Monday, 26 May 2014

New Systems 2013/14 Season Review

Let’s start with the results by system this season:

A bit of a mixed bunch here and depending on which systems you followed, you could have experienced very different returns to others this season.

I guess the starting point is system 31. Going into this season, this would have been one of the systems I’d have most confidence in and thankfully, it didn’t let me down. A profit of 56.9pts from 753 bets.  An ROI of 7.6%. It hasn’t been a smooth journey for the system this season but at the end of the day, the results are in line with previous seasons.  With the suggested bank of 80pts, this gives a return of capital of 71% which is a great return over the season.

Not for the first time, the filtering of system 31 didn’t go to plan and system 32 made a much smaller return. Unfortunately, things didn’t go to plan on system 33 and a loss was made. I think we have to keep asking ourselves this as we move through the reviews but if we know system 31 is a great system, why do we bother trying to filter it with system 32 and 33? What do we actually gain from this?

I think this season has again showed the pitfalls of filtering systems I expect and systems 32 and 33 don’t bring anything to the service. System 33 in particular doesn’t have enough bets and when you see the expected ROC targets, you have to wonder what the point of this system is.

System  41 struggled for most of the season and algorithm 4 is not as good as algorithm 3 and this has been apparent over the last two seasons. However, by season end, the algorithm had recovered from the terrible start to the season and achieved a profit of 26.7pts from 905 bets. An ROI of 2.9%. Again, the filtering from system 41 to 42 didn’t work well at all and system 42 made a loss of 14.5pts from 426 bets.

As you would expect with these single system results, the combined system results follow a similar pattern.

System 31-41 achieved a profit of 41.9pts from 618 bets. An ROI of 6.8%. I guess the question we have to ask ourselves here is why bother cross referring system 31 with system 41? We know system 31 is a great system, system 41 is not as good and when we cross refer 31-41, it filters the bets but doesn’t necessary improve the returns. For much of the season, system 31-41 struggled as system 41 was struggling whereas system 31 was doing great.

System 31-42 managed to make a profit over the season but this system struggled for the most part. Again, with system 42 making a loss, it’s great that 31-42 sneaked a profit but again, why bother cross referring system 31 with system 42 if 42 is derived from 41 which isn’t as good a system as 31?

Once we get to 32-41 thru to 33-42, we see the same issues the Est Systems had. Systems 32-41 and 32-42 made small losses but the top two combined systems had a nightmare season. Looking at the AH returns though, we can see that the AH 0.5 returns on systems 33, 33-41 and 33-42 were better than the outright returns. Therefore, system 33 suffered from too many draws and this resulted in the mess on system 33-41 and system 33-42.

As an aside, for those that followed systems 7-21 thru to 8-22, it is highly probable the same bets that caused the issue on these systems as on these New Systems. Hence, if following all these combined systems in a portfolio, a small increase in the draw strike rate has resulted in a massive loss to the portfolio of bets. This explains Steve’s issues this season I suspect.

Overall, the 11 systems made a profit of 92.6pts from 3,869 bets. An ROI of 2.4%. A poor performance but the filtering was 100% to blame. Systems 31,41 and 31-41 made a profit of 125.5pts from 2,276 bets. The other 8 systems made a loss by trying to filter the bets.

I think it’s worth looking at the split between Home and Away bets on these New Systems this season.  It’s very interesting.  Here’s the results:

As you can hopefully see, the filtering issue is on the Home bets and not the Away bets. On system 31, Home bets made a profit of 4.8% from 258 bets. System 32 made a loss of 16.4% from 64 bets and  system 33 made a loss of 54.9% from 28 bets. Only 6 winners from 28 bets at an average odds of 2.14!  As you can see, AH0.5 made a loss of only 2.6pts and therefore, draws have been a massive impact on system 33 for Home bets.

On the 4th algorithm, it is actually quite scary how bad the filtering is on the Homes!  System 41 made a profit of 11.5% from 317 Home bets. A stunning performance. System 42 took 168 bets though and made a loss of 3.2%. That’s a massive swing.

I think the combined systems are very interesting for Home bets. System 31-41 made a profit of 10.3% on Home bets. However, as you move up the combined systems, the returns get progressively worse.  It is quite amazing at how bad the filtering works on these Homes but to be fair, it is Draws causing it and for whatever reason, the filtering did a brilliant job at picking out Home bets that went onto Draw rather than win.

If we look at the results by month, I think it outlines the way the season panned out for these systems. 

