Autori
Maravall, Augustin
Gomez, Victor

Titolo
Signal extraction in Arima time series program seats
Periodico
European University Institute of Badia Fiesolana (Fi). Department of Economics - Working papers
Anno: 1992 - Fascicolo: 65 - Pagina iniziale: 3 - Pagina finale: 214

This document explains a program for estimation of unobserved components (or signals) in univariate time series. The program follows the ARIM A-model-based method first developed by Burman (1980) and by Hillmer and Tiao (1982); it originated, in fact, from the seasonal adjustment program developed by Burman at the Bank of England. The program fits, first, an ARIMA model to the series, and provides a detailed diagnosis. It identifies, next, the components present in the series; these are typically the trend, seasonal, and irregular components, although a separate cyclical component can also be estimated. The ARIMA models for the components are fully specified. Minimum mean square error (MMSE) estimates of the components are computed, as well as their forecasts. For each component, standard errors are provided for the different type of estimators (concurrent, preliminary, and historical or final estimator) and forecasts. The structures of the theoretical component and of its MMSE estimator are analysed, and compared to the estimate actually obtained. This comparison yields additional diagnostic elements. The last part of the program contains information of applied interest, concerning the properties of the different signals used in practice (mostly, the seasonally adjusted series and the trend). A detailed analysis is made of the different types of error (i.e., the magnitude of the revisions the estimators will undergo, the duration of the revision period, the error in the final estimator), and how they affect the behavior of the rates-of-growth measures used in short-term monitoring and policy making.



Testo completo: http://hdl.handle.net/1814/416

Esportazione dati in Refworks (solo per utenti abilitati)