Nonparametric Additive Model on Discrete Time Series with Periodicity [具有周期特征的离散时间序列的非参数可加模型]

作者：Li J.Y.L.; Wang S.X.; You J.H.^*

来源：Acta Mathematica Sinica (Chinese Series), 2022, 65(1): 177-204.

• B-spline
• Generalized additive model
• Locally stationary
• Mixing condition
• Periodicity

Periodicity is one of the most common factor in time series analysis. In the time series analysis of discrete-valued response variables, we use maximum likelihood estimation with penalty to establish a consistent estimator of the unknown period. Given the estimator of the period, we take B-spline to approximate the trend term and the additive function, and at the same time obtain the √n-consistent estimator of the periodic term and the initial estimators of the trend term and the additive function. Then based on the idea of back-fitting, we establish the improved estimators of the trend term and additive function, and the asymptotic normality and efficiency of them are also demonstrated. Simulation experiments and empirical analysis confirm that our proposed method performs well for the finite sample.

• 单位
  上海财经大学