Likelihood-Related Estimation Methods and Non-Gaussian GARCH Processes
Université Paris I Panthéon-Sorbonne - Centre d'Economie de la Sorbonne (CES)
Ecole Normale Superieure de Cachan
Université Paris I Panthéon-Sorbonne - CERMSEM; Lombard Odier Darier Hentsch & Cie
July 22, 2010
This article discusses the finite distance properties of three likelihood-based estimation strategies for GARCH processes with non-Gaussian conditional distributions: (1) the maximum likelihood approach; (2) the Quasi Maximum Likelihood approach; (3) a multi-steps recursive estimation approach (REC). We first run a Monte Carlo test which shows that the recursive method may be the most relevant approach for estimation purposes. We then turn to a sample of SP500 returns. We confirm that the REC estimates are statistically dominating the parameters estimated by the two other competing methods. Regardless of the selected model, REC estimates deliver the more stable results.
Number of Pages in PDF File: 33working papers series
Date posted: July 22, 2010
© 2013 Social Science Electronic Publishing, Inc. All Rights Reserved.
This page was processed by apollo5 in 0.640 seconds