I think for any new readers, it would make
sense to read the review from last September before reading this review. You
can find the previous September review here to provide some context around
these results.
The TFA systems are now into their 5th
live season and this is the review of the first month this season. The format
of the review will be similar to last season and I’ll look at the results of
the various rating algorithms and then the systems that are filtered off the
back of the ratings. I’ll deal with the UK outright bets first, then the UK
Draws and lastly, the two types of European bets. With 6 different sets of bets
now, there is a lot to discuss in these reviews but I’ll try my best to not make
the reviews too dry (easier said than done at times!).
To recap, we have 61 live TFA systems this
season. In addition, I know quite a number of subscribers are no longer using
the UK outright systems at all for their betting and they are simply using the
4 algorithms and the value ratings to derive their own portfolio of bets to
follow. Therefore, it’s important that as well as the 61 outright systems, I
track the performance of the value ratings on each of the algorithms.
As discussed in the previous blog post,
September 2014 has been the best month since my ratings went live for Home bets
and the worst month for Away bets. It’s important we keep this in mind when
looking at the rating and system results for September as we don’t want to draw
any incorrect conclusions.
As usual, if anyone has any questions, please
leave a comment or drop me an email.
Est
Systems
Here’s the results for rating algorithm 1 split
by value band:
37 winners from 95 bets on algorithm 1 and a
loss of 2.1pts (-2.3%). The lowest value bets lost 5.5pts from 44 bets. Bets
with a value of 5%+ made 3.3pts from 51 bets.
A loss of 2.3% can never be described as a good
month but last season, this algorithm lost 19.4pts from 67 bets (-29%). Based on this stat alone, it’s hard to get
too downhearted about a small loss in the month.
Here’s the breakdown between Home and Away
bets:
We can see that we hit a profit of 11.5pts from
49 Home bets and a loss of 13.6pts from 46 Away bets. Last season, this
algorithm had 46 Away bets (lost 23.3pts) and only 21 Home bets (3.9pts
profit).
Two things to take away from this. Firstly, the
changes I made in the Summer to this rating algorithm have increased the number
of Home bets as I outlined in the Summer. I expected more Home bets on all
algorithms and with the removal of the minimum odds stipulation for Home bets,
we now see the odd short priced Home bet appear on all algorithms. Both these
factors contribute to more Home bets and after this month, I think we can see
the benefit of this.
Secondly, given how difficult it has been for
Aways bets this September compared to last September, I’m amazed the Away loss
wasn’t even worse! This may be wishful thinking on my part (time will tell!)
but I believe the algorithm is a much better algorithm than the algorithm I had
last year for Aways. Obviously, it’s difficult to prove and one month doesn’t
tell us much but I strongly believe a loss of 2.3% this month on this algorithm
is a great result given the underlying results. Without a doubt, this September
is more difficult than last September and yet, we have done a lot better
overall. I think this will serve us well over the rest of the season.
Here’s the results for rating algorithm 2 split
by value band:
27 winners from 71 bets and a loss of 1.9pts
(-2.6%). The lowest value bets made a profit of 5.1pts from 33 bets and bets
with a value of 5%+ lost 7pts from 38 bets.
A bit of a strange result given the fact that lower value bets tend to
be much worse performing bets but it’s a small sample of bets and is going to
be skewed by the Home/Away split since Away bets tend to be higher value and we
know these would have had a nightmare this month.
Similar to algorithm 1, a 2.6% loss isn’t a
disaster given the underlying results. Last September, this algorithm lost
16.4pts from 54 bets (-30%). Much better
than last season and this season was tougher.
Here’s the breakdown between Home and Away
bets:
11.6pts loss from 40 away bets and a 9.7pts
profit from 31 Home bets. Last season, the algorithm had 39 Away bets and lost
19.1pts and made a profit of 2.8pts from 15 Home bets.
Similar comments to algorithm one. We are
seeing more Home bets than last season (twice as many) and this has really
helped smooth the returns this month and should hopefully help the systems and
service smooth the returns going forward. In addition, the Aways bets have
performed much better this season than last season which again gives me hope
that the changes I’ve implemented on the algorithm have been for the better!
Overall then, a small loss on both algorithms
to start the season but compared to last season, the algorithms have performed
much, much better which I’m putting down to the increased number of Home bets
and the improved performance of Away bets.
