The real difficulty in practice is how you determine that a system has stopped working. I had the experience with my first system which was profitable for 6 months, then broke down in the next 6 months, pretty much wiped out all profit made in the first 6 months. The systems I'm using now all have a predetermined set of criteria to declare the system "failed" and call a quit.
Jumping on the right side of a trend day and using filters to identify when those days are most likely to occur. It has worked for me for several years and has been documented countless times for years before that.
I found a goldmine ( for 1 yr ) ... I could have trained a bunch of chimps to scan and pick off the arb opps in 2000 ....but this goldmine turned sour starting in 2001 with the crash of the nasdaq and decimals and mergers of various ECNS