Comparison of Time Series Characteristics for Seasonal Adjustments from
SEATS and X-12-ARIM A
Catherine C. Hood
Catherine C. Hood, U.S. Census Bureau, Rm 3110-4, ESMPD , Washington, DC 20233
Key Words: seasonal adjustment, time series
Two widely-used seasonal adjustment programs are
the U.S. Census Bureau's X-12-ARIMA and the SEATS
program for ARIMA-model-based signal extraction
written by Agustin Maravall. In previous studies with
SEATS and X-12-ARIMA, we found some series where
the adjustment from SEATS had smaller revisions than
the adjustment from X-12-ARIMA (Hood, Ashley, and
Findley, 2000). Based on this previous work, I will
investigate the properties of a time series that make it a
good candidate for adjustment by SEATS or by X-12-
ARIMA. I used a version of X-12-ARIMA that has
access to the SEATS algorithm. This allows computation
of similar diagnostics for both programs — including
sliding spans and revision diagnostics — to compare
adjustments between the two programs.
In our earlier studies, we found that SEATS needs
more diagnostics before we can recommend using SEATS
for production work at the Bureau. In this paper, I show
examples of why the diagnostics in X-12-SEATS are very
useful. For example, SEATS can induce residual
seasonality into the seasonally adjusted series when the
original series isn't seasonal. The spectral diagnostics
availab le in X-12-SEATS are very important to be able to
see if the original series is seasonal or not. I also show an
example of a series with very large revisions due to the
model chosen by TRAMO . The revision history
diagnostics are very useful to see series with large
revisions.
1.
BACKGROUND
1.1
TRAM O/SEATS, X-12-ARIMA , and X-12-
SEATS
TRAM O/SEATS and X-12-ARIMA are based on
two different methods for seasonal adjustment. TRAMO
(Time series Regression with ARIM A noise, Missing
observations, and Outliers) and SEATS (Signal Extraction
in ARIM A Time Series) are linked programs developed
by Agustin Maravall and Victor Gomez to seasonally
adjust time s