How did the systems fare with trying to filter
these bets? Here’s the results for all
11 Established Systems:
121 winners from 342 bets for a loss of 41.8pts
(-12.2%). Last season, the equivalent numbers were 60 winners from 262 bets for
a loss of 92.4pts (-35.3%). Similar
comments to above I think. Never good to lose so many points in a month but
compared to last season, it’s much better.
Looking at the results by system then, one
thing that strikes me is the fact that the combined systems have had a shocker.
The filtering from system 6 and 21 were shocking and this has led to massive
losses on some of the combined systems. Those from last season will remember the
same issues were apparent all season and it’s slightly frustrating this season
has started the same way even though the algorithms have changed significantly!
System 6-21 has had a really bad month which is
a little surprising considering 6 & 21 haven’t had a disaster but as you
start to work down the combined systems, it’s not pretty at all.
In summary then, a tough month for the rating
algorithms but I think they did better than last season and I’m quite pleased
with how they performed. The combined systems are a different kettle of fish
and it’s been a terrible start for these systems. Off the back of a terrible
season last season, I think these systems now have massive question marks
against them. The long-term results are dipping quickly and the systems have
gone from looking like very solid systems to systems with a massive question
mark hanging over them. One to watch over the next few months I think and I
have no reservations about retiring systems which I feel are no longer adding
anything to the service.
New
Systems
Here’s the results for rating algorithm 3 and
split by Home/Away:
Overall, a 2.6pts profit from 84 bets (3.1%). 8
winners from 43 Away bets and a loss of 11.7pts (-27%). 25 winners from 41 Home
bets and a profit of 14.3pts (34.8%).
Looking at the results from last September, an
overall profit of 3.8pts from 84 bets. 11 Away winners from 57 bets for an
18.2pts loss and 20 winners from 27 Home bets for a 22pts profit.
Apart from the fact that both years had the
exact same number of bets (amazes me when things like this happen!), the key
difference is again the split of Home/Away bets. We have gone from 67% Away
bets and 33% Home bets to nearly a 51%/49% split. This is again due to the
changes on the algorithm in the Summer although I’m a little surprised there
weren’t more Away bets. Again, I actually think it’s good there weren’t more
Away bets as no doubt we’d have lost more money on the Away bets!
I think given the Underlying results, it is a
brilliant result for this algorithm to be making a 3.1% profit this month. I
said the exact same last season but this is what separates algorithm 3 from the
other 3 algorithms. The algorithm does a great job in controlling losses during
difficult spells which means it never needs to dig itself out such a big hole
as the other algorithms at times!
A really solid start for algorithm 3 this
season again. Let’s hope it can build on this like last season.
Here’s the results for algorithm 4 split by
Home/Away:
A loss of 8.9pts from 77 bets (-11.6%). Only 7
winners from 43 Away bets for a loss of 20.1pts (-46.8%). 19 Home winners from
34 bets for a profit of 11.2pts (32.9%).
Similar to above, looking at the same results
last season shows an overall loss of 8.9pts from 98 bets. 11 winners from 63
Away bets for a loss of 23.8pts (-37.7%) and 21 winners from 35 Home bets for a
profit of 14.9pts (42.4%).
The key thing here is the reduction in bet
number. Algorithm 4 was the algorithm that got most work during the Summer as
it was the worst performing algorithm and I felt like it needed tightening up
as well as the other changes like more Home bets. I can see I’ve tightened it up (21 less bets)
and I can see I’ve adjusted the % of Home bets (35% last year has increased to
44% this year) but in terms of the overall results, things haven’t improved at
all. If anything, you could argue I’ve made it worse as it lost an identical
amount of points in both seasons (another amazing stat!) but it did it from
more bets last year!
I guess the jury is out on algorithm 4 and the
Away performance this season has been terrible. Even allowing for the tough
Away conditions, to have a -46.8% ROI is about as bad as you can do I think and
I think this rating algorithm has a lot to answer for.
I haven’t really commented on the value bands
for algorithm 3 & 4 but it’s difficult to draw conclusions when all the
Homes bands are profitable and all the Away bands are loss making. It then
becomes a case of how many bets are in each band and if you have more Home bets
in a band than Aways, it makes the results look better!
A poor start for algorithm 4 then and I think
similar to the Est combined systems, there are a lot of question marks over
system 41.
