It's February, when a gambler's fancy turns to footy, so it's time I revealed my plans for Investors for 2011.
The short version - or the TL;DR version, if you like - is that there'll be 13 tipsters this season, the 11 heuristics from last year (revamped) plus two more, and that there'll only be two Funds, both of them new. The details follow.
From an earlier blog, you already know that the 11 heuristic tipsters will be returning this year, tailored and tweaked to handle the rigours of a seventeen-team competition. As always, they'll simply be tipping game winners, as they've neither interest in nor clue about probabilities or likely margins of victory.
Two other, far more sophisticated algorithms will be joining them this year: ProPred, the probability predicting algorithm that uses the Probability Prediction MARS outputs, and WinPred, that uses the Winner Prediction MARS outputs.
ProPred and WinPred are both forests of Conditional Inference Trees that use the most recent 365 days worth of data at the time they are run to predict the likelihood that a line wager on the home team will be profitable. As well as using their respective bespoke MARS rating inputs, they also use the following variables:
- A transformed version of the implied probability of victory for the home team based on the bookmaker's prices (the transformation is the logistic)
- The away team's venue experience, which is the number of times the away team has played at the venue in the preceding 12 months
- An interstate status flag which is +1 if the home team is playing in its home state and the away team is not, 0 if both teams are playing in their home state or both teams are playing in a state that isn't their home state, and -1 if, for some bizarre reason, the home team is playing in a state other than its home state and the away team is playing in its home state (Think Melbourne at 'home' to Brisbane at the Gabba in Round 15 of 2007)
- The for and against differential of the home team across the last 16 rounds (some of which might be byes)
- The for and against differential of the away team in the last 16 rounds (some of which might also be byes)
Both ProPred and WinPred, being forests of Conditional Inference Trees are stochastic, so they give slightly different predictions each time you run them. I'll be running each of them 100 times and then reporting their minimum, maximum and mean probability assessments. Their prediction will be based on the mean probability assessment from these 100 runs.
Last year, MAFL offered six Funds, all of which I've decided to retire.
- New Heritage finished the season at 60c having never ventured into profitable territory during the season. Indeed, its best returns were produced during the six rounds that it was precluded from wagering and, had it been revived for the upcoming season, it would, as I opined last year, have needed to be restrained from wagering on the bye. Easy decision: New Heritage, you're fired.
- Prudence rose by about 15c across the season, all of it (and a bit more) generated from Round 19 onwards and most of it reaped during the season's extended run of finals. Excluding the finals, it made only about 3% on the year. I considered giving it another season, but couldn't convince myself that its positive performance was other than due to chance.
- Hope also spent large parts of the season in the red - indeed it finished the home-and-away season down by about 2.5% - before rallying to a modest 2.5% profit by season's end. Too much effort and risk for too little return in my opinion. Gone.
- Shadow spent just one week in profit during 2010 and that profit was so small you needed three decimal places to notice it. After that brief flirtation with profitability it was loss, loss, loss for the Fund. In any case, Shadow, being an heuristic tipster, is recovering from off-season reconstructing surgery and is desperately short of a run. Anyway, it lost money last year. Next.
- Heuristic-Based, by any measure, did well last season, producing a return in excess of 21%. Under normal circumstances I'd have given it another run without question, but it too would now be based on revamped and untested algorithms, so I've decided not to risk it.
- Which leaves only ELO-Line, a Fund that also did well last season, finishing up almost 28c after climbing over $1.50 by the end of Round 19. The only reason it's not back this season is that I think I've created another line-betting algorithm that is at least its equal.
In place of these six Funds, this year there'll only be two, partly because I felt that it was too easy for me to hide behind the larger range of Funds as, almost inevitably, at least one or two of them would spin a profit, and partly because I hope it'll be simpler and more engaging for Investors if we're all of us in the same proverbial boat. Sinking or swimming.
With simplicity as the mantra, there'll also be no need for Investors to specify their own portfolio weightings this season - it'll be half of your investment in each of the two Funds.
So, let me introduce the two Funds.
(I've yet to name either of the Funds, so I'm sticking with the prosaic for now.)
This is the only Fund that will wager on head-to-head outcomes this season. It uses the same regressors as ProPred and also bases its predictions on results from the past calendar year. It's powered by yet another forest of Conditional Inference Trees, each making a prediction about the likelihood that a wager on the home team will be profitable. The head-to-head Fund wagers only on AFL Designated home teams.
The Head-to-Head algorithm outputs the minimum, maximum and mean probability of a home team victory across the 100 replications over which it is run. Its wagering will be Kelly based - though it'll be fractional Kelly for cardiac comfort - and will use the minimum probability from those 100 replicates. Wagering will commence in Round 1.
Fifty percent of Investors' Funds will be placed in this Fund this year.
This is the only Fund that will wager on line outcomes this season. It uses the same regressors as ProPred, excepting that it uses Line Prediction MARS in place of Probability Predicting MARS and it also uses the following variables:
- The start offered to the AFL Designated Home Team
- The venue experience of the home team (as well as of the away team)
It is powered by - yes, another forest of Conditional Inference Trees, using the past 365 days of results as input data. It also will wager only on AFL Designated home teams.
The Line algorithm outputs the minimum, maximum and mean probability of a home team line victory across the 250 replications over which it is run. Its wagering will be Kelly based and fractional, commencing in Round 1, but it, unlike the Head-to-Head Fund, will use the mean and not the minimum probability from those 250 replicates.
This Fund will carry the remaining 50% of Investor Funds this season.
Had the Head-to-Head and Line Funds been operating during the past two seasons, they would have turned in the following performances:
You can partly see from these numbers how much quieter and more difficult 2009 would have been in comparision to 2010. Both the Head-to-Head and Line Funds would have placed, on average, significantly smaller bets, won a smaller proportion of them, and registered lower ROIs.
What you can't see from the table is the late season surge in 2009 that allowed both Funds to record a profit. At the end of Round 12, the Head-to-Head Fund was down by about 15% and the Line Fund was down by about 10%.
We should expect then some variability in performance, but quite a bit of action and, I hope, a smattering of profit.