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 4^th 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 2^nd 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 3^rd 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 2^nd 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 3^rd 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

Established Systems 2013/14 Season Review

This is going to be the first in a series of reviews of the various systems from last season. I think the biggest change from previous Summers is the fact I intend to review a set of systems all at once, rather than individual systems or subsets of systems. I think the advantage of this is that it allows us to see all systems side by side for that particular group of systems and therefore, comparisons can be made quickly.

I’ll be doing 6 reviews in total:

Established Systems

New Systems

Misc Systems

Draw Systems

Euro Systems

Under/Over Systems

I think the Est Systems and New Systems are the most interesting for me personally given how much the systems struggled this season compared to last season. I expect these will take the longest time to produce and will probably contain the most words.

We then have the Misc Systems which also struggled over the season but I’d seen this before in their first season, so it wasn’t so much of a surprise. The Draw systems will be interesting to review as they seemed to outperform the underlying results by a long way this season which bodes well for the future I hope. The Euro systems had a much better season compared to last season although I expect the low draw strike rate may have helped these systems this season. The Under-Over system review will be a short review as these systems will be discontinued going forward.

  

Review of Established Systems for Season 2013/14

I think the obvious place to start is the results by system for the season. Here they are:

A profit of 23.3pts from 2,155 bets gives an ROI of 1.1%. The results by system were very volatile by system as quite simply, the filtering on the first algorithm didn’t work at all. System 6 made a 3% return which is a decent base to start from but system 7 took 148 of these bets and made a loss of 10.6%. System 8 improved on this a little and only lost 5.6%. A very disappointing season really for systems 7 and 8 considering the returns on system 6.

Algorithm 2 performed much better and a profit of 11.9pts (2.6%) on system 21 became a profit of 21.9pts from 94 bets on system 22. A highly impressive ROI of 23.3%.

Before I get onto the combined systems, it’s worth looking at the AH returns as I think this helps to explain the underperformance of system 7 and 8 this season. For systems 6,21 and 22, the AH returns are much worse than the outright returns and therefore, the draw didn’t impact these systems too much this season. However, it’s not the same story on systems 7 & 8. System 7 made a loss of 15.7pts if backing outright but if using AH0.5, it became a profit of 7.7pts! Less so on system 8 but the loss was extinguished at least.

A few people have asked over the past few seasons why I track performance of AH returns as well as outright betting but quite simply, it’s due to this reason. During the first season, the results of my ratings and systems dropped off a cliff at a point in time and I struggled to understand why. I guess this was before I really understood how the underlying results impact my returns and therefore, I decided to track AH returns. When your systems are making positive returns at AH0.5 and large losses for outright betting, it points to the draw being a pest.

Clearly, the draw has been a pest this season on systems 7 and 8. On system 7, 53 of the 148 bets this season have been a draw. That’s a strike rate of 35.8%. If placing 1pt on every game on system 7 to be a draw this season, it would have generated a profit of 38.69pts from 148 bets. An ROI of 26%.  System 7 would have been my 2^nd best Draw system this season and yet, my Draw systems have had a great season! :)

I guess the question everyone wants answered is…..why? Why has system 7 suddenly hit the skids this season? Well, I don’t have the answer to this I’m afraid. In a way, I would have preferred system 7 to have made massive losses for all types of returns and for the draw not to be to blame as then, it becomes quite easy to say that the filtering hasn’t worked well. However, when the bets aren’t winning but they are drawing, it becomes a very thin line between success and failure. 148 bets is not nearly a big enough sample and ultimately, that’s where the issue lies with system 7 and system 8 (and some of the higher combined systems). When you have so few bets, if you are impacted by a bit of bad luck, a few teams losing late goals who would ordinarily have won, it suddenly skews your results massively for a season.

I guess the best test for system 7 and 8 will be the overall results. 4 seasons of results is a much bigger sample and luck tends to even itself out over a bigger sample of bets. We’ll look at the overall live results later in the review.

Having seen these results for the single systems, we already know how the combined systems will look. Any combined system involving system 7 and 8 is going to be a mess this season. I think this is reflected in the results we see. System 6-21 didn’t have the best of seasons and only achieved a 2.9% return (disappointing considering the returns on 6 and 21) but system 6-22 obviously benefitted from system 22 having a good time of it this season.

