Logo image
Tests d'indépendance en analyse multivariée et tests de normalité dans les modèles ARMA
Thèses et HDR   Open Access

Tests d'indépendance en analyse multivariée et tests de normalité dans les modèles ARMA

Pierre Lafaye de Micheaux
Doctoral, Université de Montpellier
16/12/2002

Résumé

<br />Processus stochastiques Indépendance sérielle Analyse multivariée <br />Indépendance Fonction caractéristique Test lisse <br />Normalité des résidus Test d'ajustement Bruit blanc Gaussien Processus ARMA
We build a goodness-of-fit test of normality for the<br />innovations of an ARMA(p,q) model with known mean or trend. This <br />test is based on the data driven smooth test approach and is simple to<br />perform. An extensive simulation study is conducted to study the<br />behavior of the test for moderate sample sizes. Our<br />approach is generally more powerful than existing tests while holding<br />its level throughout most of the parameter space. This agrees with theoretical results showing the<br />superiority of the data driven smooth test approach in related<br />contexts.<br /><br />A semi-parametric test of independence (or serial<br />independence) is proposed between marginal vectors each of which is normally<br />distributed but without assuming the joint normality of these marginal<br />vectors. The test statistic is a Cramér-von Mises functional of a<br />process defined from the empirical characteristic function. This<br />process is defined similarly as the process of Ghoudi et al. (2001)<br />built from the empirical distribution function and used to test for<br />independence between univariate marginal variables. The test statistic<br />can be represented as a V statistic. It is consistent to detect any<br />form of dependence. The weak convergence of the process is<br />derived. The asymptotic distribution of the Cramér-von Mises<br />functionals is approximated by the Cornish-Fisher expansion using a<br />recursive formula for cumulants and by the numerical evaluations of<br />the eigenvalues in the inversion formula. The test statistic is<br />finally compared with Wilks' statistic for testing the parametric hypothesis of independence in the one-way MANOVA model with random effects.

Fichiers et liens (1)

url
Find in HALAfficher

Indicateurs

1 Consultations de la notice

Détails

Logo image