Here is another installment in the non-political arena for which Rich Mitchell called. Again, I am writing about a subjects with which I have experience – statistics – and with which I have no experience – unemployment.
We are bombarded daily with “seasonally adjusted” data. For example, last week, the Bureau of Labor Statistics of the Department of Labor published, in table A-12, and table A-15, “seasonally adjusted” and “not seasonally adjusted” unemployment figures. And the MSM announced that the seasonally adjusted unemployment rate is 8.3%.
Since we are inundated with both types of data, I thought a short, easily digested and understood introduction to the seasaonal adjustment process would prove useful. Although the “deseasonalizing” technique can be used with any time series data, this primer will focus on employment and unemployment data since it is ubiquitous. Any data collected in chronological order is a time series.
Employment and unemployment are higher in some months of the year than others. Reoccurring fluctuations in the number of employed and unemployed persons are reflected in the seasonal weather patterns that tend to be repeated, hence the name seasonal adjustment. These reoccurring variations in hiring patterns make it difficult to ascertain whether month-to-month changes in unemployment are due to reoccurring seasonal patterns or to changing economic conditions.
There are four distinct components to a time series:
Trend (t), the long-term movement of a time series, usually a straight line calculated with regression, usually measured over several years.
Cyclical (c), the reoccurring movement (crests and troughs) of a time series associated with business cycles, usually from two to five years in duration.
Seasonal (s), the reoccurring movement (crests and troughs) of a time series associated with the seasons of the year, always one year in duration. This is the component that the deseasonalizing process removes.
Irregular (i), the component of a time series that “explains” why the measured data do not exactly follow the time series. It cannot be modeled or estimated, is quite small, and is usually ignored.
So deseasonalizing data removes reoccurring patterns, thus allowing analysis data of changes due to economic factors, such as trend and cyclical factors. By deseasonalizing the time series the seasonal component is eliminated. This method, known as decomposition, was developed in the 1920’s, and is the Census I method.
The difference between “not deseasonalized” and “deseasonalized” is the seasonal component of the time series. That component may or may not be important in decision-making. Always remember, caveat emptor.
The primary purpose of deseasonalizing data is to remove reoccurring pattern(s) that can be associated with the seasons of the year, to remove factors beyond the control of employers, to neither reward or penalize employers for conditions beyond their control, to facilitate “apples to apples” comparisons by removing employment crests and troughs that can be associated with reoccurring non-economic factors, to eliminate non-economic clutter and facilitate economic analysis.
The seasonal component may or may not be of interest since the trend and cyclical components may be of more import in the decision-making process. For example, cyclical unemployment may be the result of a policy decision. Or the unemployment trend may be the result of artificial or calculation factors. In both cases “not seasonally adjusted” data analysis for decision-making purposes would be best.
As examples, consider these employment situations. Weather, particularly in the north, affects construction. Cold, rainy weather curtails construction, so unemployment rises. Similarly, in December, just before Christmas, economic activity increases, resulting in temporary hiring and a drop in employment. In neither example does the season cause changes in employment. However, by recognizing and removing the reoccurring conditions, changes due to economic reasons can be assessed.
How, then, can not deseasonalized employment data from December, January, and February be compared to data from June, July, and August? It can’t, hence deseasonalizing. But (and there is always a but) consider other factors, such as policy decisions, that may have affected employment between February and June. Other factors may be important, so proceed carefully.
By “proceed carefully” I mean that you should complete further analysis, even if you choose to use deseasonalized data. Other economic factors may be at work as well.
This article is only an introduction to seasonal adjusting. It is meant to provide a reference point for readers who encounter the concept, to provide an understanding of what it is, what the process removes, and the difference between “seasonally adjusted” and “not seasonally adjusted” data. By having this knowledge, the reader can decide for himself or herself which data to use when making a decision.