knowquestion :: Stephen Bounds

On the eve of the 2010 Australian Federal Election, I can't help but jump on the bandagon of those predicting the result. But I'm going to take a slightly different approach.

Back in 2008, there was a flurry of interest in "the wisdom of crowds" and the power of predictive markets. Despite being skeptical about their efficacy, Dave Snowden nicely outlined the key properties of an effective prediction market:

1. You need diversity of opinion, each actor needs a private opinion,
2. Their opinions should not be determined by others, i.e. they should not be following the herd
3. Participants should be able to draw on local knowledge
4. There needs to be a mechanism for aggregating the individual decisions to represent the collective.

At the time of writing, Centrebet has the ALP odds-on to win government at $1.60 to the Coalition's $2.30 -- or a 59% chance of a Labor victory. However, this doesn't tell us the likely makeup of the parliament. To get a true picture of the national position, we must analyse the odds for each individual seat.

Firstly, we assume that the odds quoted by Centrebet per seat accurately reflect local sentiment on voting propensity. Next, we use a Monte Carlo simulation to run multiple elections and track the results in each. From this we can accurately estimate the likelihood of a given outcome (or at least, the likelihood as seen by the punters).

From these figures, I've calculated the chances for various elections outcomes three times as the campaign has progressed. The ALP, while still the favourite to win, has seen its position worsen over the course of the campaign:

Technically the ALP/Green coalition scenario is also a hung parliament, but it will produce a distinctly different kind of government so I've accounted for it separately.