Comparison of two quotation charts with non-linear distortions on the X-axis

 

Have any of the intraday people noticed that often two EURUSD or GBPUSD intraday charts are similar? Not always, of course, but often yesterday's pattern surprisingly repeats today, on which you can try to profit. But...

The peaks and troughs, though repeating the pattern, do not coincide in time. For example, yesterday's mid-day dip started at 2:15pm and today's at 1pm. There are many similarity criteria - Spearman, Pearson, least squares, but I don't know of any that compare graphs subject to small distortions on the X-axis. No one knows of any such methods?

 
Breaking through a flat on a cross?
 
sv.:
Breaking through a flat on a cross?

This is a different task. I am not comparing EURUSD and GBPUSD charts, but today and yesterday of the same pair.
 
Autocorrelation? There are solutions in the code base.
 
I can suggest the following: enter a non-linear time for one of the graphs, setting it, for example, as a piecewise linear table function, dina segments and their "tempo" - parameters. Next, maximize the correlation coefficient of the two graphs using any available numerical method and selecting appropriate segment parameters. It is time-consuming, but it will work.
 
alsu:
I can suggest this: introduce a non-linear time for one of the graphs, setting it for example as a piecewise linear table function, dina segments and their "tempo" - parameters. Next, maximize the correlation coefficient of the two graphs using any available numerical method and selecting appropriate segment parameters. It is time-consuming, but it will work.


Pondering.... genetics? ....

What if we go from the graphs themselves? Approximate the graph of a polyline (this in itself allows thousands of variants), and then compare the polygons, allowing small shifts of vertices along the X-axis?

 
wmlab:


Reflecting.... Genetics? ....

L-BFGS, Levenberg-McVardt method, etc. etc.


What if we go from the graphs themselves? Approximate the graph of a polyline (this in itself allows for thousands of variations) and then compare the polygons, allowing for small shifts of vertices along the X-axis?

You can. But you will have to equalise the number of knees beforehand.
 
It is possible to approximate the graphs by polynomials of sufficiently large order... Let's say 8-10 should be enough to start with, and adjust the time transformation so that the coefficients of the polynomials match as much as possible
 

This is a particular case of finding a master/delinquent relationship. It is solved through an appropriate transformation of the TSP. And then applying the usual linear methods.

  1. Check out "History Transformation" here.
  2. And a quick calculation of Pearson's QC here.

P.S. The applicability of pattern theory should still be justified.

 
Alternatively, we should define what is "similar" and what is "identical"
(so that we can talk about similarity to a degree)

Perhaps a map of lows-maxes (or impulses up and down) should be involved.
Identify min/max of different levels - e.g. levels 1/2/3 etc.
Need to identify a reference point.

For example, if the higher level min/max sequence matches-
you could simply compare min/max lines.

Actually, if we are talking about a formal classification of days - then I did such a job.