Hi,
With respect, 20 days is a very short period. How many races in the sample ?
What period did you test the data over ? Did you use a hold out sample ? otherwise backfitting comes into play.
I recommend you test over at least a 12 month period and around a minimum of 2,000 races before arriving at any conclusion positive or otherwise. Even after testing, I would suggest to bet carefully for several months, as things don't always pan out when the real money is on the table.
Best of luck with it.
R