OK - analysis time.
I've had a chance to look at the system's performance stats for the 133 games played so far. My XLS not only calculates the qualifiers, but also records the results in various charts to help identify trends and where to adjust. There are some obvious changes that can be made, with the following observations made.
The Q-rating has 2 levels (actually 3 if you count the ones I have included using my 'judgement'). To get a 'Q', it must achieve a score of 7, to achieve a 'QQ' it must be 8+... my additionals (non-Q) tend to be what I think are strong 6's. They break down as;
This equates in real money as;
Clearly my own additionals don't compete with the system-generated picks at all. I need to keep above a 75% hit rate to ensure a profit,.. while the 'Q's and the 'QQ's are going way above that, my own choices have not. It would be more sensible for me to not to include any non-Q's that I think are good enough, or find a better selection process for those games that don't make a 7+.
I find it a little strange that the 'QQ' games that return a higher rating score than the 'Q' games have so far performed worse - this I guess means that the criteria filters just need adjusting to better define more likely winners.
Part of the checking criteria looks at the % of draws any league has to date, bearing in mind they can fluctuate up and down as the seasons progress. With the leagues that I check, they break down as;
The only thing this tells me is that 18% and under has returned a 100% hit rate. Unfortunately there have only been 17 games for that % level, so I need to tweak the filters to show that aspect more strongly within the game scores.
The game evaluation process is based on various criteria, but starts with what it thinks the score will be. Since it is originally based on home favourite wins, the matches needed to predict the scorelines of 2-0, 2-1, 3-0, 3-1 and 4-1. They break down as;
This tells me that I should only include games that predict a 3 goal margin i.e. 3-0 or 4-1. Since the other scorelines represent only a 1/5 of all the games recorded so far, I think it'll be a minor but positive change that should raise P/L, hit rate and also mean not as much money is tied up with several games being played at the same time.
I have some other auto-generated charts to dissect, but with the ones above I can make some immediate changes that ought to improve the system. To give it it's due, it has performed quite magnificently (today's results notwithstanding) considering I've made very little changes to it since I started the trial, so I believe this can only improve as I get more data and change little things along the way.