Seasonal adjustment of series with no seasonal component

[CharlieEVIEWS]

Dear all,

I have been thinking about a specific time series issue for some time, and there is an unresolved issue with some colleagues. Perhaps you can help. If a large number of series are being used in a project, is it correct to seasonally adjust all of them, as some of the papers in the literature do? In particular - the question focuses upon the case where the series do not have a seasonal component. I was once taught that even if you adjust a series which i.) has either already been seasonally adjusted ii.) or (equivalently) contains no seasonal component (i.e. interest rates), then it won't make any difference as the seasonal adjustment procedure wont 'clean' any seasonality from it.

Does anyone have any information or opinions on this matter? I'd be most interested,

Best wishes,

Charlie

[trubador]

It depends on the procedure you use. TRAMO-SEATS, for instance, will not seasonally adjust the series if it does not find any significant seasonality. It will produce a series filled with NA values. CENSUS X12, on the other hand, will still produce an adjusted series even if there is no meaningful seasonality. As expected, both series will be very close, but not identical.

These procedures are specifically designed to ease the adjustment burden for statistics agencies, which have to deal with large data sets on a regular basis. Still, most agencies spend additional time to build customized adjustment parameters for each time series.

This is because adjusting/filtering the data actually means loosing information. And it is very difficult to know ex-ante whether the relevant part will be extracted. If you are going to use these series within a multivariate context for inference purposes, then you should consider to incorporate seasonality into the model as well. Of course, the computational burden will be the main determinant here.

[CharlieEVIEWS]

Dear Trubador,

Thanks so much (as always) for your fantastic response. Your answer is more or less consistent with my understanding of the situation. In this specific project, we are aggregating data from different statistical agencies who are going to be using heterogenous adjustment methods. May I ask you for your opinion on the adjustment strategy in our .prg?

If adjusted at source, we will not 're-adjust' with our adjustment routine (x.13). If a series is not adjusted at source, but we deem it to be necessary to adjust it, we will pass it through our x.13 subroutine, and if it is unadjusted but we deem through economic theory (e.g policy rates/etc) that it will have no seasonal component, we will not adjust it.

Does this sound reasonable? Best wises (and thanks again for all your help, not just with regards to my queries, but to the valuable responses you have provided to other forum users in other posts which I think a large number of people find immensely useful).

Charlie

[trubador]

If adjusted at source, we will not 're-adjust' with our adjustment routine (x.13). If a series is not adjusted at source, but we deem it to be necessary to adjust it, we will pass it through our x.13 subroutine, and if it is unadjusted but we deem through economic theory (e.g policy rates/etc) that it will have no seasonal component, we will not adjust it.Charlie

There is nothing wrong with this strategy. Although there is a good chance that already adjusted series was better handled in the "source" than that of your batch procedure, it may be a good idea to apply the same routine to all series if it is possible to retrieve the raw data. I would try both approaches and see if there is a dramatic change in the results. You can consider this as a robustness check.

I believe it is the correct approach to leave the series untouched if theory rejects the presence of a seasonal component. If you are unsure, then let the data decide for you. I'd like to emphasize once again the importance of handling such components jointly (within a model) where possible...