![IconGems-Print](images/IconGems-Print.png)
242 – Time Series: Autoregressive Processes, Seasonality, and Unit Roots
To Revise or Not to Revise? Investigating the Behavior of X-13ARIMA-SEATS Seasonal Adjustment Revisions as New Series Values Are Added
Nicole Czaplicki
U.S. Census Bureau
Kathleen McDonald-Johnson
U.S. Census Bureau
How many historic seasonally adjusted values should we revise with each release of new time series estimates? We can revise the entire span, we can revise a minimum number of recent values, or we can choose an intermediate approach. To answer the question, we investigate how much the seasonal adjustment changes as we add new series values. We seasonally adjust using X-13ARIMA-SEATS with the X-11 seasonal adjustment method and regARIMA (regression plus ARIMA) model forecasts, using real U.S. Census Bureau series and simulated ARIMA model series. We assess the fluctuation of the seasonally adjusted estimates and how quickly they converge to a final value, particularly with respect to properties like ARIMA model coefficients and seasonal filter length. These results provide the foundation for determining the preferred number of revisions.