I spent yesterday at the Australian Conference of Economists, aptly* held at the University of Melbourne. In a session organised by Tim Fry and Sinclair Davidson, Sinclair presented a paper of theirs on trust and the 2004 election, and I presented a paper by Justin Wolfers and myself on election forecasting. It was a terrific chance to catch up with Justin and his partner Betsey, who’d flown in from Philadelphia, and to say g’day to John Quiggin (who I can confirm that is looking as fit and hirsute as always). Regular readers of the blog are probably heartily sick of election forecasting, but in case you’re not, we got a bit of coverage in the Age and AFR (over the fold).
The conference goes on today and Wednesday, but I’m teaching tomorrow afternoon, so have returned to the berra for that. On Thursday, I’ll be back on the road, speaking at the ANZSOG schools conference in Sydney and the HILDA conference in Melbourne.
* Aptly, since I think of "ACE" as a very Melbourne term of approval.
Election results: the bookies know best
AFR, 26 September 2005
Polls are useless in predicting the outcomes of Australian elections but it would appear the bookies are on the money.
Economists Andrew Leigh from the Australian National University and Justin Wolfers from the University of Pennsylvania examined last October’s national election to determine which methods best predicted its outcome.
Their research, to be presented today at the Economic Society of Australia’s Annual Conference of Economists in Melbourne, showed betting markets predicted not only that John Howard would win with an increased majority, but also foresaw outcomes in three-quarters of the marginal seats.
Analysing data from Centrebet, International All Sports, Sporting Bet and SportsAcumen, Mr Wolfers found that the bookies’ data responded quickly to campaign news as well as outside factors such as the Jakarta bombing in the 2004 campaign.
"These data also suggest that the Labor’s Tasmania forestry announcement provided the coalition with a substantial boost," their research said. The polls on the other hand, were extremely volatile and should be used with caution, the authors said.
Using the polls published by ACNielsen, Galaxy, Roy Morgan and Newspoll, the academics found their predictive power to be very low, often favouring a Labor Party win.
Betting markets suggested the likelihood of a coalition win varied only from 54 per cent to 77 per cent, but the polls predictions of a coalition win during the campaign swung widely between 0.7 per cent and 98.3 per cent over the six-week period.
"The most striking aspect of this figure is the extreme volatility of the implied probability of a coalition victory suggested by the polls."
Dr Leigh and Dr Wolfers also analysed the models that try to predict election results based on economic robustness. They found these also predicted a Howard win.