2014 - Round 24 Results : (Almost) All As Predicted

Despite a restricted schedule of just four games and tightened bet size restrictions, the three MoS Funds combined this weekend to produce the 8th-best return for a single round - the 3rd-best since Round 14 - and only the second 100% ROI of the season, the other coming way back in Round 1 when the Head-to-Head Fund snared the upset Giants' win over the Swans, and the Margin Fund bagged two collects.

Both of these Funds also had profitable weekends this time around, the Head-to-Head Fund landing its sole wager on the Roos and the Margin Fund, due entirely to Combo_NN_2's prescience, picking the correct bucket in each of the first three games of the round. The Line Fund also did well, collecting on all three of its wagers.

Combined, the Funds bumped the Portfolio value up by another 3.9% to leave it up on the season by almost 16%.

Combo_NN_2's stunning SuperMargin performance this weekend caps a five week period during which its portion of the Margin Fund has grown by almost 30% of the initial value. That's left Combo_NN_2 down by only 17% on the season, far outstripping the 42% loss from the other part of the Fund, controlled by Bookie_9's opinions. It's ironic that the losses from the portion of the Fund that was meant to provide some "insurance" exceeds that of the more speculative component.


Four wins for the four favourites and a unanimous set of Head-to-Head Tipsters made for stasis as far as this part of the MoS Tipster Leaderboard was concerned this week. (Note that the Heuristic Tipsters are no longer passing opinion on the week's activities. Many of them are still recovering from their own Mad Mondays. You really don't want to go out for a night's drinking with mates named Short Term Memory, Easily Impressed and Ride Your Luck. Home Sweet Home didn't attend, I'm led to believe.)

Margins were relatively easy to predict accurately this week too, the all-Predictor Mean Absolute Prediction Error (MAPE) finishing at 14.7 points per game. Win_7's 10.1 MAPE was the round's best, ahead of Combo_NN_2's 10.6 MAPE and Win_3's 10.8 MAPE. The three top Predictors produced the round's three worst MAPEs but their shared underperformance and the leads they had established over the remaining Predictors meant that they remained in the same order and still in the top three positions.

Win_3's performance, however, was enough to bring about a change in the orderings, slipping it into fourth place ahead of RSMP_Simple, which leaves it now ranked six places higher than its 2013 finish.

Here's the top 10 from the end of 2013:

  1. RSMP_Weighted
  2. RSMP_Simple
  3. Bookie_LPSO
  4. Combo_7
  5. Bookie_9
  6. Bookie_3
  7. Combo_NN_2
  8. Combo_NN_1
  9. Win_7
  10. Win_3

Bookie_LPSO's, Combo_7's and Bookie_9's back-to-back top five placings are impressive. Equally, Combo_NN_1's and Combo_NN_2's declines are surprising.

Still, Combo_NN_2, as mentioned, did snag three SuperMargin wagers this weekend, while the four H2H-based Predictors snared two each themselves. A few of the others landed one correct prediction leaving us with a situation where eight Predictors have season-long profitable SuperMargin records when they've predicted home team wins or draws.

 (If you're wondering why the Combo_NN_2 result shown here is for just a 7% loss and not the 17% I mentioned above it's because this table assumes level-stake wagering on each Predictor's opinions throughout the season whereas the Margin Fund imposes a variable bet size by round.)

The profile of a profitable Margin Predictor in the SuperMargin market this year has been one that's wagered in about 55% of games and selected the correct bucket at least 15 times. Combo_NN_2 has now selected the correct bucket 13 times, though from slightly fewer attempts that the profitable archetype. Bookie_9 has selected the correct bucket only 10 times and from slightly more attempts.

In the Head-to-Head Probability Predictor section of the Leaderboard you'll notice that there's been a slight downgrading of C_Prob. In my investigations to understand how it could have produced a Bookie-like log probability score (LPS) and yet lost money so relentlessly in the Head-to-Head wagering market, I discovered that C_Prob, in fact, didn't have a Bookie-like LPS. The cause? An Excel formula error.

Given the choice between finding that C_Prob's LPS was right and my Head-to-Head wagering calculations were wrong, and the opposite, I had hoped for Option A but wound up with Option B. At least my previous findings on the relationship between LPS and likely in-market profitability has been reaffirmed.

This week, the round's best LPS belonged to H2H, though C_Prob, now correctly measured, did manage a Score better than any of the Bookie-based Predictors. C_Prob remains in 4th though.

The Line Fund algorithm recorded a positive LPS for this round, its third in four weeks with the only non-positive Score effectively being a zero LPS in Round 23.