|Appendix C: Comparison with Census X-11 Arima.|
Seasonal adjustment by Arima Census x11
Census method adjusts time series by applying a centered moving average on a section of the series in retrospect. If this section contains 7 years of datas, the decomposition is considered to be final for the year that is then in the middle. Until that time provisional adjustments are used: a provisional when the new data is published, a 1st revision after 6 months, a 2nd revision after a full year and the final decomposition after three years.
The impact of such revisions is illustrated for the year 1989 of a rather stable seasonal time series. The series was corrected 5 times:
Col 3, when 1989.4 was the last known term
Col 4, 6 months later (first revision)
Col 5, 1 year later (second revision)
Col 6, after three years (final revision)
Col 2, by the Camplet method (no revision)
The scope of these revisions is reflected in the growth of one seasonally adjusted term over the preceding adjusted term.
Quarter to quarter growth is the one reason why interim figures are collected !
When a term is published, only the provisional adjustment is available. The revisions are inspected only incidentally and never for short term strategic purposes. More over, there is hardly a theoretical argument why the final adjustment should be better than any of the preceding adjustments. Each version is composed of a mixture of evolving or changing seasonal patterns.