Why are the opinion polls and the bookies odds so different?

The latest odds (Sunday 14th September) show that the implied probability of a No vote, based on the odds offered by 24 bookmakers up to midday on Sep 14th have stabilised at around 0.8. To find out how I calculate these odds, see here.

The market estimates of the probability of a No vote are consistently higher than the opinion polls suggest.  Figure 1 shows the estimated probability of a No vote from the opinion polls and the bookmakers odds for the period from 1st March 2014 to 14th September 2014. The latest data show that the bookmakers odds suggest an 80 per cent chance of a No outcome, while the opinion polls are close to evens – a 50 per cent chance of a No vote.

Figure 1: Betting Odds (Market Probability) and Opinion Poll Predictions of a No Vote in Scottish Referendum.


Why are the two methods of predicting the outcome telling such different stories?

The opinion polls suffer from a couple of important disadvantages.

  1. Whereas bookmakers odds are available many times during a single day, opinion polls are infrequent and therefore do not constitute a near continuous time series.
  2. The opinion polls are collected by a number of different polling organisations. There appear to be systematic differences between the findings of these organisations. If such differences exist then the opinion poll findings must be biased for at least all but one of the polling organisations.

Together, these suggest that the opinion poll data are likely to be much more noisy than are the market prediction data. It is therefore much easier to pick out market reaction to specific events from the betting data than from the opinion polls. From Figure 1, there are fairly clear reductions in the probability of a No outcome that followed the ramifications of Mr Osborne’s invervention early in 2014, where he ruled out the possibility of a currency union between Scotland and rUK. Similalry, there was a clear response in the betting data to the outcome of the second televised debate, which was broadly agreed to have resulted in a victory for Mr Salmond. The response of the opinion poll data to these events is much less clear than is that of the betting data.

Nevertheless, these differences cannot really explain the huge difference between the opinion poll and betting market data. Given the generally better record of betting markets in predicting outcomes, the Scottish referendum provides a very interesting test of their relative merits.

One explanation is that the wording used in opinion polls tends to focus on current voting intentions rather than attempting to forecast the outcome of the event. Dispassionate participants in the betting market are only interested in the outcome, since that is how they profit. With the bulk of the money being loaded on a No outcome, bookmakers have no option but to cut the odds on this outcome: otherwise they face a substantial loss. Thus a typical offering today (15th September) on a No outcome is 4 to 1 on.

The expectation that betting markets will be good predictors of political contests ultimately rests on a well-known piece of economic theory – the efficient markets hypothesis – which basically says that you can’t beat the market.

As with most pieces of economic theory, it rests on a number of assumptions. One of these is that market participants are not capital constrained – they can access or borrow as much as they like to place bets which reflect the maximum amount they are willing to risk on the outcome of the event.

But suppose that this assumption does not hold. Some who would like to bet more cannot because they do not have sufficient funds at their disposal. This could be happening in this contest.

Another possible explanation is that , rather than betting dispassionately, some punters are influenced by their support for one or other outcome. Further, some of our polling evidence shows that No voters are, on average, more affluent than Yes voters. They may therefore be able to place larger bets that Yes supporters, which will cause the betting odds to favour a No outcome, not because it is more likely in reality, but rather because No voters are more able to place large bets.

One interesting piece of evidence that might support this view is that the number of bets being placed on a Yes outcome substantially exceeds the number being placed on No. Currently (morning of 15th Sep) , 58 per cent of bets being placed are on a Yes outcome, while only 42 per cent are being placed on No. Yet the odds are fairly static, indicating that bets on No are typically larger than those on Yes.

Are the betting odds therefore misleading in respect of the outcome of the referendum? This is very difficult to assess, but their record on normal political contests is at least as good as the opinion polls. Whether there has been some special circumstances in this contest (such as the effect of constraints) will have to wait until the post-referendum analysis.


About David Bell

David Bell FRSE is Professor of Economics at the University of Stirling. He graduated in economics and statistics at the University of Aberdeen and in econometrics at the London School of Economics. He has worked at the Universities of St Andrews, Strathclyde, Warwick and Glasgow. His research is mainly in labour economics, fiscal decentralization and the economics of long-term care. He has been budget adviser to the Scottish Parliament Finance Committee. He is PI for the Healthy AGing In Scotland project (HAGIS).
This entry was posted in Referendum odds, Risk. Bookmark the permalink.