when backtesting a system how far back should I go? I have developed a Fairly simple system and want to backtest it to see if it is valid. I would appreciate any help.
Happy trading
The short answer is as far back as you can get good data.
At a minimum, you want to make sure you test through at least one full cycle of the various types of market conditions for the timeframe of trading your are considering.
If your backtest isn’t at 90% model quality it means nothing, literally nothing.
Most EA’s are tweaked to take advantage of past randomness. If you make an EA, and just start randomly changing Settings(like profit/Stop loss/settings on indicators) until your backtest happens becomes profitable, your backtest data means nothing. This just means you randomly optimized your EA to past conditions.
Backtesting I feel should only be used to determine to see if your EA works. If you system has sound logic, and your EA models this logic accurately, your EA should perform okay.
Foward test is all that counts. Even then you can foward test for a while and your EA just might randomly be suited for current market conditions. Thus when the overall nature of things change your EA becomes crap. Same could be said with paper systems.
Develop what you believe to be a sound system, apply good money management, make sure all the logic works out perfectly and then foward test your EA. If you have positive backtest data, I would say atleast 1-3 years, and positive foward test data, then you may have a functional worthwhile EA.
However EA’s seem to me (Having worked with many) just be to randomly profit from optimized settings based on the nature of the current market. Thus their future validity is questionable.
Remember no system that requires “human judgement” can ever be programmed into an EA. EA systems have to be purely mechanical of sorts.
Curve-fitting is the term you will often hear which describes this problem.