SEASONAL ADJUSTMENT

BEA Advisory Committee

May 11, 2012

What do users want?

• Data series that represent the concept they are said to measure to be as accurate as possible.

• In the US, they usually want a seasonally adjusted series

• In Europe and Asia, there is a healthy skepticism about the quality of SA data

Miscellaneous Thoughts

• Seasonal Adjustment is a lot more difficult than it might seem.

• Seasonal Adjustment is really forecasting. We make judgments that affect the outcome.

• There seems to be a communication gap between those who understand the inner workings of these programs and those who work with the data day to day.

Recommendations

• Need more research on situations like our “Great Recession” where we don’t just have one outlier to deal with

• X12-ARIMA doesn’t seem to be sensitive enough to detect outliers except major ones like Katrina on LA data

• Why does Tramo-Seats consistently produce a smoother seasonally adjusted series?

Recommendations

• Do benefits of concurrent seasonal adjustment offset problems it creates?

• Increased work load at the statistical agencies when faced with tight release schedules

• Additional uncertainty for users who need to forecast this month’s figure

Recommendations

• Census Bureau should be commended for the development of X13-SEATS

• We need an official release so it gets incorporated into major econometrics pkgs

• Need source code in C so it is more accessible. Most FORTRAN programmers have retired.

Recommendations

• Need greater transparency on how seasonal adjustment is done

• Very few descriptions of intervention analysis tell you how it was really done. Was it treated as an Additive Outlier, Level Shift, RAMP?

• Leaving out months seems to mean different things to different people

Conclusion

• Not clear whether the recession echo problem is caused by the tool (X12) or because the capabilities were not fully utilized or used correctly

• Even if the problem is due to user error, we need a tool that reasonably intelligent economic analysts can use without extensive training