ccf only return half cross-correlation coefficient and thus sensitive to input order

[langchen2022]

Describe the bug

I use ccf to calculate cross-correlation coefficient between two time-series arrays, but found that only half correlation was given (if I'm using it correctly). Then the maximum coefficient is sensitive to the order of the two input arrays, as the peak correlation coefficient may not be capture in one order.

Code Sample, a copy-pastable example if possible

I have a matrix called "all wav", and I want to calculate the cross-correlation coefficient for two of them. Here are what I got in different order:

Expected Output

Here is what I got directly with numpy.correlate:

Output of import statsmodels.api as sm; sm.show_versions()

statsmodels

Installed: 0.13.5

Required Dependencies

cython: Not installed
numpy: 1.22.4
scipy: 1.9.1
pandas: 1.4.4
dateutil: 2.8.2
patsy: 0.5.3

[josef-pkt]

I think we can add a keyword option to return the two-sided ccf/ccovf.

When we started this part, then we only looked at causal filters with one-sided correlation (coming from time series analysis in econometrics).
Some of those functions were or are just wrappers around scipy.signal functions that translated those to tsa in econometrics.

[langchen2022]

Now I can see why it's set up this way.
I wanted to use ccf instead of scipy.signal.correlate because I'm more interested in the normalized coefficient and I don't have to calculate myself :)