Sunday, 29 November 2020

Temporal Clustering on Real Prices, Part 2

Below are some more out of sample plots for the Temporal Clustering solutions of the EUR_USD forex pair for the week just gone. The details of how these solutions are derived is explained in my previous post, Temporal Clustering on Real Prices. First is Tuesday's solution

where the major (blue vertical lines) turns are a combination of optimal K values of 6 and 7 (5 sets of data in total) plus 2 sets of data each for K = 9 and 11 (red and green vertical lines). The price plot is
Next up is Wednesday's solution
where the blue vertical lines represent 5 sets of data with K = 6 and the red and green vertical lines 3 sets and 1 set with K = 9 and K= 11 respectively. The price plot is
Thursday's solution is
where black/blue vertical lines are K values of 9 and 6 respectively, whilst green/red are K values 10 and 7. Thursday's price plot is
Finally, Friday's solution is
where the major blue vertical lines are K = 9 over 5 sets of data, with the remainder being K = 5, 6 and 11 over the last 4 sets of data. Friday's price plot is

The above seems to tie in nicely with my previous post about Forex Intraday Seasonality whereby the above identified turning points signify the end points of said intraday tendencies to trend. Readers might also be interested in another paper I have come across, Segmentation and Time-of-Day Patterns in Foreign Exchange Markets, which gives a possible, theoretical explanation as to why such patterns manifest themselves. In particular, for the EUR_USD pair, the paper states  

  • "the US dollar appreciates significantly from 8:00 to 12:00 GMT
    and the euro appreciates significantly from 16:00 to 22:00 GMT"

Readers can judge for themselves whether this appears to be true, out of sample, by inspecting the above plots. Enjoy!

No comments: