Seasonal adjustment is a crucial step preprocessing the data for further research. Researcher however has a number of options for trend-cycle-seasonal decomposition. The most common (judging by the number of citations in empirical literature) rival seasonal decomposition methods are X-11(12)-ARIMA, Tramo/Seats (both implemented in Demetra+) and $R$'s stl. Seeking to avoid random choice between the above-mentioned decomposition techniques (or other simple methods like seasonal dummy variables) I would like to know a basic strategy that leads to choosing seasonal decomposition method effectively.

Several important subquestions (links to a discussion are welcome too) could be:

  1. What are the similarities and differences, strong and weak points of the methods? Are there any special cases when one method is more preferable than the others?
  2. Could you provide general guides to what is inside the black-box of different decomposition methods?
  3. Are there special tricks choosing the parameters for the methods (I am not always satisfied with the defaults, stl for example has many parameters to deal with, sometimes I feel I just don't know how to choose these ones in a right way).
  4. Is it possible to suggest some (statistical) criteria that the time series is seasonally adjusted efficiently (correlogram analysis, spectral density? small sample size criteria? robustness?).
  • 1
    $\begingroup$ You may be interested in this answer and in the references given there. $\endgroup$ – javlacalle Apr 10 '15 at 21:55

If you are willing to learn to understand the diagnostics, X12-ARIMA provides a boatload of diagnostics that range from (ASCII) graphs to rule-of-thumb indicators. Learning and understanding the diagnostics is something of an education in time series and seasonal adjustment.

On the other hand, X12-ARIMA software is a one-trick pony, while using stl in R would allow you to do other things and to switch to other methods (decompose, dlm's, etc) if you wish.

On the other-other hand, X12-Arima makes it easier to include exogenous variables and to indicate outliers, etc.

  • $\begingroup$ Well it means I have to learn the tricks beyond X12-ARIMA first, because most of the diagnostic tools are usually hidden in statistical packages. From practical point of view, when I tried monkey style press-the-bottom-get-the-result, I have found that Tramo/Seats works better (judging purely visually by laughing-monkey test) than X12-ARIMA, for stl I usually do the same monkey style work, so what I want is to learn the art of seasonal decomposition. (+1) for the general guides! $\endgroup$ – Dmitrij Celov May 9 '11 at 17:08
  • $\begingroup$ In X-12-ARIMA, the default .out file has pages of diagnostics, and if you read the manual and turn on a few more, you'll literally have pages and pages of information, ASCII graphs, and diagnostics. It's very logically organized and numbered and all the diagnostics refer back to the section its data came from. Walking through these diagnostics and learning what's necessary to understand them is very educational. Some of the diagnostics have ingenious heuristics. It's not hard to get most of this information put into files that you can easily import into R to manipulate and properly graph. $\endgroup$ – Wayne Sep 16 '11 at 17:06
  • $\begingroup$ For the time being (if nobody will try to give more details), I mark this one as correct, but what I would personally like to know is a practical guide, what rule-of-thumbs and graphics proved to be useful, and many other how-to-things from those who dig much deeper than me. Say I'm a bit lazy kind of person to read the manuals, but if you say do it, probably I should, thanks to the links below... $\endgroup$ – Dmitrij Celov Sep 17 '11 at 15:31

That's an answer for question 2.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.