A comparative study of the finite-sample distribution of some portmanteau tests for univariate time series models

Andy C.C. Kwan, Yangru Wu

Research output: Contribution to journalArticlepeer-review

2 Scopus citations

Abstract

Monte Carlo experiments are conducted to investigate the finite-sample properties of some portmanteau tests for univariate time series models, with special attention being paid to estimated sizes (i.e. type I errors), means, variances, andempirical powers. Our main results indicate that (i) with the exception of the test proposed by Ljung (1986), size distortions can be substantial for portmanteau tests when parameter values are close to the boundary of the stationary or invertible region, (ii) the choice of the number of autocorrelations affects the empirical performance of most tests considered, and (iii) the empirical power of these tests is significantly affected by the nature of the data (seasonal vs. nonseasonal).

Original languageEnglish (US)
Pages (from-to)867-904
Number of pages38
JournalCommunications in Statistics Part B: Simulation and Computation
Volume25
Issue number4
DOIs
StatePublished - 1996
Externally publishedYes

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Modeling and Simulation

Keywords

  • ARMA models
  • Empirical powers
  • Empirical significance levels
  • Finite-sample distribution
  • Portmanteau tests
  • Seasonal data

Fingerprint

Dive into the research topics of 'A comparative study of the finite-sample distribution of some portmanteau tests for univariate time series models'. Together they form a unique fingerprint.

Cite this