Thursday, June 30, 2016

Seat Betting Watch: The Last Few Days

I'll start a final poll roundup tonight with the release of Ipsos, and try to make a few coherent general comments about the state of the campaign then.  This separate piece continues my monitoring of the seat betting markets, because the predictiveness of betting markets is frequently a matter of discussion.  Seat betting markets performed indifferently at the 2013 election and it will be interesting to see if they do better this time around.  This election lacks the major challenge of the 2013 poll (a massive difference between national polling and seat polling) but there are new problems to make things challenging for people betting on the outcomes.  These include the large number of unpredictable minor-party challenges (especially in South Australia) and also the unclear impact of the Xenophon Team on the preference situation.

I had some interesting insight into the shallowness of some of the specialty betting markets (2PP, clean sweeps and so on) when someone contacted me about some of the odds and then threw money at them, resulting in substantial and rapid changes.  It would be interesting to have some information about how many people are actually betting on some of these markets; in some cases it is probably not very many.

 This is how things line up on the seat betting markets at present.  A seat is considered close if both parties are at $3 or less on any market:

Loss (Coalition to Labor): Barton*, Paterson*
Close Loss (Coalition to Labor): Solomon, Dobell*,  Eden-Monaro,  Macarthur,  Capricornia, Burt, Petrie, Page, Lyons
Close Loss (Coalition to NXT): Mayo
Loss (PUP to Coalition): Fairfax
Tied: Cowan (split markets)

ALP Close Holds: Batman (vs Grn), Greenway, Chisholm
Coalition Close Holds: Banks,  Gilmore, Macquarie, Cowper (vs IND), Brisbane, Flynn, Forde, Grey (vs NXT), Barker (vs NXT),  Hindmarsh, Braddon, Bass, Deakin, Dunkley, Hasluck, Swan

In the last week Lyons has moved into the Labor pile following a 45:55 ReachTEL (though on the public evidence I think Labor are still too short here).  NXT has become the favourite in Mayo and Cowan has moved to the fence.  Mark the Ballot has great detailed tracking of the Sportsbet odds; I can add that Batman has briefly been tied during the last week but hasn't stayed there for long.

Here's the colour-coded tracker:


Key to colours:

Dark blue - Coalition favoured in all markets
Light blue - Coalition favoured in some markets, tied in others
Grey - seat tied, or different favourites in different markets
Pink - Labor favoured in some markets, tied in others
Red - Labor favoured in all markets
Dark green - Green favoured in all markets
Orange - NXT favoured in all markets

While the Coalition has lost one net seat in which it is favourite, it has regained ground on my "close seat adjusted" total.  Mark the Ballot's graph of the sum of seat probabilities as perceived by the market (with longshots chopped out) shows Labor shedding about five seats over the past few weeks.  All this is consistent with the markets gradually adjusting to a small shift in national polls that began in the second week of June.

The various totals markets are expecting about the same thing.  The William Hill banded seat market expects 79 seats with 76-80 the shortest range.  The William Hill exact seat market expects 77.6 seats on average with 79-80 seats as joint favourites at $12.  Oddly the Centrebet exact seat market has 77 seats favourite at just $3.50 (and if you think Labor will win outright and can get the margin right, that's good for at least $100) - the average there comes to 78.3.  The Sportsbet 2PP market has finally fallen into line with other markets with 50-51 now the favourite range.

At the last election we saw a lot of media reporting of Monte Carlo models where people ran, say, 10000 runs of the seat-betting probabilities in the different seats and on that basis concluded that Labor had no chance of winning majority government. The same exercise gives the same result this time. I haven't seen so much of that this time around but I should point out that while such exercises are fun they provide no useful information about what markets are thinking, or indeed about anything at all.

The reason is that they depend on the assumption that seat probabilities are independent, so that, for instance, the Coalition winning Lindsay doesn't change the Coalition's chance of winning Banks.  But in reality if you know the Coalition wins a given seat, this makes it more likely that its national 2PP is good, and hence increases the chance of it winning others.  If there is a national swing or a national polling error then this changes the correct odds in almost every seat, and punters know there is a chance of this.  Such models don't even show that the chance of one side winning is such-and-such if there isn't a national swing in the last few days (or a polling error), because if we knew we had the 2PP right we could be far more confident about more seats.  The way to determine what markets "think" the seat tally will be, and with what spread of possible outcomes, is to look at markets that are directly about seat tallies.

Seat Poll Model

My seat poll model (limited to objective information as it is, but with some subjective calls regarding weightings) currently largely agrees with the betting markets - it expects about 78 Coalition seat wins based on current national polling.  If the national polling changes then so does the seat estimate.  It has Labor favourites in the following Coalition seats: Barton, Paterson, Dobell, Petrie, Capricornia, Lyons, Solomon, Cowan (which it thinks is extremely shaky, and so do the markets and so do I) and Burt.  It disagrees mildly with the markets about Macarthur (where it has a 53% Coalition retain chance assuming no further 2PP change) and Eden-Monaro and Page (neither of which have had a neutral seat poll; if I add commissioned seat polls then both flip.)  Although it has the Coalition favourite in 83 seats, I am currently overriding Mayo based on seat-polls combined with my NXT modelling attempts.  Of the rest it thinks there are enough shaky Coalition seats for the Coalition to be expected to lose about five of the seats in which it is favourite, and to pick up about one seat from Labor.  However the shortage of data about ALP seats is so severe that the model cannot single out any specific seat with a very high chance of going against the flow.  It just says it will probably happen somewhere.

