Hi Mr. Popov,
I think there is still something wrong if using the Optimizer within the Reactor and just using the "Common Acceptance Criteria". If the "Command Acceptance Critera" is use for the Generator AND the Optimizer, both with 10% OOS, no strategy should fail the validation test after the Optimizer, because the Optimizer would take into account the rules for the 10% OOS too (like ret/dd ratio minimum of 2). The Optimizer would then get the strategy from the generator and the strategy hence already fullfills the acceptance criteria (because both Generator and Optimizer are using the Common Acceptance Criteria with the exact same values and 10% OOS). During the optimization in the Optimizer, the Optimizer and if for example optimizing for best R-Squared value, the Optimizer would only find new best results if the OOS acceptance criteria have been fulfilled too - at least that´s how it always has been before you gave the Optimizer it´s own criteria. But now, I often see that the Optimizer optimizes a strategy but then it does not pass the validation of the Optimizer. Which means that the optimizer must ignore the acceptance criteria during optimization for best R-Squared (or whatever value) and only tests it AFTERWARD. But that makes no sense, previously the Optimizer always only found a new best result if all acceptance criteria have been met already during the optimization - it always worked like that until you´ve made the new change, now it doesn´t work like that anymore :-( Now ~50% of strategies that have been forwarded from the Generator are optimized, but don´t pass the validaiton of the Optimizer. Before (with the exact same settings like now), 100% have always passed the second step of the Optimizer validation (which is correct behavior), as the Optimizer must already obey the acceptance criteria during the actual optimization, not just afterward.
1) Generator and Optimizer both use the Common Acceptance Criteria (complete backtest: ret/dd of at least 4, 10% OOS at least a ret/dd of 1).
2) Both are set to 10% OOS
3) Optimizers goal is to optimize for best R-Squared, with 10% OOS
4) The Optimizer only gets strategies that have already passed the Common Acceptance Criteria from the Generator, so the strategies ALL must pass the validation-step of the Optimizer too and the Optimizer should ONLY find a new best R-Squared value during it´s optimization IF it can keep the ret/dd of 2 in the 10% OOS too - and that´s how it always was the case before that change (it has worked like that for the whole last year). But exactly that does not seem to be the case anymore, it seems to ignore the 10% OOS min ret/dd of 2 during the optimization and just optimizes the IS for best R-Squared, accepting any results that have a better R-Squared in the IS, regardless if the OOS ret/dd falls under 2. Hence it can happen that the strategy does not pass the validation-step of the Optimizer anymore - which never was the case before (always 100% passed, which is logical and it worked like that for the last year).