Let’s see how the New combined systems did with
these results on the rating algorithms. Here’s the results for all 11 New
Systems:
An overall profit of 7.7pts from 397 bets
(1.9%). Last season, the equivalent
numbers were an overall loss of 87pts from 448 bets (-19.4%).
Given this September has clearly been a more
difficult month than last September based on underlying results, the turnaround
in these systems in 12 months is amazing. I won’t have been alone in having
doubts about changing my algorithms over the Summer and refreshing them but I
think after the first month, I’m quite proud of the fact I haven’t appeared to
mess anything up! For these systems to make a profit this month is unreal and
looking back historically, any time the underlying results have been as tough
as this, these systems have suffered large losses in total.
I think the filtering on system 41 is clearly
pretty good and to turn an 11.6% loss into an 18% profit is about as good as it
gets. The filtering on system 31 is less reliable but it has been that way for
a long time now. System 32 made a loss and system 33 made a profit but I’m much
less confident about the filtering from system 31. I’ve said this before but not
sure you need to worry about filtering system 31 too much as the algorithm is
pretty good I think.
Due to the fact system 41 had a poor month, it
has been reflected in a very poor month for system 31-41 which will have
impacted a number of us. I had my reservations about system 31-41 over the
Summer as it was clear system 31 was a brilliant system and system 41 less so,
so using 31-41 always runs the risk that system 41 spoils the party and it has
really pissed on the parade of system 31 this month.
System 31-42 is a popular system amongst the
subscriber base and it’s nice to see a solid start this season after a horrible
time last season.
Overall, over the moon with the New system
results for September and if they can produce the same profits over the next
few months as they did last season, I’ll be pretty happy I think.
Misc
Systems
Here’s the results for the systems:
78 winners from 201 bets and a loss of 3.1pts
(-1.5%). Not the best of starts for these systems but given the difficulty of the
month and the other comments above about the rating algorithms and systems, a
1.5% loss isn’t a bad start to the season.
Last season, it was 58 winners from 232 bets
and a loss of 65pts (-28%). Considering we know this September is tougher than
last September, then it really helps to put this season’s results in context!
Looking at the systems then, 6-32, STOY and
STOZ made small losses and 21-31, 6-21-31 and TOX made small gains. Again,
given the fact 3 of the algorithms made small losses, I think these results
reflect this well.
Here’s the splits of the results between Home
and Away:
24 winners from 113 Away bets for a loss of
28.9pts. 54 winners from 88 Home bets for 25.9pts profit. 56% Away bets.
Looking at last season, 22 winners from 164
Away bets for a loss of 77.3pts. 36 winners from 68 Home bets for a profit of
12.2pts. 71% Away bets.
As well as the turnaround in performance from
last season, the other significant shift is the change in mix between Home and
Away bets. A big movement towards Home bets and this has really helped smooth
the P&L so far this season.
Overall then, a small loss to start the season
for these systems but compared to last season, it feels like a great month! ;)
Draw
Systems
NB.
Going into the season, these were the systems that I had most confidence in as
I barely touched the ratings over the Summer since they were only built the
previous Summer and the results last season in their debut season were
brilliant in a season when the strike rate for draws was the lowest it had been
since the 2006/7 season.
Here’s the results by system:
Only 10 winners from 95 bets and a loss of
59.8pts (-62.9%). OUCH!!!!!
Comparing this to last season then, 39 winners
from 176 bets and a loss of 38.3pts (-21.8%).
Before commenting on the results, it is worth
remembering that Sep-13 was very similar to Sep-14 in terms of underlying
draws. The strike rate was 24.5% last year over 184 bets and 24.7% this year
over 227 bets.
Clearly, we may have a problem here. I’ve no idea why the number of draws has
dropped so significantly from last season and more importantly, I can’t really
explain the sizable loss either. What you can’t see here is that the systems
suffered another large loss in the first weekend of October, so the results are
currently even worse than we’re seeing here!
It’s difficult to add much to this. 95 system
bets is a tiny sample but the filtered systems didn’t hit a winner and the
combined systems hit 2 winners from 35 bets! Clearly, it’s not been a good
start and is much worse than last season which was a bad enough start.
