Expert Advisors' Testing and Validation

[Bartosz Andrzej Danowski]

Hello everyone,

I am currently having fun, programming and developing EAs with different strategies.

And here comes a question: I would like to make some sort of "validation manual" for EA. What methods of testing do you use while programming your Experts? What does your validation process look like?

Thank you in advance for your answers, every comment will be valuable.

[Davindra Adrian Pyankaroo]

From my experience, do use the built in MT4/MT5 strategy tester to check the algo with the strategy backtester and use the best input settings you have calculated (according to your algo)

Then proceed to strategy optimisation section to derive "fine tuned" inputs through input combinations

Best of luck with your endevours!!

[Fxrobustix]

For me validation should be more than optimizing once and choosing the best settings.

First I verify the logic in visual mode, then divide the history into development and untouched out-of-sample periods. I also test several rolling periods, change important parameters by around 10–20%, increase trading costs and run Monte Carlo with different trade sequences.

If the EA works only with one exact setting or during one market period, I normally consider it overfitted. After these tests, demo forward testing and then very small live risk would be my final steps.

[Bartosz Andrzej Danowski]

Davindra Adrian Pyankaroo #:
From my experience, do use the built in MT4/MT5 strategy tester to check the algo with the strategy backtester and use the best input settings you have calculated (according to your algo)

Then proceed to strategy optimisation section to derive "fine tuned" inputs through input combinations

Best of luck with your endevours!!

Okay, thank you very much for your answer :)

[Bartosz Andrzej Danowski]

Fxrobustix #:

For me validation should be more than optimizing once and choosing the best settings.

First I verify the logic in visual mode, then divide the history into development and untouched out-of-sample periods. I also test several rolling periods, change important parameters by around 10–20%, increase trading costs and run Monte Carlo with different trade sequences.

If the EA works only with one exact setting or during one market period, I normally consider it overfitted. After these tests, demo forward testing and then very small live risk would be my final steps.

Hey, thank you very much for your answer. If you don't mind, I have one more question: do you have any methods or settings that help you do backtest that gives results the closest to out-of-sample tests?

The biggest problem I had was a huge drop of the amount of trades between backtest and out-of-sample (the amount of trades in out-of-sample was like 20% of trades made in backtest)