黃婉娟 羅雙華 張成毅



摘要:由于分位數回歸模型的損失函數不光滑,所得參數估計的效率不高,為提高參數估計的效率,首先提出復合分位數光滑經驗對數似然比,包括完全數據復合分位數光滑經驗對數似然比、加權復合分位數光滑經驗對數似然比和插值復合分位數光滑經驗對數似然比,并在一定條件下證明了它們都是服從漸近卡方分布的.其次,根據該似然比構造了回歸參數的置信區間,并證明了復合分位數光滑經驗似然估計量是漸近正態的.最后,通過數值模擬實驗說明了所得估計的有效性.
關鍵詞:缺失數據; 復合分位數回歸模型; 光滑經驗對數似然比; 漸近正態性
中圖分類號:O212 文獻標志碼:A 文章編號:1001-8395(2023)05-0628-10
1研究背景
與均值回歸只擬合一條條件均值曲線相比,分位數回歸擬合一簇曲線,能夠充分考慮到各個分位點處的信息.于是,Zou等[1]提出適當地綜合不同分位點處的信息以提高估計效率的想法,并證明了該方法能顯著地提高參數估計的效率.另外,復合分位數回歸估計方法不僅有效克服了單個分位數回歸估計效率下降的缺陷,還繼承了分位數回歸的穩健性,且被證實可以克服非正態誤差的干擾并顯著提高估計效率,是一種穩健且有效的參數估計方法.
2方法與主要結果
3數值模擬
4主要定理的證明
5結束語
本文主要研究了響應數據隨機缺失下一般線性復合分位數回歸模型的光滑經驗似然估計.由于分位數回歸的損失函數不光滑,所得估計效率不高,為提高估計效率,故考慮對響應數據隨機缺失的一般線性復合分位數回歸模型使用光滑經驗似然方法,并在一定條件下證明了缺失數據下一般線性復合分位數回歸模型的光滑經驗似然估計量的大樣本性質.通過模擬實驗說明了本文所提出估計的有效性.
參考文獻
[1] ZOU H, YUAN M. Composite quantile regression and the oracle model selection theory[J]. The Annals of Statistics,2008,36(3):1108-1126.
[2] OWEN A B. Empirical likelihood ratio confidence regions[J]. The Annals of Statistics,1990,18(1):90-120.
[3] OWEN A B. Empirical Likelihood[M]. New York:Chapman & Hall/CRC,2001.
[4] 李乃醫,李永明,韋盛學. 缺失數據下非線性分位數回歸模型的光滑經驗似然推斷[J]. 統計與決策,2015(1):97-99.
[5] REN J J. Smoothed weighted empirical likelihood ratio confidence intervals for quantiles[J]. Bernoulli,2008,14(3):725-748.
[6] 張雨婷,羅雙華. 縱向數據缺失且具有輔助信息的光滑經驗似然估計[J]. 寶雞文理學院學報(自然科學版),2021,41(4):9-14.
[7] 吉肖肖,張成毅,羅雙華. 缺失數據和輔助信息下分位數回歸的光滑經驗似然[J]. 紡織高校基礎科學學報,2020,33(2):70-77.
[8] 李忠桂,何書元. 分位數回歸的光滑經驗似然[J]. 應用概率統計,2017,33(5):497-507.
[9] LUO S H, MEI C L, ZHANG C Y. Smoothed empirical likelihood for quantile regression models with response data missing at random[J]. AStA Advances in Statistical Analysis,2017,101(1):95-116.
[10] ZHANG T, WANG L. Smoothed empirical likelihood inference and variable selection for quantile regression with nonignorable missing response[J]. Computational Statistics & Data Analysis,2020,144:106888.
[11] ZHANG T, WANG L. Smoothed partially linear quantile regression with nonignorable missing response[J]. Journal of the Korean Statistical Society,2022,51:441-479.
[12] ZHAO P X, ZHOU X S, LIN L. Empirical likelihood for composite quantile regression modeling[J]. Journal of Applied Mathematics and Computing,2015,48(1):321-333.
[13] SUN J, MA Y Y. Empirical likelihood weighted composite quantiles regression with partially missing covariates[J]. Journal of Nonparametric Statistics,2017,29(1):137-150.
[14] LI X Y, YUAN J Q. Empirical likelihood method for quantile with response data missing at random[J]. Acta Mathematicae Applicatae Sinica,2012,28(2):265-274.
[15] 郭東林. 缺失數據下幾類回歸模型的估計方法與理論[D]. 北京:北京工業大學,2017.
[16] 鞠婷婷. 缺失數據下基于經驗似然的變系數分位數回歸統計推斷[D]. 長春:長春工業大學,2018.
[17] LUO S H, ZHANG C Y, WANG M H. Composite quantile regression for varying coefficient models with response data missing at random[J]. Symmetry,2019,11(9):1065-1083.
[18] WHANG Y J. Smoothed empirical likelihood methods for quantile regression models[J]. Econometric Theory,2006,22(2):173-205.
[19] XUE L G. Empirical likelihood confidence intervals for response mean with data missing at random[J].Scandinavian Journal of Statistics,2009,36(4):671-685.
[20] WONG H, GUO S J, CHEN M, et al. On locally weighted estimation and hypothesis testing of varying-coefficient models with missing covariates[J]. Journal of Statistical Planning & Inference,2009,139(9):2933-2951.
[21] QIN J, LAWLESS J. Empirical likelihood and general estimating equations[J]. The Annals of Statistics,1994,22(1):300-325.
Smoothed Empirical Likelihood Method of General Linear Composite
Quantile Regression Model with Missing Response DataHUANG Wanjuan1,LUO Shuanghua1,ZHANG Chengyi2(1. School of Science, Xian Polytechnic University, Xian 710048, Shanxi;
2. School of Economics and Finance, Xian Jiaotong University, Xian 710049, Shanxi)
Abstract:Since the loss function of quantile regression model with missing response data is not smooth, the efficiency of parameter estimation is not high. In order to improve the efficiency of parameter estimation, the smoothed empirical log likelihood ratio of composite quantile regression model is firstly proposed in this paper, including the smoothed empirical log likelihood ratio of composite quantile regression model with complete data, weighted data and imputation, and the constructed smoothed empirical log likelihood ratio is proved to obey the asymptotic Chi-square distribution under certain conditions. Secondly, the confidence interval of regression parameter is constructed according to the likelihood ratio, and the empirical likelihood estimator is proved to be asymptotically normality. Finally, the performance of the estimators is assessed by numerical simulation.
Keywords:missing data; composite quantile regression model; smoothed empirical log likelihood ratio; asymptotically normality〖=〗
2020 MSC:62E20
(編輯 劉剛)