Markets and the election - Should we have faith in the predictive power of

rlp

Nov 6th 2016, 10:46 by Buttonwood

GAMBLERS seem to think Donald Trump is a 3-1 outsider to win the election on

Tuesday. The Predictit market has Hillary Clinton on 77% (where 100% is

certainty). But what does that really tell us? On June 21, two days before the

British referendum on the EU, Betfair was implying odds of 75% for Remain and

25% for Leave. This was despite quite a large number of polls that showed Leave

ahead. As we know, Leave won 52% to 48%.

The idea of looking at betting markets is that smart money can assess all the

information far more efficiently than opinion pollsters or newspaper

columnists. It s efficient market theory for politics. But what are the

gamblers assessing? They are looking at the same poll as everyone else. In the

EU referendum, some people seemed to believe that despite the polls, a status

quo bias would ensure that Remain won (such a thing had been seen in previous

referendums). But that wasn t the case. Instead, as I feared in my first column

of the year, voters seemed to have used the poll to protest against high levels

of immigration.

Another oddity of the pre-referendum betting was the existence of a big split;

those who gambled a lot of money thought Remain would win, small gamblers

thought correctly it would be leave. The bookies for sound reasons moved odds

to reflect the weight of money. But perhaps the number of gamblers on each side

would have been the better guide for pundits.

Perhaps. After the polls, gamblers put up the reasonable defence that they did

not say a Remain win was a certainty. Indeed, the odds implied Leave would win

on one of four occasions; this was that occasion.

This may be mathematically correct but it is not very helpful. Such elections

are one-off events; they will not be run four times. (Hands up those who would

like this Presidential election to be re-run over the next 12 months). As a

guide to this election a 75% probability is an unverifiable proposition; the

outcome can only be 100% or 0%. The gambling odds only reflect the conventional

wisdom, and that among people with enough money to bet big. But we know the

conventional wisdom already, and putting a percentage on it doesn t help much

as that number must by definition be wrong.

That brings me to the big debate of the weekend among psephologists the attack

on Nate Silver of 538 s methods by the Huffington Post. 538 has tended to show

a much higher probability of a Trump win than other sites; at the time of

writing he has Clinton at 65.8%, Trump 34.2%. I confess to being rather puzzled

how some recent polls have affected the 538 odds (since I wrote the last

sentence, for example, an NBC poll showing a four-point lead for Clinton caused

the 538 odds to drop to 64.4%). But having listened religiously to 538 s

podcasts and read the explanatory pieces, I m more than willing to trust his

judgment. Polls have been wrong in the recent past there was a shift to Trump a

week ago and the state polls have been very erratic. So if you want to pick a

probability number, 538 s guess is probably better than that of the gamblers.