Autore: Maravall, Augustin
Titolo: Unobserved components in economic time series
Periodico: European University Institute of Badia Fiesolana (Fi). Department of Economics - Working papers
Anno: 1993 - Fascicolo: 34 - Pagina iniziale: 1 - Pagina finale: 56

The paper addresses the situation in which an economic variable, for which a series of observations is available, can be seen as the combination of several unobserved components (UC). UC models have been intensively used in applied economic research; they are often found, for example, in business cycle analysis. UC are also important in short-term policy and monitoring of economic variables, and an important example is seasonal adjustment. UC used in these two fields of applications (applied econometric research and statistical practical applications) often share the same basic structure. This paper deals with UC models displaying that type of structure. First, the limitations of ad-hoc fixed filters are briefly discussed; attention is focussed on the Hodrick-Prescott filter to detrend a series, and on the X I1 filter to seasonally adjust a series. The paper develops then a general set-up for a model-based approach common to the vast majority of UC model applications. The basic feature is that the components follow linear stochastic processes. The problems of model identification, estimation and forecast of the components, diagnosis, and inference are sequentially addressed. The properties of the estimators (preliminary and historical) and of their associated estimation and forecasting errors are derived. Two examples are discussed: the quarterly series of US G n p (to illustrate business cycle analysis), and the monthly series of the UK money supply (to illustrate seasonal adjustment). The paper contains some implications for applied econometric research. Two important ones are, first, that invertible models, such as AR or Var models, cannot in general be used to model seasonally adjusted or detrended data. The second one is that to look at the business cycle in detrended series that are seasonally adjusted is a misleading procedure, since detrending plus seasonal adjustment will always induce a non-trivial spectral peak for a cyclical frequency.




Testo completo: http://hdl.handle.net/1814/475

Esportazione dati in Refworks (solo per utenti abilitati)