2014 - Round 5 Results: Regressing to the Wrong Mean

/

For the second week in a row all three of the MatterOfStats Funds lost money, the Margin Fund most determinedly by shedding 5c, but the Head-to-Head and Line Funds with only slightly lesser and approximately equal alacrity, dropping 1.4c and 1.5c respectively. Combined the three Funds brought about a loss of just over 2c for the Recommended Portfolio, leaving it now down by a little less than 3c on the season.

Next week, the Line Fund, on the understanding that its calibration for the current season will have improved substantially, doubles its standard wager. Here's hoping that its 10 and 11 record in the first five rounds has been to deliberately set the bar low.

TIPS AND PREDICTIONS

Six from nine was the best record that any of the Head-to-Head Tipsters could muster this week, this score being recorded only by the three WinPred-related Tipsters. Next best was five from nine, achieved by seven Tipsters, among them BKB, Bookie_3 and Combo_NN_1, which in recording this score elevated themselves into equal shares of second place. They sit, with Combo_7, one tip behind the current leader, Bookie_9, whose record stands at 31 from 45 (69%).

Nine Tipsters recorded the round's low score of three, including Ride Your Luck, which now sits in equal-last place with Home Sweet Home, seven tips from the lead on 24 from 45 (53%). The average score for the round across all Head-to-Head Tipsters this week was 4.125.

Amongst the Margin Predictors, low score was recorded by Win_3, its 18.77 Mean Absolute Prediction Error (MAPE) for the round sufficient to elevate it into 4th place on the Leaderboard with a season-long MAPE of 35.24 points per game. High score came from a most unlikely source, Bookie_3, who recorded a MAPE of 29.35 points per game, curiously its second-best MAPE result this year. Combo_NN_1 now leads all-comers with a season MAPE of 34.95 points per game having been 'nearest the pin' eight times this season and furthest away from it nine times. Consistency, we see again, isn't always required for success. 

In the context of the last few seasons, the mid- to high-30 point MAPEs we're currently observing for the Margin Predictors are quite high. Even those Predictors who've been most flamboyantly wrong most infrequently have still been in error by 7 goals or more in about one-third of games.

Bookie - OE, the Head-to-Head Probability Predictor whose estimates are formed by assuming Overround Equalising in the TAB Bookmaker prices, currently leads all Predictors and now enjoys a reasonably sizeable lead over Bookie - RE in 2nd place. C-Prob fell back a little this week but is still within sight of the leaders sitting in 4th place.

The Line Fund algorithm had its second-worst round of the season in terms of probability scoring and now has a negative probability score for the season taken as a whole.

SUPERMARGIN PERFORMANCES OF THE MARGIN PREDICTORS

Win_3 was the only Margin Predictor to select the correct bucket in more than one game this week, but even its two from nine performance was not enough to drag it into profitability on SuperMargin wagering for the season in full. In fact, no Margin Predictor has made SuperMargin recommendations that, if followed in every game, would have been profitable, and only one Predictor, Combo_NN_1, has made recommendations that, if followed when they suggest a Home team win or draw, would have been profitable.