A really convincing case study often has more power than a mound of statistical analysis and, for me at least, observation of the just-completed election campaign has convinced me of the correct analysis of the predictive power of betting markets, relative to polls and pundits.
To recap, the polls (which had previously put Labor just in front) showed a big shift to Labor as soon as Kevin Rudd became Labor and stayed virtually unchanged for the subsequent year, narrowing by a percentage point or two after the campaign was called. This graph from Possum’s Pollytics tells the story.
At first no-one (neither punters in betting markets, nor political pundits, nor the public in their predictions) believed the polls. But over time, they all came around, until by election day, it didn’t matter whose predictions you used, you would have been pretty much right.
Given that the polls often bounce about, it was reasonable enough not to predict a Labor victory in the immediate aftermath of the Rudd bounce. On the other hand, if the punters, taken collectively, had some special insights not accessible to pollsters, they failed to show it.
To my mind, this supports what is called the ‘semi-strong market efficiency’ hypothesis. That is, given all available information, including the polls and the analysis of pundits, the market prediction is likely to be about as good as you can get. But Kevin 07 pretty convincingly refutes strong forms of the efficient markets hypothesis such as those that were used to promote markets in ‘terrorism futures’ and similar stuff. Moreover, I think it’s safe to say that without the polls no-one would have seen this coming.
So, given that polls exist, markets do a pretty good job of taking them into account. But if we didn’t have polls, markets would be way off-beam a lot of the time.
Interestingly, the patterns of the year repeated in microcosm on the night. The Sky exit poll was virtually spot-on, and both the markets and the pundits called the result, almost simultaneously, a few hours later.
16 thoughts on “Polls, pundits and punters”
I’m not a statistician nor an economist, and I’ve probably misunderstood your point, but in comparing the outcomes from betting markets and polls aren’t you comparing apples and oranges?
In terms of the electoral betting market, don’t polls serve the same function as company statements and annual reports do in an equities market? Both are information sources which are used by market players in setting they prices they are will to pay? (Granted that the quality of the information provided by the polls is likely to be of lower quality than company financial statements – at least one would hope so.)
Absent poll information and financial statements, market players are left to make, at best, semi-educated guesses on the outcomes.
The polls in February showed Labor well in front but the question was who would you vote for today? There was good reason to think then the same answer might not hold at the election. It didn’t in 2004. It was only as the year progressed and Labor’s lead became impervious to whatever the government threw at it that it became clear that Labor would win which was when they became market favourites.
In other words with 2004 as a guide the early polls had little information content. The market ignored them quite reasonably.
I think the market prediction was entirely derivative. That’s a pun of sorts. The polls (based on statistically valid sampling methods) were the primary data source. As the polls mounted up (showing the first one wasn’t a blip) then most punters become sure enough to put their money on it.
All the â€™semi-strong market efficiencyâ€™ hypothesis seems to be saying is that markets are “reasonably smart” as they show the net effect of an agglomeration of reasonably smart people making decisions on an accumulation of evidence. This sounds true enough so far as it goes but is it not a trivial observation as Popper would say?
This was one election at which, if it weren’t for the polls, just about no-one would have predicted the change of government. As everyone kept saying, the “conditions for change” didn’t seem to exist.
I think it is indeed a trivial observation. I’m not sure why people go on about it all the time.
Well, the betting will be off with No Pokies in the Parliament, so you’ll have to rely on the polls.
Possum’s Pollytics, btw, also did a 4 poll average which had the Labor vote at 53.9 on 23 November and the final result bang on trend over the 6 weeks of the campaign.
I have no doubt that a lot of what prediction (betting) markets do is aggregate information from polls. Certainly you would be a fool to bet on an election without considering the polls. But John makes a stronger statement:
“So, given that polls exist, markets do a pretty good job of taking them into account. But if we didnâ€™t have polls, markets would be way off-beam a lot of the time.”
Testing this would require comparing the accuracy of prediction market forecasts before and after the rise of scientific polling.
