My mistakes have made my best systems...anyone else?" it's called learning -- some people see the light at the end of the tunnel, others don't
Two of the very best methods I have "discovered" and eventually automated happened when I was tying to get something to work based on when HFT would flip into their signals, the bell curve was opposite on what I was trying to do. It is seldom to do the opposite to make good results but that's what happened. Past several months of back testing for day trading has gone into only one direction of finding the hints of HFT, whereas it is getting harder to scalp and get filled at my price which is giving up one tick to get in and I refuse to give up more than that. I do see S&P500 futures chart patterns that I have from 1980s, 1990s are becoming different than those of today, I can see many doing less of day trading futures and moving into day trading options in next five years. Retail fees are truly expensive based on 5k account.
If you have been in a poker game for a while, and you still don’t know who the patsy is, you’re the patsy.
yakyakgoose : I have similar experiences. By investigation of a result or observation that is unexpected, I often gain understanding that may not have been possible otherwise. (I may have never pursued the correct question, due to my incomplete or inaccurate concept.) This is not limited to just back-testing. This "root cause analysis" of unexpected observations, seems to be a necessity for those of us that are developing without peer review. Like you mentioned: you may have entered a TYPO, but much of the time, it may be wise to investigate and understand the results you observe prior to "correcting" the TYPO.
IMO: I prefer to examine if some trading strategy could have been successful in the past before wasting time (and/or money) with it in forward testing. -- Backtesting provides rapid feedback, however, for forward testing you must wait!
That certainly happened to me. I backtest using PHP so everything has to be written from scratch. When I made a mistake in my code, sometimes I do discover pockets of good return. But when the mistake is corrected, the real return can barely cover slippage. I don't think it's a surprise that you need to think outside the box. You can't really expect profit using something simple like follwoing three moving averages, otherwise everyone would be using it already.
The number of possible market scenarios is virtually endless. If your system profits often enough for you to stay net positive in random historical data long-term, it's worth finalizing and running live. Otherwise, it's mere curve-fitting.
I can only backrest back 5 years...but there's at least 100 trades a year. My return also increases every year....not sure if this is because of the gradual increase of computerized trading.
I would consider my system pretty simple compared to what I've seen others talked about. I read in a few places that a simple system can be better because it works across more market scenarios but it's just something I read so I take it with a grain of salt.