Matter of Stats

View Original

2013 : Round 5 - Results

Until late into time-on in the Lions v Dees game on Sunday afternoon I thought I'd be writing this week about our making a small profit, having extended our run of successful wagering to four weeks in a row, but then the Dees kicked successive goals at the back end of what's appropriately known as "junk time", the first turning a 1.4c profit into a breakeven result for the game, and the second consigning us to a 0.9c loss.

Then in the last game of the round, the Hawks' inability to win by either of the requisite margin ranges made sure that all three Funds recorded a loss for the week. The Head-to-Head Fund dropped 0.4c, the Line Fund 1.5c, and the Margin Fund 0.3c, which translates into a 1.1c loss for the Recommended Portfolio.

(You can see the details of each Fund's wagering performances so far this season in the navigation menu space at the left of the screen, which this week includes a mini-graphic to show weekly Fund returns.)

TIPSTERS AND PREDICTORS

The season continues to be a remarkably predictable one, with seven more favourites collecting the chocolates this week. Favourites have so far this season won about 69% of contests.

On Head-to-Head tipping, however, a slew of MAFL Tipsters have found even more of the results to be predictable, especially Combo_NN2, which remains 3 tips ahead of Bookie Knows Best and 1 tip ahead of all other Tipsters. Bookie_9 was the week's best-performed Head-to-Head Tipster, registering 8 correct tips from 9, though its commendable accuracy in both Head-to-Head and Margin prediction unfortunately did not translate into success for the Margin Fund.

The week's best Margin prediction performance came from H2H_Unadj_3, which turned in a remarkably low 17.35 mean absolute prediction error (MAPE) for the round. This was insufficient though to drag it any further up the Leaderboard than fifth-last where it currently sits with a season MAPE of about 29.7 points per game.

Other strong performances were attributable to the remaining H2H-based Predictors, ProPred_7, the two WinPred-based Predictors, Bookie_9 and RSMP_Simple, which all had sub-20 MAPEs for the round. RSMP_Simple currently lead all-Predictors based on season-long performance.

The portion of the table relating to the Margin Predictors has two new columns this week, the first headed with a golfing flag symbol that records how often the Predictor has been "nearest the pin" (ie has predicted the margin closest to the actual result amongst all the Margin Predictors) and the second - in a tortured attempt to extend the metaphor - headed by a bunker that records how often the Predictor has been furthest away from the actual game margin.

It's interesting to note how weakly a Predictor's season-long MAPE is related to its nearest and furthest from the pin performances. RSMP_Simple, which has the lowest MAPE of all Predictors has never been nearest to or furthest from the pin, while Combo_NN_2, which is 3rd based on MAPE, has been nearest for 8 games and furthest away for 6. Win_7 has a record broadly comparable margin proximity record to Combo_NN_2's, finishing nearest 8 times and furthest away 8 times more, but has the second-worst MAPE of all Predictors.

I mentioned earlier that Bookie_9's strong performances in head-to-head and margin prediction hadn't translated in successful Super Margin wagering. This was due partly to the fact that it was in error by just a single bucket in three contests, but also because the one game in which it selected the correct bucket - the Port v Eagles game - was one in which Combo_NN_2 was tipping an away team win, thus preventing Investors from benefitting from Bookie_9's brilliance.

Across the whole season though, Bookie_9's SuperMargin record is still poor in the context of all other Margin Predictors, as the following table shows.

Returning to a review of the MAFL Leaderboard, on line betting the two RSMP-based Predictors currently have the best positive records.

Amongst the Head-to-Head Probability Predictors it's still the Bookmaker-related Predictors filling the first three places, with the Predictor based on a Risk-Equalising interpretation of the overround implicit in the TAB Bookmaker's prices doing best.

The Line Fund algorithm has a relatively poor round and has slipped back into having a negative probability score for the season to date.