B0735
Title: Predictive model selection for jump diffusion models
Authors: Yuma Uehara - Kansai University (Japan) [presenting]
Abstract: A model selection problem is considered for jump-diffusion models based on high-frequency samples. The terminal time is supposed to diverge (ergodic setting), and the interest is to select drift and diffusion coefficients and jump distribution among candidates. Unlike the diffusion case, the stochastic flow approach cannot be directly used in order to evaluate the transition density when the jump term is parametrized in some way. To validate such an approximation, new transition density estimates are presented. From the estimates, an explicit predictive information criterion is proposed, constructed by the quasi-likelihood function. The choice of the threshold is also discussed which distinguishes the existence of jumps within an interval. After that, the simulation result is shown by using YUIMA package.