Germany's Seasonal Adjustment Failure

Most of the time seasonal adjustments are useful because one really wants to know what is happening from month to month rather than from year to year - and seasonally adjusted series let you see that. But from time to time they go wrong, because they adjust only slowly to structural shifts in the pattern of economic behaviour (spending, saving, investing, ordering etc). And sometimes, particularly under the pressure of economic or financial crises, those structural shifts can be abrupt.

For much of the world, the period since 2008 has been one of very rapid structural adjustment. As a result, some seasonally adjusted data is proving more fallible than usual. 

Take, for example, this morning's September retail sales data from Germany. According to the seasonally adjusted data, sales rose an impressive 1.5% mom - far stronger than anyone had expected.  But if you look at the raw data, sales fell 3.1% yoy, which was far weaker than anyone expected. What's more, sales fell 1% mom in September, which contrasts with the 1.9% mom averaged in the previous five Septembers. That's a disappointment which is 0.9 Standard deviations below the series. 

So which is right? Were Germany's retail sales stronger than expected in September, or weaker than expected?  Clearly it matters, and your judgement on it will depend on your trust in Germany's seasonal adjustment process. 

Now, at its most basic, the idea of seasonally adjusting is to reallocate monthly sales in such a way as to compensate for the sometimes violent seasonal fluctuations (the peak in December, the crash in January, for example). This has the consequence that the acid test for whether the process is working properly is whether over a 12 month period, the adjustment process becomes statistically invisible. In other words, the 12ma for a seasonally adjusted series must be near-identical to the 12ma for the non-adjusted series. If it isn't, there's something wrong.  Here's how the two series yoy 12ma for German retail sales, adjusted and non-adjusted look. 

The graph makes the point: there's something wrong - Germany's seasonal adjustment process doesn't meet even the most basic test of coherence. Although Germany's seasonal adjustment process was working fine all the way up to mid-2010, since then it has increasingly diverged from raw data, with the seasonal adjustment producing systematically stronger-than-justified readings during mid-2010 to mid-2011, and subsequently producing seriously weaker-than-justified readings from mid-2011 to August 2012.

By happy chance, the 1.5% mom reading for September - far stronger than seems justified in its own right - has brought the 12m average back to the correct level.  But given the adjustment process's failure since mid-2010, it would require a leap of faith to believe this is anything other than a happy coincidence.

The conclusion? On this series, at the moment you can not  trust the seasonally adjusted data.  And a second conclusion? Germany's September retail sales were not surprisingly strong, as today's headlines proclaim: they were (almost) shockingly weak.