Sportpunter's AFL Report Card 2005

Welcome to the Sportpunter AFL Report Card 2005. Here we will closely analyse and discuss, sometimes in detail, how we went in the year 2005, in previous years and how we expect to go into the future. We will be looking at data extensively and we hope to have similar Report Cards for every sport that we offer our clients. Hopefully by looking deep into the data we can maintain our money better and make more from the predictions in the years to come.

The report card will be split up into the following contents:

1. Betting History of AFL 2005

2. Analysis of different probs/odds/overlays

3. Optimal Kelly Fractional.

4. 2005 Bank Analysis

5. 2006 and Beyond: Predicting the future.

1. Betting History of AFL 2005

As shown below we have the betting history for AFL 2005 and previous years.

Firstly looking at head to head betting, we made a profit of just over 15% ROI for the year, which is slightly above the average on previous years. Our profit was just under $3,000 which is largely comparable to previous years also. Line betting received a lower %ROI (10%) but a comparable, but slightly lower profit for the year. There were 8 more bets in line betting and a large amount of money placed per bet which is expected given the Kelly system. (e.g. not betting smaller amounts on larger underdogs). Margin betting provided a 20% ROI which was lower than in 2004. Likewise the profit was at $1.75k which is considerably lower than in 2004 but still very desirable.

The graph shown here shows our betting history throughout the year shows a big increase in banks from rounds 3 to 8. Following this, we recorded a drop and then an increase until the end of the year. Those clients who jumped on from round 8 should have recorded a bank of around the same as starting, whilst those at the start of the year would have made most of their money from these first 8 weeks. Line betting seems to have been less consistent than H2H betting throughout the year, increasing and decreasing at greater values, whilst interestingly, margin betting was probably the most consistent of all three bet types despite finishing on a lower profit. This could be due to smaller bet sizes for margin betting.

2. Analysis of different probs/odds/overlays

Shown here is the analysis of different probs/odds/overlays/home team analysis as well as record at different stages throughout the year. This analysis is provided for head to head betting only. It is only a small sample for 2005, so don't look too much into it, but we will also look at the overall betting since 1999 as well.

First analysis shows that we made most of our money betting on teams with probabilities of under 60%. This is backed up in the odds analysis that shows that we made the majority of our income betting on underdogs. This proves the models great ability to find underdogs who are undervalued. However significant money was also made on favourites. Analysis of overlays is interesting. Whilst we fortunately made a lot on very small overlays, and unfortunately lost on midrange overlays, overlays greater than 35% we profited on very well. Interestingly these overlays seemed to mimic the %ROI very well, which shows the power of the model (more on this in section 3).

When looking at home ground advantage, most money was made on betting on teams hosting interstate traveling teams, as well as on teams traveling interstate, which shows that the bookmakers have probably not correctly evaluated not only the teams, but the nature of interstate travel.

Shown here is the analysis of the model for head to head betting since 1999. This is probably more accurate than just an analysis of 2005, but alas it was interesting to see how we did last year anyway. Here we see a very broad range of profits for different probabilities, as well as odds. Once again the %ROI seems to closely mimic the %overlay, which, I might add, is very remarkable, which proves that our probabilities are a LOT closer to the true probability of a team winning than the bookmakers odds. Little money was made betting on interstate traveling sides. This was probably boosted from this years stock, but could be explained by the sudden dominance on non Victorian sides in the last 4-5 years. Where we win throughout the year, is proven that this year's early wins was no fluke. Early wins are more often occurring than later in the year, but this hasn't meant in the past that the 2nd and 3rd thirds of the year are not profitable. No significant patterns were found looking at individual home team analysis.

3. Optimal Kelly Fractional.

Little has been done about finding the optimal Kelly fractional despite being a large talking point. The Simulator that is provided creates a great idea of what type of Kelly fractional to use, but still doesn't provide what the optimal Kelly fractional should actually be.

We noted before the remarkable result that the %overlay of bets seemed to closely mimic the %ROI. This is quite extraordinary and shows the extreme power of the model in predicting. We all knew that is predicts better than the general public (based on the bookies odds), - as we make a constant profit over the years - but this analysis of it's predictability has left me even very surprised.

The way to look at the optimal Kelly fractional is to work out the expected return of investment over the years and compare this to the actual return of investment since 1999 (13.3%). Since 1999, we have gained an expected %ROI of 23.1%. This value is calculated by comparing what we expect to gain from every bet based on the overlay, odds bet and amount bet, adding this up and dividing by the total amount bet to obtain the expected %ROI.

