Good question Vortech - it depends on the type of filters and the extent of the backfitting. Backfitting in some form or another is present in ALL systems otherwise how else does a system get developed? You back test certain filters, analyse the results and make adjustments then you back test again - this is how a system is developed. I prefer to tailor the results to suit the filters as opposed to tailoring the filters to suit the results.

What I mean by that is working with filters that I know are quality filters (based on backtesting - there's that word again) and then making adjustments around that. What alot of people do which I believe is the wrong way to go is that they manupulate filters to suit their results and end up with filters of this nature:

- days last start 9-37 (no logic here but it suits the results)
- last start finish 2-9 (not knocking Barny specifically here but why look at a horse that finished 9th last start especially when you have no idea of the margins)
- win percentage 23-82 (again no logic here but it suits the results)

These are just a few examples but there are many more. If I were using these types of filters (and I do) I would go about it in this way:

- days last start 7-21 (or something of this nature, you know the horse is likely to have a good level of residual fitness and the days are a logical number)
- last start finish 1-3 or 4+ (this depends on the type of system I am doing, one that looks for consistent horses would be 1-3 and one to find value horses would be 4+ but I would generally use this in conjunction with a margins filter)
- win percentage 30-100 (this is far more logical and not tailored around results)

Again these are just a few examples but I believe it all comes down to the individual being honest with themselves about why they are including a certain filter. If there is no logic about a filter that adds value then I suggest it will fail longterm but if the filter has logic (real logic, not punter's logic) then I suggest it has a chance of working longterm.

Here is a system that I use, I have added a couple of extra filters but here are the base filters:

7 days to last start
No change in distance
No change in weight
No change in prizemoney
Same track as last start

Basically this system is looking at horses racing under similar conditions to last start backing up after 7 days. It is a simple system with simple and logical filters that finds some great value winners with an acceptable strike rate and average price winner around $10.

Just my thoughts on another issue to is the definition of "longterm". I don't test back any further than 2010 due to the changing nature of the industry and the distasterous effect of EI in 2007 which I believe still had consequences up until 2009 and even early in 2010. I have found that many systems have a drastic change in results when testing beyond 2010.