Wednesday, July 3, 2013

Seat Betting Favourites Watch - July 3

This is a series that may run on here sporadically from now til the federal election, the date of which is now unknown but considered most likely to be in either late August or late October. At this stage I'm thinking of posting an update in a fresh article weekly around the middle of the week, but the posting time will vary according to my other commitments.  Also it may well be that someone else starts covering the same thing better, in which case I may just link to them and leave them to it.

Following the first week of Rudd-return polling, six polls by five companies have shown Labor somewhere in the range of 48-51% two-party preferred, with five of the six showing the government just behind.  Seat models point to a very close election if the an election was proverbially "held today".  The re-fired Bludgertrack projects 74-73-3 (ALP first), Mark the Ballot gives a uniform swing (or lack thereof) at 73-75-2 and Pollytics reports "My election simulation produces a similar result 76 seats to the ALP vs. 72 to the Coalition, with 2 Independents." (This is not the same as the 77-71-2 simulation I questioned as a slightly unsound extrapolation from its given data at the bottom of my previous post, and I assume it incorporates more data.)

A model that is not exclusively poll-based, the Pottinger projection, which incorporates data from betting odds as well as polls and past election polling-to-election day patterns, is a bit less optimistic of Labor's prospects, assessing a median outcome of 67 seats and giving only a 2.9% chance of a Labor outright victory and a 15.6% chance of the outcome that would amuse me endlessly, another hung parliament.  (That model assumes an August 24 election; I suspect that with a later election assumed Labor's predicted chance would rise.)

When we look at pure betting odds, as this series will do, it turns out that at present there is a big discrepancy between the polling and the betting.  Although if current polling was reflected on polling day the election would be on the wire, it's not likely now that the government's polling will just "random walk" from here to the line without anything in particular pushing it one way or the other.  There is a bounce effect from the change of Prime Minister, but we don't know yet where it will peak, and we don't know how far it will go down after it peaks.  Anyone trying to model where voting intention will fall to by polling day will be finding that at this stage this one is even harder than last time, when the bounce was rather small.

So this election is an excellent opportunity for the betting markets to demonstrate they really do know something about the outcome without needing to be told it by the polls or by the more orthodox assumptions of election modelling. That's something I have a lot of doubts about, so I've decided to monitor it and see how it goes. 

Here then is my opening record of the state of seat betting markets following their resetting a few days ago.  Most of the markets may not have yet seen that much action but a few (notably Kingsford Smith) have moved since relaunching and there's been time for anything obviously silly to be snaffled up by someone:

The colour-coding:

Medium blue: A seat in which the Coalition is favoured in all betting markets.
Pale blue: Coalition favoured in some markets, level in others.
Grey (none currently): All markets tied or both parties ahead in some markets and behind in others.
Orange: Labor favoured in some markets, level in others.
White (tried red but it made the text look smudgy): Labor favoured in all markets.

Presently two markets are used, Sportsbet and Sportingbet/Centrebet.  (There is some seat betting on Betfair as well but not much action apparent at this point.)

The starting position is that the Coalition has 75 seats it notionally owns (including New England and Lyne) and Labor has 73 if it recaptures Melbourne.

From this, the seat betting markets currently have Labor favourite in 58 seats, with 15 expected losses. The seat number betting markets also - on average - think Labor will win about this many seats.  As is sometimes the case the "rake" in seat number betting markets is hideous; it's been up around 40% this week. Also notably a very wide spread of outcomes are at relatively short odds:

With odds like that either the bookies are scared or the markets are confused; nothing between a massacre perhaps slightly worse than under Gillard, and the even result shown by current polling is longer than $5.00.  There's a slight central tendency but perhaps it just reflects the line of thought that says anything within a certain range could happen so it is slightly more likely to fall in the middle. I will say here that I just do not believe that the probability of a sub-41-seat result for Labor is anywhere near as high as 17%.

These early Rudd-return seat markets don't currently predict Labor is winning any specific Coalition seat, and they also aren't expecting a swing to Labor in any state.  If punters generally thought there was enough juice in the Queensland swing idea shown in early polling to account for the LNP margin plus sophomore effect in all those low-hanging targets there, then we would see Labor favourite in some of these.  However, at present the Coalition is favourite in Labor's most marginal Queensland seat, Moreton.

There are much more sophisticated ways to model what the markets are collectively "thinking", for instance probabilistic models which keep track of the chances of moves both federally and at state level affecting multiple seats at once, like those done by Pollytics for Crikey at previous elections.  For the time being the seat-favoritism picture is enough to show that there is a sharp difference from current polling and betting market expectations.

However I suspect such a model would project a Labor a bit less than 58, since there are currently twelve seats including Melbourne where Labor are relatively close favourites (defined as both parties under $3) compared to only six such seats for the Coalition.

So we have a contrast between a tight race in current polling and a betting market that overall still thinks Labor is heading for a moderate defeat (and this is true not only for the seat markets but also the headline betting).  Seeing how this discrepancy is resolved will be a good test of whether or not the markets are capable of getting it right when they are uninformed by predictively reliable polling and apparently confused.


  1. I think that spread market is a pretty good reflection of (my) current expectations, and the expectation that there will be a brief peak in polls that will then collapse to a very uncertain degree, based upon how well or poorly Kevin (and Labor more broadly) handles the transition to power.

    Significant Labor loss to Marginal Labor Victory are now all possible.

    Locally I'd expect that the change makes Franklin into a likely hold, but that Bass, Braddon and Lyons will still switch.

    1. When the ReachTEL came out my impression was that the polled gap was so large Lyons would probably not be held even with a leadership switch. I'm less convinced of that now as I think between the switch, the natural variation of such small samples and the possibility of a ReachTEL house effect, there's enough there to explain away nine points, provided that most of the bounce lasts. I still think the seat is at significantly greater risk than Franklin.

      I did expect the leadership bounce effect would be somewhat less in Tas than elsewhere. The swings in the two available post-switch samples were smaller than the national swing as expected but their sample sizes were tiny so more data needed.

  2. You might be interested to know that the same model with a 21 September election date gives the probabilities as: ALP win = 9.3%, Coalition win = 72.1%, hung parliament = 18.6%.

    The median number of ALP seats in this treatment is 70, with a mode of 72.

    Some of the differences between our models will also relate to the treatment of poll bias. Mine is calibrated off the last election. To the extend that polling agencies have changed their methodology significantly then this will affect the outcome.

    I'll quote my bias estimates in the next update.