Fortunately, such data exists – Koleman Strumpf and Paul Rhode have put together data on US political prediction markets back to 1880. It would be interesting to see whether (as John conjectures) the political prediction markets became dramatically more useful subsequent to the rise of polling. My (very casual) glance at the data suggests not, but it is surely worth a second look.
A broader question: Is anyone aware of the existence of political betting in Australia from an earlier (pre-Centrebet) era? I’m sure it was occurring in pubs everywhere, but wonder whether it was organized enough that one could collect coherent data on it.
Had a quick look through the paper – it claims semi-strong efficiency and notes some apparently successful insider betting, violating strong efficiency.
Unfortunately for our discussion, the rise of polls coincided with a downturn in betting markets due to policy changes and adverse sentiment.
Coming back to the main point, the polls clearly outperformed the markets this time, and I think this will make any inference claiming superiority of the markets in the current era pretty fragile.
The market was reliant on the polls in great measure and the market tried to influence the polls.
AS regards Jill Rush’s point at 8, there was a certain amount of conjecture that the market in Bennelong was distorted in that every time Howard’s odds slipped, “someone” would lay a big bet so as not to have the obvious, odious headlines – after all, the polls appear to have called that one pretty well too.
Another point is that several folks, myself included, used the centrebet freebet – lay down $30 on an odds on cert (labor winning) and they give you $100 to bet on anything. So, for the sake of $30 of personal cash I made zip difference to the odds in the national election but may well have at least contributed to shortening Maxine’s odds.
Obviously the polls pre-date the punters. Punters take the polls into account, while the ordinary folk being polled do not ordinarily consider the betting market.
OTOH they may well be influenced by the MSM pundits, which is where things get more interesting: it’s no accident that the pollsters in the Murdoch stable were (again) overly optimistic of Coalition chances. It’s no accident that Murdoch opinion-makers continually pushed these polls as proof that things were not quite as bad as they otherwise seemed.
And that sort of thing inevitably has some influence on perceptions, which shape polls, and hence markets.
I note with some satisfaction that Murdoch pundits like Milne, Sheridan, Shanahan and Albrechtsen are copping a lot of abuse for their shameless post-election columns. I hope we will soon see some more informed, objective and balanced voices taking over their role.
I would like to thank Milne, Sheridan, Shanahan and Albrechtsen etc at the Government Gazette for talking up Howard’s chances of retaining Bennelong right up to the jump. They helped me clean up on Maxine Mckew on Betfair at 2/1.
The problem is that polls are not always representative and sometimes market players are smart enough to figure this out. In the case of the Australian federal election the national polls were a pretty good indicator but it isn’t always the case.
Polls are indeed variable and, at times, flawed; however, taken in aggregate and looking at trends they are really very good now. This year has well and truly put paid to the “markets know best” theory, in my view. The betting markets were clearly out of step with what was happening for many months, and only became realistic in the final few weeks. Labor’s (and Maxine’s) odds were ridiculously long for much of the year – a lot of people who could see what was really happening have done very nicely, thank you.
From the vantage of the Great White North, i was tracking the various seat projections (as we do in several jurisdicitons). It was not clear ’til during the actual campaign that a Labor victory was a sure thing:
We’ve also posted a scoreboard for the nine most prolific modelers.
cheers from the Yukon!
Has anyone noticed that now we’re into counting pre-poll and postal votes, the ABC has reduced the definite number of Labor seats to 79 (it was 83 at one stage). They’re still expecting Labor to win 85, but looking at the 8 close counts on the AEC site, I’d say that Labor has Buckley’s Chance of winning all but two of them.
On a more cheering note, Maxine is maintaining her lead in Bennelong. Maxine for President!
I dont see how we can infer much at all about the relative performance of polls versus betting markets from the election result.
Just because the betting market converged to a position closer to the polls doesnt make the polls right. Both are bets on the liklihood of a future event. The only possible comparison is the relative performance of polls versus betting markets as they were just before the election.