Behaviour of gold that I am unable to explain. Pls help.

New comment
 

I was optimizing my EA for GOLD. For the time filter I was trying to find out which session was the best. I got mixed answer for London and Asian session. However few even pointed out a 24hr session for gold without specific reasons an so were not so convincing.

I all this I was biased for Asian session being the best for gold.

On a 5 year test from 2016-2017 for 4 time sessions to check the performance of gold- 1.Asian 2.London 3 London+Asian and 4.24hrs.

I found that 2.London session & 3. London+Asian session have always been random and always given bad results.

However for 2016 a 24hr session was good while for 2018-2020 Asian session was good.

In fact during 16-17 Asian session was pretty bad for gold and that's what scares me.

I want to know why this happened and what factors might have contributed.


I would also love a few tips to introduce in an EA that trades gold.


thanks.

 
I would estimate this shows on which types of markets your algorithm is working.

Most (almost all) algorithms are specific to certain market structures. And I would state this is the proof.

My guess is, this should enable you to see what makes your algorithm or trading strategy in fact works on.

Now you could go ahead with this analysis and pick sample timescopes from these tests and optimize other parameters of your EA to better fit to the (obviously) incompatible market structures.

I would say, optimization is about finding the weakness of your Algo. Not about making it fit to a certain market type. -> Move away from curve fitting.

Seems to me very reasonable results.
 
Dominik Egert:
I would estimate this shows on which types of markets your algorithm is working.

Most (almost all) algorithms are specific to certain market structures. And I would state this is the proof.

My guess is, this should enable you to see what makes your algorithm or trading strategy in fact works on.

Now you could go ahead with this analysis and pick sample timescopes from these tests and optimize other parameters of your EA to better fit to the (obviously) incompatible market structures.

I would say, optimization is about finding the weakness of your Algo. Not about making it fit to a certain market type. -> Move away from curve fitting.

Seems to me very reasonable results.
Let me start by saying that by deep optimization I mean changing the static values that are fixed to decrease the number the number of optimization parameters of my EA and are always same for a specific instrument say eurusd and prevent curve fitting of my EA. 
And by simple optimization I mean the parameters (usually 2-3) that are slightly optimized every 6 months or so.

Now for a new instrument (in this case XauUsd) I deeply optimized at first. Simple optimization don't highly affect the proportionate results from time filter so they were set at an average value.
This was done since our sole scope was to compare the results during different trading sessions and not to get higher profits. Moreover time filters don't need changing, maybe during dst changes but I keep it on for the common period between the dst thus getting a lesser time slot. Changing time filter parameters may be done for a scalper EA maybe.

My EA is based on indicators so it should work on all markets if properly tuned with static for particular instrument. 

And the only doubt is xau showing different properties in few years gap. Specifically which market session really affects it. I'll let all know about it when I have an answer. Perhaps a gold enthusiast will have the proper explanation.
 
Occam's razor: your EA probably performs exactly the same on all sessions, backtest results just have some random variation.
 
Jere Katainen:
Occam's razor: your EA probably performs exactly the same on all sessions, backtest results just have some random variation.
If it does then it might prove that gold is not dependent on any time particular session and this might be true....perhaps a 10 year test in slots of 4 or 6 months will give a clear view. You may be right. I'll let you know. 👍🏻
New comment