Seasonal adjustments or unadulterated data? Imputations and trend-cycle analysis or extrapolating trends from raw data?

After beating consensus estimates for two months in August and September, seasonally-adjusted retail sales came back down to earth in October. One of the primary drags on growth was a rather large decline in automobile sales – a topic we have covered extensively in our ABOOK and in other postings. We have also covered what we feel is some extraordinary or superfluous statistical adjustments that may be obscuring what is really important here. What we are trying to discern, more than anything, is the general trend in retail sales to get a better sense of consumer spending demand.

There is no denying the appeal of seasonal adjustments, turning monthly changes into an apples-to-apples comparison. Looking at the unadjusted data below, the seasonal fluctuations are absolutely clear and it stands to reason that statistical agencies should be able to smooth out those expected variations. However, in doing so we might be losing sight of the overall big picture in vain pursuit of perhaps artificial precision.

Forgetting the attempts at precision, what we want to find are any obvious changes in the overall trend direction. Changes in trends, inflection points, may not be as easy to see in the adjusted data series because of attempts to “explain” variations as the normal course of intra-year flow. So instead of focusing on the minute or minutiae, instead we might want to just look at the big picture of where the raw data seems to fall out of what are really hardened and solid upward trends of growth.

It’s really amazing that, for the most part, the growth periods exhibit robust trends that are remarkably durable and consistent. This is particularly true when focusing on the all-important Christmas shopping season and the immediate January withdrawal. In that respect, wherever that consistency is lacking it becomes a potentially important inflection. There was a definite positive inflection in the Christmas season of 1997 that began a distinctly separate upward movement at a noticeably higher growth rate. It ended in the Christmas season of 2000, as December 2000 retail sales clearly violated the trend and began the descent into the 2001 recession. Similarly, there was a violation of trend in the Christmas 2007 season, even though it appeared minimal at the time.

Separate from the Christmas seasonal variation, there is also a “back-to-school” seasonal fluctuation in the August-September period. This forms another important baseline that we can use to measure validity. In 2007, the August-September results appear to have been the first major trend violation that would continue through that Christmas season and into 2008.

The dropoff is not very large, but it is not insignificant. With the benefit of hindsight, we know that these inflections were the beginning of the Great Recession.

Looking at the unadjusted trends in the post-crisis period, I think we see the same process playing out.

In 2012 so far, there was first a significantly weaker summer period, followed by a noticeable violation of trend in the August-September period. To get a better sense of just how much 2012 might be falling below trend, we know that retail sales dropped 4.92% from August 2011 to September 2011. That was nearly identical to the 4.55% decline seen in the August-September 2010 period. The decline in the August-September 2012 period, however, is estimated at a much larger 7.44%.

What does that tell us? It might be too early to still confirm a 2007-style trend violation, but it is clear that the summer setup has left the retail sector behind “schedule”, so to speak. That would mean we would need not only a Christmas season that followed the 2010 & 2011 trend, we would need a robust shopping season that also made up for the summer and back-to-school shortfall.

We can also infer from this analysis that, at least so far, something is different about 2012 than 2010 & 2011. It may not be a startling difference, certainly nowhere near the disaster in 2008, but that isn’t exactly a high standard. Again, it may be too early to draw any definitive conclusions, but this potential trend change does fit with a lot of other diverse data series for 2012 (a few examples below). It also fits the overall picture painted by slowing growth rates as the “recovery”, such that it was, has become increasingly exhausted, shallow and painfully devoid of a genuine engine of economic growth.