Take Bookmakers Over Opinion Polls

The general election [of 2010] has thrown up many interesting issues for investors, but none was so marked as the continuous disparity, up to and including polling day itself, between the outcome in seats implied by the opinion polls and endorsed by many pundits, and the different outcome that was forecast throughout in the odds quoted by bookmakers and on the betting exchanges.

It has been an article of faith of mine for years that, given a choice between the polls and the betting market, or between the odds and “expert opinion”, the punters are the group to follow. That approach rarely lets you down, and the result this time round again vindicates it, with the Conservatives emerging as comfortably the largest party, having traded between 300 and 330 seats on the betting exchanges throughout the campaign, including the Clegg mania phase.

Until now however, I have never given much thought to why betting markets should be better at the prediction business. It seems obvious that the rapid growth in the betting market, and in particular the arrival of the betting exchanges, which allow significant sums to be wagered on events such as elections without the traditional bookmaker’s take, has transformed the depth and validity of the informational content implicit in the odds.

But while that must be true, the change may still only be a matter of degree rather than a purely new phenomenon. Three years ago Professor Leighton Vaughan Williams of Nottingham Business School, who researches both gambling and political forecasting, published an interesting paper which suggested that bookmakers have long held an edge over both opinion polls and expert opinion when it comes to predictive power. (Click here for a summary).

His research, drawing on betting records, showed that in all the US Presidential elections between 1868 and 1940, the era before opinion polls, only once did the candidate who was favourite in the betting a month before the poll fail to win the election.

Since then, despite the arrival of the opinion polls and the much greater media interest that is customarily accorded to them, in almost every fair contest the betting markets have comfortably outshone all competitors in the predictive business. In 2004, one of the two main predictive exchanges that are allowed to take political bets in the United States, “went stratospheric in predictive accuracy”, according to Prof Vaughan Williams, by predicting the winner in every single state while a number of pollsters and pundits continued to favour Kerry over Bush right up until polling day.

“Just think if you had an accumulator bet on that!” Prof Vaughan Williams enthused last week, as we chewed over the odds before the result was known. In his ranking of methodologies, Prof Vaughan Williams puts the betting exchanges at the top and, gratifyingly for another long held prejudice of mine, econometric models at the bottom. He believes that Betfair and other exchanges which do not limit the amount that can be wagered on political outcomes, as happens in the United States, are potentially more accurate as a result. The heavier the bet, the greater its predictive power.

There are good reasons why, absent manipulation, betting odds should often do so well. Unlike say the Grand National, general elections have very few potential winners and the odds are ultimately set by those with superior professional knowledge – which means principally those who have access to detailed information on the ground, such as canvas returns.

With opinion polls, however, there is the permanent problem that those invited to offer an opinion are not selected because they are likely to know something; an opinion poll has to be random to be statistically valid. At the same time, there is no penalty for giving an inaccurate or false reply.

Taken collectively, the opinion polls in a general election are often accurate within their margins of error about the percentage of votes cast. Translating that into seats won requires more specialist or local knowledge than a simple read across, assuming a uniform national swing, allows. This is the other main reason that the results derived from the polls by the media and the general public are so often misguided – the obvious question being why, if this is now so well known, misinterpretation continues to leave money on the table for those who know how to read the runes more accurately. Even the most contrarian of investors should rarely bet against the oddmakers.