I guess the issue this season lies with systems 7-21 thru to 8-22. All 4 systems made a substantial loss and therefore, this has to go down as a massive disappointment. It is actually quite scary that system 22 had 94 bets and a profit of 21.9pts but system 7-22 ended up with 49 of these bets and an 11pt loss! Clearly, system 7 had a massive influence on this and the AH returns for all 4 combined systems 7-21 thru to 8-22 were better than the outright returns.

Clearly, algorithm two had a good season, algorithm one had an OK season but the filtering between system 6 and 7 is impacted by draws and yet, this translates into a large loss on systems 7-21 thru to 8-22. This is the crux of the issue on these systems this season. The fact that some people followed all 4 systems at the same time means they have taken a small issue (too many draws on system 7) and compounded the issue MASSIVELY, resulting in crazy losses and a nightmare scenario.

We all need to learn from this and we need to ensure we don’t end up in a position whereby if one little thing goes wrong like a system hits too many draws in a season, it suddenly follows that we’re going to lose a massive proportion of our betting bank. I’m fed up using Steve (daily 25) as an example but his returns this season show what can happen if you follow a sample of bets whereby the returns and risks are cumulative.

I think if someone had offered me a 3% return on algorithm 1 and a 2.6% return on algorithm 2, I would have accepted this at the start of the season. It’s a great base to build from and after this, it becomes a case of trying to filter the best bets if you don’t want to follow all bets on system 6 or 21.

I think the hardest thing about the season for these systems has been the distributions of returns by month. Quite simply, the systems have been loss making from day one and that alone always makes things feel much worse. Here’s the returns by month:

A 92.4pt loss in September, followed by a 33.5pt loss in October meant the systems were down 125.9pts after two months. Not the start to the season you ever hope for! I think the fact the systems made a profit of 149pts from 1,553 bets over the next 7 months shows there isn’t too much wrong with the systems. I think the galling thing for me was that before March appeared, I was confident the systems would have had an OK season overall. However, March put paid to that and even though April was better than anyone could have expected, the season still was well below par.

If I had been offered this season at the end of October, would I have taken it? Yes, without a doubt. I was sure the systems would have made a loss this season at the end of October (even if I was putting on a brave face!) and therefore, I have to be happy with where it ended up.

I think the next step of this review is to put the season results into the overall live results and take a look at these. At the end of the day, this season is only one season out of four live seasons for these systems and therefore, if we want to draw any conclusions, we should look at all live results to date.  Here they are:

Given the results these systems have had this season, it’s somewhat reassuring for me to see the overall results again! I’m fed up looking at losses on these systems. :)

I guess what we want to ask ourselves here is if there are any major issues with the ratings, the filtering or the combined systems. I tend to ask myself that each Summer for each of my sets of systems.

If we take the ratings first, algorithms 1 & 2 now have returns of 4.5% and 7.2% respectively. Given this is over sample sizes of 2.6k and 2.1k bets, it’s safe to say I’m happy these algorithms have an edge. Algorithm two looks stronger than algorithm one though and to be fair to myself, I’ve been saying that since I first developed the algorithms. I learnt a lot from building algorithm one and therefore, I was always confident algorithm two would prove more profitable long-term. I think we have seen that come to fruition now.

In terms of the filtering, I think this is where there appears to be an issue. In an ideal world, what we want to see is that as we move from each single system to the next system, we see an improvement in ROI.

Probably easier to start with system 21 and moving to system 22. This looks perfect and clearly, the filtering between system 21 and system 22 is working very well. A 7.2% ROI on system 21 becomes 12.3% on system 22. This is exactly what I’d be hoping for and the fact it worked so well this season is just another sign that everything is working as expected.

If we look at system 6 to system 7 then, clearly, the impact from this season can be seen and what we see is a lower return on system 7 than system 6.  Looking quickly, we can see that system 7 is the only system where the AH0.5 returns are greater than the outright returns. On every other system, the AH0.5 returns are a fraction of the outright returns.  Clearly, system 7 is catching more than its fair share of draws.

3.2% ROI over 869 bets on system 7 is probably about as low a return as I would expect. I think this means we now have a massive doubt over system 7 going forward and ultimately, I would place system 7 (and systems 7-21,7-22) on the watch list for next season. 

