El problema de las condiciones iniciales en los algoritmos de estimación recursiva de modelos lineales

  1. Sotoca López, Sonia
Journal:
Documentos de trabajo de la Facultad de Ciencias Económicas y Empresariales

ISSN: 2255-5471

Year of publication: 1992

Issue: 20

Type: Working paper

More publications in: Documentos de trabajo de la Facultad de Ciencias Económicas y Empresariales

Abstract

This article proposes two solutions to allow the final recursive estimates of a regression model to be independent from arbitrary initial conditions. The first solution uses a recursive algorithm which discounts the effect of arbitrary initial conditions on the final parameter estimates. The second solution uses filters based on propagating the information matrix rather than the covariance matrix. By construction, this algorithm has exact initial conditions and is numerically robust.