On one hand, part of me thinks this is just
really, really bad variance and because it was a tough month for draws, the
systems have struggled badly. The other part of me looks at the results last
weekend, sees a very high strike rate for Draws and realises that my systems
made another large loss and the filtered systems couldn’t hit a winner again!
Clearly, panic is setting in just a little for
me personally as about 25% of my betting portfolio is made up of UK draw bets
this season and this was an intentional ploy on my part given the results last
season. I won’t be alone here and anyone who has taken on a draw system or two
after last season must be scratching their head!
I guess if I look at the overall live results,
the ROI is still decent enough and therefore, there is probably nothing to
worry about but I’ll feel a bit better when I see these systems hit a few
winners. I think the jury is out on these systems and another few weeks like
this and I think I’d maybe have to really start looking at betting banks etc.
for the systems and how much has been lost and whether the banks need revisited
etc.
Let’s hope they turn the corner soon as if not,
my fingers will be burnt along with a few others I suspect.
Euro
Systems
Here’s the results so far:
Only had 1 bet this season and it was a winner.
8.6pts profit from 8 bets (107%) This was the first season the Euro bets had a
bet in September as the bets didn’t tend to start until October! So, it’s
better than last season then. :)
Euro
Draw Systems
Here’s the results so far:
First season for these bets, so not any history
to go on but 5 winners from 11 bets isn’t a bad start! A profit of 4.8pts from
11 bets (43.6%),
My only reservation about the Euro Draw systems
is that as my ratings have been waiting for the first 6 games to pass in every
league, the number of Draws in most leagues have been amazingly high!
Therefore, these bets are going to kick in at a peak which added to my comments
about how difficult it is historically to make money betting on draws in these
leagues fills me with a little fear. Given the run the UK draws have had and my
stakes on these, it’s an easy decision for me but I’ve reduced stakes on the
Euro Draw systems this season before the bets even started! Happy to take a backseat
approach to these systems like I sort of did last season with the UK Draws
(still overstaked these IMO!) and when I see results, I’ll start to adjust
stakes upwards once I get comfortable.
Overall, 83.5pts lost from 1,054 bets so far
this season.
Graeme, before questioning some of you're systems I really think you should look at performance from a less correlated way. For example, you're draw systems. How many unique picks are we talking there (30?), and you have 5? unique winners there (10 are needed for a profit). I know you are a man who know's the numbers, then you can't be serious when you think 5 out of 30 winners (when 10 are needed for a decent profit) is bad, bad variance, I mean you ain't seen nothing yet....this is really just complete standard variance and it's impossible to draw any conclusions at all from it imo. Yes it's a bad start, but nothing more...
ReplyDeleteMartin/Pokersunited
Martin,
DeleteWelcome to TFA.
My compliment for a great guest post and comments that clearly shows you are a trained statistician like myself.
Hopefully you will continue to make many more comments in the future.
dmitri
It can be difficult at times to understand exactly what the many numbers in the reviews represent.
ReplyDeleteI will walk you through an example to help you understand some basics about systems, portfolio of systems and money management/staking strategy.
My example is based on TFA UK Draw picks on Oct. 4 2014.
System D1 had these picks:
Date League Game Count
10/4/2014 Bsq Prem Chester v Welling 1
Macclesfield v Forest Green 1
League 1 Scunthorpe v Doncaster 1
League 2 Mansfield v Accrington 1
Wycombe v Northampton 1
Championship Middlesbrough v Fulham 1
Wolves v Wigan 1
Brentford v Reading 1
SPL St Johnstone v St Mirren 1
Premiership Hull v Crystal Palace 1
Grand Total 10
System D6 had these picks:
Date League Game Count
10/4/2014 Bsq Prem Chester v Welling 1
Nuneaton v Lincoln 1
League 1 Scunthorpe v Doncaster 1
League 2 Mansfield v Accrington 1
Wycombe v Northampton 1
Championship Wolves v Wigan 1
Brentford v Reading 1
SPL St Johnstone v St Mirren 1
Premiership Hull v Crystal Palace 1
Grand Total 9
D1 has 10 unique picks and the staking strategy is level staking, placing one unit on each pick.
Similarly D6 has 9 unique picks and is also using level staking.
