facebook app skip to content
Primary navigation

Changes in Seasonal Adjustments in the CES Program

by Nick Dobbins
February 2018

Beginning with the January 2018 release, seasonally adjusted Current Employment Statistics (CES) estimates at the state and area level are being generated by a new process as the Bureau of Labor Statistics moves from the traditional method of projecting seasonal adjustment for the upcoming year to a concurrent monthly method.

The CES program produces two types of employment counts for state and metropolitan statistical areas. One is a set of unadjusted, direct estimates, meant to represent the best appraisal of our current job market. The other is a set of seasonally adjusted estimates, which reduce noise in unadjusted estimates by controlling for regularly recurring employment movements. For example, the traditional increases in retail employment over the holiday months are usually so large that they obscure any actual employment trends in the industry groups. By removing the part of a monthly change that is regularly attributable to that recurring seasonal variation, adjusted estimates provide a clearer look at the trends that are currently shaping the job market.

Historically seasonality has been addressed in CES by calculating seasonal factors for each month at the beginning of the calendar year. Those factors are then applied to the unadjusted estimates. For example, the seasonal factor for the retail trade sector in December of 2017 was 1.0287. The unadjusted estimate for retail in December was 312,516. Dividing that by the seasonal factor got us to our original adjusted December estimate of 303,800, which was unchanged for November estimates, controlling for the regular spike in retail jobs for the holiday season. This figure was eventually revised in the benchmark re-estimation process.

Starting with 2018 estimates BLS will be generating seasonal factors concurrently with the unadjusted estimates. Instead of generating all seasonal factors at the beginning of the year, thereby missing any new data that may indicate changes to seasonal patterns, seasonal factors will now include the most up-to-date estimates available. Factors will be updated for the current month’s preliminary estimates as well as the previous month’s final estimates. This new method, which has been used for national estimates since 2003, will improve the quality of seasonally adjusted data we provide by assuring that we’re using the most timely data available.

Concurrent seasonal adjustment is considered a superior estimation method to projected adjustment. However, time and resource constraints and a preference among some for pre-publishing seasonal factors have delayed its implementation for CES at the state and area level. To confirm the effectiveness of the new method, extensive testing was done for the CES data series, comparing concurrent and non-concurrent seasonal adjustment with the final sample data estimates as well as the more complete ‘universe’ Unemployment Insurance data to gauge accuracy. The testing showed that estimates are more accurate when seasonal factors are generated concurrently. However, the differences between concurrent and projected estimates were generally very small, with estimated gain or loss from concurrent adjustment measuring less than 0.1 percent in 92 percent of cases. In those cases where the gain or loss was greater than 0.1 percent, concurrent adjustment was more likely to produce a better result. In series where concurrent adjustment moved the estimation by 1 percent or more, the change was an improvement in 80% of cases. Smoothness of the series was also shown to improve. While not a direct measure of accuracy, less volatility in monthly estimates is a positive side effect on what would otherwise be unavoidable revisions to the CES data eliminated preemptively by the inclusion of more up-to-date information.1

A short release from the Bureau of Labor Statistics addressing the change can be found here: https://www.bls.gov/sae/saeconcurrent.htm

1Steve Mance, Bureau of Labor Statistics, https://www.bls.gov/osmr/pdf/st150110.pdf
back to top