The fact that this is very close to the actual %ROI is very startling. This shows that the optimal %ROI since 1999 is actually 23.1 / 13.3 = 1.74. Or 1/1.74th Kelly The expected profit from 2005 was exactly 25% which is once again comparable to previous years.

This of course tells us a few things. Firstly, the supreme accuracy of the AFL model I cannot talk enough about. Secondly it is most likely that most punters were operating on 1/3rd to 1/4th Kelly for the AFL in 2005, which according to this optimal Kelly fractional is very conservative. Of course despite these new findings, I don't intend on betting at 1/1.74th Kelly, as I believe such betting is far to high a risk of bankruptcy (to check use AFL simulation). However it has been proven mathematically that such a Kelly Fractional will maximise your bank balance.

Obviously it is up to the individual about maximising bank balance and offsetting this against the amount of risk that they want to take on board.

4. 2005 Bank Analysis

Firstly lets assume that we have a $10,000 bank. How would we have gone in 2005 given different Kelly Fractionals and constant of moving bank. By moving bank, we will change the bank at the end of each week.

Below shows tables for how much our bank balance would have been at the end of the year given different Kelly fractionals and either constant/moving banks.

As shown above, with moving banks, results were comparable, with 1/4th Kelly providing lower profits than other years. With a constant bank, lower fractionals provided high profits than high fractionals. However constant banks provided significant more profits than moving banks.

This graph shows the moving and constant banks for 1/2nd Kelly This shows that in the "bad weeks" from rounds 10 to 13, the moving bank method lots appreciative ground, whilst the constant Kelly method dipped only slightly. On the counterside however, the moving bank increased more rapidly when in winning form. This if course is exactly what we expect from it. We expect a moving bank to increase more over the long term. It might lose more when on losing runs, but will gain more when on top. And considering that there are more winning runs than losing, it should be more beneficial over the long term, despite not being in 2005.

This graph shows the long term relationship for using 1/1.74th, 1/2nd, 1/3rd and 1/4th moving bank Kelly's for the data since 1999. Despite the massive profits which would surely be limited by betting limits, it is good for comparison between the Kelly fractions. This graph shows the 1/3rd and 1/4th Kelly betting more closer.

These graphs clearly show the long term benefits of these lower Kelly fractionals (although don't go too low!). However the risk is obviously there two. In the last year, we clearly see that the 1/1.74th Kelly lost 3/4 of it's bank in a matter of a few weeks (no doubt between weeks 8 and 13 this year). As the Kelly fractional increases, so too does the end profit, but also too the amount that is lost in this period is less (in both monetary value and percentage of bank). Of particular note is the exponential relationship of these graphs, which is clearly a characteristic of the moving bank.

When comparing the moving bank to the constant bank as shown in this graph since 1999 (using 1/3rd Kelly), the difference is obvious. The moving bank is far superior, and whilst the constant bank looks like a straight line is isn't as shown here but lacks the exponential effect. In other words, it is not utilising it's previous successes to create more success in the future. It is merely just dwarfed by the massive profits of the moving Kelly

However this last graph is of particular importance. It would be a godsend to any serious investor. We've all seen graphs of the stock exchange, gradually increasing and then starting to head to a very weak exponential trend. However, nothing is more constantly growing than the last graph mentioned. It's constant increase in bank balance is a clear indication of the models success. And whilst there are small decreases from time to time, the constant increase in it's nature shows that one should not be troubled by these small losses and should really be focused on the long term success.

5. 2006 and Beyond: Predicting the future.

The last graph shows that there is no reason to say why another constant profit should not eventuate in 2006. Last years profits were in line with other years, and hence the value in each of the bets is still there and very very real. The sheer accuracy of the model as outlined by the very small optimal Kelly fractional indicates that it is, beyond doubt, Sportpunter's most accurate sports model.

But as we all know, predictions are not enough by themselves to get you over the line. Knowing where to bet (at good odds) and how much to bet is critical in any gambling endeavor, and hopefully this report as outlined a bit more about the how much to bet aspect so as to increase your profits in the future.

As what does the future hold? Well as mentioned before, the following years should emulate the previous one's as there is no sign of decay at all. But as bookmakers get smarter (only once punters get smarter - as Sportpunter is!), then less value might occur in the future. One cannot say when this will occur, or if - but it means that there is no better time to subscribe to Sportpunter's AFL model whilst the going it good.

If you have any comments about the AFL 2005 Report Card then please feel free to make them in the forum, so that all may discuss the analysis and implications for next years betting.

And once again,

Best of punting luck!

Jonathan Lowe

www.sportpunter.com