INTRODUCTION TRAMO (Time Series Regression with ARIMA Noise, Missing Observations and Outliers) and SEATS (Signal Extraction in ARIMA Time Series) are two closely coupled programs. The latest versions, which reside within the TSW (TRAMO-SEATS for Windows) program, have become virtually fused.
The TSW program, of which there is both a single-user version and a networked version, is freely available from the web site of the Bank of Spain at http://www.bde.es/servicio/software/tswe.htm. It is accompanied by a reference manual, which is illustrated with numerous screen shots and currently has 37 pages. The manual contains instructions for installation and use of the program, along with a glossary and index of the program's input parameters.
The TSW program has been implemented by Gianluca Caporello working under the supervision of Augustin Maravall. The original inspiration for the SEATS program was the work of Burman (1980), who built a program for the seasonal adjustment of economic time series while working for the Bank of England. The previous versions of the TRAMO and SEATS programs, which appeared in 1996, were the joint work of Victor Gomez and Augustin Maravall.
It is fortuitous that TSW has reached its present state of maturity at the same time as the Windows version of the rival STAMP program of Koopman et al. (2000) has been released. Although the two programs share the same objectives, they are underpinned by different philosophies, and they obtain their results in quite different ways. We shall highlight some of these differences in the course of this review.
The essential objective of TSW is to estimate a seasonal ARIMA model and to decompose it into additive components. The estimation is accomplished by TRAMO, and the decomposition is by SEATS. This review will concentrate primarily on SEATS, which is where most of the distinctive characteristics of TSW are to be found. The three regular components into which SEATS resolves