Just guessing. Paper trading will tell me how right/wrong I am but the general idea would still apply.
Of that first set, I see an obvious link between the right-hand bars for 1-2-3, 4-5-6, and 7-8-9 -- but that's all I see. Nothing jumps at at me for which I might infer any link to the left-hand side -- whether the dark blue line, the light blue line, or their difference. As far as the in-sequence similarities of the right-hand bars, my forex days bring to mind those pairs that move in-step with others, and without looking, I suspect that the groups you posted were specifically from those-that-move-together? Sorry, Noobs. No help here.
Thanks! I'm not so much looking at the answer as to whether there is a link, but more like what steps would you take next to find out _why_ certain currencies work well, and others don't? My distribution was just a guess as to what could be related. It does seem to be sort of related but I don't even know how to verify that, to be honest! And to answer your question, I just posted all of them in random order, no distinction on ordering. The forum software did not let me upload more than 10 per post so I had to break them out
How do you know the 'works / doesn't work' is pure chance? What is the annualised sharpe ratio of the best and worst currency, and how many years does the backtest last for? GAT
How do I know it *is* or *isn't* pure chance? I am 99% sure that it is pure chance as usual. Backtest is 10 years, but I use that just for validation, hypothesis was tested on trailing 1 year. No sharpe ratio yet, currently working on reshaping the distributions and propagating that back to the prices to see if that makes a difference. Like all of the other stuff I've posted here I'm sure it won't work!!!
I meant chance in a statistical sense. What I meant was, are the returns of the worst and best currency statistically different? If not then you should just trade them all and not worry about why one works or doesn't. GAT
Ho-lee-fuck-ing-shit... Rearranging the distributions and propagating the transform back to the prices does the right thing. Which means I'm definitely doing it wrong lol. Before: After:
Sadly I can't wholesale apply the transform (gives worse results than before) which means I have to identify the currency's distribution first, then transform only that one. Why? Who the fuck knows ... So now I'm into overfitting territory and will have to pick up Robert's book again. FML.