報告承辦單位: 長沙理工大學數學與統計學院
報告內容: Automated Estimation of Heavy-tailed Vector Error Correction Models
報告人姓名: 凌仕卿
報告人所在單位: 香港科技大學數學系
報告人職稱/職務及學術頭銜: 教授/IMS Fellow
報告時間: 2019年5月21日周二下午3:30
報告地點: 理科樓A-419
報告人簡介: 凌仕卿教授于1997年取得香港大學統計學博士學位,1997年至2000年西澳大學經濟學系博士后,2000年至2006年香港科技大學數學系助理教授,2003年至2006年受聘于西澳大學經濟學系和數學與統計系兼職副教授,2006年至2010年香港科技大學數學系副教授,2010年至今香港科技大學數學系教授。凌教授的主要研究方向為:大樣本理論、經驗過程、非平穩時間序列、非線性時間序列及計量經濟學。現為《Journal of Time Series Analysis》聯合編輯《Statistics & Probability Letters》、《Bernoulli》、《Electronic Journal of Statistics》、《Journal of the Japan Statistical Association》國際期刊的副主編。2003年和2013年分別榮獲澳大利亞和新西蘭MSS委員會頒發的Early Career Research Excellence Prize和Biennial Medal,2005年當選為國際統計學會會員;2007年榮獲計量經濟學期刊(Econometric Theory)頒發的Multa Scripsit Award的獎勵,2013年當選為澳大利亞和新西蘭MSS的Fellow。2015年當選為ITTI的Inaugural Distinguished Fellow。2019年當選為IMS Fellow。
報告摘要:It has been a challenging problem to determine the co-integrating rank in the vector error correction (VEC) model when its noise is a heavy-tailed random vector. This paper proposes an automated approach via adaptive shrinkage techniques to determine the co-integrating rank and estimate parameters simultaneously in the VEC model with unknown order $p$ when its noises are i.i.d. heavy-tailed random vectors with tail index $\alpha\in (0.2)$. It is showed that the estimated co-integrating rank and order $p$ equal to the true rank and the true order $p_{0}$, respectively, with probability trending to 1 as the sample size $n\to\infty$, while other estimated parameters achieve the oracle property, that is, they have the same rate of convergence and the same limiting distribution as those of estimated parameters when the co-integrating rank and the true order $p_{0}$ are known. This paper also proposes a data-driven procedure of selecting the tuning parameters. Simulation studies are carried to evaluate the performance of this procedure in finite samples. Our techniques are applied to explore the long-run andshort-run behavior of prices of wheat, corn and wheat in USA. Our results may provide a new insight to the Lasso approach for both stationary and non-stationary heavy-tailed time series.