Strategies that work on multi pairs are worse than strategies generated on single pairs!

Just got this in my inbox from my well respected friend Daniel. His findings are kinda shocking to me, but are in relation with what I´ve found out the last months too after doing some heavy testing by generating strategies that work on up to 20 pairs at once, then doing pseudo OOS and surprisingly came to the same conclusion as him. I didn´t know he was testing the same right now, but “great” to see he found out the same so that I know I am not alone with my findings and that they are not wrong. Have a read and see his results:

http://mechanicalforex.com/2016/05/are-multiple-pair-strategies-better-than-single-pair-strategies.html

How weird, I’m literally reading that same thing now. Then saw your post. 🙂

I have never believed the “it should work on every currency, stock, commodity, metal in the world, on every timeframe, for every broker” rubbish.

It’s clear that for example, USDCAD is nothing like USDJPY, one being a commodity currency and the other being a “safe haven” currency.  Each pair has a unique character and strategies that succeed find some nuance on that pair.  What you find will not work on other pairs, unless they have similar characters, economies and safe haven risk on/risk off status.

A lot of Daniel’s findings I disagree with as it contradicts a lot of finds from professionals that I know are successful, but maybe if using his methodologies, his optimizations, his filters, his data miner etc, maybe its true then.

Certainly for manually (not generated) building strategies that take advantage of real patterns, its pretty important in futures and FX for it to work on a general level across them all unless its taking advantage of an asset specific pattern like shale rig count.

One problem I have had with testing on multi pairs is the parameters should be different. For example at 10 pip stop works great on eurusd but would fail on gbdjpy. So I was thinking maybe a good strategy is to build on one pair. Then test on WFM. If it passes then do WFM for other pairs. This way it will cycle through various parameters and “catch” ones thst work for this new pair. Then we could have a true test if it work on other pairs matching the specific behavior of thst pair. What do you think?

Sent from my HUAWEI MT7-TL10 using Tapatalk

Notch. Yes it’s been talked about. But jot what I am proposing. Second, from my experience, Atr is not the golden solution. The multiple of atr changes between pairs. So just using atr by itself with a multiple of 1 is not usually the solution. So this, the variable still will change on pairs. Hence my idea about using WFM as a potential solution that could yield the answer if a strategy will work on multiple pairs. Thus signifying it is robust.

Sent from my HUAWEI MT7-TL10 using Tapatalk

Okay,Notch . Thanks for the patience in explaining. It sounds like we are saying the same thing. So I guess my epiphany is an idea that you are already using. This makes me happy because it validates the idea as something I will start using.

Sent from my HUAWEI MT7-TL10 using Tapatalk