A massive loss of 87pts in September was recouped by the end of December and things looked back on track. January was then a small loss but February was a great profit and at the end of February, it looked like these systems were on course to have an OK season which would have been some result given the losses in September. March then appeared and rewrote the record books for losses and a loss of 204pts meant the systems were facing up to their first ever losing season.

April then came along and rewrote the record books for profitable months and a profit of 157pts recovered the position somewhat before the season ended with a small loss. If you have your worst ever month by a long, long way, followed by your 2nd worst month in the same season, then to post any sort of profit is amazing. The season felt tough as it started badly and just when it looked like it would turn out OK, March came along. Had easier seasons!

Let’s try to put the season results in some form of perspective. Here’s the results for the first 3 seasons for systems 31-33:

I think it’s fair to say that we don’t see any benefit of trying to filter system 31. System 31 can produce a very decent return over a large number of bets and as we try to filter these bets, the returns don’t improve at all. I think it’s interesting how consistent the results are across these 3 systems and therefore, I think this points to systems 32 and 33 as having large question marks against them now.

If we split the results by Home and Away for these bets, we get the following picture:

I think this is quite alarming if I’m honest. The filtering of the Aways works perfectly and we can see an increase in return as we move up the systems. Unfortunately, the Homes show the opposite trend and as we filter the Homes, we see things get worse. Clearly, the Homes that appear on systems 32 and 33 aren’t as good as the bets that don’t appear on these systems which also sit on system 31. 

Again though, Draws are to blame. The Home bets that appear on system 32 and 33 are profitable if using AH0.5 but all Homes on system 31 make a loss to AH. As we filter the Home bets, we seem to pick up a helluva lot of draws! The Home bets are much more profitable if backing them to draw than backing them to win on systems 32 and 33.

Here’s the results for the last 2 seasons for all the New Systems then:

I think over the last two seasons, it’s hard to be too hard on the New Systems tbh. The first season was an amazing season and last season was a disappointing season even though it was profitable. Adding the two seasons together and we can see the results don’t look too bad at all.

I think the question that’s quite difficult to answer is what do the Combined Systems bring to the party?  I think if we look at it closely, they don’t bring too much unfortunately.

System 31-41 doesn’t improve on system 31
System 31-42 doesn’t improve on system 31
System 32-41 doesn’t improve on system 32
System 32-42 does improve on system 32 and system 42
System 33-41 is in line with system 33 (only 3 bets on 33 that aren’t on 33-41)
Systems 33-42 doesn’t improve on system 33

Overall, it appears to me that trying to filter system 31 or trying to cross refer the algorithms doesn’t improve upon the results of system 31. Therefore, I’m not sure what is the point of systems 32,33,42 or any of the combined systems!

Looking at the ROI on the combined systems, it’s hard to sit here and say they are all a waste of time but ultimately, they are. If we all just followed system 31, we’d be much better off!

I’ve already said I’m going to give all systems another go next season but I expect if we see no improvement in the filtering or the cross referring, we’ll retire a bunch of these systems at the end of next season.

So, if we can’t use the systems to filter the bets, can we use the value ratings? Well, here’s the results this season for the 3rd algorithm (System 31):

Unlike the Est Systems, there doesn’t appear to be an easy conclusion to draw here from looking at the value bands. 2 of the top 3 bands make a small loss and the low value bets appear to be profitable here. Given the high bet number that appears in the low value band, you could increase your ROI to 10% by ignoring these bets with less than 5% value but I think this is risky.

Here’s the same results for system 32:

Again, difficult to draw concrete conclusions due to the very small sample sizes.

System 33:

Given the difficulties the system has had this season, not sure these results mean too much.

Overall then, on algorithm 3, it’s difficult to draw conclusions from the value ratings. I’m not sure there is an easy way to reduce the bet number and increase the ROI if using the value ratings.

Here’s the results by value band for algorithm 4 (system 41):

This is much more interesting for me. Both the lowest bands are loss making and therefore, it points to something similar to the Est Systems. If we exclude all the bets with less than 10% value on this algorithm, we’d have made a profit of over 15% last season. Not bad!

Here’s the results for system 42:

This really jumps off the page here but we could have missed off the bets with less than 10% value on system 42 and turned a big loss into a big profit. Over a 20% ROI for all bets with more than 10% value last season.

