Questions and Answers on Upcoming Changes to Seasonal Adjustment for Model-Based Estimates
On this Page:
How does LAUS estimate employment and unemployment for states?
LAUS state estimates are produced using time series models to estimate the true values of household employment and unemployment from state Current Population Survey (CPS) series, which contain sampling error. The estimates consist of trend and seasonal components that sum to the not seasonally adjusted (NSA) data series.
During estimation, constraints are imposed requiring the sum-of-state NSA model estimates exactly equal the national CPS estimates of employment and unemployment. This process, called real-time benchmarking, makes the models more robust to sudden shocks to the overall economy.
What are the problems with the current approach to seasonal adjustment?
Benchmarking introduces distortions into the model estimates. First, since the constraints are imposed each month, irregular variations in the national CPS are transmitted to the state model estimates. Second, since the national CPS is highly seasonal and is used to benchmark the trend estimates, benchmarking has a tendency to add a small amount of seasonality from the national CPS to the model trend estimates. In addition, benchmarking artificially increases instability in the seasonal components.
How has LAUS accounted for these problems in the past?
To reduce the distortions to the benchmarked trend estimates, LAUS applied a combination of a trend filter and seasonal suppression filter. This simultaneously smoothed out irregular variation and removed any seasonality present.
Why is LAUS changing its approach to seasonal adjustment?
The trend plus seasonal suppression filter effectively eliminated the residual seasonality in the benchmarked trend. However, the filter was designed to remove strong seasonality, and it tended to over-adjust the series having a small amount of residual seasonality. Moreover, no attempt was made to reduce the benchmark distortions to the estimates of seasonality.
What is the new LAUS approach to seasonal adjustment?
LAUS will no longer attempt to simultaneously benchmark and seasonally adjust all of the state model estimates within each census division. Rather, seasonal adjustment will be performed externally after the joint benchmarking/estimation step.
The estimation process will continue to create benchmarked NSA data. However, the directly-produced benchmarked seasonally adjusted series will not be used. In their place, LAUS will directly adjust the benchmarked NSA data using an X-11 type of seasonal adjustment filter. All seasonality will be effectively removed from the adjusted series without over-adjustment. The estimates of seasonality will also be less distorted by the benchmarking process.
As before, a trend filter then will be applied to the seasonally adjusted data to smooth out irregular variation introduced by the monthly benchmarking.
Will there be any other changes to the seasonally adjusted series?
LAUS is switching the trend filter used to remove the non-seasonal volatility introduced by real-time benchmarking. A Reproducing Kernel Hilbert Space (RKHS) filter will be used. This filter places less weight on the current month’s estimate than did the Henderson-13 trend filter used previously. This decreases the month-to-month volatility in the seasonally adjusted estimates.
When will these changes to the seasonal adjustment procedures be implemented?
LAUS will implement the new seasonal adjustment procedures in early 2018, effective with Regional and State Unemployment 2017 Annual Averages news release scheduled for February 27. All seasonally adjusted data will be replaced with data created using the new procedures. There will be no discontinuities in the seasonally adjusted series due to this methodology change.
Last Modified Date: December 21, 2017