Improved Portmanteau Diagnostic Check for ARFIMA Time Series Models

• Activity Number: 265
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Jinkun Xiao*+ and Ian McLeod
• Companies: University of Western Ontario and University of Western Ontario
• Keywords: Time Series ; Long Memory

Abstract

The multivariate ARMA diagnostic test based on the generalized variance of the residual autocorrelation matrix was introduced by Peña and Rodriguez (2002) and further developed by Mahdi, E. and McLeod, A.I. (2012). In this work we suggest an approximate method for computing the p-value for this test and we investigate the accuracy as well as the power of this test by simulation. Our simulations compare the approximate p-value method with the more exact Monte-Carlo test. We also compare this new test with previously suggested Box-Ljung portmanteau test for ARFIMA that was derived in Li and McLeod (1986). Since the simulations were extensive, serial farming on SharcNet was utilized.