2015 - Round 24 Results: Two Home Teams and Just One Favourite
/This week's wagering was the antithesis of last week's - though, regrettably, entirely in keeping with the majority of the season - with all four bets providing no return, leaving the Head-to-Head Fund down by 1.5c, the Line Fund down by 1.3c, and the Overall Portfolio down by 1.4c. Across the season then the Portfolio is now down by 34.1c.
(It's not lost on me, by the way, that the Head-to-Head and Line Funds managed to wager this week on the only Home teams that failed to collect in the head-to-head and line markets.)
TIPS AND PREDICTIONS
As if the weekend's wagering wasn't bad enough, on Friday evening just prior to game time while idly reviewing the round's predictions (and having already had to adjust them earlier in the week for the two Elimination Finals due to incorrectly inputting the venue information) I noticed that my input data in the other games had the Eagles' and Dockers' playing at Docklands rather than Subiaco. I then spent a frantic hour before game time assessing how much of a difference this error had made.
The most dramatic impact was on C_Marg and C_Prob, the correction being enough to have them favouring Eagles and Dockers wins, though there were also some effects on other algorithms.
Amongst the Head-to-Head Tipsters it saw C_Marg, as I just mentioned, swapping over to West Coast and Fremantle, but it also had the oft-eccentric Combo_NN2 swapping off West Coast and on to Fremantle. One of the joys and mysteries of non-linear models is that they sometimes react to data changes in the most counter-intuitive ways.
The changes for the Margin and Probability Predictors I thought were interesting enough to share with you and appear in the table at right.
Some of the Margin Predictors changed not at all or only a little because they either largely or completely ignore venue information other than that provided by the fact that a particular team is the first-named team in a contest. Others, however, did react to varying degrees.
C_Marg added 13 points to the Eagles, since it swapped the 6.5 points Interstate Advantage from the Hawks to the Eagles, and it added 6.5 points to the Dockers, since it previously had no Interstate Advantage for either team and now granted this filip to the Dockers. (Note that ChiPS does not recognise an HGA in a Final unless the Home team is also the higher-Rated team. This was not true for the Eagles or the Dockers.)
Generally smaller changes were made by other of the Margin Predictors, though Combo_NN1, like Combo_NN2, did make a contrarian adjustment to the Eagles' expected margin, reducing it by 13 points. It did also, however, like Combo_NN1, make a small, positive adjustment to Fremantle's expected margin.
Amongst the Head-to-Head Probability Predictors it was C_Prob that reacted most, adding 14% points to the Eagles' chances and 8% points to the dockers'. The Line Fund algorithm also significantly increased the Eagles' chances in the line market, though it still had them as narrow underdogs. It also increased the Dockers' chances though only very slightly (though enough that, had I used the correct inputs before locking in the weeks' wagers, would have seen us with a Dockers line bet. Small mercies, I guess.)
Anyway, please note that the results shown below are based on the corrected predictions, not those from Friday's blog post.
Most of the Head-to-Head Tipsters landed only the weekend's sole victory by a favourite, superior tipping only emerging for the (corrected) C_Marg predictions, as well as for Combo_NN1 and Combo_NN2. C_Marg's 3 from 4 allowed it to extend its lead at the top of the Leaderboard, which now stands at four tips.
The all-Predictor average Mean Absolute Error (MAE) amongst the Margin Predictors came in at 19.3 points per game per Predictor this week, the second-lowest average for the season, behind only the 16.6 points per game per Predictor in Round 13.
Best was C_Marg's 16.4 points per game, ahead of Combo_NN1's 17.6 points per game and RSMP_Weighted's 17.7 points per game, though no Predictor's performance was sufficiently good or bad that it led to a re-ordering.
C_Marg remains the standout in terms of line market wagering performance it now having selected the right team in 64% of contests. Combo_NN2 and Bookie_3 have less-impressive but also profitable season-long records, as does H2H_Adj_7, but only barely.
Head-to-Head Probability Prediction proved difficult this week, with none of the Predictors registering a positive probability score. Bookie-OE had the round's best score, just slightly ahead of Bookie-RE and then C_Prob. C_Prob though still heads the Leaderboard.
The Line Fund algorithm had a poor week of probability scoring, its result the worst since R21.
