With the International break, I decided it was worth taking the time to do one of the mundane and boring jobs that needed to be done at some point.
I can’t recall if I mentioned it on the blog or not if I’m honest but I’ve definitely mentioned it on the SBC forum. After coming up with the Similar Games Model systems (TOX, STOY and STOZ) and discussing them in emails with a few people, the overwhelming feedback was that the system had too few bets and therefore, this really limited the potential of this system in terms of others following the bets. In addition, it would have taken about 10 years of bets before we knew if the systems really had an edge or not.
After a bit of thought, just before the season started, I stripped out a lot of the filters on this system which meant I was including many more games than before. It was always going to reduce the returns but it was a price worth paying.
Based on my previous backtesting, I expected these systems to have about 30-40 bets each this season. We’re halfway through November and they have already had 43,37 and 41 bets respectively, so clearly, the systems aren’t exactly the same systems which had the backtested results I was showing on the blog.
I said last week that I would go back and backtest these systems next Summer if the systems did well this season. If they flopped this season, then they’d be dropped anyway as they do take a lot of time to find the bets for each week.
Anyway, due to the fact I’ve had a couple of hours spare at night due to the fact there have been no games, I’ve gone back and backtested the systems with the reduced filters and updated all the historical results in my spreadsheet. I’ve also updated the blog with these system results. Just so you know, it was easier for me to include this season’s results too in the analysis as it was too much work to strip them out and then put them back in etc.
You can find the updated backtested results here if you scroll down:
For those that are only interested in the summary of results by season, here they are:
Now, the first observation anyone can make here (you don’t need to be a shi* hot analyst to notice this!) is that this season’s results look nothing like the historical results. People will automatically jump to the conclusion that these 3 systems are backfitted to the max and won’t work and in a way, they are correct. The systems are backfitted to the max as it’s the way the Similar Games Model works.
However, they are backfitted on data from 2000-2006. All of the historical results I am showing here are simply backtested results. There are NO backfitted results in these 5 years of results in the table above. I could show the results from backfitting (50%+ ROI in every season) but they are meaningless. Backfitting is just a means to an end.
Clearly, some people reading this won’t believe this fact and will consign these systems to the scrap heap along with the other thousands of backfitted systems that fail when they go live. Of course, these systems may well end up on the scrap heap as I say above as I’m not going to track them beyond this season if they are loss making!
However, if you analyse these systems in more depth, there is lots of hope here I believe and I’ve said it before on the blog and I’ll state it again now but long-term, I think these 3 systems may well be the best systems I actually have. The fact they are loss making after 40 bets is irrelevant really.
The systems are actually making a good profit on the Home bets and yet, are losing on the Away bets. This contradicts every one of my other systems as Aways have made all the profits and Homes have really struggled this season so far. Therefore, these systems are able to identify the best Home bets that my ratings algorithms can find. This is definitely worth knowing for future reference.
Secondly, when you study the drawdowns on these systems this season, it is perfectly in line with the historical drawdowns. Therefore, these systems are more than likely suffering from short-term variance and at some point, they will start producing profits hand like they have in the past I suspect. The Away bets are the bets that cause the variance on these systems but they are also the bets that drive the substantial pts profit each season, so as soon as the Away bet strike rate picks up, the systems P&L will pick up.
I thought it was worth sharing this on the blog. I know people are very sceptical about systems and backfitting in particular but if there is one thing I’ve shown so far on this blog, it’s that I am able to build a system, test it via backtesting and then launch it in a live environment and achieve results close to the results in backtesting. The fact it hasn’t happened for these 3 systems yet isn’t down to the systems in my opinion. This season just hasn’t been in line with past seasons so far………