Message Board

Dear readers, authors and reviewers,you can add a message on this page. We will reply to you as soon as possible!

2020 Volume 42 Issue 5
Article Contents

Ting CAI, Hong-chang HU. Complete Convergence for the Weighted Sums of NSD Sequences and Its Application[J]. Journal of Southwest University Natural Science Edition, 2020, 42(5): 126-131. doi: 10.13718/j.cnki.xdzk.2020.05.017
Citation: Ting CAI, Hong-chang HU. Complete Convergence for the Weighted Sums of NSD Sequences and Its Application[J]. Journal of Southwest University Natural Science Edition, 2020, 42(5): 126-131. doi: 10.13718/j.cnki.xdzk.2020.05.017

Complete Convergence for the Weighted Sums of NSD Sequences and Its Application

More Information
  • Corresponding author: Hong-chang HU
  • Received Date: 05/09/2017
    Available Online: 01/05/2020
  • MSC: O211.1

  • In this paper, we investigate some limit theorems for weighted sums of sequences of NSD random variables. By using the truncation technique and the properties of sequences of NSD random variables, we obtain the complete convergence for weighted sums of sequences of NSD random variables. Applying these results to the linear regression model containing the least square estimation of parameter β, and to the estimation of the nonparametric regression model of the weight function about g, we obtain their strong consistency.
  • 加载中
  • [1] HSU P L, ROBBINS H.Complete Convergence and the Law of Large Numbers[J].Proceedings of the National Academy of Sciences of the United States of America, 1947, 33(2):25-31. doi: 10.1073/pnas.33.2.25

    CrossRef Google Scholar

    [2] 章茜, 蔡光辉.WOD随机变量序列的完全收敛性和矩完全收敛性[J].数学物理学报(A辑), 2019, 39(5):1183-1191.

    Google Scholar

    [3] 黄海午, 邹航, 易艳春.行为渐近几乎负相关随机变量阵列加权和的完全矩收敛性[J].数学进展, 2019, 48(1):110-120.

    Google Scholar

    [4] SUNG S H.A Note on the Complete Convergence for Arrays of Dependent Random Variables[J].Journal of Inequalities and Applications, 2011, 2011:76. doi: 10.1186/1029-242X-2011-76

    CrossRef Google Scholar

    [5] HU T Z.Negatively Superadditive Dependence of Random Variables with Applications[J].Chinese Journal of Applied Probability and Statistics, 2000, 16(2):133-144.

    Google Scholar

    [6] ZHANG L R, ZHANG J J.Strong Consistency of Estimators under Missing Responses[J].Journal of Applied Mathematics and Physics, 2019, 7(1):93-103. doi: 10.4236/jamp.2019.71008

    CrossRef Google Scholar

    [7] FUK D K, NAGAEV S V.Probability Inequalities for Sums of Independent Random Variables[J].Theory of Probability & Its Applications, 1971, 16(4):643-660.

    Google Scholar

    [8] LIANG H Y, JING B Y.Asymptotic Properties for Estimates of Nonparametric Regression Models Based on Negatively Associated Sequences[J].Journal of Multivariate Analysis, 2005, 95(2):227-245.

    Google Scholar

    [9] HU H C.Bahadur Representations of M-estimators and Their Applications in General Linear Models[J].Journal of Inequalities and Applications, 2018, 2018:123. doi: 10.1186/s13660-018-1715-x

    CrossRef Google Scholar

    [10] DU J, ZHANG Z Z, XIE T F.A Weighted M-estimator for Linear Regression Models with Randomly Truncated Data[J].Statistics & Probability Letters, 2018, 138:90-94.

