A quick update on where I'm at with the new ratings.
I've stumbled across a bit of an issue with the DNB ratings in the sense I can't seem to get something that fits the data from 2002 to 2007 and then fits the last few seasons. Part of it is due to the results in 06/07 and 07/08 as these results seemed to throw up the best results for traditional H/A betting but it isn't working when I'm trying to test these seasons with DNB.
I've looked at the results and one way around it is to exclude a few seasons from my backfitting completely and just have less backfitted results to work from. This keeps things from getting too narrowed and too backfitted in places and should help the results in more recent years make more sense.
I'm going to start again and backfit the early data again for DNB and exclude seasons 06/07 and 07/08 from my backfitting completely. However, I'll bring in season 09/10 into my backfitting by sampling 50% of the games in this season.
What should happen is that I'll end up with 4 seasons of backtested results now (instead of the 3 seasons I have with my current systems) and of course, I'll need to be careful when I interpret season 09/10 as it won't be a realistic return compared to the other seasons.
The other interesting observation I'm finding is that although I'm trying to maximise DNB returns, I still can't get anywhere near beating the traditional H/A returns. Simply, no matter what way I look at the data, backing the draw is loss making using my ratings. Again though, the purpose of DNB systems isn't to provide a better return, it's simply to try to smooth the journey hopefully.
Anyway, that's a very quick update of my last couple of days or so.
I should be in a position to have the first cut of ratings later this week and I'll share the results on here when this is done.
The blog is going to be a bit boring for the next wee while as it is basically me working away in the background doing the donkey work for next season! I'll try to provide updates of how my work is going though to keep this place ticking over.
No comments:
Post a Comment