TipBetPro AFL 2017 Launch!
The AFL 2017 season is only 4 weeks away! Behind the scenes the TipBetPro crew have been working hard refining our model for the 2017 season. In the interests of being transparent with our loyal followers – below we’ve given a brief run-down of the TipBetPro approach and changes to our model from the 2016 season.
The TipBetPro Approach
The TipBetPro Model incorporates numerous team based and player based factors into our algorithm for forecasting AFL matches. The key factors are as follows:
AFL Player Ranking (long term performance – 2 year time horizon)
TBP injury adjustments are applied to the AFL Player Rankings as players who have not played at least 40 games over a two season time horizon are unfairly punished by the AFL Player Ranking system in our opinion
Short term performance (measured by weekly Champion Data ranking points)
Our record to date
*Excludes Money-back specials
Changes from 2016
The 2016 season whilst not horrible was clearly the worst performing season for the TipBetPro AFL model as the success of defensive powerhouses Sydney and the Western Bulldogs highlighted a significant trend away from factors traditionally associated with winning teams (attacking, experienced sides).
We don’t over-react to one season anomalies, however we also don’t ignore trends so we’ve back-tested the last three seasons data to make some small tweaks to our model to reflect the way the game is changing.
A summary of these changes are as follows:
Reduction in player experience weighting. Whilst the more experienced teams still tend to be stronger, the really talented youngsters are excelling consistently earlier in their career now. The young Western Bulldogs proved in 2016 that experience is no longer a pre-requisite to win premierships
Higher weighting to short term performance vs long term performance. The use of the AFL Player Rankings as an input to our own TBP Player Rankings has been most successful over the years but there is a lag on recognising short term out-performance. Whilst that is not such a bad thing (a consistent performer over two seasons tends to out-perform a player who has had two good weeks more often than not!) – testing suggests the lag is just a little too high and a higher weighting to short term performance is required
A higher weighting to team defensive statistics. Individual player statistics are terrific for recognising attacking players involved in key scoring chains but not so great at recognising the more defensive players or teams with great defensive systems (i.e. superior coaching). So we do this on a team level based not just on “points against” but also on the capability of teams to restrict the influence of opposition key play-makers. As coaching and strong defensive structures takes more and more importance so too does the weighting TBP apply in our model.
Probabilities and Minimum Confidence Levels
Regular followers of TBP AFL would know that we apply a minimum 35% win probability and a minimum expected profit of 5% (10% early in the season) before recommending a head to head bet. So, for instance:
The reason for this is that the majority of our historical tips and back-tested data are in the 35% to 65% probability range and so we have confidence that our probabilities for this range is accurate – i.e. long term, approximately 40% of our 40% win probability tips will come in. Outside of this range, we simply haven’t had sufficient data to have confidence in our probability ranges. Empirical evidence has suggested that we tend to under-estimate high end probabilities and over-estimate low end probabilities which is why we still make bet recommendations on >65% win probability recommendations but not on <35% win probabilities.
As our AFL model is our longest running model and data continues to be collected, we now have sufficient confidence to reduce our minimum bet probability to 30%. So what that means is that you can expect a few more high odds, low win probability bet recommendations in the 2017 season. Certainly this is a higher risk/higher value approach (as by definition we only expect 30% of these bets to come in) so for those more risk adverse punters, we will always publish our expected win probability and you are always welcome to apply your own minimum probability filters.
There have been other minor changes to the TBP model for the 2017 season. A short summary of these changes are provided below:
Greater confidence required for line bet recommendations early season when data is scarce
Greater confidence required for line bet recommendations on higher line bets when results tend to be more volatile
A more exponential shape to our expected probabilities to reduce under-estimating bias on high probability tips
Any questions are welcomed - I hope you are looking forward to the 2017 AFL season as much as we are!