We make two complementary contributions to e¢ ciently estimate dynamic factor mod- els: a frequency domain EM algorithm and a swift iterated indirect inference procedure for ARMA models with no asymptotic efficiency loss for any …nite number of iterations. Although our procedures can estimate such models with many series without good initial values, near the optimum we recommend switching to a gradient method that analytically computes spectral scores using the EM principle. We successfully employ our methods to construct an index that captures the common movements of US sectoral employment growth rates, which we compare to the indices obtained by semiparametric methods.

A spectral EM algorithm for dynamic factor models / Gabriele, Fiorentini; Alessandro, Galesi; Enrique, Sentana. - In: JOURNAL OF ECONOMETRICS. - ISSN 0304-4076. - STAMPA. - 205:(2018), pp. 249-279. [10.1016/j.jeconom.2018.03.013]

A spectral EM algorithm for dynamic factor models

Gabriele Fiorentini;
2018

Abstract

We make two complementary contributions to e¢ ciently estimate dynamic factor mod- els: a frequency domain EM algorithm and a swift iterated indirect inference procedure for ARMA models with no asymptotic efficiency loss for any …nite number of iterations. Although our procedures can estimate such models with many series without good initial values, near the optimum we recommend switching to a gradient method that analytically computes spectral scores using the EM principle. We successfully employ our methods to construct an index that captures the common movements of US sectoral employment growth rates, which we compare to the indices obtained by semiparametric methods.
2018
205
249
279
Gabriele, Fiorentini; Alessandro, Galesi; Enrique, Sentana
File in questo prodotto:
File Dimensione Formato  
A spectral EM algorithm for dynamic factor models.pdf

Accesso chiuso

Tipologia: Pdf editoriale (Version of record)
Licenza: Tutti i diritti riservati
Dimensione 927.88 kB
Formato Adobe PDF
927.88 kB Adobe PDF   Richiedi una copia

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1107913
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 7
  • ???jsp.display-item.citation.isi??? 5
social impact