B0569
Title: Robustifying Gaussian quasi-likelihood inference
Authors: Hiroki Masuda - University of Tokyo (Japan) [presenting]
Shoichi Eguchi - Osaka Institute of Technology (Japan)
Abstract: Gaussian quasi-likelihood analysis is considered for non-ergodic stochastic process models observed at high frequency. The parametric estimation of the continuous part is addressed, leaving other characteristics as nuisance elements. The estimation strategy is based on some robust divergences which mitigate the well-known fragility of the Kullback-Leibler type Gaussian quasi-likelihood against the non-Gaussian variation in a short time. The contrast function is fully explicit and provides us with a simple interpretation. The theoretical properties of the proposed estimator are presented, followed by illustrative simulation experiments.