Home > Economics - General, Oz Politics > Polls and punters

Polls and punters

April 15th, 2014

I’ve written a few times about the idea that betting markets provide a more accurate guide to political outcomes than do polls or ‘expert’ judgements or statistical models (which usually incorporate polls along with economic and other data). The problem is that, close to an election, they all tend to converge. So, the best time to do a comparison is early in the election cycle. Right now there’s quite a sharp contrast. The polls have had the (federal) ALP and LNP just about level for months, but the betting markets have the LNP as strong favorites.

One possible explanation is that governments generally do worse in polls than in election, so that the polls underestimate the government’s support. I’ve heard this claimed, but never seen any systematic evidence to support it. Another possibility is that market participants know something that’s not reflected in the polls. I’m sceptical on this.

The final possibility is that betting markets this far out from the election are thin and inefficient. If that’s right, then the odds for Labor look very favorable. I’m not going to bet myself (I did OK on my one foray into the US Republican primaries, but the hassle involved was too much to make it worthwhile), and I’m not giving betting advice.

Still, I’d be interested in responses from those among my fellow economists who’ve claimed efficiency properties for betting markets. I guess Andrew Leigh is precluded from commenting, and Justin Wolfers is a long way from the action in Oz, but I’m sure there must be others willing to jump in

Categories: Economics - General, Oz Politics Tags:
  1. Doug
    April 15th, 2014 at 15:26 | #1

    An interesting recent case may be the South Australian election. Any one have the details on that?

  2. April 15th, 2014 at 16:47 | #2

    In SA, the Coalition were 1.01 favourites, and Labor was paying around 15.00 as I recall. I don’t recall any pundits predicting a Labor win, but Kevin Bonham said they were “still competitive” two days out based on Newspoll and the fact that the Coalition probably needed at least 52% on two-party-preferred to win enough seats for government (in the end they missed out even with 53%!): https://twitter.com/kevinbonham/status/444107117671112705

    The Reachtel poll had the Liberals leading 55-45, so on average the polls tipped a Liberal win, but a simple “apply polling to the pendulum” would have given more of a chance to Labor than the betting markets.

    I don’t know about the main “overall winner” betting, but individual seat markets are definitely very thin — I picked off arbitrage opportunities for months in some federal seats last year. The limiting factor was the agency that limited me to one bet on each seat per week; prices eventually reached an equilibrium but it was an extremely slow process and it was obvious that very few people were betting on those markets. Things sped up in the weeks before the election itself, but there were still plenty of arbitrage opportunities which would last a day or two.

    Early in the election cycle, I expect that polls are almost uncorrelated with the next election’s outcome, while the markets usually just back the incumbent. In Australia, the incumbent usually wins, so I think that the markets will beat polls the “early in the electoral cycle” test.

  3. April 15th, 2014 at 19:06 | #3

    Polls this far out are incredibly uninformative. So one view is that betting markets (and informed expert opinion) may have the same point estimate as the polls, but a much much larger standard error. At a minimum, I suspect we can agree that there’s an enormous amount of non-sampling error to account for. Andrew and I have written about this in the past, pointing out that taking polls too literally through the cycle can yield embarrassing predictions.

    Second, there’s a view in the US that a campaign causes polling to “firm up” and to systematically head toward “fundamentals” (the economy; ideological divergence, etc). I’m not aware of similar work for Australia, but it would be interesting to know if it were true.

  4. Robert
    April 15th, 2014 at 19:29 | #4

    I’n not a betting market efficiency kind of guy, but in this case I think they are closer to the truth. First, although the polls are close, they aren’t showing the kind of lead that would result in a change of government. Second, as Justin says, polls this far our tell us next to nothing. When you update over the prior that first-term governments get reelected, this leads to very little change in expectation.

  5. paul walter
    April 16th, 2014 at 04:44 | #5

    It’s just the more depressing for me, the latest result, because the barbarians are in the castle the horse has already bolted and the damage will be done before they can be removed, if ever.

    Sorry, I know the prof was looking for specialist comment, I’m just a mug who looks on, disturbed at the horrible war against Australians currently conducted.

    Last Mondays Stephen Long 4 Corners on the auto industry was particularly painful, because I grew up in Elizabeth and have watched for a long time as a once healthy if rough and ready community has been crushed down from factory fodder, to welfare, and now the unnecessary decimation of Holdens, the fundamental plank upon which the local economy was based… If a fraction of the $three billion recently announced to be wasted on drones to hunt down asylum seekers had been spent on the auto industry, how much suffering could have been saved.

