High/Low Correlation

Hello ,

it's a long time that I do not write on a forum in the meantime I studied a lot everything related to machine learning and data mining.
I write this thread to ask an explanation about the strange correlations that I found on EURUSD (against other currencies will have similar values). I explain briefly what I did:

I have calculated the first differences between the high, low, close, open; for example assuming to take into account the 3 subsequent high as for example 1.3000 --- 1.3050 1.3020 ----- the first differences are 0, 0050 and -0.0030
I calculated the correlation - https://en.wikipedia.org/wiki/Correl...and_dependence - considering different lag values values (for simplicity here are reported only lag 1, that represents the correlation of the first differences between current and old values . Through this calculation I found some high correlation values:




This means taking as an example the first row that High at time t is very high correlated with close at time t-1, medium correlation with low at time t-1 and low correlation with high at time t-1. For close you can see that it is virtually uncorrelated with the past.

As for High / Low values seem strangely high, moreover this would imply that the prediction of the sign is possible for high and low while the close is practically impossible (see line diff_close).

These values are calculated by taking into account different time periods (typically annual) values taken from 1h chart, but the same values can be founded on different timeframe and even on different currencies.

I wanted to know if anyone can explain these so high values for high/low and in the future maybe hypothesize a practical application.

Thank you,
Andrea182

0 means neutral? and the open is not 100% correlated to the previous close due to gaps?

0 means that there is no correlation with that data.

Open is high correlated with previous close 0.97 but not 1 due to gaps.

Andrea182

Why not try dynamic time warping https://en.wikipedia.org/wiki/Dynami...g#cite_note-10 or some similarity analysis like euclid, manhattan,mahalanobis etc... distance measures?

Hi Patara,

sorry but i can't understand your answer, how is correlated the DTW or the other similarity analysis with the correlation that i've found?

Thanks,
Andrea

I would like to highlight a very simple thing, that can be easily verified . Take A candle with close greater than open, you will notice that, it's more likely that the future high will be higher than the present high (the same concept can be extended to the low). If you combine this concept to this one:
a trend is represented by higher high and higher low....

that means only one thing....

Another piece of information....even this paper support my hypothesis
https://www.snb.ch/n/mmr/reference/w..._2011_11.n.pdf
High and low are predictable as demonstrated before.

Hello Andrea182,

Thanks for the research paper. Looks interesting.

I also see that the next high is higher than this high...

Could you please elaborate on

"If you combine this concept to this one:a trend is represented by higher high and higher low...."

I mean how do we combine with Higher High and Higher Low?

Thanks

With my findings (supported by an approved paper) you easly know that future high and low are predictable.

A trend is represented by higher high and higher low so if you know future high and low you know future trend.