Backtesting problem

[Hooloovoo]

Hi all, I have developed a computer program that read statistical data in the form of CSV-files from the site www.football-data.co.uk. The backtesting currently covers over 22.000 matches from 60 seasons in 5 countries. It calculates ROI, match ratings and produces probability distribution tables. I then continue to scatter plot these distributions according to the PDF "ratings.pdf", also available at mentioned site. This i do to find the "best-fit" relationship with the match ratings. Then a best-fit line can be superimposed on these data in the graph, and the equation for this line can help me find the fair odds for a particular outcome and a particular match rating. Now to my problem: When i put the equation into the backtesting program to calculate the fair odds, then it should only bet on the matches where the bookies odds (as given by the csv-files) are higher than the fair odds. Doing so should yield a profit in the long run, if the forecast is accurate. The problem is i can't replicate the positive yield from the pdf file, altough a have the same criterias. (The rating system in the pdf is a simple one based on goal superiority). I noticed that in their graph for home wins, they get a R2-value of 0.8623, which means that the best-fit relationship provides a good fit for the real data. But in my graph for home wins, i only get a value of 0.688, which means the relationship is weaker. :unsure Could this be the source of my problems? Any input / ideas / suggestions would be greatly appricieated :)

[Hooloovoo]

Re: Backtesting problem Maybe the first post was too much blah-blah.... I have a set of criterias to produce match ratings. An example: After running through 20.000+ matches, a match rating of 10 shows that home wins occur exactly 50% of the times. The fair odds for home win in an upcoming match with a match rating of 10 should be exactly 2.00 So if i'm offered home win odds above 2.00, for example 2.15, we have a value bet. Betting only these value bets should produce a profit in the long run. Since backtesting does NOT produce a positive yield, what am i missing?