Overcome over-fitting strategy with reoptimizationn

[Luandre Ezra]

Hi all, I'd like to hear your opinion about over-fitting and reoptimization. When we tried to find the best parameter for our strategy we often found an over-fitting situation and we all know that over-fitting could lead us to loose money.

One thing that I know is that the over-fitting strategy will loose at certain time in the future, it will not make us out of the market with a single trade (there's always a SL for every trade). Because the over-fitted strategy will loose it's power in the future, Now comes a question, what if the strategy be re-optimize to find the next best optimization parameter?

Doing this cycle of finding the best optimization and re-optimize it frequently does it means that we overcome the over-fitting in our strategy?

Appreciate your opinion.

[Sergey Golubev]

There is one article with curve fitting (over-fitting) explained and how to avoid it:
How to Test a Trading Robot Before Buying

[Lorentzos Roussos]

Luandre Ezra:
Hi all, I'd like to hear your opinion about over-fitting and reoptimization. When we tried to find the best parameter for our strategy we often found an over-fitting situation and we all know that over-fitting could lead us to loose money.

One thing that I know is that the over-fitting strategy will loose at certain time in the future, it will not make us out of the market with a single trade (there's always a SL for every trade). Because the over-fitted strategy will loose it's power in the future, Now comes a question, what if the strategy be re-optimize to find the next best optimization parameter?

Doing this cycle of finding the best optimization and re-optimize it frequently does it means that we overcome the over-fitting in our strategy?

Appreciate your opinion.

Well you don't know if the "overfit" is partially sitting on the multi dimensional space that is "your strategy working" .

So if you can imagine that for any given strategy there's a space that the strategy "flows", it performs at its best , not profitable per se but at its maximum ability.

You want to be "stepping" in that space as much as possible .

One way is to run an initial heavy optimization up to a date , you will limit the date inside your test not on the tester .

Then for the remainder of the test period (after the learning cutoff date) you want the strategy to be as stable as possible . Then you should give the genetic algorithm a reward that takes into account both profitability and stability .

[Lorentzos Roussos]

Sergey Golubev #:
There is one article with curve fitting (over-fitting) explained and how to avoid it:
How to Test a Trading Robot Before Buying

That is not what he asked i'm afraid 

[Fernando Carreiro]

Your topic has been moved to the section: Expert Advisors and Automated Trading
Please consider which section is most appropriate — https://www.mql5.com/en/forum/172166/page6#comment_49114893