Matter of Stats

View Original

2014 - Round 19 : Ratcheting Down

Even though the Line Fund has found six games worthy of an investment this week, the Head-to-Head Fund two games and the Margin Fund six games, Investors have only about 14% of the Recommended Portfolio at risk, down by more than one-third compared to the imperilment they faced in Round 18 and by more than 40% on that which they endured in the round before. This week's activity takes us past the point where every dollar in the Recommended Portfolio has been wagered twice over.

Five of the six Line bets are on teams giving start - not the subtype of Line wagers in which the Line Fund has tended to excel so far this season - and only one, on the Roos getting 8.5 points start, is of a sort that history would suggest should fill Investors with hope.

Both of the week's Head-to-Head wagers are on mild underdogs and both are of a size large enough to cause non-trivial pain if they're unsuccessful. That said, the success of either would be sufficient to secure a net profit for the Head-to-Head Fund for the round.

In three of the games in which we've SuperMargin wagers, Combo_NN2 and Bookie_9 differ in their opinions to such an extent that we've a Chasm of Despair, and in one of those games we've an Extended Chasm with our preferred ranges separated by over 20 points.

The Roos, on whom we've a Head-to-Head and a Line wager, are comfortably the team with the greatest upside for Investors, a win by them capable of lifting the Recommended Portfolio by over 4c. Six other teams - Fremantle, Sydney, Adelaide, Gold Coast, Hawthorn and Collingwood - are each capable of adding about half that amount to the price of the Portfolio. 

The Roos are also capable of inflicting the largest loss on Investors. This loss, of 3.5c, would be realised if the Roos were beaten by more than 8 points. The six other teams mentioned in the previous paragraph represent the second tier of potential losses, each capable of shaving 1.7c off the price of the Recommended Portfolio.

A best-possible week would add over 18c to the Portfolio price while a worst-possible week would knock just under 14c from it.

TIPS AND PREDICTIONS

In only two games do we find unanimity amongst the Head-to-Head Tipsters this week, that being the case for the Tigers, taking on the Giants, and the Hawks, taking on the Dogs. Four other teams (Fremantle, Sydney, Adelaide and the Gold Coast) enjoy the backing of at least 23 of the 28 Tipsters.

The highest levels of contention are reserved for the remaining three contests, in which there's 16-12 majority support for the Cats over the Roos, 17-11 support for the Dees over the Lions, and 19-9 support for Port Adelaide over Collingwood.

Short Term Memory I, Easily Impressed I and Follow The Streak have the week's highest Disagreement Indexes, each different from the other Tipsters on average 47% of the time.

In only three games do the Margin Predictors offer forecasts with differing signs, eight tipping the upset win in the Roos v Cats game, two in the Dees v Lions game, and six in the Pies v Port Adelaide game.

The range of margin predictions spans less than five goals in every game except the first of the round where it spans just under seven goals. Combo_NN2 has the round's highest Mean Absolute Deviation (MAD) from the all-Predictor average, a consequence of being the Extreme Predictor in six contests. Combo_NN1 has the next-highest MAD, though it's the Extreme Predictor in only two contests. Combo_7 is the Predictor with predictions most similar to the all-Predictor average, its MAD coming in at just 2.4 points per game. It's not the Extreme Predictor in any contest.

There's some prospect for re-orderings at the top of the Margin Predictor Leaderboard, the evidence for which is summarised in the matrix that follows. In it, entries below the main diagonal record the gap, in prediction error terms, between the Predictor listed at the head of the row and that listed at the head of the column. So, for example, the gap between Win_3 and Combo_7 (C7) is 36 points. The rows have been ordered so that the Predictors with the smaller season-long Mean Absolute Prediction Error are listed higher, so the gaps shown are always how far behind is the Predictor listed in the row from the Predicted listed in the column (so, for example, Win_3 trails Combo_7). 

We can think of these numbers as the amount by which one Predictor needs to predict margins closer to the actual results compared to another Predictor over the course of the remainder of the season in order to catch that Predictor.

Entries above the diagonal record the pairwise aggregate absolute differences solely for the predictions in Round 19. The distance between Combo_7 and C_Marg (C_M) is, for example, 54 points. We can think of these values as the maximum closure of the gap between two Predictors that could be achieved solely on the basis of outcomes this week.

If, for example, C_Marg's prediction was between the actual final margin and Combo_7's prediction in every game, it would reduce the gap between itself and Combo_7 by 54 points. That would still see it trail Combo_7 by 312 (ie 366-54) points.

The possibility of a change in the ordering of Predictors is signified by the number above the diagonal being larger than the equivalent number from below the diagonal. Win_3 could therefore overtake Combo_7 this week since the season-long gap between the two Predictors is 36 points but the distances between their predictions for this round sum to 61 points. Win_7, it turns out, could also overtake Combo_7 since the season gap is 53 points but the aggregate distance is 61 points. The only Predictor that C_Marg could overtake on the basis of this week's predictions alone is Combo_NN2, which C_Marg currently trails by just 28 points and compared to which its predictions this week differ by 115 points.

PROBABILITY PREDICTIONS

In three contests the Probability Predictors find themselves on either side of a 50% assessment for the home team. In the Roos v Cats contest they're split 4-4 while in the Dees v Lions game C_Prob is the lone dissenter and in the Pies v Port Adelaide game opinion is split 5-3 in favour of the Power.

ProPred, though Extreme Predictor in only two games, has the week's largest MAD from the all-Predictor average. WinPred is Extreme Predictor in five contests but by only 2 or 3% in each case, so it has only the fourth-highest MAD for the week. C_Prob this week sits 5th of 8 amongst the Predictors in terms of MAD.

The Line Fund algorithm has, in general, made less sharp probability predictions this week than in recent weeks. Its most extreme assessment is just 67% for the Roos, followed by 64% for the Hawks, 60% for the Crows, 57% for Fremantle, 56% for Sydney, and 55% for the Lions.

ChiPS PREDICTIONS

The ChiPS System has selected to win the team with the higher Rating in every contest this week except in the Roos v Cats game where the Roos' inferior form and mildly negative HGA at Docklands has been enough to see its 1.4 point RP-based advantage become a 3.6 point deficit once these mitigating factors are accounted for.

At current TAB prices, no team is deemed worthy of a wager by ChiPS.