Monday, 9 September 2013

Election Betting Not Always A Good Predictor

It appears election betting odds are not always as good a predictor of voting intentions as I thought.
On the morning of the federal election, I ran a simulation which produced the most likely outcome of 98 seats for the Coalition, 49 for Labor and 3 independents, although I did comment that this probably overestimated the Coalition’s chances because the betting in seats like Werriwa was unrealistic. Ten days prior, I got 91 – 57 – 2.
Now, with most seats in the lower house determined, the currently most likely result is 90 – 55 – 5. Bob Katter and Andrew Wilkie were returned as most people thought they would. Cathy McGowan appears to have secured Indi with a very strong campaign against unpopular Liberal incumbent Sophie Mirabella. The generally well heeled, but pious, lefty flock of inner city Melbourne have asked for three more years of sanctimonious preaching from middle class hypocrites, returning the Greens’ Adam Bandt with only a 1% swing against him. In a disturbing upset, self promoting windbag Clive Palmer looks almost certain to win Fairfax. Perhaps the good people of Bundaberg believe the carbon tax will fade their curtains.
So where did the simulation go wrong?
Clearly, Adam Bandt paying $2.35 and Clive Palmer paying $5 significantly underestimated the chances of both candidates. Shorter odds would have seen the simulation produce many outcomes with 4 or 5 independents.
The odds for Kingsford-Smith, Parramatta and Werriwa in Sydney and Wayne Swan’s Brisbane electorate of Lilley were also clearly wrong, but these predictions were discounted when scaling back the forecast number of Coalition seats from 98 to 94 or 95. Interestingly, they were more or less correct two weeks ago when they favoured the ALP, although Werriwa was only won by about 3500 votes. Some nongs must have done their money with sizeable bets on the Liberal in these seats.
Four of the late flips: Page, Petrie, Hindmarsh and Lyons did end up falling to the Coalition. McEwen, Capricornia and Lingiari are all tight, as per the betting. However, Blair was easily retained by the ALP, with a +1.5% swing. The Liberals’ Teresa Harding must have been a pretty ordinary candidate.
Speaking of poor candidates, the Liberals virtually threw away Greenway by standing Jaymes “we have a solar panel” Diaz. What a fucking dunce! Yet he was paying about $1.20 to win the seat.
So, wrong odds for Adam Bandt and Clive Palmer and a few seats with very poor Liberal candidates. If a candidate is paying $1.20, they will win in most simulated scenarios. A handful of those types of errors and your simulation overestimates by 5 or 6 seats.
Of course, the bookies did well from the misestimated payouts: the odds for the Coalition only came in in these seats because people placed (losing) bets on them.

Saturday, 7 September 2013

Election Betting Says Coalition By 40+

Ten days ago, I wrote that using implied probabilities from Sportsbet and Centrebet odds to simulate the result of all 150 federal seats, the most probable outcome would be a 32 seat Coalition majority in the House of Representatives. At that time, the most probable outcomes were 91 or 92 seats for the Coalition, 56 or 57 for Labor and 2 or 3 for independents.
I ran the simulation again early this morning, using the latest available odds on both sites, just prior to the suspension of internet betting as the polls opened.
Now the most probable outcome in the simulation is 98 seats for the Coalition, 49 for the ALP and 3 independents, giving the Coalition a 46 seat majority after winning an extra 25 seats.
Why the 7 seat change over the last 10 days?
Firstly, there are several seats which have flipped to the Liberals in the betting: Kingsford-Smith, Parramatta and Werriwa in Sydney, Lilley (Wayne Swan’s electorate) and Petrie in Brisbane, Hindmarsh in Adelaide, Brand in Perth and Lyons on the east coast of Tasmania. The only one of these which I strongly doubt is Werriwa, Gough Whitlam’s old seat. There might be a few aspirational Liberal voters in Denham Court and the newly developed suburbs, but I can’t see it falling to the Libs. Even discounting Werriwa, that’s 7 in which the Liberals are now favoured. The outcome could easily be Lib-ALP 5-2 instead of the 2-5 it looked like a couple of weeks ago.
Secondly, a number of seats in which Labor were paying around 1.40 are now even money: Page in NE NSW, McEwen just north of Melbourne, Blair and Capricornia in Queensland and even Lingiari (all of the Northern Territory outside of Darwin).
Note that the simulation is predicated on the bookmakers’ payouts being accurate representations of the true probabilities of each candidate winning. In fact, they are as much determined by how much has been bet on each candidate as any statistical analysis of booth returns and polling. Having said that, I wouldn’t expect the money to be flowing one way and votes the other. The best we can say without being able to train the model on betting data from previous elections is that payouts are highly correlated with outcomes.
For example, in Kevin Rudd’s seat of Griffith, the Libs are paying 2.75 and Rudd 1.35. That translates into a probability of 30% of Rudd losing his seat. While I’d like to see that happen, I believe $1.35 is pretty generous odds for Rudd to retain his seat.
Bob Katter and Andrew Wilkie are now almost certainties to be returned. Independent Cathy McGowan is now paying 1.90 to Sophie Mirabella in Indi and the Greens’ Adam Bandt has shortened to 2.35 in Melbourne. That’s why the simulation produces a much higher frequency of outcomes where we have 3 independents (Greens included). Even Clive “Skywhale” Palmer has come in to $5 from $8 a week ago (although I think that significantly overestimates his chances).
So, if I discount the seats where I think the betting is skewed, such as Werriwa, the Coalition is still probably up 4 or 5 seats on their position ten days ago.
I reckon that most likely puts them on 94 or 95, with Labor 53 and 2 or 3 independents ie. a 40 seat majority.