tsa.stattools.acovf is slow and is missing nobs options about statsmodels

[bashtage]

@josef-pkt This ancient issue can be closed. Both are fixed in master (fft on its way to becoming true, and nobs)

[josef-pkt]

What about the "auto" options?
I was looking at this issue earlier today and wasn't sure about "auto"

[bashtage]

I think auto isn't worth much since the difference between the methods when the time series is slow is in the us range while the difference when the series is long is in the s range. Always using fft is the simplest way to ensure good performance when it matters.

Additionally, the auto can't really exist since it is platform dependent and it is possible that fft may be faster or slower for any (small) size due to the implementation of the fft and whether it makes use of uarch features of the cpu being used.

[josef-pkt]

thanks, closing then