The problem with the national polling my model is based on is that yet again it is all several days old.  If things have changed in the final week we will not get a hint until tonight, and will then need more polls to confirm, by which time it will all be over!

Updates or extra detail, especially re notable betting movements (if any) will be added to this site tomorrow and tomorrow night.  If you see a seat "flip" in betting, feel free to let me know.

Updates:

Split Market In Batman: As of 4 pm Friday Bhathal (Green) has become favourite in Batman on at least one market but not yet all of them.  Also the seat of Melbourne Ports is tightening with the Liberals in to $4.50 from $7 a few days ago.  (The problem in Melbourne Ports is that Michael Danby is being targeted by the Greens, who could push him into third, at which point they might struggle to get enough preferences given Danby's anti-Green stance.)

Libs Favourite In Cowan: Betting markets now favour Liberal Luke Simpkins in Cowan, making the Coalition favourite in 80 seats.

Murray: I've pretty much ignored this intra-Coalition contest but the Nationals have hit the lead.

Cowper Flips: As noted in comments, Cowper has flipped and Rob Oakeshott is favourite at $1.75 Sportsbet $1.65 Crownbet while William Hill have pulled the market on it.  That means the current list of seat favourites has Coalition 79 Labor 64 and Others 7.

Final (2am election day):

Loss (Coalition to Labor): Barton*, Paterson*, Dobell, Solomon
Close Loss (Coalition to Labor): Eden-Monaro,  Macarthur,  Capricornia, Burt, Petrie, Page, Lyons
Close Loss (Coalition to NXT): Mayo
Close Loss (Coalition to IND): Cowper
Loss (PUP to Coalition): Fairfax
Tied (ALP vs Green): Batman (split markets)

ALP Close Holds:  Greenway, Chisholm
Coalition Close Holds: Robertson, Gilmore, Macquarie, Brisbane,  Grey (vs NXT), Barker (vs NXT),  Hindmarsh, Braddon, Cowan



Key to colours:

Dark blue - Coalition favoured in all markets
Light blue - Coalition favoured in some markets, tied in others
Grey - seat tied, or different favourites in different markets
Pink - Labor favoured in some markets, tied in others
Red - Labor favoured in all markets
Dark green - Green favoured in all markets
Orange - NXT favoured in all markets
Purple - Ind favoured in all markets

Total of favourites: Coalition 79 Labor 64.5 Green 1.5 Ind 3 NXT 1 KAP 1 
Close seat adjusted: Coalition 79.6 Labor 63.6 Green 1.5 Ind 2.7 NXT 1.6 KAP 1

5 comments:

  1. Betfair will show you the total matched per market, but given it's not really the main bookmaker in Australia, it might be tough to interpolate from that figure.

    ReplyDelete
  2. Ta. I hadn't looked at Betfair yet, but as it happens a lot of seats there have no bets or trivial totals (Mayo $9), while some do have modest amounts matched (eg Eden-Monaro over $1000). Their headline rate around the $9 range seems consistent with others given the way they work.

    ReplyDelete
  3. Granted that national polling tends to overestimate the Greens vote.
    BUT the experience of the Vic and NSW state elections is that seat polling UNDERESTIMATES the Greens vote.

    On that basis, wouldn't you be tempted to put Batman as a probably gain for GREENS and loss for Labor?

    ReplyDelete
    Replies
    1. ReachTEL had some shockers with the Green vote in NSW seats at the last state election, and had a similarly bad experience in Melbourne (federal seat) early in its existence. There is a lot of enrolment churn and transience in inner-city electorates and up-to-date databases are very important. However the one neutral poll of Batman I recall has been by Newspoll/Galaxy which has a better record with such cases (albeit from limited evidence).


      Concerning Victoria I remember the Greens putting out commissioned polling saying they would win both Melbourne (state seat) and Richmond, but only winning Melbourne and missing the margin in Richmond by six points. So I don't recall seat polls for the Victorian election underestimating them.

      Subjectively I think the seat is at high risk because of the inner-city intensification thing we saw in both those state elections, plus the Feeney disasters and the possibility Liberals will buck the card. Objectively the evidence for a 10 point swing to a party with only a modest national gain needs to be stronger than a poll showing Feeney three points ahead. So it's a tricky one. I'm tempted, but for different reasons.

      Delete
  4. You should also have a look at the betting market in Cowper. Sportsbet started reducing Oakeshott's odds dramatically this afternoon. He was equal favourite (at 1.87) earlier in the evening and is now favoured at 1.75. William Hill did not react so quickly and have stopped taking bets on Cowper after leaving Oakeshott at 3.00 or so for too long (they reacted too late to avoid at least some arbitraging).

    ReplyDelete