Univariate Time Series Decomposition

[Brajesh Singh]

Hello,

I'm trying to find out if there's any implementation that allows one to decompose a given Time Series into Seasonal, Trend and Level components like what's been done in 'decompose' in R or in the 'forecast' package of R.

Regards,

Brajesh

[Skipper Seabold]

For seasonal trends, not at the moment no, but I've been working on
and off on this recently, and we should have STL, etc. at some point "soon".

For non-seasonal but still periodic trends, you should look at the
available filters -- Baxter-King, Christiano-Fitzgerald, and
Hodrick-Prescott for now.

http://statsmodels.sourceforge.net/devel/tsa.html#other-time-series-filters

Skipper

[Andreas Hilboll]

feel free to use my simple rpy2-based wrapper to call R's STL from Python:

https://gist.github.com/andreas-h/7808564

Cheers, Andreas.

[Skipper Seabold]

Nice. Thanks. What other goodies do you have to throw up on github?

Skipper

[Andreas Hilboll]

Thanks :) So far I've only played around a tiny bit with rpy2, so I fear
that the STL wrapper is my only goodie so far. But soon (?) I hope to be
doing the same for strucchange_.

Cheers, Andreas.


.. _strucchange: http://cran.r-project.org/web/packages/strucchange/