Analysing the SPIN ‘321’ markets

A few days ago, Sporting Index (www.sportingindex.com) opened a number of new election spread-bet markets. Amongst these is a set of ‘321’ markets, which they describe as follows:

Points are awarded on the following basis:

3pts for each constituency this party wins;
2pts for each constituency this party finishes second;
1pt for each constituency this party finishes third.
0pts for any other result.

Note: The maximum theoretical result = 1950 from 650 constituencies.

However, they also state that no Northern Ireland seats count towards any of these markets, so in practice there are 632 seats in all. That means the total number of points available for all parties is 3792.

At the time of writing, the prices on the headline seats markets are:

Con 332-337
Lab 222-227
LD 56-59

The prices on the ‘321’ markets are:

Con 1405-1435
Lab 1290-1320
LD 870-900

This raises the question: are these sets of prices consistent? In other words, what should the 321 markets be priced at, to be consistent with the headline spreads?

To answer this, we need to estimate which party would be in 2nd and 3rd place in each constituency, if the final result were (as SPIN are currently implying by their midpoints) something like Con 334, Lab 224, LD 57, Nats 15, Others 2.

As a first approximation, I used www.electoralcalculus.co.uk and played around with the vote shares to get close to the above figures. Typing in 40/29/21 gets you pretty close, and you can extract the vote shares in all seats into a spreadsheet, and rank the parties 3/2/1.

I then made some adjustments. Firstly, Electoral Calculus treats 'Others' as a single party, so it erroneously makes them appear to be 3rd place in a number of seats where in reality Others would be split between Greens, UKIP, BNP etc. I therefore went through all the seats where Others were placed in the top three, and checked with the 2005 results to see whether there was a single dominant contender. Where there was not, I downgraded Others to 4th, leaving 8 seats with a minor party in third place: Glasgow NE (Ind Soc), Rotherham, Dagenham & Rainham, Keighley (BNP), Newton Abbot, Torbay, Totnes, Devon East (UKIP). The exact constituencies don't matter, of course - we're interested in the totals, and that seems a reasonable guess.

I also fixed up a few specials: Buckingham (Speaker 1, Others=UKIP 2, Others 3), Wyre Forest (Con=1, Ind=2), and Blaenau Gwent (Lab=1, Ind=2).

Finally I adjusted some of the Scottish and Welsh seats to make the total for SNP/Plaid Cymru up to 15, rather than the 10 outright victories predicted by Electoral Calculus. This makes the figures consistent with SPIN’s SNP and PC spreads.

These adjustments are all fairly small in the overall scheme of things, and don’t make a big difference to the final conclusion. But they do give a result which is internally consistent, which is what matters.


Results

These are the tables for the 632 UK mainland seats (including Buckingham) which I ended up with this method:

Con:
1: 334
2: 214
3: 77
Total points: 1507

Lab:
1: 224
2: 236
3: 155
Total points: 1299

LD:
1: 57
2: 162
3: 360
Total points: 855

SNP and PC:
1: 15
2: 16
3: 27
Total points: 104

Speaker and Minor parties (Respect, Greens etc):
1: 2
2: 1
3: 8
Total points: 16

Others (Independents, UKIP, BNP etc):
1: 0
2: 1
3: 9
Total points: 11

Total number of 1st places: 632
Total number of points: 3792


Conclusion

To be consistent with SPIN’s current headline figures, it looks to me as though the 321 market spreads should be around Con 1507, Lab 1299, LD 855. That leaves 131 points for other parties, divided into SNP/PC 104, Others 27. Note that these results are consistent with what happened in 2005, where the SNP, PC and Others got 121 points in all.

Thus the conclusion is that the Con321 market looks substantially underpriced at 1405-1435, and the LibDem market somewhat overpriced. The Lab price looks about right within the spread, perhaps a touch high.

This conclusion – especially the underprice on the Con321 market - looks fairly robust even if the headline spread of Con 332-337 falls significantly.