I think the next move then is to look at the value ratings for all live results.  We may be able to make more concrete conclusions on bigger sample sizes.  Here’s the results for algorithm 3:

Amazing returns for the highest band here but a little disappointing that the 2nd highest band is loss making to be honest. The other 5 bands are profitable though and I think it’s difficult to draw any conclusions. It would be folly to try to filter bets by the value ratings I think and you’d run the risk of missing out profitable bets.

Interestingly, if I look at system 32, we see something interesting:

We can see that when the lower value bets make their way onto system 32, they don’t tend to do well. A loss of 14.2pts from 141 bets isn’t great. It’s a small sample size and therefore, I wouldn’t like to jump to too many conclusions but this is quite interesting I think given the issues system 32 has with filtering system 31.

Here’s system 33:

More difficult here I think and 3 bands are loss making with the other 4 in profit. I wouldn’t be trying anything here to filter the bets.

I think all we could possibly do on algorithm 3 is remove the bets that appear on system 32 with less than 10% value but even then, with only 141 bets, it’s not really concrete proof of any issues on this algorithm with the value ratings.

I do think we can draw much more concrete conclusions on algorithm 4 though. Here’s the results over the last 2 seasons for system 41:

A 55.2pts loss from 661 bets. It’s not concrete proof (we’d need a bigger sample) but it is a helluva big pointer that the low value bets on this algorithm aren’t worth following. Algorithm 4 is the weakest of my 4 algorithms but seeing these results, I can sort of understand it. If we simply removed all these low value bets, we’d reduce the bet count by 1/3 (bringing it more in line with my other algorithms in terms of bet numbers) but more importantly, we’d increase the ROI from 3.6% to over 10%.  Hence, this algorithm would jump from my worst algorithm to my best algorithm!

Again, I need to question myself whether or not these low value bets should be removed from the system and the service completely for algorithm 4. I’m inclined to wait another season for more proof although by highlighting this now, people can use the information if they wish next season.

Here’s system 42:

On this system, I would go as far to say that dropping any bets below 10% value may not be a bad idea. Both low value bands make a loss on this system.

Overall then, I think it’s easier to draw conclusions on algorithm 4 than any other algorithm. I think we have to question whether or not the low value bets on algorithm 4 are worth following at all.  making losses over 2 seasons doesn’t mean there is no edge but it’s hard to believe there is too big an edge considering they are meant to be the lowest value bets anyway and therefore, to see them losing so badly means it’s difficult to see them ever making a 5% return if I’m honest.

Similar to what I did for the Est Systems, here’s the results by League by season and also a table showing results split by Home and Away by season. These are for information only and anyone wanting to analyse these will need to download the results from the site.

OK, so I think the final part of the review is to try to look ahead now to the future.  This table shows the live results, the betting banks and the targets for next season. I know it’s fairly obvious but if anyone wants to remove low value bets or do any tweaks to simply staking 1pt on every bet on a system, they will need a bespoke betting bank and will have their own targets which vary from what I’m showing here.

I think the one thing that is really standing out for the New Systems now is the fact that once you move away from system 31 or systems 31-41 and 31-42, I’m not sure there is much else worth following here.  All the other 8 systems have ROC targets of less than 32% and IMO, this highlights the fact that there are a lot of systems here that carry too much risk for the return they provide.

I wrote earlier about the fact that system 31-41 doesn’t improve upon the return of system 31 but the advantage that system 31-41 brings is seen by the ROC both systems have achieved. The average drawdown on system 31-41 is a little lower than 31 and this manifests itself in a lower betting bank. Over the last two seasons, system 31-41 has achieved a ROC of 69% on average against system 31 over the last 3 seasons with an average ROC of 66%.

System 31-42 has done great over the last two seasons with an average ROC of 53% but beyond this, there are too many impressive ROCs.

I think this completes the review of the New Systems this season. A lot to take in again but I think there are some clear conclusions we can draw on these systems now if we are going to use them next season.

In summary:

·         Algorithms 3 and 4 were profitable again this season even though the underlying results were a bit more difficult than previous seasons
·         System 31 was the star performer again for the 3rd season in a row and has achieved a 197% ROC over the first 3 seasons in total which is very impressive
·         The filtering between 31 and 32/33 didn’t work well at all this season and we have to question what systems 32/33 bring to the service
·         System 41 had a profitable season even though all bets with less than 10% value were badly loss making
·         Looking at the last two seasons, bets with less than 5% value on system 41 have lost 55pts from 661 bets
·         System 42 follows a similar pattern to system 41 and the low value bets are loss making
·         Systems 31,31-41 and 31-42 have respectable ROC targets going forward, the other 8 systems have very low ROC targets


  1. Hi Graeme,

    Could you provide the stats for each independent selection to 1pt level stakes for the season? If you could then provide the same data broken in to home, away and draw and also with and without the less than 5% value selections? If you could provide the same data for previous seasons and all time that would be great to. I really believe readers would like to see these stats. If you can generate 500 bets a season at 3%+ ROI for any length of time you can stop right there. That would be 3% using Pinnacle, Exchanges (after commissions) and Asians.



