Friday, 21 June 2013

Introduction to the TFA Draw Systems

Well, here goes, a first look at the new TFA draw systems. Not going to go into any great depth about the rating building element as I don’t think anyone really cares too much about how I build the ratings and systems these days! All I can say is that the methodology is in line with the way I’ve built all the UK systems and I would hope that these systems are as robust as the other UK systems.

The first question someone asked me earlier this week about these draw systems was how confident I was in these draw ratings compared to the Euro ratings I built last Summer and I’ll restate my answer here. 

Last Summer, I had lots of reservations about the Euro ratings due to the fact I was playing in leagues that I didn’t know anything about, hadn’t built anything before in these leagues and therefore, my confidence wasn’t as high as it was usually.  I’ve now built draw ratings in the UK leagues where I’ve never had an unsuccessful set of ratings, never built a system to make a loss, have 3 seasons proofed results behind me and therefore, I have to feel more confident than I felt last season about the Euro systems.

If there is one thing I’ve learnt over the past few years it’s that my confidence in my system building is usually a good indicator of the potential success of the systems. Admittedly, in a season, it may not shine through as well as I hope (TOX, STOY and STOZ in their first season springs to mind!) but as we saw last season, my confidence in these systems wasn’t misplaced. Given the ratings look fairly strong I’ve built during backtesting and the fact the systems I’ve built look like they tick all the boxes I look for, I’m much more confident in these draw systems than the Euro systems.

Am I as confident in the draw systems as I was in the Home/Away systems? Well, no. Simply, I don’t know how easy it is to find an edge on draws and given the fact there aren’t lots of people running algorithms in the UK leagues on draws, my concern is similar to the Euro systems in a way. If it was easy as it appears to back draws in these leagues and make a profit, surely everyone would be doing it?

In terms of my advice regarding these systems, it’s much in line with my advice regarding all my systems over the last couple of seasons. Why would anyone choose to follow a new system with no proofed results where there are another 30-40 systems with great results over a couple of seasons and a proven edge?  Of course, I want people to pay attention to the systems and I’m sure a few people will follow with small stakes but if given a choice between following a system with 3 season’s live results and no live results, you have to go with the proven system.

I’ve deliberately not included the 2006/7 and 2007/8 results even though I have them as the data for these years was all used in the backfitting of the ratings. Season’s 2008/9 and 2009/10 also won’t be representative of what to expect from these systems as a random selection of 50% of the results in these seasons were used in the backfitting of the ratings. So, the final 3 seasons of results should be the most reliable but as I always say, backtesting doesn’t mean a thing when it comes to the future results. One season’s live results is better than 10 season’s backtested results!

So, here’s the results by system and season for each of the systems. 

I think you can see the impact in the first two seasons of including some of these games in the backfitting of the ratings as the returns are a good bit higher than the last 3 seasons which are all 100% backtested.

The most interesting thing has to be last season’s results. The base ratings did really well (D1 and D6) but the filtering of the systems didn’t go well at all. So much so that as you work down the ratings, the opposite trend occurs to what you would expect! The ROI reduces as the bet number reduces. 

Obviously, when I’m testing the ratings and the results for profitability, I look at the base ratings and it’s then only at the final stage when I build the prep systems which are then used to build the final systems you see here that I see these sort of things appearing. I guess the next thing that leads me to want to look at is the performance of the systems for these seasons combined to see if the same pattern emerges.

Now, this is much more like what I expected to see overall. As you filter the bet number, the ROI increases across all the systems. The most interesting aspect here is the fact that some of the higher combined systems have phenomenal ROI’s. Therefore, it would appear that last season was a bit of a blip and the underlying trends are all OK.  Those reading this who remember the backtesting for the New Systems last Summer will remember that some of the combined systems were loss making in their final backtesting season and yet, they produced amazing results in their first live season last season.  You can’t read too much into one season of backtested results I think.

Here are the results by month:

One thing that stands out here is the fact when a bad month arrives, it does tend to be quite bad! Likewise, the decent months look very good and therefore, these systems may be quite volatile on a monthly basis. It makes me chuckle a little looking at the results in the first 7 months of results. An average of over 100pts profit a month. What we’d give for these systems to start like this when they go live!

The results by League are:

Now, we’re talking! As most subscribers and readers of this blog know, my ratings have had real issues with League Two bets, especially League Two Aways. I had picked up the fact that losses are minimised or even a profit created by covering the draw (basically dutching the selection and the draw or backing the selection AH0.5) but this table starts to get my brain ticking over. Likewise, the Bsq Prem hasn’t been the best league for me at times and both leagues appeared to be affected by the draw with regards to my other ratings. Well, both these leagues are key leagues for these new draw ratings and therefore, I would expect a lot of overlap between selections from my other ratings and these new draw ratings in these leagues.

The ROI’s of over 26% across over 2,500 bets in each league is pretty good and I can’t recall too many leagues having such an impressive set of figures during backtesting in the UK leagues for my other ratings.  It will be very interesting to see how these leagues do next season with these new draw ratings.

Here’s the league performance by system:

Plenty to like about these results too I think and the systems look decent enough across all the leagues with no real issues jumping out at me.

In terms of next season, I have set some targets as I always do for the systems but with no live results, the expected bet number could be way off, the ROI could be way off and therefore, the ROC targets could be way off.  Always nearly impossible to set targets and based on past experience, I wouldn’t be surprised if the bet number on the base ratings D1 and D6 are much higher than backtesting as we’ve seen that with the other systems when they went live.

Here’s the targets:

Not going to comment too much but I think my concern with the higher combined systems is the lack of bets and it’s a general point about all the combined systems. There aren’t a lot of bets overall, which means it will take longer than a season or two to get something statistically significant and therefore, variance could impact these systems. I suspect looking at the results from last season, the higher combined systems just suffered a little from variance as they clearly had a massive edge in every other season, so it’s just bad luck they only did a little better than break-even last season.

Assuming the ratings work like I hope they too, D1, D6 and D1-D6 probably have the potential to maximise ROC a lot more than the other systems simply due to the expected bet number but how much you give up in terms of ROI is the key.

In terms of how we use these Draw systems, my idea is that by using a few of the draw systems alongside some of the other systems, it may allow you to smooth the P&L a little as traditionally, when my main systems struggle, it’s usually due to the draw % being higher than expected and higher than the long-term average. Of course, I don’t envisage anyone following an equal number of outright and draw systems as you may end up offsetting the profits/losses and not making much headway overall but importantly, the draw systems are built independently of all my other ratings and appear to have their own significant edge.  Therefore, you can hopefully follow a few outright systems, include a draw system or two and improve your overall P&L whilst at the same time, minimising the drawdowns overall. 

That’s the hypothesis and the reason I’ve spent 3 weeks on these ratings and systems. Will it work? No idea really but we’ll learn more next season.

Next project for me is the value ratings work I started a few months back. Hoping to complete that in the next week or so before I head off on holiday for a bit. I’ll provide an update on my progress next week at some point.



    Hope you are keeping well.

  2. Hi Cassini.

    I’m good. Busy as always but between my work and football analysis and betting, not time for much else these days! Hope you are keeping well and well done on another successful season with the draws.

    I had read your blog post before I went to bed last night and I was planning on leaving a comment today. Will go and do that now before I forget!