thesis

Approche fréquentielle de la causalité évolutive entre processus aléatoires bivariés non-stationnaires

Defense date:

Jan. 1, 1990

Edit

Institution:

Montpellier 1

Disciplines:

Economics

Authors:

Directors:

Abstract EN:

In the study of causality between stochastic processes, the assumption of stationarity (at least of second order) is often tacitly made. It's possible, indeed, essential to release such assumption (case of nonstationary processes) using evolutionary spectral analysis and timedependant parameters models. Evolutionary causality (i. E time-dependant causality) which is issued results partly from frequency-domain analysis (evolutionary causality test) and from the combination of a set of evolutionary prediction models (arma models and transfer function models with changing parameters).

Abstract FR:

Dans l etude de la causalite entre processus aleatoires, l hypothese de stationnarite (au-moins du second ordre) est souvent tacitement faite. Il est possible voire essentiel de s affranchir d une telle hypothese (cas des chroniques non-stationnaires) en prenant en compte l analyse spectrale evolutive ainsi que les modeles a parametres dependant du temps. La causalite evolutive (dependante du temps) qui en decoule resulte d une part de l analyse frequentielle (test de causalite evolutive) et d autre part de la combinaison d un ensemble de modeles evolutifs de prevision (modeles arma et modeles de fonction de transfert a coefficients variables).