Leng and I use betting odds to make our predictions. None of this is new — academics have been using them for years — but they’re pretty new to the mainstream media, so we thought it’d be good to talk through some of their strengths and weaknesses.
Why use betting odds?
- They’re more accurate: Previous studies have shown that they often yield comparable, if not better, predictions than polls. Betting odds reflect the market’s estimate of the underlying probabilities of victory for each party, implicitly combining information from many sources. Unlike poll respondents, bettors are putting their money on the line, so their views are likely to be better informed than those given to pollsters.
- They include more information: Participants will include not just their own views, but those of their friends, colleagues, pundits, polls and other sources in forming these views. This means betting markets tend to have a larger “effective” sample size, compared to polls.
- They’re naturally designed for making predictions: While pollsters ask participants about their current voting preferences, betting markets implicitly ask participants, ‘who will win?’, so their results are more useful in making predictions.
- They allow us to answer questions national polls can’t: Elections are won by winning seats. But we don’t have separate polls for all 150 electorates. Betting markets provide us with electorate-level data that allows us to make direct predictions about who will win the election. An added bonus is that we can also predict answers to many other interesting questions that are difficult to answer with national polling data: how many seats will the ALP lose in western Sydney? How likely is a hung parliament? What are the odds of the Liberal party winning the bellwether seat of Eden-Monaro, but losing the election? Provided we have good betting odds, we can use our general framework to answer all kinds of interesting questions that can’t be answered using national polling data without making lots of assumptions.
What are the weaknesses of using betting odds?
This blog is about exploring the potential of betting odds for making predictions and, while we believe these predictions are likely to be accurate close to the election, we are just as interested in exploring the weaknesses of these techniques as their strengths. Here are a few:
- Betting odds can yield poor predictions when not many people are betting, particularly in the early stages of the campaign. In the extreme case, where no-one has placed a bet on an electorate, the betting odds will reflect the views of a few people in the betting agency and are essentially useless (although we’re able to exclude some of them from our analysis by looking for electorates where the odds have not changed from their opening values). This is why we’re cautious about making predictions at this early stage in the campaign.
- We can’t quantify the uncertainty on our predictions: Unfortunately, the betting agencies don’t make available information on market size, let alone individual betting behaviour, so we can’t put error bars on our predictions.
- They have their own biases: for instance, they tend to overestimate the probability of victory of underdogs (this is called “longshot bias”).
Overall, we think electoral betting odds are a really rich source of data for making predictions about elections, as long as they’re interpreted with caution. As we get closer to the election, we’ll start rolling out more of our analyses.