A1246
Title: Quasi-Akaike information criterion of SEM for diffusion processes
Authors: Shogo Kusano - Osaka University (Japan) [presenting]
Masayuki Uchida - Osaka University (Japan)
Abstract: A model selection problem is considered for structural equation modeling (SEM) for diffusion processes. SEM is a statistical method that describes the relationships between latent variables. Statisticians often have some candidate models for SEM. In this case, selecting the optimal model among the competing models is necessary. Thus, many researchers have studied the information criteria of SEM for IID models. Recently, SEM for diffusion processes has been studied. The method enables the analysis of the relationships between latent processes based on high-frequency data. On the other hand, to the best knowledge, there are few studies on the information criteria for the SEM. Therefore, the quasi-Akaike information criterion (QAIC) of the SEM is proposed, and the asymptotic properties are investigated. The situation where a family of competing models includes some misspecified parametric models is also dealt with. It is proved that the probability of selecting the misspecified models by QAIC converges to zero. Furthermore, examples and simulation results are given to show the performance of the proposed information criterion.
