Thread: One For Barny
View Single Post
  #52  
Old 11th December 2012, 12:22 PM
mattio mattio is offline
Member
 
Join Date: Jan 2011
Posts: 268
Default

Quote:
Originally Posted by evajb001
Barny, I don't think mattio is saying his aren't backfitted at all, he's saying the following:

If you have a bunch of data that your backtesting on that goes back say 5 years. You've added in some rules, such as the following:

Days Last Start - between (and inclusive) 5-24
Win Strike Rate - above 24.635%
Placing last start - 1
Placing 2nd last start - 5
Won at track 3 times
Won at distance 3 times

^^ Clearly the above is backfitting, your changing numbers to specifics to maximise your POT, which is great to show a high POT but the chance of that perfect storm repeating in the next 5 years is minimal at best.

Mattio (i think) is simply saying that your aim should be to have logical filters that aren't specified down to minute detail so that you avoid substantial backfitting. So in reference to the above example your filters may be this instead:

Days Last Start - between (and inclusive) 7-21 (logical as its ran on the same day previously, i.e. not starting on a saturday now and some random weekday previously)
Win Strike Rate - above 25% (obvious one, you can't backfit to a specific % its just illogical)
Placing last start - 1 (fine to leave as 1 if you want a last start winner, nothing wrong with this filter)
Placing 2nd last start - 3-x (instead of specifying it had to have finished 5th last start, it needs to be a range otherwise its again being too specific to your backfitting)
Won at track 3 times (you could possibly use this filter but i'd suggest saying won more then 3 times at track, as saying exactly 3 times is too specific again)
Won at distance 3 times (you could possibly use this filter but i'd suggest saying won more then 3 times at distance, as saying exactly 3 times is too specific again)

I think there is room to have maybe 1 filter that is more specific that gives you the edge, but as soon as you start tinkering took much its just over-backfitting too your data. I think thats all mattio is trying to get at Barny, he's not having a shot at you and neither am I, were just suggesting people are careful with how they approach creating a system.


Mate that is the general jist of what I was getting at, thanks for your support. Unfortunately there are people that always have to believe they know exactly what other people are thinking so that will be all I have to say on this topic - hopefully the information will be found to be useful by someone.
Reply With Quote