Title: Bootstrap method for misspecified ergodic stochastic differential equation models
Authors: Yuma Uehara - The Institute of Statistical Mathematics (Japan) [presenting]
Abstract: Ergodic stochastic differential equation models driven by Levy processes under model misspecification are considered. A Gaussian quasi-likelihood based method serves as a good device to estimate drift and scale parameters in the models. However the correction of misspecification bias makes it difficult to construct a consistent estimator of the asymptotic variance of the Gaussian quasi-likelihood estimator, and thus confidence intervals and hypothesis testing. For such a problem, we propose a (blocking) weighted bootstrap method to directly approximate the asymptotic distribution of the estimator. We show that the approximation theoretically works well, and present some numerical experiments.