Sesgos en estimación, tamaño y potencia de una prueba sobre el parámetro de memoria larga en modelos ARFIMA

• Published in: Lecturas de economía Vol. 73; no. 73; pp. 131 - 148
• Main Authors: Karoll Gómez Portilla, Santiago Alejandro Gallón Gómez, Elkin Castaño Vélez
• Format: Journal Article
• Language: English; Spanish
• Published: Universidad de Antioquia 01-12-2010
• Subjects: Hypothesis testing; time-series models
• ISSN: 0120-2596

Castaño et al. (2008) proposed a test to investigate the existence of long memory based on the fractional differencing parameter of an ARFIMA (p, d, q) model. They showed that using an autoregressive approximation with order equal to the nearest integer of p* = T1/3 for the short-term component of this model, the test for the short memory null hypothesis against the long memory alternative hypothesis has greater power than other long memory tests, and also has an adequate size. We studied the estimation bias generated on d, and the effect on the power and size of the test when the short-term component is ignored and when the used models do not approximate itadequately. Additionally we analyze whether the obtained results by Castaño et al. (2008) can be improved employing a different autoregressive approximation.