Open jorgecarleitao opened 6 years ago
Thank you for reporting this issue. I did some tests a year ago between different packages (R/Python) and the results were inconsistent - in some cases the PyFlux implementation gave superior results, other times other packages gave superior results. But in the case you give, it looks like a problem with the PyFlux optimization. Will try to have a look when I get some time.
Consider the code below that
I am getting that pyflux does not estimate the MA coefficient right. In opposition, statsmodels does. Anyone gets the same result?
How to reproduce:
If I run
_test_ARMA(1, 0)
or_test_ARMA(2, 0)
, the test passes. If I run_test_ARMA(0, 1)
the test does not pass. Specifically, the MA coefficient is computed to be ~0, while the expected result is 0.5.I took the same test and ran it with
statsmodels
usingand the test passes (ignoring the normal scale coefficient that is not estimated by the statsmodels).
Any ideas?