    Google Scholar

    [11] BAO X H, LIN J J, WANG X J, et al.On Complete Convergence for Weighted Sums of Arrays of Rowwise END Random Variables and Its Statistical Applications[J].Mathematica Slovaca, 2019, 69(1):223-232. doi: 10.1515/ms-2017-0216

    CrossRef Google Scholar

    [12] WANG X H, LI X Q, HU S H.On the Complete Convergence of Weighted Sums for an Array of Rowwise Negatively Superadditive Dependent Random Variables[J].ScienceAsia, 2016, 42(1):66. doi: 10.2306/scienceasia1513-1874.2016.42.066

    CrossRef Google Scholar

    [13] 邱德华, 陈平炎, 肖娟.END随机变量序列加权和的矩完全收敛性[J].应用数学学报, 2017, 40(3):436-448.

    Google Scholar

    [14] YANG W Z, XU H Y, CHEN L, et al.Complete Consistency of Estimators for Regression Models Based on Extended Negatively Dependent Errors[J].Statistical Papers, 2018, 59(2):449-465. doi: 10.1007/s00362-016-0771-x

    CrossRef Google Scholar

    [15] WANG X J, WU Y, HU S H.Strong and Weak Consistency of LS Estimators in the EV Regression Model with Negatively Superadditive-dependent Errors[J].Advances in Statistical Analysis, 2018, 102(1):41-65. doi: 10.1007/s10182-016-0286-8

    CrossRef Google Scholar

  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Article Metrics

Article views(2424) PDF downloads(316) Cited by(0)

Access History

Other Articles By Authors

Complete Convergence for the Weighted Sums of NSD Sequences and Its Application

    Corresponding author: Hong-chang HU

Abstract: In this paper, we investigate some limit theorems for weighted sums of sequences of NSD random variables. By using the truncation technique and the properties of sequences of NSD random variables, we obtain the complete convergence for weighted sums of sequences of NSD random variables. Applying these results to the linear regression model containing the least square estimation of parameter β, and to the estimation of the nonparametric regression model of the weight function about g, we obtain their strong consistency.

  • 设{Xnn≥1}为NSD随机变量序列,ani为双下标下三角常数列,即当in时,ani=0.考虑加权和:

    完全收敛性[1]是极限理论中的重要研究内容.在实际的应用中,更多的是随机变量序列加权和的情况,所以对相依序列加权和的完全收敛性问题的研究成为人们所关注的焦点.

    定义1 [1]设{Xnn≥1}是定义在概率空间(ΩAP)上的随机变量序列.若存在常数θ,对∀ε>0,有

    则称{Xnn≥1}完全收敛于θ.

    很多学者对完全收敛性展开了广泛的研究[2-8].本文研究NSD[5]随机变量序列加权和的完全收敛性,并将其结果应用于含参数β的最小二乘估计[9-10]的线性回归模型中及关于g的权函数非参数回归模型估计中,得到了强相合性定理[11-15].本文结论改进了文献[5]中的相应结果,下面给出NSD随机变量的概念.

    定义2 [5]函数Φ${{\mathbb{R}}^{n}}\to \mathbb{R} $称为超可加的,如果对任意的xy${{\mathbb{R}}^{n}} $

    其中:记号“∨”表示两者之间的最大值,“∧”表示两者之间的最小值.

    定义3[5] 随机向量X=(X1X2,…,Xn)为负超可加相依(NSD),如果满足

    其中ϕ(·)是超可加函数,Y1Y2,…,Yn相互独立,且对任意的iXiYi同分布.

1.   相关引理
  • 引理1 [5]X1X2,…,Xn是NSD随机变量序列,f1f2,…,fn均为非降的函数,则随机变量f1(X1),f2(X2),…,fn(Xn)仍是NSD随机变量序列.

    引理2 [7]设{Xnn≥1}是NSD随机变量序列,$ \mathrm{E}{{X}_{n}}=0, \mathrm{E}{{\left| {{X}_{n}} \right|}^{p}}<\infty , p\ge 2, {{S}_{n}}=\sum\limits_{i=1}^{n}{{{a}_{ni}}{{X}_{i}}}$,则存在一个仅与pρ(o)有关的正常数C=C(pρ(o)),有

2.   主要结果及证明
  • 定理1 设{Xii≥1}为NSD随机变量序列,$ \mathit{\boldsymbol{E}}{X_i} = 0, {{S}_{n}}=\sum\limits_{i=1}^{n}{{{a}_{ni}}{{X}_{i}}}$.对于1<P≤2,存在$ S\in \left( \frac{1}{P}, 1 \right]$,有

    对于P>2,存在$ S\in \left( \frac{1}{2}, 1 \right]$,使得

    则对∀ε>0,

    分以下两种情况讨论.