    Which is of course the one thing the vindictive, sado economic Abbott “government” would never countenance.

  6. John Quiggin
    April 16th, 2014 at 05:31 | #6

    @Justin Wolfers

    Polls this far out are incredibly uninformative. So one view is that betting markets (and informed expert opinion) may have the same point estimate as the polls, but a much much larger standard error.

    I thought about this argument, but I don’t think it works, at least from a standard Bayesian perspective, unless there is some more reliable information about fundamentals. But what would that information be?

  7. John Quiggin
    April 16th, 2014 at 05:34 | #7

    ” In Australia, the incumbent usually wins, so I think that the markets will beat polls the “early in the electoral cycle” test.”‘

    But it’s also true that the winning party usually enjoys a honeymoon in the polls, so there’s not much to go on for the case when the election lead is lost almost immediately. It happened in 2010 of course, and I think in 1993, and we know how they turned out.

  8. Fran Barlow
    April 16th, 2014 at 07:09 | #8

    I can’t imagine what body of data punters could use to guide their betting outside of polling by major organisations. Even if some punters had access to some better source of data, most wouldn’t.

    So it’s hard to imagine how people betting on the outcome could improve on polls.

    It’s true that parties that win elections typically have a “honeymoon”. This simply reflects the fact that people don’t like admitting to themselves that they have been conned, again, and are invested in believing they have made the right choice. So cognitive dissonance applies.

    Even those who voted for the losers feel a certain obligation to appear non-tribal and to avoid insulting those who voted for the winners. Also, it’s depressing to think that the winners are as bad as you thought, so again, a combination of cognitive dissonance and cultural taboo underpins the new regime.

    This was much less strong in 2010 because the regime was not a majority government but a minority and seemed to have merely squeaked over the line. The supporters were relieved but ambivalent, given that they needed us Greens and some Indies to support them and the Murdoch Press was howling for blood, especially ours and that of the Indies.

    In 1993 the regime had had four terms and probably should have lost, but won because they managed to wedge and embarrass the then LotO on a populist canard. The win was tactically clever but left the country in the hands of people most probably thought ought to be in opposition. The result in 96 was therefore not surprising, particularly when Howard ran as Keating-lite and disavowed the G&ST.

    Curiously, had the Liberals won in 1993, there would probably have been no Tampa, no Howard, a better Telstra sale, and far more progress on decarbonising.

  9. Fran Barlow
    April 16th, 2014 at 07:12 | #9

    Oops … Forgot the word tab0o is tab0o … Damn you spam filter!

    I can’t imagine what body of data punters could use to guide their betting outside of polling by major organisations. Even if some punters had access to some better source of data, most wouldn’t.

    So it’s hard to imagine how people betting on the outcome could improve on polls.
    It’s true that parties that win elections typically have a “honeymoon”. This simply reflects the fact that people don’t like admitting to themselves that they have been conned, again, and are invested in believing they have made the right choice. So cognitive dissonance applies.
    Even those who voted for the losers feel a certain obligation to appear non-tribal and to avoid insulting those who voted for the winners. Also, it’s depressing to think that the winners are as bad as you thought, so again, a combination of cognitive dissonance and cultural tab0o underpins the new regime.

    This was much less strong in 2010 because the regime was not a majority government but a minority and seemed to have merely squeaked over the line. The supporters were relieved but ambivalent, given that they needed us Greens and some Indies to support them and the Murdoch Press was howling for blood, especially ours and that of the Indies.
    In 1993 the regime had had four terms and probably should have lost, but won because they managed to wedge and embarrass the then LotO on a populist canard. The win was tactically clever but left the country in the hands of people most probably thought ought to be in opposition. The result in 96 was therefore not surprising, particularly when Howard ran as Keating-lite and disavowed the G&ST.

    Curiously, had the Liberals won in 1993, there would probably have been no Tampa, no Howard, a better Telstra sale, and far more progress on decarbonising.

  10. Ikonoclast
    April 16th, 2014 at 07:32 | #10

    Thoughts that occur.

    1. Do election punters constitute a valid sample of the electorates? Are they distributed evenly geographically? Are they distributed evenly through the classes? Are women well enough represented?

