My armchair has Stata installed

Dennis Shanahan in today’s Oz:

While armchair critics out of the political loop cite professional betting as an election prediction, they forget betting markets drift about according to the polls and political news and commentary.

To which a natural rejoinder would be:

While commentators out of the academic loop often claim that betting markets merely follow the polls, they forget that such a hypothesis has been formally tested on a number of occasions, including in Australia. Moreover, a spate of careful studies show that the mean forecast error of election betting markets is lower than that of opinion polls. Therefore, journalists can better serve their readers by placing more emphasis on stable prediction markets and less on volatile opinion polls.

(Hat tip: Brenton Caffin. If the title of the blog post confuses you, this may help.)

This entry was posted in Australian Politics. Bookmark the permalink.

7 Responses to My armchair has Stata installed

  1. Leopold says:

    Given the way the betting markets have lurched to Labor following the last two polls, I think you are flogging a dead horse Mr Leigh.

  2. derrida derider says:

    That’s a classic confusion of correlation with causality, Leopold. The government’s recent bungling has driven both the betting markets and the polls towards Labor. It doesn’t necessarily mean that the betting markets are following the polls; they might only be responding to the bungling.

    You need the panoply of formal statistical techniques (eg IV) to test this – which is exactly what Andrew and his confreres have done.

  3. Andrew Leigh says:

    Thanks DD. Although one ought never criticise one’s defenders, I should probably point out that we didn’t use IV – we merely tested whether lagged polls predicted future betting market outcomes. I think this is the contention that Leopold and Shanahan are making.

  4. derrida derider says:

    Whoops, that’ll teach me to read the bloody paper before I shoot my mouth off about it (will I never learn?).

  5. Sinclair Davidson says:

    To be fair, this far out the polls and betting market are less effective. Once we’re actually into the campaign the betting market will be more effective than polls (that’s what has happened in the past, and I can think of no reason why that won’t happen next time).

  6. Sinclair Davidson says:

    Actually, that is a fantastic headline (stick it up ’em stuff).

  7. Leopold says:


    I might have been confusing correlation with causality, but in this case the timing is clear – was trying to be brief in the above comment. If you take a look at the sequence of events (thanks to Bryan Palmer for his marvellous website), the betting markets had swung to the government in the week before the ACN poll was released to the extent that by the weekend of 10/11 March they were favourites again (52.0% probability of being returned). Clearly, the punters thought the government’s position was improving during that week; then ACN came out on Monday 12th March with ALP 61-39, and the markets promptly jerked back the other way to a 48% probability of Labor being returned; Newspoll this week has been followed by a further movement to the government.

    Now, I don’t doubt there are a whole range of things that feed into the betting markets, and these movements (5% probability shift) are not really that significant. I also have no doubt that Andrew has run the relevent tests and found only limited correlation. But in this case, the betting market has been very clearly following the polls (unless something else happened early morning of Monday the 12th that shifted the odds). 😉

    By the way Andrew – did you include the Morgan poll in your analysis? Any punter with more than one active brain cell would ignore it, so that could foul up your data. Just a thought.

    Disclaimer: I am having difficulty accessing Mr Palmer’s website at present, so the above is from memory. I am 99% certain that is the correct sequence of events.

Comments are closed.