Warning/ENH: coint - Augmented Engle Granger cointegration test
24 views
Skip to first unread message
josef...@gmail.com
Aug 9, 2016, 3:01:28 PM8/9/16
to pystatsmodels
https://github.com/statsmodels/statsmodels/pull/3146
coint is a hidden function that is in use in the wild. Several
options are ignored and replaced by hard coded values.
I'm finally replacing it by a cleaner function, after having learned
enough about cointegration during the summer.
It is modeled after the Stata user contributed function egranger, and
should produce identical results.
I'm not sure whether there should be small sample corrections.
The signature is a bit longer than needed to make space for future
extensions, e.g. adding Phillips-Ouliaris cointegration test, and
returning a results instance. (That's left for another day.)
It uses adfuller, with default maxlag and autolag, so it's slower
unless the number of lags is fixed (autolag=None)
Josef
coint is a hidden function that is in use in the wild. Several
options are ignored and replaced by hard coded values.
I'm finally replacing it by a cleaner function, after having learned
enough about cointegration during the summer.
It is modeled after the Stata user contributed function egranger, and
should produce identical results.
I'm not sure whether there should be small sample corrections.
The signature is a bit longer than needed to make space for future
extensions, e.g. adding Phillips-Ouliaris cointegration test, and
returning a results instance. (That's left for another day.)
It uses adfuller, with default maxlag and autolag, so it's slower
unless the number of lags is fixed (autolag=None)
Josef
josef...@gmail.com
Aug 17, 2016, 10:10:01 PM8/17/16
to pystatsmodels
but I forgot that I wanted to add a warnings.warn('behavior has changed")
Josef
>
> Josef
Reply all
Reply to author
Forward
0 new messages