This is the first season review I’ve posted and there are
going to be 8 season reviews in total. Last Summer, I probably spent too long
on the season reviews aspect as I reviewed all systems individually. Given the fact there are twice as many
systems to review now, I couldn’t spend anywhere near as long on each system as
I have in the past.
The 8 season reviews are going to cover:
1). 6,7,8,21,22
2). 6-21,6-22,7-21,7-22,8-21,8-22
3). 31,32,33,41,42
4). 31-41,31-42,32-41,32-42,33-41,33-42
5). 6-32,21-31,6-21-31,TOX, STOY and STOZ
6). E1,E2,E3,E6,E7
7). E1-E6,E1-E7,E2-E6,E2-E7,E3-E6,E3-E7
8). Under/Over 2.5 Goals Systems
In terms of layout, I’m going to look at the following
aspects in order:
Season 2012/13 Performance
Live Results to Date
Statistical Significance (P-value)
2013/14 Targets
Detailed Analysis
I’ll be concentrating on ROI, ROC, Drawdowns and updated
betting banks in the main. The 2013/14 targets are going to be what subscribers
should be most interested in I think as it shows what we’re aiming for next
season.
I’ll finish off each review by looking at a variety of
things including performance by season, league and by Home and Away bets. I can’t imagine that every review is going to
be interesting to everyone but hopefully some readers will pick up something
that interests them from each review. The point of spending time reviewing the
past is to give us any insight into the future to help our betting going
forward. I don’t think it’s any coincidence that most subscribers with the
service make more money the longer they are with the service. The more
understanding each subscriber has, the more likely they are to make more money
in future.
I will be releasing a spreadsheet for download after I’ve
completed each of the season reviews and this will have all the tables in it
I’ll be using in the reviews. Therefore, people can do their own analysis if
they wish and the spreadsheet should be used in conjunction with reading the
system reviews.
So, I’ll start with the review of systems 6,7,8,21 and 22……………
This is the 3rd season that these systems have
been live and all results in this review are live results. I will no longer be
discussing the backtested results from 2006/7 to 2009/10 and if people are
interested in these, they will find the results in the historical results
section on the blog.
Season 2012/13
Performance
I think it’s a pleasing 3rd season for these
systems if you look at the results now the season has finished. The systems
started the season very slowly and at various points this season, the systems have
found it tough going but class always shines through in the end I think.
Systems 6 and 21 are always key as these are my first two
algorithms I developed 3 years ago. System 6 has struggled this season if I’m
honest but a return of 4.4% from 725 bets isn’t a disaster.
On the other hand, system 21 had a really, really good
season again. An ROI of 9.8% across 598 bets is pretty phenomenal for a set of
ratings and to make near 10% return from backing each of the games blindly
makes this algorithm one of the best in the business I suspect.
If I was being critical, I would say that it’s very
disappointing that system 7 and system 22 couldn’t improve upon the returns of
6 and 21. System 7 really struggled and turned a 4.4% profit into a 2.9% profit
but system 22 did a better and at least managed to maintain an ROI in line with
system 21 (9.2%). I think once you see
the overall live results, the picture looks a bit clearer and it’s probably
just a bit of variance this season impacting 7 and 22 as they don’t have a lot
of bets.
In terms of ROI, the star of the show this season was
system 8. A 14.2% ROI from 103 bets is a very good return. It’s a little
strange that system 8 had such a good season when 6 and 7 struggled so much but
again, I suspect it’s variance again but this time, it’s positive variance!
Anything can happen over 103 bets and all it takes is a couple of games to go
the system’s way and it can bump up the ROI a fair bit.
Overall, a profit of 125.8pts from 1,805 bets this season
for these 5 systems means a respectable 7% ROI and you have to say it has been
a good season for the Established ratings again this season.
All Live
Results (last 3 seasons)
I think the first thing to point out is the trend of ROI
as you move up systems. System 6 has an ROI of 4.9% from 2,010 bets but the ROI
improves to 6% on system 7 and 10.8% on system 8. The same story applies for
the second algorithm where the ROI increases from 8.5% on system 21 to 10.4% on
system 22. Therefore, the filtering of
the ratings by the systems appears to work as I intended it to and clearly, the
systems are able to pick out the better bets. As this season shows, over a
smaller sample of bets, this phenomenon may not occur but long-term, the trend
remains.
