Seasonality in Data
"To be interested in the changing seasons is a happier state of mind than to be hopelessly in love with spring." - George Santayana
I love Mark Perry's "Carpediem" blog -- often it's the first to get an interesting link up, and the rest of the econoblogosphere follows. He covers, I think, innovation better than anyone else out there, particularly in the healthcare, energy, and education markets.
Today, Tuesday, though, he posted a brief article about the American Staffing Association index, arguing that it and temporary hiring numbers "suggest that conditions in the labor market will continue to improve going forward," which I don't disagree with. But I think he would have made his long-bullish argument better if he detrended the obvious seasonal swings in the ASA index data.
In September 2011, another time he did this, I sent him a personal email, because the fact that the index tends to rise predictably throughout the year "obscure[d] an underlying bearish trend" which could be seen in year-over-year change. That has since changed, as economic growth accelerated between then and now, and since I have a blog, I'd thought I'd put up the data, showing what the ASA index looks like without the seasonal trend. (Admittedly, it is a rough-cut effort, as I didn't have a lot of data and there are large swings in the signal that interfere with trying to collect and cut out the statistical noise, but it's better than the raw data, which is confusing to interpret and has proved misleading in the past.)The ASA index without seasonality I find far more informative than the raw graph.
The detrended data quality suffers for several weeks in December, and I imagine a more accurate observation could be made from extrapolation off the November and January data. We see that the decline in the index began in January 2008 and lasted until July 2009 -- remember, the measures of the labor market tend to lag general economic indicators by one to two quarters -- and that there was that weakening I wrote to Prof. Perry about from January to September 2011, although there has been substantial strengthening since then. Ignoring the erratic data in the last week of December, we can say that the ASA index is actually at a post-recession high.
If you're curious, the reason why the holiday data is so poor is because the average struggles to capture the magnitude of the drop that occurs in late December because the seasonal noise is so strong and variable in that period, the simple (mean) average can't figure out the proper adjustment. It seems too conservative, which is why you should disregard the sudden jumps and subsequent dips in the late December data.