Fun with Seasonality

There are many data series that vary in a regular seasonal pattern.  That is why all of the best sources do seasonal adjustments.

Here at "A Dash" one of our missions is to find methods that help the average investor to beat the "experts."  Sometimes we are surprised at how easy this might be.

The cottage industry engaged in disparaging government data includes the seasonal adjustment.  Many sources  play upon the inherent suspicion of government.  Even when sources engage in statistical methods that are routine — absolutely standard practice for experts — some pundits call them into question.

Let us take a closer look.

Why Consider Seasonality?

To begin our consideration, here is the take from Wikipedia, a neutral source:

One famous example is the rate of unemployment which is also
presented by a time series. Particularly this rate depends on seasonal
influences. This is why it is important to free the unemployment rate
of its seasonal component. As soon as the seasonal influence is removed
from this time series the real trend of the unemployment rate is
visible. Seasonal adjustment is mostly used in the official statistics
implemented by statistical software.

When seasonal adjustment is not done with monthly data, year-on-year
changes are utilised in a naive attempt to avoid contamination with
seasonality. However, each year-on-year change is the sum of twelve
monthly changes. This moving window (with a width of 12) is often a
poor way to understand a series. As an example, Bhattacharya, Patnaik and Shah
find that by using point-on-point changes of seasonally adjusted data,
the analyst is able to better obtain early warnings in the inflation
time-series.

Briefly put, looking at year-over-year changes complicates your task.  You are not going to spot a significant change in real time.  You wind up making many adjustments in your head.  Did this month's YOY change look better than last month's?

A Compelling Example

Let us consider the recent report on Mass Layoffs from the Bureau of Labor Statistics (BLS).  These are layoffs involving 50 or more employees.  We have taken the entire data series, using the unadjusted raw data, and presented it in a form that will clarify the pattern.  Our team (Thanks, Feray) took the data and presented each year separately.

The seasonal pattern is quite apparent.  The big diversion is in September, 2005, the time of Katrina.  The move in July, 2009 is consistent with the regular seasonal pattern.

A Misleading Analysis

Taking a very different approach, one of the authors writing as "Tyler Durden" sees the July move as a spike!  In an article entitled "Mass Layoff Events Spike" the author suggests this chart as evidence.

By using data without seasonal adjustment, the chart presents a misleading view of the mass layoff pattern, showing a deceptive "spike."  In fact, the BLS seasonally adjusted data showed the opposite conclusion.  "The number of mass layoff events in July decreased by 606 from the prior month, and the number of associated initial claims decreased by 72,440."

Our Take

There are many popular sources that present data with a consistently bearish spin, completely ignoring seasonality.  Other sources — particularly those covering housing — frequently cite the seasonality in the data, but make no seasonal adjustment.

Market followers who are looking for important concurrent changes in data should look to seasonally adjusted results as the best source.  The alternative is to do a complex calculation about how the year-over-year data are moving.  The seasonal adjustment software does what you want to do.  It does it better and more accurately.

We acknowledge that there can be some unusual factors affecting the seasonal results.  Concerning employment, some have noted that layoffs in the auto industry occurred sooner this year, so that the July seasonality is somewhat overstated.

Readers should note that seasonal adjustments balance out with raw numbers over the course of the year.  Those emphasizing one side of this question always highlight unusual factors when the result serves their viewpoint, ignoring it when it does not.  The data must eventually balance.

Meanwhile, the seasonally-adjusted approach provides a better starting point for those seeking good information on economic changes.