I think with 3 seasons behind these systems now, you have
to be pleased with these results. It is now clear that algorithm two is better
than algorithm one but I’ve known that for a while. I learnt a lot about
building a successful rating algorithm after I built the first algorithm and
therefore, it was always likely to be the case that the second algorithm would
turn out to be better than the first algorithm.
In terms of Return on Capital (ROC), the results are
variable by system but all 5 systems have decent results.
I think the ROC figures are key. I know people get sucked
in by the ROI some of my systems have but as I’ve been saying for a long-time
now, ROC is the key at this game. System
21 leads the way with a ROC of 97.5% this season and a ROC of 233.6% over 3 seasons. To nearly double your betting bank in one
season is a great achievement and an average ROC of 77.9% across 3 seasons
makes this a very, very powerful system in its own right.
Not surprisingly, system 6 is the second best and this
system managed to achieve an ROC of 53.7% this season even though it hasn’t had
the best of seasons. This is actually
very consistent with the long-term performance though and over the first 3
seasons, it has achieved an ROC of 164.6% with an average ROC of 54.9% per season.
The other 3 systems aren’t really made for achieving a
high ROC but even so, if you look at the performance of these systems over the
3 seasons, it isn’t bad. An average ROC of 24%, 25% and 30% across 7,8 and 22
are highly acceptable for any tipster I think but in the TFA portfolio, this
puts them down as systems that people shouldn’t really be following.
Statistical
Significance
Here are the P-values for each system.
Not surprisingly, all 5 systems are highly statistically
significant but the chances of system 21 achieving the results it has achieved
by luck (an 8.5% ROI over 1,649 bets) is 1 in 50 million I think! The least
significant are systems 8 and 22 but even then, a P-value of 0.5% and 0.3% is
still highly acceptable and show that these systems aren’t achieving these
results by luck!
Overall, I think you have to be pleased with these
results for the 5 Established Systems and that’s 3 successful seasons in a row
for these systems.
Targets for
2013/14 Season
I think for many reading this, the most important aspect
of the reviews are going to be the targets for next season. I will state the
official betting banks for each system as well as the target ROI and ROC that
the systems will be looking to achieve next season. No guarantee that any system will do what it
sets out to do but you need to have targets in this game and with so many
systems to choose from, these targets can be used to help subscribers (and
myself!) decide which systems to include in our portfolios next season.
I think the first thing to point out is that I am
tweaking the betting banks on system 8 and 22. The betting banks are basically
overstated as the losing spell that sets the banks occurred in a bad run 3
seasons ago and with the bet number these systems have nowadays, the betting
bank looks very conservative based on the last two seasons performance. System 8 is now going to have a 40pt betting
bank and system 22 is going to have a 50pt bank. Given the low bet number, the
systems could never lose the full betting bank in a season I think but I’m
happy to go with these new betting banks.
Not surprisingly, system 21 has the highest target ROC
next season and will be aiming to grow the betting bank by 80%. We then have system 6 with a target ROC of
47% and then the other 3 systems all have a ROC target of less than 30%.
Importantly, the targets I set are always based on past
performance and if the systems perform as they have done for the first 3
seasons, these targets should be well within reach. As I said above, no guarantee that any system
will achieve what it sets out to do but at least those following can look at
the historical performance and that should help set expectations about the
upcoming season.
I expect system 21 may actually prove to be a popular
system next season given the performance over the first 3 seasons, the P-value
and the target ROI/ROC next season.
Wouldn’t put anyone off only following system 21 in isolation if wanting
to follow a single system from the Est Systems.
Detailed
Analysis
I’m not going to go into a great deal of depth on each
aspect but I’ll paste the tables and people can do their own analysis if they
like.
Performance by
Season
You can see clearly that this season hasn’t been as good
as the 2011/12 season but then again, it is better than the 2010/11
season. If you look at the AH returns, you
would say that the only reason the first season was worse than this season was
due to the fact the systems struggled a lot more with draws, so in many ways,
this season has been the toughest of the 3 seasons to date. I think that
probably backs up my own thinking about these systems this season and it has
been a tougher season than the first two seasons but it has still been a decent
season.
Performance by
League
This isn’t a surprise to anyone who has been following
the systems for a while but the poorest performing leagues are League Two
(where a loss has been made over the first 3 seasons) and the Bsq Prem (only a
3.2% ROI). Again though, if you look at the AH returns, League Two has hit a
fair number of draws and if using AH0.5, a profit is made, so I’ve not given up
on this league yet!
