OLMIS

by Tracy A Morrissette
Published May-24-2010

Introduction The Purpose of Seasonal Adjustment The Characteristics of a Seasonally Adjusted Series How Unemployment Rates are Seasonally Adjusted Conclusion

Seasonal events, such as weather changes and holidays, recur at approximately the same time each year. Seasonal events often influence economic activity, creating "seasonal fluctuations" in economic data measured by month and quarter. Seasonal fluctuations are increases or decreases of a similar magnitude that occur in a data series around the same time each year. The seasonal adjustment process attempts to estimate and remove seasonal fluctuations from time series data.

Two versions of Oregon's unemployment rate are published for each month. One version is "seasonally adjusted," the other "not seasonally adjusted." This article discusses these two versions of the unemployment rate, explaining the purpose of seasonal adjustment and providing a brief overview of the process.

Over the course of each calendar year, Oregon's unemployment rate fluctuates with the seasons. Unemployment is higher in winter months and lower in summer months. Graph 1 is Oregon's not seasonally adjusted unemployment rate. This version of the unemployment rate contains a component that makes a complete cycle each year. This 12 month cycle is the "seasonal component."

Unemployment increases each year in January and February due to the end of most seasonal retail jobs and various weather-limited jobs.

Unemployment declines from March to May as the weather gets warmer and drier and many outdoor and tourism-related jobs increase. A slight increase occurs in June when students enter the labor force to look for summer jobs. After June, the unemployment rate declines through October, before increasing slightly in November and December. This January-through-December cycle repeats itself to a similar degree each year.

The seasonal adjustment process adjusts Oregon's unemployment rate for seasonal increases and decreases, producing a series without the seasonal component (Graph 2). The seasonally adjusted line cuts through the center of the unadjusted line, making it easier to observe the trend in the unemployment rate. The trend is the long-run upward or downward tendency in a data series, hidden among seasonal and other types of fluctuations. Increases and decreases in the trend are associated with changes in macroeconomic conditions.

Seasonally adjusting the unemployment rate reduces the influence of seasonality on month-to-month changes, which can be large depending on time of year. However, seasonal adjustment filters commonly leave smaller fluctuations in the seasonally adjusted series. For example, the U.S. seasonally adjusted unemployment rate series is characterized by small but frequent up and down shifts around the center of the series (Graph 3). The small fluctuations do not cause problems for observation of the long-run trend, since they "cancel each other out" over longer spans of time.

These small but frequent fluctuations, or "volatility," however, can create problems in interpreting month-to-month changes in seasonally adjusted unemployment rates. Volatility obscures detection of changes in trend, which are often regarded as the most interesting part of the analysis associated with the business cycle. Volatility in a data series can be reduced using "smoothing procedures," or "filters" such as moving averages. Smoothing procedures reduce the irregular influences in a data series thereby making the trend more noticeable. One such procedure is "Smooth Seasonal Adjustment," which is discussed in the next section.

Classical seasonal adjustment methods separate a time series into trend-cycle, seasonal and irregular components. The trend-cycle component is the underlying tendency of the series. The seasonal component accounts for seasonal fluctuations in the series. The irregular component is what remains after the trend-cycle and seasonal components are identified. In general, the three components are identified using moving average or model-based seasonal adjustment methods. The seasonal component is subtracted from the original series if the mode of adjustment is additive, and the series is divided by seasonal factors if the mode of adjustment is multiplicative.

Oregon's labor force data, which includes the unemployment rate, are seasonally adjusted by a model-based method designed by the Bureau of Labor Statistics (BLS), and then smoothed by the Henderson Trend Filter (H13). A signal-plus-noise model describes Oregon's Current Population Survey (CPS) employment and unemployment data over time as the sum of "true" values and survey error. The model of the signal is a combination of trend, seasonal, and irregular components. The noise model accounts for error patterns in CPS data. The H13 procedure uses moving averages to smooth the model-based seasonally adjusted series; a symmetric moving average filter with 13 terms is used to smooth historical data, and an asymmetric moving average filter with 7 terms is used to smooth data in real time. Seasonal adjustment and H13 smoothing results in a series called "Smoothed Seasonal Adjustment," which is a less volatile version of the model-based seasonally adjusted series.

Seasonally adjusted unemployment rates for Oregon's local areas are produced using X-12-ARIMA, a program developed by the U.S. Census Bureau. Seasonally adjusted unemployment rates for local areas are not official BLS data series. Although ARIMA (Auto-Regressive Integrated Moving Average) models are part of the procedure, X-12-ARIMA retains the moving average approach of its predecessors to produce seasonal factors. The X-12-ARIMA program produces a seasonally adjusted series and a number of diagnostics used to determine whether the program successfully identified and removed the seasonal component. Seasonal adjustment is adequate if three conditions are met: the original series is seasonal, the seasonal component is generally "stable," and the seasonally adjusted series contains no residual seasonality.

Seasonal adjustment is a process used to identify and remove the influence of a seasonal component from a time series. Seasonally adjusted data make the trend more noticeable. Oregon's unemployment rate is published in a "seasonally adjusted" and "not seasonally adjusted" form. Seasonally adjusted data series are subject to volatility, but smoothing procedures can reduce this volatility to provide a better indication of the underlying direction of the series.