Looking at systems 7 to 8 then, we move from a 3.2% return on system 7 to a 8.4% return on system 8. Again, I think this is highly acceptable for the filtering and there doesn’t appear to be too many issues here. My issue with system 8 these days is just the low bet volume and 419 bets over 4 seasons is not enough turnover. As we saw this season, all it takes is a few results to not go your way and you end up with a losing season.

Overall then, I think there is potentially an issue with filtering system 6 but the rest of the filtering appears to work as expected.

That probably leads us onto the combined systems. Does cross referring the algorithms add anything to the individual systems?

System 6-21 improves on the returns of 6 and 21

System 6-22 improves on the returns of 6 and 22

System 7-21 improves on the returns of 7 and 21

System 7-22 improves on the returns of 7 and 22

System 8-21 improves on the returns of 8 and 21

System 8-22 improves on the returns of 8 and 22

I think some reading this may look at the above and say that although the ROI improves, the lower bet volume means the combined systems don’t add as much as it appears and I would agree with this to an extent. It’s difficult to say whether 6-21 with an ROI of 7.7% over 1,751 bets is that much of an improvement on system 21 which has an ROI of 7.2% over 2,111 bets.

I personally take more solace from the fact that two algorithms agree with a bet although as we have seen this season and before, when too many algorithms and systems agree, it isn’t necessarily a good thing and you end up with diminishing returns or even negative returns.

I think whether or not someone follows system 21 or 6-21 or system 22 or 6-22 is up to them at the end of the day. I guess if people believe the 2^nd algorithm is much better than the first algorithm, then what benefit do we get of cross referring with an inferior algorithm?

I have deliberately started with the system results rather than the algorithm results for the season review. As most of you will be aware, I always started the monthly review with the algorithm results. I guess the change in tact is because most people followed the systems and didn’t make full use of the ratings and therefore, I didn’t want to concentrate too much on the ratings as people may think I’m deflecting attention from the systems. However, as I think we’ll see with the rating results this season, there are some positives to take from the season and it’s how we use these going forward that will determine the season's learnings I think.

Here’s the results for rating algorithm 1 this season:

I’ve been moaning all season about the fact it’s dangerous to draw any conclusions from the ratings with the small sample sizes and to an extent, the same is true here. However, given it’s a season review, I better put my neck on the line and try to draw some conclusions. We can add the data this season to the previous seasons and try to draw more concrete conclusions later.

Looking at these results then, I think the one thing that stands out is the poor performance of the bets with less than 5% value. I identified these bets last Summer as a potential weak link on the algorithm and we have definitely seen that this season.  I guess if I was trying to not be too harsh, I would say that the draw has clearly impacted these low value bets more than the other value bands this season but I guess that could be because they are low value bets.

I think the surprising performance is probably the 5%-10% bets. An ROI of 15.4% is well above expectations for this group of bets. 

As we move down the value bands (increasing in value), then the data gets quite thin and we small losses and profits in each band.  The outlier is obviously the highest value bets. These are never going to be a big group of bets as I don’t believe bookmakers really make these sort of errors too often but if my value ratings work as intended, the biggest profits long-term should fall in the 30%+ value band.  This season, we have seen a profit of 21.1pts from 14 bets for an ROI of 150.9%!

I think over a season, we can’t look too much into individual bands but I guess we hope to see a pattern of the lowest value bets making lower profits with the higher value bets making higher profits. We can see this on the first algorithm.

Here’s the results for rating algorithm 2 this season:

If the low value bets stood out on algorithm 1, they jump off the page here for algorithm 2. A loss of 32.3pts from 200 bets. Wow!  To think this algorithm made a profit of 2.6% and yet, there was a very simply way to improve the ROI massively by just ignoring 200 bets that appeared on this algorithm!

As we move up the value bands, we can see the data gets quite thin but the results look fairly decent. The top two value bands achieved a ROI of 55.7% and 44.6% respectively.

Overall, I have to say the results on algorithm two look better than algorithm one in terms of the value bands and the returns. I think the fact the low value bets are so loss making is actually a huge positive as it means my ratings can split the better value bets out from the poor value bets.

