I will test this strategy on demoaccount.
Your tutorial suggests using two currencies to test a strategy in order to minimize the chances of curve fitting. Could this be used to test all related currencies at the same time by selecting multiple currencies? For example you suggest testing the EURUSD and the second currency of GBPUSD. Is there any detriment to also add the AUDUSD and NZDUSD pairs?
using more currencies is an option – the more currencies the strategy work on, the more robust it is. but it is also much more difficult (and it takes more time) to find a good strategy that works on multiple currencies.
EURUSD and GBPUSD behave in a similar way, it is relatively easy to find a strategy that works on both without any parameter changes.
Other currencies might not work as good, it depends on how much correlated are the pairs.
AUDUSD should be correlated to NZDUSD, so it should be also relatively simple to find a strategy that works on them both.
so we can make a list of corelated currencies.
I think here are in the forum are many professional trades who have the knowledge to say what strategies are good correlated?
At the moment we have.
Who have more suggestions for correlated currencies?
last week I’ve tried
EURJPY-GBPJPY EURJPY as the one for finding strategies and GBPJPY to crosscheck
USDJPY-GBPJPY / EURJPY USDJPY as the one for finding strategies and GBPJPY / EURJPY to crosscheck
I stopped those as the results wasn’t that good. As Mark already sad some posts ago, it seems quit complicated to find good strategies on Yen pairs…
Note: I am using Mark’s setup file and history data from foreshistorydatabase (bought version).
for currency correlation I found this
The next question is.
1) We need currencys who have a stable correlation. I mean the currencys must have a good correlation over a long period.
So we have to make a snappshoot of this table more than one time and compare this tables over a long period.
The currency pariss who have a good correlation over a long period will be a good candidate.
2) Make it sense to build an indicator of this correlation matrix. For example. If some currency´s have good correlation I will start my eas
Make it sense ? Have someone has experience with this ?
I think you make it too complicated if the goal is only to find currencies that can be tested together.
I look at big picture – EURUSD and GBPUSD correlate because they are a part of EU economy, any change in UK causes also change in EU and vice versa.
NZ and AU are also connected economies, when one grows the other grows too.
You can trade correlation between pairs, I was playing with that before.
I didn’t make anything useful, but I’ve seen a successful robot doing that, but it is quitedifficult.
The concepts are totally different than trading just one currency. And in MT4 it is not possible to backtest multi symbol EA, the robot was also for another platform.
I also think just correlation doesn’t necessarily mean that the same strategy will work on both currencies, I guess EURUSD and GBPUSD are quite similar,
but some other correlated currencies might have very different candle ranges, which could cause the strategies to fail on one of them.. But it is just my idea.
Hey Mark and others, I’m currently following your post from 18.07.2013.
I am at the beginning, means the system is generating strategy almost a week now. The next step is to stop the system and sort the generated strategies by there OOS performance.
The question is now, as strategyquant is always doing this genetic evolution among generating new strategies I was asking myself if there is a special point of time where to press the Pause/Stop button or is the software doing that automatically?
you can klick stop or pause every time you want.
But the system react not every time immediately at you action.
This is a problem of the java gui-system.
It is not so easy to solve this problem.
But the normal process of generating random or genetic calculation takes
one or two days. I think the problem is not so important.
You haven’t said whether your in random or genetic mode. If in random I let it run for as long as I’m accumulating profit, lowest dd%, lowest stagnation%, highest profit factor and lowest dd for robustness.
So its not just oos profit to look for in either random or genetic mode. Set your views to items you feel best describe characteristics your looking for, those happen to be mine.
Afterwards I move entire databank to initial population bank to geneticize. I uncheck robustness tests here because it seem to drag down speed the most in genetic mode, I may be wrong on that, but its what I noticed.
I’m not sure I was any help for you, just noticed you letting it run for so long so spoke up. Feel it better to let runs run as long as you get significantly observed above factors improved results arriving, not just let it run & run, may be wrong but results seem to get to a plateau of minimal improvements at some period, of course it depends on how many years and periodicity and whether open-bar or otherwise.
as Thomas and Jerry explained, the program can work as long as you stop it. Also in genetic evolution mode it can repeat the evolution indefinitely.
This is a feature – you can let it run until it generates sufficient number of strategies for further filtering.
It could take hours to days – depending on your settings and filtering options.
I usually let it generate at least 1000-2000 strategies, then I stop it and I’ll continue with the process described in the article:
I’m posting one more strategy that is mentioned in Walk-Forward Matrix article.
This one is for GBPUSD/H1, it passed the robustness tests and I currently test it on a small live account.
The chart shows equity of this strategy during walk-forward optimization (blue) compared to original strategy without reoptimizations (gray).
Just a note – this strategy is strictly for SQ owners, don’t publish it anywhere else.
i retest this strategy and its not so good, did i make any mistake? i tested 2007-2013 and DD is bigger than profit….
But thank you for sharing this.
did you test it on the same data with the same settings?
When you get different results it is usually problem with different settings.
Could you save the strategy with your test results and send it to me or attach it here?
I can check what is wrong.
You must be logged in to reply to this topic.