I'm trying to get some thoughts on an approach I'm going to tackle.

I've managed to obtain some data from 1999 right through to 2012.

Lets say if I run every result I get a loss of 15%. Base it on unitab

Now if I run one filter (Runs this Prep)
Results may be as follows

1st up runners 27% LOT
2nd up runners 17% LOT
3rd up runners 5% LOT
4th up runners 8% LOT
5th up runners 17% LOT
6th up runners 18%LOT
7th up runners 13% LOT

Lets assume that the samples may change from each runner but there is enough data to say these figure will remain constant into the future.

Now we run another filter say sex of horse

Colts - 16% LOT
Fillies - 14% LOT
Mares - 13% LOT
Stallions - 17%LOT
Geldings 15% LOT

Now if we combine Geldings and 3rd up runners together and have (15% LOT) and (5%LOT) and the results come back as a 6% LOT.

Should you expect the LOT to be 10% and if it is less than 10% you are seeing this combination underbet?

The same can be said if I combined Colts (16%) and 1st up runners (27%) I should see 21.5%POT. Lets say the results come back as 23% LOT. This may mean the combination is overbet to its true value.

I think if you have enough data then filters on there own will never change in the long term. You have seen this with LOT of favourites.

When you start with a combination of filters you run the risk of having not enough data to support the long-term results. Hence if you have 8 filters you really run into trouble as you might be starting to bet when the system is on a high and about to drop below its true average value.

Maybe this only works with strike rates on filters and the Profit / Loss on turnover is always out of our control

If I got too much sun today at golf and this makes no sense please ignore

Cheers