We can make a portfolio of D1 and D6, and this portfolio had these picks:
Date League Game Count
10/4/2014 Bsq Prem Chester v Welling 2
Macclesfield v Forest Green 1
Nuneaton v Lincoln 1
League 1 Scunthorpe v Doncaster 2
League 2 Mansfield v Accrington 2
Wycombe v Northampton 2
Championship Middlesbrough v Fulham 1
Wolves v Wigan 2
Brentford v Reading 2
SPL St Johnstone v St Mirren 2
Premiership Hull v Crystal Palace 2
Grand Total 19
The portfolio has 11 unique picks and the staking strategy is no longer level staking.
On some picks we place one unit and on other picks we place two units.
This is called variable staking.
So it’s important to know if we are looking at a single system like D1 or a portfolio of systems.
They use different staking strategies.
Now let me look at all TFA UK Draw picks in 2013.
According to TFA reviews the profit from these picks were 229 units from 2535 picks, a healthy 9% yield.
These numbers are based on a portfolio of a total of 11 draw systems and variable staking.
Therefore some picks get one unit and some picks get 11 units.
There are actually only 653 unique picks among the 2535.
If we used level staking on the portfolio instead of variable staking we would place 653 units and the result would be a loss of 20 units!
One way to look at these numbers, and I think this is the ‘official’ TFA way, is to credit the systems for the profit. By making a portfolio of 11 systems TFA has made a healthy profit.
Another way to look at the numbers, and this is my way, is to credit the staking strategy for the profit. By using variable staking instead of level staking on the portfolio TFA has made a healthy profit.
At this point I have to refer to a mathematical theorem.
If a betting system is not profitable using level staking NO staking strategy will make a profit in the long run.
Of course variable staking can make a profit in a short period like 653 bets in a year.
But the loss of 20 units using level staking in 2013 does not prove the portfolio system to be profitable, on the contrary maybe.
dmitri
Good post Dmitri! What about the 4 main algorithms (6, 21, 31, 41)? Are they all in profit using unique bets level staking or are some of them only in profit using variable staking?
ReplyDeleteThank you!
Delete6, 21, 31 and 41 are actually not rating algorithms. They are systems obtained by filtering four unseen rating algorithms.
7 and 8 are also systems. They are obtained by filtering system 6 and 7. So the unique bets for system 6 includes all bets of 7 and 8.
Therefore we say that system 6, 7 and 8 are built on algorithm 1.
Likewise other rating algorithms 2, 3 and 4 are used for (21, 22), (31, 32, 33) and (41, 42)
Yes, 6, 21, 31 and 41 are all in profit using level staking.
When you read the reviews all systems be it 6, D6 or D1-D6 use level staking.
What can be confusing is f. ex. when the 11 UK draw systems are aggregated to a single profit/loss number. This number is the result of a portfolio of the 11 draw systems, and are obviously not using level staking.
A much, much better number would be the result for the portfolio using level staking.
You are asking if the systems are in profit.
Although this is a relevant question a more important question is: Are these systems profitable?
You want to know what happened in the past.
I want to know what to expect in the future.
A simple Excel sheet can answer your question. It’s exactly what the reviews does.
To answer my question you need to apply statistical methods.
And the answer is that until now none of the 4 systems can definitely be called profitable.
If system 21 and 31 perform well in the next 1-2 seasons they might make the test.
I can’t see system 6 and 41 qualify in the foreseeable future.
Let me repeat an important theorem:
If a betting system is not profitable using level staking NO staking strategy will make a profit in the long run.
It’s so easy to use false logic here.
Looking at the positive results for a portfolio of draw systems in 2013 does NOT show that this system is profitable.
You have to start with the portfolio using level staking.
And as I showed the evidence is very clear. Such a portfolio is NOT profitable.
So now you are warned if you continue to bet on these draw picks!!
dmitri
Thanks for the explanation Dmitri. Good stuff!
DeleteSorry for mixing up algorithms and systems. Its a complicated service. You need to study a lot to understand it completely ;)
I pray to God this latest comment is not from a subscriber as if it is, I give up!
ReplyDeleteGraeme
Hi Martin.
ReplyDeleteYeah, I was a little sensitive about the start the Draw systems have experienced this season but like you say, there were not a lot of bets at all. Unique bets aren’t really the way to judge the draw systems (as Dmitri points out) but system D1-D6 is the core system and it has started the season poorly. I’m starting to think it’s whatever system I put into Cassini’s league that is the jinx!
