The betting market
How the betting market predicts
Australian economist, Andrew Leigh, has argued the odds in the betting market are a better predictor of election outcomes than the opinion polls.
According to Leigh, the economic theory is that markets are efficient aggregators of information, and that equilibrium prices in betting markets reflect an efficient estimate of election probabilities. I will try and explain this process in plain English.
The bookmaker has an incentive to set the odds so that the amount bet is always larger than the contingent or likely payout, regardless of the outcome. This is achieved with a little trick known as the bookmaker’s over-round.
If punters discern a factor they believe will change the likely electoral outcome, there is an incentive to get a bet on quickly before the odds change. This incentive makes the betting market highly responsive to the underlying dynamics of the electoral race even if these dynamics get little coverage in the media or the opinion polls.
As the bookmaker’s odds are constantly adjusted they reflect the aggregated public perception about the election outcome. With a bit of simple mathematics, the fixed decimal odds of the betting market can be adjusted for the bookmaker’s over-round and expressed as a probability of a win for the Coalition and the Labor party.
On this page I will track Centrebet’s fixed decimal odds and the derived probabilities of a Labor win. (Although I do not graph it, the probability of a Coalition win is simply the mirror image of the probability of a Labor win reflected around the horizontal 0.50 probability line).
I aim to check the odds each morning during the campaign. (Admittedly, sometimes it only gets done in the afternoon, and I have missed the odd day).
Andrew Leigh maintains a longer time series graph of the Centrebet probabilities.
Since Centrebet has not been posting its odds for the last few days (prior to 23 August 3004), I have decided to also track the derived probabilities from bookmakers listed through Sportspunter.com.
In the last three weeks of the campaign, Centrebet has opened a book on each of the marginal seats. This book allows an analysis of the individual seats that may change parties at the 2004 Federal election. The analysis can be presented graphically as a scatter-plot of seats. On the x-axis is the two-party preferred Coalition vote the seat attracted in the 2001 Federal election. On the y-axis is the probability of a Coalition win derived from Centrebet’s odds. The seats in the blue quadrant on the graph are likely Coalition gains. The seats in the red quadrant on the graph are likely Coalition loses.
Note: The above image is animated. If the image is not moving, you may have animations turned off on your web browser. For Internet Explorer check Tools > Internet options > Advanced > Multimedia > Play animations in web pages.
The seats (from left to right) are as follows:
| Seat | % TPP Coalition Vote 2001 |
|---|---|
| Bendigo Vic | 46.4 |
| Ballarat Vic | 46.8 |
| Banks NSW | 47.1 |
| Chisolm Vic | 47.3 |
| Bass Tas | 47.9 |
| Swan WA | 48.0 |
| Bonner Qld | 48.1 |
| Hasluck WA | 48.2 |
| Stirling WA | 48.4 |
| Kingston SA | 48.7 |
| Brisbane Qld | 49.0 |
| Solomon NT | 50.1 |
| Canning WA | 50.4 |
| Dobell NSW | 50.4 |
| Adelaide SA | 50.6 |
| Hindmarsh SA | 51.0 |
| Parramatta NSW | 51.2 |
| Paterson NSW | 51.4 |
| Herbert Qld | 51.5 |
| Deakin Vic | 51.6 |
| Eden-Monaro NSW | 51.7 |
| Richmond NSW | 51.7 |
| Hinkler Qld | 52.2 |
| McEwan Vic | 52.2 |
| Gippsland Vic | 52.6 |
| Moreton Qld | 52.6 |
| Page NSW | 52.8 |
| McMillan Vic | 52.9 |
| Bowman Qld | 53.1 |
| La Trobe Vic | 53.7 |
How did the betting market perform?
Very good, but not as good as in 2001 is the short answer.
Like most of the commentators, the derived probabilities from the betting market on marginal seats underestimated the extent of the 2004 Coalition win.
The betting market did not predict the following Coalition wins.
| Seat | Predicted probability of a Coalition win | Coalition 2004 two-party preferred outcome (%) |
|---|---|---|
| Bonner (Qld) | 0.22 | 50.5 |
| Kingston (SA) | 0.29 | 50.1 |
| Solomon (NT) | 0.41 | 52.8 |
| McMillan (Vic) | 0.43 | 54.8 |
| Stirling (WA) | 0.45 | 52.0 |
The betting market did not predict the following Labor wins.
| Seat | Predicted probability of a Coalition win | Coalition 2004 two-party preferred outcome (%) |
|---|---|---|
| Adelaide (SA) | 0.56 | 48.7 |
| Richmond (NSW) | 0.69 | 49.8 |
There were 28 seats on Centrebet’s betting market that received between 46 and 54 per cent of the two party preferred (TPP) vote in 2001, where the market had predicted either a Coalition or Labor win at 7am on Saturday 9 October 2004. [This was the last time I checked the betting market].
Of these 28 seat predictions, the betting market got 7 wrong and 21 right. This is a strike rate of 75 per cent.
In 2001, according to Andrew Leigh, 43 of the 47 favourites won. That is a strike rate of 91 per cent. However Centrebet’s 2001 set of seats included a lot more safe seats than those on which Centrebet provided odds in 2004. Consequently, in comparison with 2004, the 2001 strike rate may not be as good as first thought.
Why did the betting market do poorer in 2004? I do not know for certain, but I have a couple of hypotheses.
First, the activity on the betting market was highly publicised in the 2004 election campaign compared with 2001. It is possible that the 2001 prediction was more accurate because proportionately more people were trading in the betting market with inside information. These would include party hacks with access to party polling results and the like. In 2004, because of the high level of publicity, it is possible that proportionately fewer informed bets where placed, and those with inside information did not have sufficient capital to invest against the uninformed bets. The market was less accurate as a result.
Second, if you look at the animated scatterplot of derived probabilities, the marginal seat market was largely moving from the Coalition to Labor for the period 21 September to about 4 October. Between 5 October and 9 October it appears to move back to the Coalition. It is possible that this movement represents actual shifts in voting intention. However, because this change in momentum occurred so late in the campaign, it was not fully captured by opinion polls and other analytical techniques, and was therefore underestimated by the punters.
Note 1: the betting market had the seat of Hindmarsh (SA) on even money to go either way. As it turns out it went to Labor with a margin of 50.06 per cent. So I think we can say the betting market got this 29th seat pretty well right too.
Note 2: Centrebet also took bets on three safe Coalition seats that were not in the 46 to 54 per cent range for TPP vote in 2001: Dunkley (Vic), North Sydney (NSW) and Wentworth (NSW). These three seats were not included in my graphs during the election campaign. If we add these seats to the total, we have thirty two seats, 31 of which made a definitive prediction. Of these 31, the favourites won in 24 — giving a strike rate of 77 per cent.