I have been refining my code for the individual seats market at Sportsbet.com.au. Before we get to the charts for every seat, let me take you through the changes I have made.
Occasionally, Sportsbet suspends odds for a particular candidate in a seat, and sometimes it even removes an entire seat from the market for a few days. This morning, there were no odds at all for the seat of Franklin in Tasmania. Furthermore, the odds for a candidate in each of the seats of Bendigo (Vic) and Lyons (Tas) were suspended. In each of these cases I use the most recent odds available to calculate seat probabilities. However, to be transparent, when this occurs I add a grey bar to the seat probabilities chart.
Second, I have changed the way I manage the favourite-longshot bias problem. Before I explain these changes, I need to explain why I do this. Counting favourites from each seat does not give a good picture of the Parliament that is likely to emerge. A better picture of the likely Parliament comes from summing the probabilities. But when I sum the simple probabilities, because of the favourite-longshot bias problem, I do not get enough Labor and Coalition seats (the typical favourites), and I get far too many seats for "any other candidate" (the long shot).
To address this bias, I treat odds up to \$7.50 normally (ie. for probabilities greater than roughly 15 per cent after the bookmaker's overround). I inflate odds between \$7.50 and \$25 by multiplying them by their square root, this reduces the associated longer shot probability. Finally, I ignore odds over \$25. In many seats, the odds for any other candidate are set at \$26 (fairer odds in many of these cases might approach \$1000).
With that aside, let's look at the likely Parliament according to the punters.
And each of the seats (in reverse alphabetic order) follow.
And because you have made it this far, the overall winner odds and probabilities.
No comments:
Post a Comment