Polls pundits and pollsters
My article on this topic, from Thursday’s Fin, is over the fold. It benefitted from earlier discussion here. Feel free to comment more
Politicians routinely say ‘the only poll that counts is on election day’, particularly when they are behind, though that doesn’t stop them paying very close attention the results. But polls are not the only basis for predicting election outcomes. Economists including Andrew Leigh and Justin Wolfers have made a strong case that betting markets provide a more accurate guide than polls. And opinion pages like this one allow pundits to make their own predictions on the basis of information or insights presumably not available to the average punter or poll respondent.
Most of the time, these indicators all point in the same direction. In 2007, however, polls, pundits and punters have given very different predictions of the likely outcome of the election due later this year. Ever since the election of Kevin Rudd as Opposition leader, the polls have given Labor a commanding lead, stable for some time at 58 per cent of the two-party preferred vote, plus or minus the sampling error of 2 to 3 per cent.
Most pundits and political insiders have until recently dismissed these poll results and predicted a resurgence for the government and John Howard. Over the last weekend, however, there seems to have been a sudden collective shift, with a string of pundits suggesting that the government is on the ropes, and Howard himself dramatically playing the underdog card. The trigger for this change of view seems to have been the absence of the much-predicted ‘budget bounce’ in the polls, though in the past such bounces have been the exception rather than the rule.
By contrast, there has been no real shift in the odds set in betting markets. The Coalition is still virtually even money to win, though, thanks to the house spread, the odds on Labor are slightly shorter.
The slowness of pundits to believe the message coming from the polls reflects the experience of the 2001 and 2004 election campaigns, when Howard came from behind to win. It might be argued that such a small sample is not a good basis from which to draw conclusions. After all, back in 1987, Howard seemed to be in a winning position, but was unable to recover from the ‘Joh for Canberra’ fiasco.
That leaves the markets. The idea that markets can provide better predictions than surveys or expert opinions is an example of the general claim that financial markets give the best available estimate of the value of financial assets such as shares. This claim is called the efficient markets hypothesis.
The efficient markets hypothesis comes in several versions, commonly called ‘weak’, ‘semistrong’ and ‘strong’. The weak form says that investors efficiently use all the information available from observing past share prices. In the current context, this says that the betting market should get the outcome right at least as often as someone who bases their judgement on past election results. The most obvious strategy of the latter kind is one of betting on the incumbent government to win, as has happened in 19 of the 24 federal elections since the current party system emerged with the foundation of the Liberal party in 1944.
The ‘semistrong’ form of the hypothesis states that the market makes the best available use of public information, such as opinion polls and the published judgement of presumably expert commentators. The polls have done a pretty good job of picking the changes of government, and most pundits correctly predicted the outcomes in 1972, 1975, 1983 and 1996. On the other hand, at least some of the time, polls and pundits have predicted changes of government that didn’t happen. The most obvious case is 1993, when John Hewson was also a short-priced favorite in the betting markets as well.
The most controversial, strong form of the efficient hypothesis states that market prices reflect both public and private information. It was this idea that led to proposals, rapidly squashed by public outcry to establish futures markets in terror attacks and other political crises.
Tbe current divergence between the results of polls, and those of betting markets, provides what may be a crucial test. If Labor loses, it would be hard to see the point of continuing to report poll results, except in the period immediately before an election. It would seem better just to watch the betting markets.. On the other hand if Labor wins the victory long predicted by their surveys, the pollsters can celebrate.
In the end, though, there is still only one poll that counts.