free’ bond portfolio that is offered at a better price than ‘risk free’ bonds in the market. Did you ever fix that spreadsheet?

Comments on this entry are closed.