    (1) 对于1<P≤2与S必存在$ {{S}_{0}}\in \left( \frac{1}{P}, S \right]$,使得

    其中$ X_{i}^{\left( n \right)}\triangleq {{X}_{i}}{{I}_{\left( \left| {{x}_{i}} \right|\le {{n}^{{{s}_{0}}}} \right)}}$,则对∀ε>0,

    EXi=0,(1)及(2)式可得

    因此当n充分大时,由(4)及(5)式知

    即证以下(6),(7)式成立.

    由(1)式得

    由引理2知

    i→∞时,ρ(2i)→0,故存在n0>1,in0时,有

    则对∀n≥1,

    q=2,则

    将(11)式代入(9)式得II<∞,故(3)式完全成立.

    $\left( \text{ii} \right) $对于P>2,对于情况$\left( \text{ii} \right) $中的s,存在${{S}_{0}}\in \left( \frac{1}{2}, s \right] $,可得ps0>1.由$ \left( \text{i} \right)$的证明可得(3)式成立.结合$\left( \text{i} \right) $$\left( \text{ii} \right) $两种情况,可得定理1成立.证毕.

    由(3)式可推得$\underset{n\to \infty }{\mathop{\lim\limits }}\, {{S}_{n}}=0\ \text{a}\text{.}\ \text{s}\text{.} $,故由定理1可推出文献[5]中的定理1.

    推论1 设{Xii≥1}为NSD随机变量序列,$ \mathrm{E}{{X}_{i}}=0, {{S}_{n}}=\sum\limits_{i=1}^{n}{{{a}_{ni}}{{X}_{i}}}$,且$ \underset{i\ge 1}{\mathop{\text{sup}}}\, \mathrm{E}{{\left| {{X}_{i}} \right|}^{p}}<\infty $对于P>2.若$ S\in \left( \frac{1}{2}, 1 \right]$,使得$ \underset{1\le i\le n}{\mathop{\max }}\, \left| {{a}_{ni}} \right|=O\left( {{n}^{-s}} \right)$,则

    由(3)式,可得$ \underset{n\to \infty }{\mathop{\lim\limits }}\, {{S}_{n}}=0\ \text{a}\text{.}\ \text{s}\text{.}$,故从定理1可得推论1成立.

    注1 由推论1可得文献[5]中的推论,即推论1对比文献[5]放宽了对加权系数ani的限制.

    定理2 设{Xii≥1}为NSD随机变量序列,$ \mathrm{E}{{X}_{i}}=0, {{S}_{n}}=\sum\limits_{i=1}^{n}{{{a}_{ni}}{{X}_{i}}}$,若存在bi≥0,使

    若存在$ {{S}_{1}}\in \left( \frac{1}{r}+\frac{1}{2}, 1 \right]$s2>2对于r>2,有

    则对∀ε>0,

    i≥1,令$ X_{i}^{\left( n \right)}={{X}_{i}}I\left( \left| {{X}_{i}} \right|<{{b}_{i}} \right), S_{j}^{\left( n \right)}\sum\limits_{i=1}^{j}{\left( {{a}_{ni}}X_{i}^{\left( n \right)}-{{a}_{ni}}\mathrm{E}X_{i}^{\left( n \right)} \right)}$.则类似于定理1的证明可得定理2的证明.

