Is there a better way to calculate time-shifted correlation matrices?

[Frank Einstein]

TL;DR

I want to know if there is a better way to obtain the correlation matrix of time-shifted series than just removing observations.

Hello everyone.

I have four thermometers which measure the temperature in four different positions. The data is distributed as a matrix, where each column is a sensor, and each row is a measurement. All measurements are made at exactly the same times, one measurement each hour. I have calculated the correlation matrix between all four positions.

Now I am interested in the calculation of the time-shifted correlation matrix. The only method I can think of is to remove the first n rows of the sensors 1 and 2 and the last n rows of the sensors 3 and 4 to see how the correlation changes.

I was wondering if there is a better way to do this than just removing rows.

Any help is appreciated.

Best regards.
Frank.

PS. I am using Python, so I have just used the function np.cov(Tdata_shifted2) and np.cov(Tdata) to obtain the shifted an non-shifted matrices.

 

[andrewkirk]

This stackexchange problem seems to match yours.
Most answers seem to only address calculating autocorrelations of each sensor with itself, not cross-sensor delayed correlations. It looks like you do want those latter. The answer by jboi (Feb 17, 2018 at 22:38) seems to provide those.

 

[Frank Einstein]

Thanks for the answer