One thing I haven’t really looked at too much this season is how the value bets look on systems 7, 8 and 22. Here are the results on system 7:

As you would expect, the data gets fairly thin but the point is that the low value bets are still loss making but unfortunately, due to the filtering that takes place on system 7, many of the high value, high priced bets don’t make it onto system 7 and therefore, the results aren’t great.

Here’s the results on system 8:

The data is far too thin here but clearly, the low value bets don’t improve as we move up the filtered systems and therefore, the value ratings can’t really be used alongside the filtering of the ratings.  I think using the value ratings and the filtering of the ratings is a step too far based on the fact you’ll have no bets left to place!

Here’s the results on system 22:

I did find these results very interesting and clearly, filtering system 21 to 22 actually manages to remove the worst performing low value bets this season. However, it also removes most of the high value bets that were profitable on system 21 and therefore, similar to my last comment, I’m not sure using the filtering of the ratings and the value ratings together is a great idea.

I think to try to draw more concrete conclusions from the value ratings, we really need to look at all live results to date. Here are the results for algorithm one over the last 4 seasons in total:

I think the most pleasing aspect of these results (if you know what I mean) is that it is clear that the low value bets don’t really bring much to this algorithm. Over 4 seasons, the bets with a value of less than 5% have lost 20.8pts from 648 bets.

Here’s the same results for algorithm 2:

Same story again here with a loss of 8pts from 527 bets for the low value bets.

Looking quickly at systems 7,8 and 21, we have the following results:

All systems experience  a loss for low value bets overall. Therefore, the obvious place to turn to now is….what would have happened if we just ignored the bets where value was less than 5%?

Here are the results by season and the overall results for the 5 single systems by season if we exclude all bets with a value of less than 5%:

To end this analysis, here’s the results by system over the first 4 seasons for all bets and the equivalent table for all bets exc bets where value is less than 5%:

Overall then, by removing all bets where there is less than 5% value, it reduces the bet numbers by 22% and increases the overall ROI by 46%.

What I haven’t done is bother recalculating the returns for the combined systems after removing the bets with less than 5% value but clearly, the returns would improve on all combined systems too.

I guess the obvious solution after seeing this analysis is just to remove the bets with less than 5% value from the rating algorithms. I’m not a fan of this as at the end of the day, if you start reducing the number of bets on the ratings, we can’t tell how good the ratings are at finding value or understanding the returns of the value bands as we’d have no lower value bets to compare to. I think the low value bets present a good control group to judge against.

I do think there is clear evidence that if people want to improve their ROI and reduce their turnover, they can remove lower value bets from the systems they follow. Of course, it may be the case that it’s only Home/Away bets that create this loss or it may be certain leagues where the ratings don’t seem to be as powerful but anyone who can manipulate data in Excel can look into these things.

Here’s some additional tables looking at performance by league by season and also splitting all results by season and Home/Away bets.

Again, I’m not going to overanalyse this. My own view is that performance by league is volatile and a few people are still licking their wounds from dropping League Two Aways this season and looking at the results, I can see why!

In terms of Homes and Aways, I think I’ve covered this before that Away bets have a higher ROI and therefore, are the most profitable bets long-term but given the higher average odds, there is more volatility in these bets. What is interesting is that the Home bets last season had a nightmare season overall on the combined systems but they actually started the season very well and helped to minimise the system losses as Aways created a massive hole early on in the season. 

As the season progressed, the returns switched and Homes went from being profitable to being loss making on these systems. Again, it’s a little surprising that Homes were loss making in a season when the Underlying results were favourable for Homes but it is due to the type of Home bets that were profitable. Odds on homes were very profitable but the bigger value homes (odds of 2.50+) were badly loss making and it’s these home bets that tend to make their way onto my ratings fairly often and therefore, I didn’t catch as many odds on Home winners as I would have liked.

I think the way to end these season reviews is to show the overall live results, the betting banks for next season and the targets that the systems will be aiming for next season. Here are these numbers for the Established Systems:

A lot to take in here but I think it’s interesting to concentrate on the relationship between ROI and ROC. As I’ve discussed lots of times now, ROI is not as important as ROC and I think as time is going on, we are seeing that more and more with these systems. System 6-21 has the lowest ROI out of the 6 combined systems but has the highest ROC achieved per season. I think the impact of this season is also now seen on the returns on systems 7-21 thru to 8-22 as well as systems 7 and 8. The target ROC for all 6 of these systems next season is only 14% on average which basically puts these systems at the lowest end of the target ROC for next season out of all my systems. I think unfortunately, unless we see a massive upturn in performance next season, these higher combined systems aren’t worth tracking any longer even though they have a very good ROI. They don’t have enough bets to generate enough of a ROC. In other words, they carry too much risk for not enough return.

