Dear Tom:
I have a basic question about seasonal adjustment. For US, BEA publishes nominal and real seasonally adjusted quarterly GDP at regular intervals. For China, only nominal data is published. In general, X11, X12 ，X13-arima or tramo seats are used to do it, which use smooth technique. In real-time context, while new data gotten, data of previous date changed, due to change of sample. How can I seasonally adjust these quarterly GDP. I wonder the UC model with filter can hold historical data stable. Ignoring classical method, eg x12 or tramo seats, it seems not wise. How BEA do it.
Best regard
Hardmann
a question about seasonal adjustment in real-time
Re: a question about seasonal adjustment in real-time
If you google "BEA seasonal adjustment" you can find information on that---in particular, you might want to look at the Kornfeld power point. Much of the data that goes into GDP is seasonally adjusted by the source agency, so they are adding seasonal adjusted components to get seasonal adjusted totals. For very important series like GDP, there are a lot of judgment calls by the staff to prevent outliers from contaminating the estimates. In particular, the outlier adjustments for earlier years are usually fixed through a calendar year rather than being re-estimated month by month.