Robustness after Optimization

Good morning all,

i have a strategy, built with indicators, for which i have a master file that i can optimize to work with different asset classes. It shows good results on many pairs, so it is for sure cross market and cross time frames too (tested it).
But when i come to robustness test, it is very rare that it could pass the “randomize strategy parameters” test at 50% ratio between Ret/DD at 95% and original Ret/DD.

The thing is: should i rethink the 50% threshold, given that is not a “building strategy” robustness test, but it is done on a reoptimized strategy?
I assume it’s normal for a reoptmized to stay in the upper part of the robustness test cloud, so it’s very difficult i could find it at 50% with 95% confidence.
Consider that the strategy usually shows positive net profit and positive ret/dd even at 100% confidence.

Any thoughts/suggestions?

I hope you do not relay only on that one strategy to trade your portfolio,
if you have more strategies in your portfolio beside that strategy that you are talking about then you are good to go and you do not need to perfect each and every strategy,

just diversify as much as you can..

Automatic trading is just part of my trading..
I obviously have different strategies, but this one has a robust logic and is not builted in SQ, it comes from my past experience.
The only problem is the one i told above, but is it almost always profitable.. it shows profit in 100% confidence, in every run of WF.. but there are not so much that reach that 50%. Just that.
Don’t worry, the strategy is good 😉

20% Probability
20% change

Yes, i agree with you.
i’m considering lowering my expectations about the ratio ret/dd 95% against original ret/dd.
Right now, if i see a strategy where the ratio is at least 30%, and where 100% confidence shows good positive net profit and acceptable drawdown, i keep it and go forward

As example: would you delete a result like this?

With probabibily of 0.95 you should not have the drawdown larger than 21% while the return is tenfold of drawdown. If WF analysis seems also promisingly I would bet on this 🙂

massidm wrote:

As example: would you delete a result like this?

It would depend how it gels with the other components of the portfolio. Additions to my portfolio typically have an equity curve and robustness stats as per the pic below:

This example looks be overfitted or curvefitted.
The equitycurve is too linar.
thomas