Overall then, I think that completes the review of the Established Systems this season. A lot to take in and ponder here and probably worth reading through the results and making your own observations if you are going to use these systems next season.

In summary:

•             Algorithms one and two were profitable again for the 4th season in a row although returns were lower than previous seasons due to more difficult underlying results

•             The filtering on algorithm one didn’t work well from system 6 to system 7 and created the losses on systems 8 as well as the issues on systems 7-21 thru to 8-22

•             A higher than expected number of draws on system 7 was basically to blame for the underperformance of the Established Systems this season

•             The combined systems did appear to improve on the single system results again although it is debatable about how much value cross referring the algorithms created

•             The value ratings have been useful in the sense we can clearly identify the fact that the lower value bets on each algorithm were loss making last season and have created no profits over 4 seasons now

•             Filtering the systems doesn’t work as well as the value ratings and using both the filtering and the value ratings doesn’t leave too many bets on the systems, so it points to using the value ratings as a way to filter the bets on the algorithms rather than using the systems to filter the bets

•             It is fair to assume that removing low value bets from the combined systems would also improve their results although filtering the systems and using the ratings isn’t going to leave too many bets to follow!

•             Performance by league last season has highlighted how difficult it is to draw conclusions from small datasets and whereas League Two aways were loss making before this season, that is no longer the case

•             Homes underperformed over the season even though it was ‘easier’ to make money backing Homes this season – the profits were due to odds on bets being very profitable but my ratings tend to exclude many of these low bets

Underlying Results for 2013/14 Season

One of the best additions to the service IMO this season has been the work I’ve done on Underlying Results. I probably have another big piece of work to do on this for the Euro leagues but I can hopefully tackle that later this Summer. I think understanding the relationship between the underlying results and the system results is important due to the fact it allows us to take a more objective view of how the systems are performing.

At the end of the day, we all know the systems are judged on their P&L and this is the only way the systems could be judged that is meaningful. However, when you have as many system bets as my systems have each season, being able to understand why we’re seeing some of the results we’re seeing is important for a couple of reasons. Early on last season, I was at a loss to explain what had happened in September 2013 as quite simply, the performance of the Aways bets were so poor and although it was offset a little by the improved performance of the Home bets, due to the make up of the bets, it resulted in a substantial loss on all 4 algorithms. It’s easy with hindsight to look back now after the work I’ve done and say it was simply due to the below performance of the Away bets and it mirrored the results in Feb-12 or Apr-11 but at the end of September, I was pulling my hair out along with everyone else.

The same can be said for the Mar-14 results although having been through the pain in September, going through it again so soon after wasn’t quite as bad as going through it the first time round.

I guess the point I’m getting at is that by tracking the underlying results, it allows us to understand why some of the trends with the system results may be occurring. It doesn’t change the results and if I was some rouge tipster trying to explain away lack of an edge, I think this info is maybe dangerous in the wrong hands but given the fact we know my ratings have an edge long-term, the underlying results allow us to understand some of the variance in the results we’re seeing.

I personally have loved the fact I’ve been tracking these results since the turn of the year. I actually get as much fun updating the underlying results as I do the system results now as I want to try to understand if my results are being influenced by the underlying results. Some weekends when there are a flux of away winners on the systems, I love looking at the underlying results and seeing that I managed to hit an amazing weekend as there were very few away winners overall. On the other hand, I do get frustrated if I see a massive bias towards away winners and we’ve been on the Home bets or we’ve missed all the nice Away winners.

Going forward, I will probably try to send the underlying results out every month with the monthly reviews. As I said in the intro, I probably need to do a piece of work for the Euro leagues too but this can wait until later on this Summer or I can try to do it during the season next season. I’m more relaxed about the Euro leagues tbh although with the Euro Draw systems next season, I may get more interested!