    2. What proportion of election punters would tend to bet the way they intend to vote?

    You will see what I am driving at. If election voters form a valid or semi-valid sample and tend to bet the way they intend to vote then they might roughly approximate a standard poll. Given the error margins possible in standard polls, is a betting poll any worse as a predictor?

    Apologies if this theory is not original.

  11. Paul Norton
    April 16th, 2014 at 08:20 | #11

    I remember when Miss Finland was favourite for the 2007 Melbourne Cup. I also remember when Weekend Hussler was favourite for the 2008 Melbourne Cup. Then there was the early market for the 2010 Melbourne Cup.

    QED “betting markets this far out from the election are thin and inefficient”.

  12. Ikonoclast
    April 16th, 2014 at 08:47 | #12

    @Paul Norton

    There is a difference or two. Punters are not horses in the Melbourne cup. But punters are “horses”, that is to say, voters in a Federal Poll. See my theory re samples at comment 10. In a horse race, one horse wins (excluding ties). In a Federal poll, the party is a proxy winner, but the real winners are the majority voters. (This last point is numerically true but heavily ironic in political terms in a bourgeois democracy.)

  13. April 16th, 2014 at 10:22 | #13

    @David Barry
    I wish I’d put a bet on now.

  14. derrida derider
    April 16th, 2014 at 13:16 | #14

    There’s another important doifference the OP blurs. Betting markets are about who the punters BELIEVE will happen, current polls are about what the punters WANT to happen.

    A majority of those polled say they would vote for Labor (on 2PP) now if an election was held tomorrow – but I bet many fewer would answer “yes” if asked “do you expect Labor to win the next actually existing election?”. IOW what the betting market and the polls predict cannot be compared because these polls are not tryng to predict anything.

  15. April 16th, 2014 at 18:12 | #15

    You should be able to test some of the discussion here… fix a beta prior estimated from all past first-term govt results; fix a binomial likelihood based on the polls; this generates an (analytically tractable) beta posterior. Then see how the posterior compares with the betting markets (and what results the posterior actually gives). I imagine that with the polls tight and the beta prior so strongly in favour of the incumbent govt, the posterior would give probabilities in favour of the govt that are quite similar to the betting market.

    This would be an interesting proposition to test over time: compare say quarterly updates of the posterior (based on the likelihood obviously) with the markets, and see if the markets are really just a form of naive Bayesian estimation.

  16. Robert
    April 16th, 2014 at 18:34 | #16

    JQ,
    This Peter Brent post might also be of interest. He makes the point that Labor may need a bigger lead than they have to win the election (apologies for linking to the Australian):

    http://blogs.theaustralian.news.com.au/mumble/index.php/theaustralian/comments/how_much_for_victory/

  17. TerjeP
    April 16th, 2014 at 19:06 | #17

    The polls are asking “who would YOU vote for if an election was held TODAY”.

    The betting markets are implicitly asking “who will the MAJORITY vote for when the election ACTUALLY OCCURS”.

    We should expect different questions to get different answers. Maybe there would be more convergence if the pollsters asked the question that is implicit in the betting markets.

  18. Ikonoclast
    April 16th, 2014 at 20:00 | #18

    Is anyone making a book on the chances of Campbell Newman having to resign like Barry O’Farrell? Seems like the tentacles of this scandal might get to Newman too. One can only hope.

    As for Barry O’Farrell’s “massive memory fail”…. LOL! More like a massive ethical fail.

  19. Ron E Joggles
    April 16th, 2014 at 21:39 | #19

    @Ikonoclast Polls purport to canvass voting intention, while betting is an expectation of who will win. Perhaps people are more thoughtful when their money is on the line. Maybe also, people who bet on elections are a subset of bettors generally, most of whom would only bet on the horses, and that subset may well be more experienced and more objective in their predictions.

  20. Historyintime
    April 16th, 2014 at 22:11 | #20

    The analogy here is TAB tote odds (polls) and fixed price markets (fixed price election odds). They both provide some information at points in time but the fixed market is always a more reliable predictor of the final odds (both tote and fixed) until the event closes at which time both markets will have converged.

  21. Historyintime
    April 16th, 2014 at 22:19 | #21

    Also it is my experience that political markets are slightly inefficient (a few percentage points of value) but only in the last week or so before an election when the polls become very reliable. This is because of a “two horse race” effect where the underdog is often too short in the market

Comments are closed.