    推论2 设{Xii≥1}为NSD随机变量序列,$\mathrm{E}{{X}_{i}}=0, {{S}_{n}}=\sum\limits_{i=1}^{n}{{{a}_{ni}}{{X}_{i}}} $,且$\underset{i\ge 1}{\mathop{\text{sup}}}\, \mathrm{E}{{X}_{i}}<\infty $r>2,若存在${{S}_{1}}\in \left( \frac{1}{r}+\frac{1}{2}, 1 \right] $,使得

    则对∀ε>0,有

    在定理2中取bi=is,易得定理2的条件满足,结合定理2的证明即得推论2成立.

3.   线性回归模型中的应用
  • 其中{Xij}为已知的设计点列,Λ≥=(β1,…,βn)T为未知的回归系数向量,ei为随机误差,记β的最小二乘估计$\overset{\Lambda }{\mathop{{{\mathrm{ }\!\!\Lambda\!\!\text{ }}_{n}}}}\, =\left( \overset{\Lambda }{\mathop{{{\beta }_{n1}}}}\, , \cdots , \overset{\Lambda }{\mathop{\beta _{np}^{\text{T}}}}\, \right) $.则由文献[8]知

    其中$\mathrm{A}_{n}^{\left( j \right)}=\sum\limits_{i=1}^{n}{{{\left( a_{ni}^{\left( i \right)} \right)}^{2}}=\frac{1}{V_{ij}^{\left( n \right)}}}, {{\left( V_{ij}^{\left( n \right)} \right)}_{p\times p}}={{\left( \mathrm{X}_{n}^{\text{T}}{{\mathrm{X}}_{n}} \right)}^{-1}}, {{\mathrm{X}}_{n}}={{\left( {{x}_{ij}} \right)}_{n\times p}}$.

    于是,对固定的j∈{1,2,…,p},令$ {{a}_{nk}}=\frac{a_{nk}^{\left( j \right)}}{A_{n}^{\left( j \right)}}$k=1,2,…,n.则由定理2及(13)式得以下定理3.

    定理3 设(12)式中随机误差{eii≥1}为NSD随机变量序列,满足Eei=0.若存在bi≥0,使

    对于r≥2,若存在$ {{s}_{1}}\left( {{s}_{1}}>\frac{r}{2}+1 \right)$s2(s2>2),使得

    则对于有j=1,2,…,p$\underset{n\to \infty }{\mathop{\lim\limits }}\, \overset{\Lambda }{\mathop{{{\beta }_{nj}}}}\, ={{\beta }_{j}}, \ \text{a}\text{.}\ \text{s}\text{.} $.

    2) 非参数回归模型中的应用

    p是一个正整数,A$ {{\mathbb{R}}^{p}}$中一个紧集,考虑以下回归模型

    其中X1(n),…,X(n)nA为已知的非随机设计点列,g为未知的实函数,ε1(n),…,εn(n)为均值为0的随机向量.取g(x)的权函数估计为

    下面给出gn(x)在NSD序列下的强相合性,现作如下基本假设

    $ \left( {\rm{i}} \right){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \mathop {\lim }\limits_{n \to \infty } \sum\limits_{i = 1}^n {{W_{ni}}} (x) = 1;$

    $\left( {{\rm{ii}}} \right){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \sum\limits_{i = 1}^n | {W_{ni}}(x)| \le C < \infty ; $

    $ \left( {{\rm{iii}}} \right){\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} \mathop {\lim }\limits_{n \to \infty } |\sum\limits_{i = 1}^n {{W_{ni}}} (x)(g({x_i}) - g(x))I(|x_i^{(n)} - x| > a)| = 0,\forall a > 0{\kern 1pt} .$

    定理4 设模型(14)基本条件$\left( \text{i} \right) $$ \left( \text{ii} \right)$$ \left( \text{iii} \right)$成立,{εi(n)i≥1}为NSD随机变量序列,且当r>2时,

    若存在某个正数$S\in \left( \frac{1}{2}, 1 \right] $,使

    则∀xc(g),其中c(g)为g的连续点集,有

    由于

    由文献[13]引理3知

    由于

    再由推论2知

    因此(18)式得证.

Reference (15)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return