I am trying to reconstruct the US SAAR housing starts series from the US Census Bureau using State-level data which is nsa and not annualised. The seasonal adjustment used is based on the X12 algorithm but I am unsure about the parameters used.
I have tried several specifications and used x12(mode=a,filter=msr) (I cannot use a multiplicative adjustment because of NAs in the Washington DC series).
The match is pretty good:
- housing1.JPG (42.14 KiB) Viewed 1849 times
but my saar series (state_tot) is more volatile than the reported one(usa_houst):
- housing2.JPG (41.19 KiB) Viewed 1849 times
I am not very familiar with the X12, could you please suggest a specification that results in a smoother saar series?
Thanks,
Javier