I have some minute-bar data which my professor suggested I resample to 5 minute bars and then separate it into timeseries per bar period. For example, I get one time series for 12:00, another one for 12:05 etc. Then a single timeseries would have changing date component but the time would be the same:

2019-12-02 12:05:00, 10.10
2019-12-03 12:05:00, 10.35
2019-12-04 12:05:00, 11.18
2019-12-05 12:05:00, 11.90
2019-12-06 12:05:00, 10.30

He then suggested I do a linear regression between these timeseries suggesting this would help me find patterns. Before I speak to him further about this, I would be interested to learn more on how one uses linear regression to uncover patterns.


You should come up with some relationship you want to explore. For example maybe you think that the t+1 observation depends linearly in some way on the last (t) observation. Then you would regress the t+1 observation on the t observation and examine the fit. Maybe you think that the sum of the last 5 observations has a relationship to the t+1 observation. Then you would regress the t+1 on the sum of the last 5, etc.

  • 1
    $\begingroup$ Hi: You should not use linear regression on prices. ( prices are far from stationary and regression assumes stationarity of the thing being predicted ). Assuming your professor is not aware of this, then you're probably better off speaking to someone who is more familiar with the field of pattern discovery in finance. $\endgroup$ – mark leeds Feb 15 '20 at 16:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.