Every so often things crop up that unexpectedly test the statistical models I've constructed. For example a client will ask for a model to be used for a practical purpose such as forecasting, or they'll ask me to rerun a set of scenarios that I've created using a model, but this time with slightly different inputs.
What makes these situations a test for the models - and, I'll be honest, frequently a 'hold-your-breath-as-you-run-the-numbers' moment for me - is when there's an obvious way to validate the results using pure commonsense. Forecasts should look like a plausible extention of the actual time series that you modelled, and scenario outputs based on different inputs should usually differ from the earlier outputs you produced in logical ways.
Last week's drawn Grand Final, which meant that the MAFL Models faced the unprecedented situation of having to forecast the result for exactly the same matchup two weeks running, provided just such a test.
I'm pleased and not just a little relieved to report that, to my mind, the MAFL models have 'passed' by behaving sensibly - although in the case of the New Heritage Fund, that's a relative rather than an absolute assessment.
Both the New Heritage Fund and Prudence Funds, chastened a tad and accordingly recalibrated by the Pies' inability to win last week, have stuck with the Pies, now priced at $1.55, but have made smaller wagers. As well, ELO, which is based on MARS Ratings, now forecasts a slightly smaller win for the Pies, but a win still large enough to encourage ELO-Line to wager its customary 5% on the Pies giving 11.5 start.
(I should note that a minor glitch had me reporting ELO as tipping the Pies to win by 51 points last week. It should have been by only 40 points. This week the forecast margin is 38 points.)
Chi, HAMP and LAMP have also made only small but nonetheless downward revisions to their predictions of the Pies' victory margin. Some validation for their 2-4 point adjustments is provided by the corresponding change in the bookie's margin - by 5 points, from 16.5 to 11.5 points.
Here's the detail:
In short, everything's pretty much as it was last weekend, though we stand to collect more this weekend if the Pies win than we would have had they won last weekend, and we also stand to lose less this weekend if they don't.
For the sake of completeness I should let you know that, this week, line bets are inclusive of extra time. So, in the unlikely event that the Pies and Saints again finish level at the end of normal time, our line bet will not be lost at that point. If the Pies then go on to kick two goals clear of the Saints by the time the hilarity finishes ensuing, we'd collect - on all the bets as it happens.
Based on the Margin Tippers margin predictions, HAMP will finish the season with a sub-30 MAPE provided the final margin lies between a Saints win by 6 and a Pies win by 36 points, and LAMP will do the same if the Saints win by 33 or fewer, or if the Pies win by 59 or fewer. BKB will finish with a MAPE below 29.5 if the Saints win by 19 or fewer, or if the Pies win by 42 or fewer.
Lastly, here's the week's full Ready Reckoner:
And that definitely is the last pre-round MAFL blog for 2010.
No comments:
Post a Comment