高維度時間序列之模型選擇

Model Selection for High-Dimensional Multivariate Time Dependent Models

黃堃綾(Kunling Huang)

指導教授 ： 銀慶剛

國立臺灣大學/社會科學院/經濟學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201600544

摘要

無資料

關鍵字

高維度時間序列 ； 模型選擇 ； 稀疏性 ； 貪婪演算法 ； 模型選擇一致性

並列摘要

We use the multivariate orthogonal greedy algorithm (MOGA) to help select variables for high-dimensional time dependent models. Under weak moment assumptions, we develop the optimal convergence rate of the prediction error, and propose an information criterion whose penalty varies with the imposed order of moments. The resulting regression estimate is shown to reach the oracle property.

並列關鍵字

high-dimensional time series ； oracle property ； sparsity ； vector autoregressive ； sub-Gaussian ； greedy algorithm

參考文獻

[3]David F. Findley and Ching-Zong Wei. Moment bounds for deriving time series CLT’s andmodel selection procedures.Statistica sinica, pages 453–480, 1993.

1]Hirotugu Akaike. A new look at the statistical model identification.Automatic Control, IEEETransactions on, 19(6):716–723, 1974.

[2]Tzu-Chang F. Cheng, Ching-Kang Ing, and Shu-Hui Yu. Toward optimal model averaging inregression models with time series errors.Journal of Econometrics, 189(2):321–334, December2015.

[4]Ching-Kang Ing and Tze Leung Lai. A stepwise regression method and consistent model selec-tion for high-dimensional sparse linear models.Statistica Sinica, pages 1473–1513, 2011.

[6]Ching-Zong Wei. Adaptive prediction by least squares predictors in stochastic regression modelswith applications to time series.The Annals of Statistics, 15(4):1667–1682, 1987