"

Autori
Maravall, Augustin
Gomez, Victor

Titolo
Initializing the Kalman filter with incompletely specified initial conditions
Periodico
European University Institute of Badia Fiesolana (Fi). Department of Economics - Working papers
Anno: 1993 - Fascicolo: 7 - Pagina iniziale: 1 - Pagina finale: 20

We review different approaches to Kalman filtering with incompletely specified initial conditions, appropiate for example when dealing with nonstationarity. We compare in detail the transformation approach and modified Kalman Filter (KF) of Ansley and Kohn, the diffuse likelihood and diffuse KF of de long, the approach of Bell and Hillmer, whereby the transformation approach applied to an initial stretch of the data yields initial conditions for the KF, and the approach of Comez and Maravall, which uses a conditional distribution on initial observations to obtain initial conditions for the KF. It is concluded that the later approach yields a substantially simpler solution to the problem of optimal estimation, forecasting and interpolation for a fairly general class of models.



Testo completo: http://hdl.handle.net/1814/448

Esportazione dati in Refworks (solo per utenti abilitati)

Record salvabile in Zotero