How to measure the concentration of MC retest method’s result?

We know that we should not only focus on a certain confidence level MC retest method results, but also check the divergence of all simulation results.

What are the suitable metrics for sorting (databank column) or automatic filtering?

Hello,

that’s what cross-check filtering conditions are used for. But many traders evaluate these based on visual discretion (not using any strict or specific value). They just want to see there are very few or ideally none results that diverge from the the rest significantly

Thank you, tomas262,

I think it only measures the value at a certain confidence level, not to assess the distribution of simulation.

Maybe we can try this method,

1, Calculate the difference between the one simulated equity and the original.

2,  Calculate the sum of squares of each set of differences, and then calculate its square root.

3,  Divide the result of step 2 by the original equity.

4, Sum the result of step 3 of all simulation.  For example, if we call it Monte Carlo distribution.

Are there similar criteria in SQ?  If not, then how to implement my ideas in SQ?

imagine one simple thing – you will have for example strategy entering on some OHLC building block, for example OPEN DAILY

this building block doesnt have parameters at all – it could have only shift, nothing more – so OPEN DAILY is for MC tests always the same OPEN DAILY, nothing will be changed in the MC tests

and here comes the conclusion – MC tests with these non parametric building blocks will always be looking better than for strategies using for exmpale EMA with parameters

so is it comparable? i am afraid not…MC tests are for me only some projection for the future, it tells me nothing about quality or future profitability

I use “Randomize starting bar” and “Randomize strategy parameters”,  most strategy have 2 or 3 conditions with parameters.

The MC test can evaluate the robustness of the strategy to some extent, I think.

Good day all,

Great question Mr, eastpeace I really like it and I think a lot of users have the same concern.

unfortunately admin reply not aiming to the question and go to traders ethics and myths.

I think the informed that we need from support unit is the specific way to use the program , specialty in generation stage selecting building blocks and the proper way for using ranking in all process.in my view bad using for ranking will destroy all your work mostly what is the perfect number to choose for each ranking condition and what is the proper condition to select.

You are right, there are mathematical tools that could be used to evaluate “quality” of robustness test results (curves) how close the curves are and how ideal it all looks.

As Hankeys mentions the question is what does it say about future results? What I evaluate visually in a fraction of a second will be hard to express in some meaningful formulas. I have seen some many perfect looking strategies with great RT test results to fail miserably while others performing poor in those tests make money year after year

I will check with Mark what he thinks about this

filtering by MC tests leading only to lower diversification

thanks a lot for your genuine reply.we hope with time we can catch the proper way to do it.

Is that WFM is more important than MC TEST?