  2. Matt

    Very similar to the strategy Ive employed over the last 3 seasons.1pt level stakes on every unique bet on each algorithm.All bets placed at betfair,sbo or pinny. Bets placed 1hr to kick-off.

    Returns vary by season but easy enough to make 5% to 7% return on each algorithm.Play one bank for all algorithms and easily managed to grow bank by over 75% for last 3 seasons with very little trouble.

    Dont read this blog too often as dont agree with graham's approach to filtering bets but anyone not making money with these ratings should not be betting.


  3. Thank you very much Greg for posting. I bloody knew someone else got the idea maybe even everyone does, I don't know. Stakes are easily scalable this way especially leaving it to 1 hour before kick off for the odds rebound. The season I subscribed to TFA I spent a while every week "highlighting duplicates" in excel etc etc. One simple list, throw in a piece of best price broker software (maybe with a few more of the Asians added in as well) even and it is 1 hours work for solid gain and you will be able to get stakes ranging from £2,000 up to £10,000+ (much larger still if Graeme can figure out a handicap model) across the various leagues. I ran my own model alongside Graeme's and ended sticking with mine (which is optimized for Asian Handicaps) but his still had great results and with no fuss at all and a fairly smooth diversified ride. Hopefully he will add a system which is just the unique bets and all can see just how good the real benchmark is (I am still interested to see the numbers with the sub 5% value bets stripped out).



  4. Hi Matthew.

    Apologies for the delay in replying, been busy the last day or so.

    As I’ve discussed on the blog before Matthew, it’s not an easy analysis to do as with each algorithm having its own value ratings, you get bets that are less than 5% value on one algorithm but big value on another. The only way to do it properly is to treat each algorithm individually. If I do this, the results are actually in the first two reviews but I’ll restate them again.

    All Bets

    Alg 1 – 2,620 bets, 4.5% ROI

    Alg 2 – 2,111 bets, 7.2% ROI

    Alg 3 – 1,486 bets, 8.9% ROI

    Alg 4 – 1,847 bets, 3.6% ROI

    Bets with value > 5%

    Alg 1 – 1,972 bets, 5.9% ROI

    Alg 2 – 1,584 bets, 10.1% ROI

    Alg 3 – 1,545 bets, 8.5% ROI

    Alg 4 – 1,186 bets, 10.3% ROI

    These are the results for all live bets Matthew. Easily ticks the box you suggested in your post. I guess these returns might explain why TFA was voted number 1 footie tipster last year and yet, Steve (Daily 25) lost £40k this season! Cherry picking bets is a dangerous game but as I’ve come to realise I think, I’m not sure I need to do much filtering looking at the above returns.

    Cheers for the comment again.


  5. Hi Greg.

    I thought you’d given up on TFA mate, not heard from you all season. Hope things are well?

    There is no right or wrong way to play TFA as I’ve said before to you and many others. Before last season, I don’t think anyone had ever lost with TFA since every system was profitable every season but I now appreciate that there are ways to minimise the risks of making a loss and still achieving a decent return. The season past has changed my way of thinking to something much closer to what yourself and Matthew have suggested for a while, along with a number of others including Tage.

    Interesting that you now place your bets 1 hour to kick-off. How do you find the odds at this time and do you track what % of my quoted odds you achieve on average? I placed my bets last season on a Saturday morning mainly due to the higher liquidity and the odds rebounding a little. Would be keen to hear your experience?

    Drop me an email mate or reply on here.


  6. I would like see a breakdown of value ratings for homes and aways respectively for all 4 algorithms.

  7. Hi Anon.

    You can download the historical results from the website and look at it yourself if you can manipulate the data. Otherwise, I’ll look at this for you later this Summer when I have more time. I’m working on the next review whilst updating some ratings at the moment, so don’t have too much time for one-off requests. Remind me later this Summer and I’ll do it for you.