A strategy that only works on certain currencies?

[R1234]

I've got a swing trading strategy I've been working on that I hypothesized should work, and it seems to work in backtests on certain currencies. I've also traded it manually a few times to get a feel for it. The problem is that it doesn't work on all currencies and I don't know how to determine when to trade a currency with this strategy.

I began looking into it from the perspective of distributions of various price-related data and found what appears to be some sort of a relation between cumulative returns and this distribution.

In the images that follow, the left chart is cumulative return (dark blue) vs price (light blue). The right chart is a price-related distribution. To me, there appears to be some sort of a pattern with the currencies where it does work, and currencies where it doesn't.

I don't understand how to determine what really affects it.

Note: these results are without leverage. I would likely do 5x leverage in live trading.

Based on comments from others on this site, I am considering using deep learning to see if there is a way to back out a relationship that I don't know exists but I'm not sure I want to invest the time TBH.

What next steps would you take here?

View attachment 202754 View attachment 202755 View attachment 202756 View attachment 202757 View attachment 202758 View attachment 202759 View attachment 202760 View attachment 202761 View attachment 202762 View attachment 202763

on a side note, when you backtest forex do you account for the rollover swap rates? For example if you are long $10k of USDMXN you get debited about $2.00 or if you are short USDCHF you get debited about $1.00. For this reason I prefer to backtest on backadjusted CME currency futures where it is all built-in

 

[nooby_mcnoob]

on a side note, when you backtest forex do you account for the rollover swap rates? For example if you are long $10k of USDMXN you get debited about $2.00 or if you are short USDCHF you get debited about $1.00. For this reason I prefer to backtest on backadjusted CME currency futures where it is all built-in

haha nope.

 

[athlonmank8]

Why do you want to trade every currency? Why not just use the strategy on the ones it works best on and toss out the rest.

 

[nooby_mcnoob]

Why do you want to trade every currency? Why not just use the strategy on the ones it works best on and toss out the rest.

Because unless I understand WHY it isn't working, I wouldn't trust it. Based on the distribution transformation, I have some idea as to why but I'm not sure if it is overfitting to apply that transform only for the funky distributions.

 

[nooby_mcnoob]

Subscribe to this thread for the "Found the bug" post that will come in the next week.

Perfect. I found an edge last year.

Although.. Not the first time returns have been here, it looks like the MACHINE LEARNING ALGOS have found the same edge.

 

[tommcginnis]

I'm sensing an inverse correlation to the number of Presidential tweets.

 

[nooby_mcnoob]

I'm sensing an inverse correlation to the number of Presidential tweets.

holy shit

 

[Stockolio]

Holy edge you need is Hong Kong Dollar is about to poppppp! HK CB might defend one or two more times, but it's clear they are out of US Dollars, can't defend much more, they have raised Interbank rates quite drastically which will collapse their banking system, just to attract few extra US Dollars... Might have missed the boat on it, but worth pointing out

 

[nooby_mcnoob]

I'm sensing an inverse correlation to the number of Presidential tweets.

I'm still trying to figure out what the correlation there would be. Will look into volatility, that's the only thing I can think of.

 

[nooby_mcnoob]

Returns are more stable if you include specific currencies, from this lens, doesn't look too far from the norm.