Optimization hints for v1.080

[kenorb]

The optimization of the EA is a very challenging task. Here are few tips:

Disable noise and complex rules

• Disable boosting (Boosting_Enabled set to 0) and action (Account_Conditions_Active set to 0) sections when running optimization process, so you can get pure technical results (tests are quicker and more reliable). Boosting and actions can be always tested at the final test, whether they add some value on top of it or not.

Make all trades equal

All trades along the test should be equal as possible, so they're not dependent on the previous success or rely on the current balance, so:

• Set LotSize to the minimum (e.g. 0.01), so they are not automatically increased with the balance.

• Set ProfitFactorMinToTrade and ProfitFactorMaxToTrade to 0, to make sure strategies aren't disabled due to their failure or success.

• Use CloseOrderAfterXHours to close trades after 24h, 48h or 72h, so profits from orders opened after a day or few are not taken into the account.

  If the trades are open for too long, that means something isn't right. Secondly, the conditions (based on technical indicator data) which were used to open the trades, after a day, they cannot be trusted anymore (support, resistance and pivot points are different each day). Otherwise, you're relying on success which were barely luck and won't repeat anymore.

Make it simple

• Always test one strategy at a time. Then merge them at the final test.
• Avoid enabling any unnecessary params which can affects other params.

Use large datasets

• When optimizing params, always test on the large dataset giving your trades as much as possible scenarios and different market conditions. For that reason, it's good to set fixed and minimum lot size (0.01).

Final test

• At the final test, try testing with boosting on/off or/and actions on/off (or few specific actions).
• Consider tweaking params which affects all the strategies, such as: CloseOrderAfterXHours, RiskMargin, RiskRatio, RiskRatioDecreaseMethod/RiskRatioIncreaseMethod and similar.

Troubleshooting

• If SignalMethod is too much filtered (less than <100 trades), that's sometimes bad, as you don't have enough dataset to optimize other params (in other words, the optimization results could be not reliable).
• For very slow optimization tests, use Control points (not recommended) to have some rough idea which value is better, but the results cannot be reliable.