The Premiership continues to be the best performing
league by far with an ROI of 28.6% across the first 3 seasons to date.
As a side note, I know one person last Summer told me
they weren’t following any more League Two bets and I think I mentioned this on
the blog last year. Was it a good idea?
Performance by
League and Season
As you can see, League Two actually created a small
profit of 8.9pts from 482 bets this season. Not great but a good step forward.
The poorest league in the 2012/13 season was the SPL by far. A loss of 25pts
from 107 bets.
Performance by
Home and Away
Nothing too surprising here. Twice as many away bets than
home bets but the average odds are much higher for Away bets (3.64) than for
Home bets (2.16). This is reflected in the strike rates where Homes have a strike
rate % of 51.2% and 31.4% for Aways.
In terms of the returns, a 7.6% ROI for Away bets and
6.2% for Home bets. Again, given the
average odds, we’d expect the Away bets to achieve a higher return.
Performance by
Home and Away by League
I think this is maybe the most interesting table from
this section. There are a lot of trends on show here and whether or not they
will continue is uncertain but it maybe highlights some weaknesses within the
ratings for these systems as well as some strengths.
Aways are better than Homes in the Bsq Prem and Homes are
only break-even in 3 seasons in this leage.
Championship Aways
are very strong whereas the Homes are only marginally profitable.
League One Aways are very, very profitable whereas Homes
are loss making.
League Two Aways are very, very unprofitable whereas
Homes are very, very profitable.
Premiership Aways and Homes are both very profitable.
SPL Aways are very profitable but the Homes are loss
making.
Now, what do we make of all of the above? Well, firstly,
the sample sizes aren’t great but I think there is a clear issue in League Two
where the draw is a pest for the Aways. A -10% ROI for backing outright but a -1.3% ROI for
backing using AH0.5 is a stunning turnaround for backing the draw and simply,
if we were to only be backing the draw in these games where we’re on the Away
team, we’d have a very healthy profit.
The strange thing for me is that League One and League
Two are right next to each other on the football pyramid but clearly, my
ratings work great for Aways in League One but not for League Two but likewise,
my ratings work great for Homes in League Two but not in League One!
It’s hard to fathom and I can’t help but feel it’s just
variance but I wouldn’t be against anyone deciding to vary stakes based on some
of these trends as at the end of the day, we have 3 seasons worth of data and
some of these trends have been around since the beginning and aren’t changing!
First of all, congratulations for another fantastic season!
ReplyDeleteRegarding the statistical significance, I would like to comment two things:
- You should take into account the bookie's overround when calculating the expected winners/losers
- How are you calculating the p-value?. I've made the chi-square test in Excel, not considering the overround, and the results are different, althought they also indicate that systems clearly have an edge. For example, I get a p-value=0,008808 for system 8 (0,88%) and a p-value=0,000827 for system 7 (0,082%)
Hi Diego.
ReplyDeleteThanks for the comment and the congratulations.
I always tend to steer clear of statistical tests on the blog as people always have differing opinions on them and how useful they are but the reason I’ve included the P-values is so we can compare systems with each other. Hence, as long as it is a consistent measure, I’ll be OK.
I agree I should really take into account the bookies over-round (thus improving the P-value) but I’m ignoring the over-round on all the systems, so as I say, it will be consistent.
I’ve had a quick look and I think there are two reasons for the discrepancy with your figures. Firstly, the average odds are not rounded in my calc and therefore, the amount of expected winners/losers isn’t exactly what I have shown in the table.
The biggest issue though is the fact that I’ve divided my answer by two and you haven’t. Hence, I’ve done Chisq.Test(x1,x2,y1,y2)/2. I think your answer is twice as high as mines (corrected for the fact the numbers aren’t round in my calc).
I’m not a statistician or mathematician unfortunately, so I don’t know what is correct but given I’ve supplied the figures for the actual and expected wins/losses, those of you smarter than me can do your own analysis!
As you suggest, it doesn’t change the conclusion of the analysis although the actual figures aren’t factually correct anyway since I’ve not taken out the overround.
Hope this makes sense.
Graeme
PS. If someone else can confirm I shouldn’t be dividing by two, I’ll restate all the P-values…..will take me roughly 2 seconds!