Early days and hopefully things improve for the Draw systems but as I said in the post, the long-term results are OK and we are only into the 11th live month of results for the Draw systems. I’d like to see another season’s worth of bets before drawing any conclusions although Dmitri has already jumped the gun and implied these systems are useless.
We shall see but I still have faith that long-term, these systems will do OK.
Cheers,
Graeme
Dmitri,
ReplyDeleteWe all know that backing each ‘unique’ draw pick is not profitable and therefore, it’s not really a method to judge the Draw systems by. The draw systems D1 & D6 clearly contain a lot of dud bets that shouldn’t be there and this makes them very inefficient. Therefore, what we need to do is find a method to filter out these dud bets as they will clearly pull down the profitability of the systems.
We have witnessed it live last season that the way around the deficiencies on D1 & D6 is to ensure that you only bet on teams that appear on both systems. I would go as far to say that if a Draw pick only appears on one of the systems, I think it’s probably a lay or you use this information to find a Home/Away team to back but that’s for another day. I haven’t quite got my head around backing the opposite to what my systems tell me to but I know others do things like this.
Just because unique bets on D1 & D6 are not be profitable DOES NOT mean that D1-D6 will not be profitable. The level stakes profit on D1-D6 last season provides some evidence on the contrary maybe.
Graeme
I agree with most of what you say.
DeleteMaybe I was not clear enough in my posts and my warning against draw bets.
So let me give it another try!
When judging D1, D6, D1-D6 separately as three different betting systems it’s correct to use the numbers from the reviews. Because results are based on level staking.
The problem occurs when someone gets the idea to make a portfolio P1 of these three systems.
P1 will contain all unique selections from D1 and D6, but will stake these selections by variable staking, not level staking.
My point is, that to judge P1 it’s necessary to first calculate the results using level staking.
Is it possible that one of your members could get the idea of using P1 as his/her betting system?
Let’s take a look at the only numbers this member has about the three systems D1, D6 and D1-D6, the results from the only live season, 2013.
D1 made 17.54 units (3.49% yield) from 503 bets.
D6 made 3.37 units (0.63%) from 539 bets.
D1-D6 made 43.88 units (11.25%) from 390 bets.
I know that not all members can determine from these numbers alone if it will be a good idea to pool the three systems into a portfolio P1. Because I can not. Nor can you!
But the fact is, that P1 made -20.52 units (-3.14%) from 653 bets using level staking.
This number is not shown in the review.
You would show that P1 made 64.79 units (4.52%) from 1432 bets, because you use variable staking.
P1 is not profitable. The level stake result shows this very clearly.
Therefore the fine result for P1 using variable staking shall be seen as a lucky punch. In the long run P1 will lose money.
Let me add that for other systems it will be a good idea to make the portfolio.
It essentially depends of the correlation between the individual systems in the portfolio.
As for system D1-D6 it’s way too early to make a judgement of the profitability.
As of this date it has a yield of 9.28% from 419 bets.
If it can continue to make 9-10% yield it will need about 1400 bets before it qualifies.
dmitri
Your last paragraph is interesting imo.
DeleteSo much that I have decided to offer another analysis.
I will look at system D1.
This is the first draw system you constructed on your new draw algorithm.
I’m sure you have spent many hours on this work and you can only be satisfied to see it make 17.54 units (3.49%) from 503 bets in its first season (2013).
If you had stopped here, there would be no problem, no questions.
But you made another algorithm and from this a new system D6.
This gave you the possibility to split the 503 D1 bets in a GREEN group consisting of the 390 bets that also are in D6 and the 113 RED bets that are in D1 but not in D6.
The GREEN bets happened to be in profit, 43.88 units (11.25%) while the RED bets made a loss, -26.34 units (-23.31%)
Remember, both the GREEN and the RED bets come from the exact same algorithm.
If we assume that D1 will continue to make a profit going forward, and why should we not (?), then the most probable scenario will be, that the RED bets will do better and the GREEN bets will do worse in the future.
But the GREEN bets are the bets in system D1-D6!!
So this analysis raises a warning signal for D1-D6. Be careful.
We will follow these bets with interest.
So far we have 39 new D1 bets in 2014.
Not a great sample of course.
The RED bets are up 3.00% and the GREEN bets are down -17.28%.
dmitri