Methods of carrying out a rolling forward - page 2

New comment
 
Комбинатор:
Erm walk-forward serves as a check, not a choice of something.
And what other criterion can you suggest to figure out which version of the code is better?
 
Комбинатор:
Erm walk-forward is for checking, not selecting something.
I agree - we select by backtest and check by walk-forward. And on the other hand, if we have failed forward, we will come to the point where we need to change the criteria/method for backtest selection. So it's still the forward that affects the selection)
 
Youri Tarshecki:
And what other criterion can you suggest to understand which variant of the code is better?
You probably don't mean a variant of the code, but a variant of the EA's settings?
 
Youri Tarshecki:

Choose by backtest - do not understand the essence of the volking-forward.

Then explain - how to select settings for backtesting tomorrow.

Here we have backtest for example from 14-10-2015 to 14-04-2016. Which of 10,000+ options to run?

I valking forward is understood this way - that in addition holding ennumber of other backtests + running them forward, to develop criteria / method by which to select the backtest, which statistically gives good results in the future (ie on the forward test).

As a result, applying this method to the backtest completed today (i.e. from 14-10-2015 to 14-04-2016), I can hope that the EA will trade in profit next month.

 
elibrarius:

Then explain - how to choose the settings to run the EA for tomorrow.

Here's a backtest for example from 14-10-2015 to 14-04-2016. Which of 10000+ options to run?

I understood the backtest in the following way - in addition to having conducted an enumeration of other backtests + backtests, I want to develop a criterion/method of selecting backtest results which statistically gives good results in the future (i.e. on the forward tests).

elibrarius:
You probably don't mean a variant of the code, but a variant of the EA settings?

Volking is an IMITATION of FUTURE, so to speak. Its task is really a test one. If I have two options - I run them through volking and see which one has worked better for me in an unfamiliar situation. I choose, make corrections, run them through, compare them again - you get an evolution without adjusting. The idea of volking is the market invariance, that's why for trading I, of course, take the most actual set, but as far as I've defined the optimal periodicity for the variables update - in our case it is the duration of back. I find the duration experimentally, again using volking. Thus overoptimization is more or less avoided.

I.e. expert settings should be as fresh as the step of verification interval of your volking, i.e. back.

Naturally it all makes sense if there are many of such sections. And that, in turn, is usually determined by capacity.

 
Youri Tarshecki:

Volkking is an IMITATION of FUTURE, so to speak. Its purpose is really a test. If I have two options, I run them through volking and see which one earned me more in an unfamiliar situation. I choose, make corrections, compare them again - you get an evolution without adjustment. The idea of volking -INERITY of the market, so for trading I take, of course, the most actual one, but as far as I have determined the optimal periodicity for variable updates - in our case it is back duration. The duration I find experimentally, again with the help of volking. Thus overoptimization is more or less avoided.

I.e. settings for Expert Advisor should be as fresh as the "step" of verification interval of your volking allows.

And the most interesting thing is what method to use to select the only variant out of 10000+ which will be used in real trading? It seems that my option - max profit at <20% drawdown, is not very good.(
 
elibrarius:
And the most interesting thing - what method to choose the only one out of 10000+ variants that will be used in real trading?
I.e. choose the method of selecting variants, which provided you with the best result in terms of forward sum when you passed the whole test.
 
elibrarius:

I chose the selection criterion not max profit, but max profit at a drawdown<20%. Who has - what are the options for selecting that one result from the optimisation that you run in real work?


And you just try both of them. Which method has the best sum of OOS profit - that variant is better. And then you will advise people. The same thing with the step of testing, it may be different for each case.
 
Youri Tarshecki:
And you just try both. Which method has the best sum of OOS profit - that variant is better. And then you will give your own advice to people. The same thing with step of testing, it may be different for each case.

And what are the options for trying?

You've already made a good point: that "And this, in turn, is usually determined by capacity."

I've spent a few weeks on an endless number of optimizations, haven't seen any interesting results yet. It's a pity I didn't do a forward for all results at once and didn't save them and results of back-optimizations. It could take a long time to optimize and optimize....

That's why I'm asking you to share your best practices for optimizing results.

 
Youri Tarshecki:
And what other criterion can you suggest to figure out which version of the code is better?
OK, I can't suggest anything other than that.
123456789...13
New comment