Heuristics + Patience + Home Teams = Profit

At the start of the season I wrote about the impressive aggregate performance of the (revamped, bye-accommodating) MAFL Heuristics over the period 2006 to 2010. This year the MAFL Head-to-Head Fund has recorded a significant loss - would any of the Heuristics have done better?

Yes, they would. In fact, depending on the restrictions placed on their wagering freedom, somewhere between a few and most of them would have spun a profit.

The table at left records the details. The first column provides the ROI that would have ensued from slavishly following the tips of an Heuristic, wagering from the first game of the first round and regardless of whether the Heuristic's prediction was a Home or an Away win.

In the next column we impose the constraint that no wagers are placed in Rounds 1 to 5, the logic being that no-one, not even a rule-of-thumb, knows much about the relative merits of the teams in the first five rounds of the season. The financial benefits of imposing this discipline is something I've chronicled on MAFL many times in the past - if only I'd applied it to the MAFL Funds this year ...

For the third column we revert to commencing wagering from game 1 but we now constrain ourselves to wager only on Home teams. This too is a restriction we've addressed many times on MAFL and is motivated by the observation that it's home teams that are more often mispriced.

In the final column, both constraints are imposed: delayed gratification and home team allegiance.

Since I was going to the trouble of performing this analysis for the MAFL Heuristics, I thought I might as well extend it to include the MAFL Head-to-Head Probability and Margin Predictors. To do this, I needed to convert probabilities and margin predictions to head-to-head tips, which I did in the obvious way by deeming a Probability Predictor to have tipped the Home team if it has assigned a probability of greater than 50% to that Home team, and by deeming a Margin Predictor to have tipped the Home team if it has predicted a margin for that Home team that is greater than zero.

In the table, which has been sorted based on the final column, the results for all the Heuristics are shaded gray. Completely unconstrained, only Ride Your Luck, Follow The Streak and Consult The Ladder eke out a profit. Waiting until Round 6 before commencing wagering drags only one more Heuristic into the black, Bookie Knows Best, thereby confirming that not even the bookmakers are good judges or form - and hence favouritism - in the early rounds.

Wagering only on Home Teams but starting in Round 1 pushes Silhouette, Shadow, Short-Term Memory I and Short-Term Memory II into profitability, but plunges Bookie Knows Best back into the red. Finally, imposing both constraints gets Bookie Knows Best back into profitability, leaving only Home Sweet Home and Easily Impressed I and II unable to make a profit.

In summary for the Heuristics then, this season has so far continued the long-term efficacy of level-stake wagering on their predictions.

The Probability Predictors, which are H2H_Un, ProPred and WinPred in the table above, generally perform unspectacularly, though H2H_Un and ProPred both generate modest positive ROIs when both constraints are imposed, and WinPred makes the narrowest of losses.

Amongst the Margin Predictors, those based on neural networks (CNN1 and CNN2) and those based on the TAB Sportsbet bookmakers' market prices (B3 and B9) all perform well, especially when both constraints are imposed. The other Margin Predictors that spin a profit when doubly constrained are those based on the Head-to-Head algorithm (HU3 and HU10, as well as HA3 and HA7), the other ensemble-based Predictor (C7), and the Predictors based on the ProPred algorithm (PP3 and PP7).

In fact, amongst the non-Heuristic based models, only the Predictors based on the WinPred algorithm find themselves unable to record a profit in the last column.

(It pains me a little to point out that the current ROI for the MAFL Head-to-Head Fund is -7.3%, a performance that's better only than those of Shadow, Easily Impressed I and Easily Impressed II under the constraint of Home team only wagering - the constraint which applies to the Head-to-Head Fund. In other words, the Head-to-Head Fund has outperformed only a simple heuristic based on the imagined behaviour of a non-existent dog, and two other heuristics that can only remember as far back as last week or the last time they tipped a team and it won.)