;)
Tuesday, March 15, 2011 at 9:46PM You're watching the footy with a mate who leans over and says he reckons the Cats will win by 15 points. How much leeway should you give him to make it a fair even money bet? Surprisingly - to me anyway - the answer is 24 points either way. So, if the Cats were to record any result between a loss by 9 points and a win by 39 points you should pay out.
Tuesday, March 8, 2011 at 8:11PM I've been leery of neural networks for some time because of their perhaps undeserved reputation for overfitting data and because of the practical difficulties that have existed in using them for prediction. Phil Brierly's Tiberius software includes an implementation of neural networks that has, at least for now, converted me. As a consequence, I'm adding one final margin predictor to the mix for 2011.
Saturday, March 5, 2011 at 10:43PM We've fresh tipsters for 2011, fresh Funds for 2011, so now we need fresh margin predictors for 2011. This year, all of the margin predictors are based on models that produce probability forecasts, which includes the algorithms powering ProPred, WinPred and the Head-to-Head Fund and the "model" that is the TAB Sportsbet bookmaker. The process for creating the margin predictors was to let Eureqa loose on the historical data for seasons 2007 to 2010 to produce equations that fitted previous home team margins of victory as a function of these models' probabilities.
Friday, February 25, 2011 at 8:52PM In the previous blog we considered the logarithmic probability score on ProPred, WinPred and the TAB bookie and found that the TAB bookie was the best calibrated of the three and that relative tipping performance was somewhat unrelated to relative probability scores. For the Head-to-Head Fund, whose job in life is to make money, the key question is to what extent do its probability scores relative to the TAB bookie's shed light on its money-making prowess.
Friday, February 25, 2011 at 5:52PM Almost 12 months ago, in this blog, I introduced the topic of probability scoring as a basis on which to assess the forecasting performance of a probabilistic tipster. Unfortunately, I used it for the remainder of last season as a means of assessing the ill-fated HELP algorithm, which didn't so much need a probability score to measure its awfullness as it did a stenchometer. As a consequence I think I'd mentally tainted the measure, but it deserves another run with another algorithm.