OK, so I think that’s an overview of why I’ve started looking at the Underlying results. I’m not going to repeat everything I’ve said in the previous posts on underlying results as I think we all know how the season has panned out but I’ll go over some of the highlights again now we have a full season’s worth of results.

Here’s the underlying results by season and for the 4 algorithms:

The Underlying results show that Homes have had their best season so far in terms of ROI, Draws and Aways have had their worst season.  Overall though, the season is only the 2^nd toughest as 2011/12 had a larger loss overall. I guess this season has maybe felt worse as the Aways started so badly in September, improved throughout the season and then had a massive loss in March before rebounding again in May. Hence, the best results the Aways have had all season is at the end of the season. I think this makes the season feel worse than it ultimately was.

Looking at the results of the 4 algorithms then, all 4 algorithms made a profit. I was facing up to the fact at the end of March that it was likely to be the first time over a season that I’ve had a losing algorithm but the results over the last 6 weeks of the season were very good and ensured the season ended in profit on each algorithm. I’ve said this before by my starting point each season is to hopefully make a profit on each algorithm. Each algorithm has around 500-900 bets a season and therefore, to achieve a profit over this sample of bets is no mean feat. The fact each algorithm has managed it since its inception shows the edge the algorithm has.

In terms of the returns, algorithms 1,2, and 4 had their worst season to date which fits in with the underlying results. Algorithm 3 bucked the trend and to make an ROI of 7.6% this season is an outstanding performance. Overall, a respectable performance from the algorithms given the Underlying results.

Here’s the breakdown by month of the same results:

Not going to spend too long on the monthly breakdown but I think the pleasing thing for algorithms 1 and 2 is that during a very tough season, the only losing months were September and March which were basically impossible months. Algorithm 3 somehow escaped September’s nightmare but then got caught up by the January losses and also had a strange May month too. Throw in March and that’s 4 months with losses greater than 10%! However, with the 29.7% return in February and an amazing 44.3% return in April, the overall season was great! A very strange season for this algorithm but at the end of the day, with the profit achieved, I can’t be too disappointed. Algorithm 4 had nightmares for 3 of the months of the season and then an amazing April and May to finish very strongly.

Hopefully some of you reading this will be able to look at your own P&L this season and will be able to validate their results this season with the underlying results by month and start to make a little bit more sense of what happened to their P&L this season and why.

Here’s performance by League:

Given the issues I’ve had with League Two Aways over the years I’ve been doing this, my systems were maybe due a run of League Two Away winners? Not sure tbh but I feel bad for anyone who dropped League Two Aways this season as it was the easiest league by far for Away winners and my rating results reflected this. The SPL is a league my ratings can’t get a handle on and it was another tough season for my bets in this league. Again, not helped by a poor performance from the Aways in the Underlying results.

Bsq Prem has been a nightmare for Aways this season and my Away bets have taken a battering in this league this season. It has been well document by me and others that the Premiership Draws had a very low strike rate this season but it was also the case League One, Bsq Prem and SPL too. Hence, when we come to review the Draw systems this season, it will be interesting to see which leagues generated the profits. I expect League Two had a good season on the Draw Systems.

I think although the underlying results don’t appear too bad this season, it masks the fact that the bias towards certain types of bets is against my ratings.

Here’s the analysis for all bets based on an odds range:

As we know, the majority of my system bets fall into the odds range of 2.51 or greater. Therefore, the bias we have seen this season towards shorter priced teams has actually impacted my ratings and systems a fair bit. As you can see, if blindly backing all bets (Home & Away) priced at 2.51 or greater, you would have lost 6%. Quite simply, Home underdogs really struggled this season, along with Away underdogs and therefore, for anyone running a selection of algorithms that specialise in finding underpriced teams, the results are never going to be great. The fact that my ratings made a profit in this odds range this season shows the edge they have over the market.

Overall then, I think that’s enough of an update on the Underlying Results at the end of the season. I feel like given I’ve talked through it a few times on the blog already this season, it can become quite repetitive. I’ll keep the analysis of the Underlying results going next season and we’ll see if it helps us explain the results of the ratings. I expect it’s a useful analysis tool when things go wrong but when things are going our way, I’m not sure me sitting here telling everyone we’re being lucky is going to go down too well!