AndrewKent Posted April 20, 2011 Share Posted April 20, 2011 Hi guys, I've been busy working on improving a number of betting systems using Excel/Access/VBA. I've got a pretty comprehensive database and program that works really well and I am now starting to see profit appearing from my work. As rewarding as this is, the ROI is only between 5%-6% and I am sure I can get more out of this. The systems I am finding most accurate apply weightings to match characteristics and with my program, I have built a looping routine that will run analysis on a number of scenarios. Each scenario applies a different weighting to the match characteristics (goals, shots ect) allowing me to see which combination works best. The issue I am having is although I have a base system (Xpts for a Home Goal, Ypts for an Away Goal ect ect) I am not sure how best to optimize this. Which weightings should I increase/decrease, how should I know which ones to touch and which ones to leave alone? At the moment I'm randomly applying different weightings but there MUST be a logical way to do this. Any advice? Andy Quote Link to comment Share on other sites More sharing options...
Vilamoura04 Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization Back test against your data to test which weightings give you the best profit Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted April 20, 2011 Author Share Posted April 20, 2011 Re: Weightings & Optimization Back test against your data to test which weightings give you the best profit Sorry I think I must not have been detailed enough in my original post. I have everything set up to back test the data, and the whole point of this exercise is to find which weightings give the best profit! What I mean is for example, if I said the following: Home Team Goal Scored = 2pts Home Team Goal Conceeded = -1pts Home Team Shot on Target = 1pts ...I am trying to get some advice about how go about adjusting these logically. It's easy enough to increment, but I was looking to see if anyone had done this kind of thing before. I have about a dozen different match characteristics I need to weight, what would be a good approach to finding out which ones effect the Strike Rate the most? Andy Quote Link to comment Share on other sites More sharing options...
Mulkis Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization Ok, you have a dozen of different factors... Maybe you could provide more info what kind of model do you feed all these parameters. Is it Poisson, Bayesian hierarchy or what? Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted April 20, 2011 Author Share Posted April 20, 2011 Re: Weightings & Optimization All these parameters are fed into a match rating system. No, it doesn't use Poisson or Bayesian. The ratings system looks at the points scored in each game (depending on the weightings) and produces a rating. The how my system works isnt important in this question (I don't think) what is important, is how to weight numbers against numbers. Short of saying: "I am going to test every combination of numbers for 12 parameters from +5 to -5" the scenarios would run into the millions. I just wondered if there was a more sensible way of doing things. Quote Link to comment Share on other sites More sharing options...
Gooner80 Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization hey Andrew you probably know more than me but I have been looking at financial forecasting and to optimise their forecasts they use solver in excel and try to reduce the mean square error for weighted exponential smoothing which might help you Quote Link to comment Share on other sites More sharing options...
Mulkis Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization It seems like most important (statistically) factors are team's defence ability and attack ability. Both are based on goals scored/conceded. All other factors like shots on targed, corners, crowd presence, etc. are somehow minor. Probably the best way to proceed with your work would be back-testing, as already suggested. But I still wonder, what are you trying to atchieve with your system - is your goal to find "true odds", or you are looking for "hidden winners"? Quote Link to comment Share on other sites More sharing options...
Vilamoura04 Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization I think you may well find that a teams league position is the biggest indicator. But as I previously said there isn't a prescribed golden rule weighting that you would apply to each statistic, especially as your model is probably different to others. Trial and error I'm afraid. Quote Link to comment Share on other sites More sharing options...
Sir ratholer Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization I'm no means an expert on this but could you compare each variable to the sp of each event? I suppose then you could see what is more relevant than other factors. Looking at your original thoughts if you are going to analyze through shots on target then it should have 1/3 of the value of goals because on average it takes 3 shots on target per goal. Shots off target is about 10 per goal on average. Quote Link to comment Share on other sites More sharing options...
sodabrab Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization Andrew I have a very profitable system similar to what you're working on and return is much higher about 25% for singles. Unfortunately there is no quick answer just lots of hard graph with different variables. If you have a bogus email address I would be happy to send you an email with a few pointers but be prepared to work hard. Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted April 20, 2011 Author Share Posted April 20, 2011 Re: Weightings & Optimization Andrew I have a very profitable system similar to what you're working on and return is much higher about 25% for singles. Unfortunately there is no quick answer just lots of hard graph with different variables. If you have a bogus email address I would be happy to send you an email with a few pointers but be prepared to work hard. I've been working on the VBA program and have refined about 6 systems now in the last 8 months so hard graft isn't an issue. A few pointers would most appreciated. Are you able to email me at "andrewkent 1981 @ gmail .com" obviously without the spaces. Andy Quote Link to comment Share on other sites More sharing options...
sodabrab Posted April 20, 2011 Share Posted April 20, 2011 Re: Weightings & Optimization Andrew email sent excuse the typos Quote Link to comment Share on other sites More sharing options...
Donkson Posted April 21, 2011 Share Posted April 21, 2011 Re: Weightings & Optimization Try to look at the correlation with each stat (r^2 value). You obviously want the stats with higher correlation to have more weight. http://en.wikipedia.org/wiki/Coefficient_of_determination lol at 25% being attainable long-term. Quote Link to comment Share on other sites More sharing options...
Gooner80 Posted April 22, 2011 Share Posted April 22, 2011 Re: Weightings & Optimization I have been thinking over your problem mainly because I too can yield the same as your system but believe there is more I can do. It suddenly clicked in my head that there is a principle called the pareto principle which basically states that 80% of your yield will come from 20% of your system and this rule will repeat itself through all of your system and the key is to keep refining your system wiki link on the principle attached http://en.wikipedia.org/wiki/Pareto_principle basically the 80/20 rule is finding the 80% that is waste and refining it Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.