Generalized divergence criteria for model selection between random walk and AR(1) model
Résumé
We investigate a general class of divergence measures among distributions for model selection. As alternative to the
classical test of model choice, we introduce kernel type estimators of \alpha-divergence for continuous distributions
based on model selection criteria in general non parametric case.
We introduce the Divergence Indicator DI method by proposing a test for choosing between a random walk and a
regression one, using a unified divergence measure. Under the assumptions of standard type about model densities, the
asymptotic properties estimator of the expected divergence between the true unknown model and the candidate model
are established. From the point of the resulting statistics divergence estimator, the performance of the discrepancy
criteria is discussed and illustrated in various settings in model selection test.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...