As I noted in the previous blog, when I rebuilt the heuristics to cater for the bye, I made a few other what I thought were subtle changes to some of the building block rules on which they're built. My assumption was that these changes would have had minimal effect on the tipping accuracy of each heuristic and, to be honest, I didn't even consider whether it would effect their wagering performance.
So, let me address both those aspects here.
Firstly, tipping accuracy. Here's a table showing each heuristic's tipping performance for each of the last 11 seasons.
Relative to their performance under the previous rule definitions, the heuristics have seen the following changes:
- The rebuild had no effect on BKB and CTL - a favourite's a favourite, and a higher ladder position is a higher ladder position - so their performance is unchanged and remains impressive. Across the 11 seasons, tipping the favourite yielded a 67% accuracy rate, and tipping the team higher on the ladder yielded a 64% accuracy rate, making these the number 1 and number 2 heuristics in our family.
- Ride Your Luck's performance was also barely altered. It's preferred rule is to tip the team with which it's had the longest string of tipping successes, and the only change to that rule is that draws now break the streak. Overall it's tipped at just over 63% across the 11 seasons and it's also narrowly out-tipped CTL over the five most-recent seasons.
- Silhouette's season-to-season performance has varied quite a bit, altering by up to 4 correct tips in either direction within a season, but these variations have tended to cancel one another out, leaving it at just over 62% and fourth amongst our heuristics.
- Shadow is next best, also a tick above 62% across the 11 seasons. It's benefited enormously from the rule redefinitions, leaping by as many as five tips in some seasons while falling by only one or two tips in the worst seasons.
- Follow The Streak is unchanged and on 62%, in sixth.
- Short Term Memory II is next best at just under 62%, and it has, like Shadow, benefited considerably from the rule redefinitions.
- Short Term Memory I come next at a tick under 61%, another beneficiary of the rule redefinitions, though only a minor one.
- Easily Impressed II is in 9th with 60.5% and suffered a small decline in performance as a result of the rule redefinitions.
- Home Sweet Home, despite benefiting most of all from the rule redefinitions, which in its case had it tipping the AFL Designated Home Team, still languishes in 10th on just over 58%. But for the 2002 to 2005 seasons, its performance would have been worse still.
- Easily Impressed I ranks last on tipping accuracy, also at just over 58%. Rule redefinitions did little to change its results.
On balance then, the rule redefinitions have tended to be beneficial to the heuristics' accuracy. Nine of the 11 heuristics have tipped at over 60% across the 11 seasons, and these same nine have tipped at 61% or over across the last five seasons. Nice.
Whither the wagering?
The old versions of the heuristics generated profit from level-stake wagering their home team tips commencing in Round 6. Here's how the new versions perform under that strategy and under a few more:
The first row shows how unprofitable it would have been to level-stake bet any heuristic's tips across seasons 2006 to 2010. Skipping the first five rounds of the competition wouldn't have helped much, unless you were following Home Sweet Home, in which case you were automatically overlaying a 'bet only on home teams' strategy.
Row the third demonstrates the general efficacy of following an heuristic's prognostications only if they suggest a home team victory, and row the fourth confirms that commencing such a strategy in Round 6 would be even more efficacious. BKB's unprofitability under such a strategy tells us that the profit is coming from home team underdogs and not home team favourites, which is a phenomenon that MAFL Investors have known about for a few years.
Lest you think that the profitability opportunity from this general strategy disappeared a season or two ago, I've included one more row in the table, which shows the return for each heuristic for season 2010 alone. CTL, Silhouette and Shadow dip into red ink, but the remaining heuristics, BKB aside, are all in black.