Profitability for Investors this weekend could conceivably depend on my rounding policy.
All of the Margin Predictors, Combo_NN2 included, predict real number victory margins. (Though, god knows, their predictions do sometimes seem to have an imaginary part too. Pure maths humour: job done.) This is largely an irrelevance until you have to choose a bucket for a SuperMargin wager and you find that your Predictor has tipped:
- the Dogs to win by 9.7 points, which rounds to 10 points and therefore lands us in the 10 to 19 points bucket - which is a range that doesn't include the actual prediction
- the Pies to win by 29.6 points, which rounds to 30 points and, in a similar fashion, lands us in a bucket that doesn't actually contain the prediction, in this case the 30 to 39 points bucket
Here the rounding issue is a minor mathematical curiosity, though it is one with financial implications, but it illustrates for me the broader point that there's so much more to the practice of statistical modelling in a real world context than just getting the data, fitting the model and taking the model-indicated action. I do wonder how many of these small, practical, need-to-make-it-work-in-the-real-world decisions are hidden in the forecasts and actions of businesses, implementers of policies, and others who rely on statistical models as inputs to what they do.
Anyway, these two troublesome SuperMargin wagers have been joined by two more of the same ilk as well as a single Line bet and a pair of Head-to-Head wagers.
The week's most lucrative result would be a Collingwood victory by 20 to 29 points, which would secure the Line and the SuperMargin wagers and nudge Overall Portfolios up by over 5c.
This same contest also harbours the greatest risk for Investors should Collingwood contrive to lose, to draw or to win by 15 points or fewer. Such a result would torpedo both wagers and chisel 3c off the Overall Portfolio price. No other game sees more than 1.4c of Investor money at stake, yet all but the Fremantle v Essendon game carry an upside of at least 2c, though usually this is tied to a fairly unlikely outcome.
Amongst the Head-to-Head Tipsters, three games are hotly disputed. Collingwood, Essendon and the Roos are all favoured only by the barest possible majority of Tipsters in their respective games, in the Roos' case despite being over $3 underdogs on the TAB.
It's perhaps worth noting that our highest-ranked Head-to-Head Tipsters, ProPred and WinPred, have opted for the favourites in every game except the Fremantle v Essendon game where they've selected Fremantle, who were narrow favourites when the TAB market opened on Monday, but narrow underdogs by Wednesday noon.
The Fremantle v Essendon game is also a source of contention amongst the Margin Predictors, with nine of them predicting victories by Fremantle of between 2 and 12 points, and the other four predicting victories by Essendon of between 1 and 10 points.
In the only other contests where there's less than unanimity there's only a single dissenting voice. In the Sydney v Geelong game, Combo_NN1 is that dissenter in predicting a Geelong win by 8 points, and in the Roos v Crows game it's H2H_Unadj_10 in tipping a 1-point Roos win.
The TAB Bookmaker aside, the Head-to-Head Probability Predictors are all relatively confident about Fremantle's chances of defeating the Dons on Saturday at Subiaco. As a group they also think that the Roos are better prospects than the TAB does - though none go as far as to rate them outright favourites - and they mostly rate the Swans, Pies and Dees as shorter favourites than does the TAB.
In the Dogs v Lions game, WinPred and H2H roughly agree with the TAB's assessment of the teams' relative chances, but ProPred thinks that a Lions upset is more likely - though still less likely than not - than does any other Predictor.
The Line Fund algorithm rates four of the games as being very close to parity after the handicap has been applied, and is only barely confident enough in the Pies' chances to warrant a wager. It also feels that the Lions and GWS have been given too much start in their respective games and rates them about 65% chances to win on line betting.