• From: France
  • Registered: 2018-07-26
  • Posts: 80

Topic: What are we fighting against ? Over-Optimization only ?

Hello,

I'm in this "game" for many years now without success so I've decided to return to the basics!

Why an EA thats show great equity curve on backtest fail on live trading ?

- Over-Optimization is the most critical part of why EA's Fail on live trading
- Another part is in backtest DATA, The backtest Data and live data should "match"
- Quality of the data, we do not want big gap or anything weird in the data


I think if we generate EA's thats not been Over-Optimized and have been created on good data, Live trading should be very good and show good profit ! but we all know thats not the case !
It is "easy" to generate EA with good data and that have not been Over-optimized (Thanks OOS)
So why EA's keep failing !?

An EA that passed OOS of lets say 6 month should easily perform on live data for at least 2 month isnt it ?
So why these EA fail too ? what are we fighting ?

Thank you

Sorry for my english, its not my first language

  • Popov
  • Popov
  • Lead Developer
  • Offline
  • From: Bulgaria
  • Registered: 2006-11-30
  • Posts: 7,972

Re: What are we fighting against ? Over-Optimization only ?

I think the only way to find some reasonable answer is to have measurable data.

A long time idea is to make a system that compares (optimises) the generating -> optimising -> validating workflow.

For example, we can generate Collections with different methods and settings and then find which workflow makes bigger or more reliable profit on newer data.

One possible scenario is to generate collections one the same data and to store them together with the settings.
Then to test them on newer data and to sort them on different criteria. in that way, we will know which settings and workflow have better results.

We can make a social platform for finding the best workflow based on EA Studio.
Let's say traders generate collections on a defined market (for example Premium Data H15 UTC from 01-01-2019 to 01-01-2021). The collections can be generated with various settings of the Reactor with various tools switched on.

The traders to be possible to "Push" the collections with the settings to a Database. Then we can have a service that calculates the Collections on new data and show stats.

It is possible to find which tools or options and settings make difference.
For example, if the Optimiser makes a better profit on newer data or overoptimizes.

We can even have this as a continuous process and have data every month. Let's say, to have the settings that produce a greater profit. Or what OOS per cent is better. Or if the Multi Market makes difference.

What do you think?

Popov's Website

  • From: France
  • Registered: 2018-07-26
  • Posts: 80

Re: What are we fighting against ? Over-Optimization only ?

I think it is a great Idea!
this will allow us to go forward and have reliable data to rely on

EA Studio is so powerful and have so much good tool that we get lost with our testing.
We want to try everything but we get lost on what work really.
A tool like that will help us monitoring what we are doing and what really work.
For me thats a genuine idea

  • Registered: 2021-05-08
  • Posts: 1

Re: What are we fighting against ? Over-Optimization only ?

Popov wrote:

I think the only way to find some reasonable answer is to have measurable data.

A long time idea is to make a system that compares (optimises) the generating -> optimising -> validating workflow.

For example, we can generate Collections with different methods and settings and then find which workflow makes bigger or more reliable profit on newer data.

One possible scenario is to generate collections one the same data and to store them together with the settings.
Then to test them on newer data and to sort them on different criteria. in that way, we will know which settings and workflow have better results.

We can make a social platform for finding the best workflow based on EA Studio.
Let's say traders generate collections on a defined market (for example Premium Data H15 UTC from 01-01-2019 to 01-01-2021). The collections can be generated with various settings of the Reactor with various tools switched on.

The traders to be possible to "Push" the collections with the settings to a Database. Then we can have a service that calculates the Collections on new data and show stats.

It is possible to find which tools or options and settings make difference.
For example, if the Optimiser makes a better profit on newer data or overoptimizes.

We can even have this as a continuous process and have data every month. Let's say, to have the settings that produce a greater profit. Or what OOS per cent is better. Or if the Multi Market makes difference.

What do you think?

hello
I wanted to know what "newer data" is?
Do you mean we should generate strategy on our broker's data and after generating we select the good ones then we run them on data from another broker to see if they really are profitable experts??
Could you please explain more about that? I'm a bit confused.
Thank you so much.