RE: Newsletter Article - "Robust Optimization for Strategy Building"

[Red]

Mike,

Thank you for the article.

I have always had issues with both the walk-forward and Monte Carlo methodologies of verifying the "robustness" of a trading strategy.

Walk-forward:  You take a given strategy and verify the robustness of parameter(s) over a rolling periods of in-sample/out-of-sample data sets. The implication is that the selection of future parameters are based upon learning from the previous selected parameters?  I guess if the rolling periods all have "good" results, the strategy is robust.  Now, which set of parameters does one utilize?

Monte Carlo:  The re-arranging of strategy trades for verifying the consistency of results seems reasonable.  However, do we take the samples with or without replacement?  In addition, if the next data bar results are somewhat dependent upon the previous data bar, isn't this methodology unable to measure that attribute? 

Obviously, I am oversimplifying these methodologies and don't fully understand the statistical analysis of each - too complicated for me.  Personally, I like R-Squared.

Red

[MikeBryant]

With respect to walk-forward optimization, the selection of future parameters is not based on the parameters from a prior period. Each period is a separate optimization. As I noted in the article, walk-forward optimization does two things: (1) it gives you the optimal parameters values, and (2) it validates the optimization method. You use the optimized parameters from the most recent segment; the results on the prior segments validate that you can reasonably do that.

For Monte Carlo analysis, the choice of with or without replacement should not usually make a big difference unless you want to expand the number of trades, in which case you need to use sampling with replacement, but that's more of a practical choice than a theoretical one. It's true that if there is dependency in the data, then the MC results will be more conservative, but that's not necessarily a bad thing. If you're using MC for position sizing, for example, it implies your position size will end up being smaller as a result.

Mike Bryant

[Jeff Kirkson]

Good morning Mike.

 

I found some "Top Strategies" in the project ("RobustOpt.gpstrat" ​​file)  attached to the last newsletter,

but I don't understand how they were chosen since there are no selected metrics in the appropriate area ("Conditions for selecting Top Strategies").

 

Instead, the project process and other parts/metrics are all pretty clear.

 

Thanks for the help.

Jeff

 

Da: adaptrad...@googlegroups.com <adaptrad...@googlegroups.com> Per conto di MikeBryant
Inviato: mercoledì 18 settembre 2024 23:24
A: Adaptrade Builder <adaptrad...@googlegroups.com>
Oggetto: Re: Newsletter Article - "Robust Optimization for Strategy Building"

--
You received this message because you are subscribed to the Google Groups "Adaptrade Builder" group.
To unsubscribe from this group and stop receiving emails from it, send an email to adaptrade-buil...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/adaptrade-builder/c2a2f778-0c81-41be-98c9-d47e5f91b453n%40googlegroups.com.

[MikeBryant]

Those were just some strategies I saved during the build process for comparison purposes. I probably should have deleted them before posting the file.