On how many timeframes/pairs must an EA show good results to be considered any good?
Would EA coders please share their experience for how many pairs and/or timeframes do you need to see sound backtest results before you consider the EA worth demo/live testing? I'm sure there are many people who optimize their EAs on e.g. eur/usd M1 chart and are happy to go live. I'm generally not happy with an EA until I see reasonable results on at least 2 different currency pairs for example eur/usd and aud/usd on the same set of parameters and timeframe because I believe there's then greater chance that the EA logic isn't overly curve-fitted.
What's your experience? I don't want to talk about how much historical data you test an EA on but more about on how many timeframes/pairs you want to see good test results. Obviously, the more the better but what's your minimum?
As many as possible . . . if I had consistent results on 3 pairs I would test on more, if I had consistent results on 6 I would test on more, I have tick data on 15 pairs, that would be a good place to stop if I had consistent results on all 15 pairs.
I've written a few EAs, some have had very good results on some pairs for some date ranges, but the story soon changes when a non-correlated pair is used. Even a coin toss EA can have a good run, it can also be made to have a 95% win rate.
As many as possible . . . if I had consistent results on 3 pairs I would test on more, if I had consistent results on 6 I would test on more, I have tick data on 15 pairs, that would be a good place to stop if I had consistent results on all 15 pairs.
I've written a few EAs, some have had very good results on some pairs for some date ranges, but the story soon changes when a non-correlated pair is used. Even a coin toss EA can have a good run, it can also be made to have a 95% win rate.
As I said above, the more pairs the EA is successful on the better but the thing is, where's your threshold for going live? In other words, on minimum how many non-correlated pairs/timeframes you want to see good and consistent test results before you consider the EA good enough for demo/live testing? If your EA looks good on 2 or 3 major pairs, is it good enough for you to put money on it?
As I said above, the more pairs the EA is successful on the better but the thing is, where's your threshold for going live? In other words, on minimum how many non-correlated pairs/timeframes you want to see good and consistent test results before you consider the EA good enough for demo/live testing? If your EA looks good on 2 or 3 major pairs, is it good enough for you to put money on it?
Thanks for the reply. What you said makes sense: if there's consistency of results across pairs, fewer pairs may be sufficient than when positive results are more random. I tend to focus my testing on 8 major pairs since in theory they should behave more technically just because more people or EAs trade them. Also spreads are narrower. When I can get good consistent results across 3-4 majors (e.g. eur/usd, gbp/usd, usd/jpy, aud/usd) it's time for demo. In my example are all usd pairs but they are not strongly correlated in their moves (unlike for example eur/usd and usd/chf).
I'd appreciate if anyone else would share their opinion/experience.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Would EA coders please share their experience for how many pairs and/or timeframes do you need to see sound backtest results before you consider the EA worth demo/live testing? I'm sure there are many people who optimize their EAs on e.g. eur/usd M1 chart and are happy to go live. I'm generally not happy with an EA until I see reasonable results on at least 2 different currency pairs for example eur/usd and aud/usd on the same set of parameters and timeframe because I believe there's then greater chance that the EA logic isn't overly curve-fitted.
What's your experience? I don't want to talk about how much historical data you test an EA on but more about on how many timeframes/pairs you want to see good test results. Obviously, the more the better but what's your minimum?