Topic: Monte Carlo

Hi,

I finally made it to the members club

Is it possible to run more than 100 monte carlo during the reactor build stage? I want to run at least 500 on a strategy to test robustness and I want that to be part of the selection criteria. I see that you can select 500 tests within the strategy itself.

And if I do use the MC function under the strategy tab, how do I tell whether the strategy has passed the test? Is this yes/no answer or should I be referring to the confidence table? If so, what kind of thing am I looking for for a decent result?

Thank you,
Matthew

Re: Monte Carlo

Is it possible to run more than 100 monte carlo during the reactor build stage?

100 is the maximum in the Reactor because we think that everything above 50 is a loss of time and electricity. Monte Carlo works perfectly well with 20 tests.  You may check it by running the Monte Carlo several times and to see if the results are consistent.

I see that you can select 500 tests within the strategy itself.

It must be some ranges in any parameter. 500 is a number we set because it is a way above the useful values. Anyway, you are correct, I have to make the ranges equal.

how do I tell whether the strategy has passed the test?

The program shows something like "Validated: 20 of 20". It means that 20 tests of 20 calculated have passed the test. It is good. However, it is as good as the tests itself. If you make very strict validation criteria, no strategy will pass. Everything is a matter of balance.

I suggest you to start with the default settings and to monitor the performance. If many strategies fail in Monte Carlo, you have to make the validation rules less restrictive. The opposite is also possible.

Probably around 10% of Passed validation is a good starting value.

It is a matter of workflow. Some traders prefer to set heavy validation rules and to ruin the Generator for days to produce 2-3 good strategies. The opposite approach is to run the Reactor with more relaxed rules and to make a full collection for an hour. Then it is more fun to check and evaluate the strategies.

I usually monitor the Collection until the Reactor works. I love to remove upcoming strategies with ugly balance curves when they appear.

should I be referring to the confidence table?

The Confidence Table serves a marketing purpose. It is here because some other software have it. They tell their user that the confidence in the table is the probability to make a particular profit. We never say that but if we don't show the table, our users will start asking for it.

Re: Monte Carlo

Just make 500 tests available in the Reactor's Monte Carlo. It, of course, doesn't mean you have to use it every time.

Re: Monte Carlo

Hi - thank you, that's a very clear answer and I see that 500 has been added.

I'm surprised to hear that 500 is above the usual number of tests, but I am still learning. So, between 20 and 50 tests should be Ok.

The other robustness test - multi currency - how many ideal other markets do you think we should aim to pass? Im currently going to 3 additional markets but that may be too many.

I have a feeling Im setting my parameters too strict so Im seeing fewer results than I want.