A1736
Title: Statistical inference for multinomial-valued time series
Authors: Wangzhengtian Su - Waseda University (Japan)
Yan Liu - Waseda University (Japan)
Yan Liu - Waseda University (Japan) [presenting]
Abstract: Statistical inference for multinomial-valued time series, which commonly arise in economics, epidemiology, and social sciences, is studied. Classical time series models usually fail to account for discreteness, non-negativity, and dependence across multiple categories. A multinomial count time series model is considered and a quasi-maximum likelihood estimation framework tailored to this setting is developed. The proposed estimator is a multinomial quasi-maximum likelihood estimator constructed from the multinomial quasi-likelihood function. Consistency and asymptotic normality of the estimator are established under general regularity conditions, and its asymptotic distribution is derived, including non-standard limits when parameters lie on the boundary of the parameter space. A local asymptotic normality representation of the multinomial quasi-likelihood is obtained, which enables a precise characterization of these boundary asymptotic behaviors. Monte Carlo simulations demonstrate that the finite sample performance of the estimator is consistent with the theoretical asymptotic results. The simulation results show accurate parameter estimation and support the validity of the derived asymptotic distributions. A real data application illustrates the practical usefulness of the proposed method.
