Message Board

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

2022 Volume 47 Issue 7
Article Contents

CHEN Aigu, PENG Zuoxiang. On Penultimate Approximation for Distribution of a Class of Hill-Type Estimator[J]. Journal of Southwest China Normal University(Natural Science Edition), 2022, 47(7): 65-69. doi: 10.13718/j.cnki.xsxb.2022.07.010
Citation: CHEN Aigu, PENG Zuoxiang. On Penultimate Approximation for Distribution of a Class of Hill-Type Estimator[J]. Journal of Southwest China Normal University(Natural Science Edition), 2022, 47(7): 65-69. doi: 10.13718/j.cnki.xsxb.2022.07.010

On Penultimate Approximation for Distribution of a Class of Hill-Type Estimator

More Information
  • Corresponding author: PENG Zuoxiang
  • Received Date: 09/10/2021
    Available Online: 20/07/2022
  • MSC: O211.4

  • In this paper, an appropriate middle distribution sequence has been defined, which can better approximate the distribution of a class of moment statistics and the induced Hill-type estimator than the normal distribution under the second-order regular variation.
  • 加载中
  • [1] RESNICK S I. Extreme Values, Regular Variation and Point Processes[M]. New York: Springer-Verlag, 1987.

    Google Scholar

    [2] GOMES M I, MARTINS M J. Generalizations of the Hill Estimator-Asymptotic Versus Finite Sample Behaviour[J]. Journal of Statistical Planning and Inference, 2001, 93(1-2): 161-180. doi: 10.1016/S0378-3758(00)00201-9

    CrossRef Google Scholar

    [3] HILL B M. A Simple General Approach to Inference about the Tail of a Distribution[J]. The Annals of Statistics, 1975, 3(5): 1163-1174.

    Google Scholar

    [4] MASON D M. Laws of Large Numbers for Sums of Extreme Values[J]. The Annals of Probability, 1982, 10(3): 754-764.

    Google Scholar

    [5] DEHEUVELS P, HAEUSLER E, MASON D M. Almost Sure Convergence of the Hill Estimator[J]. Mathematical Proceedings of the Cambridge Philosophical Society, 1988, 104(2): 371-381. doi: 10.1017/S0305004100065531

    CrossRef Google Scholar

    [6] HAEUSLER E, TEUGELS J L. On Asymptotic Normality of Hill's Estimator for the Exponent of Regular Variation[J]. The Annals of Statistics, 1985, 13(2): 743-756.

    Google Scholar

    [7] DE HAAN L, RESNICK S. On Asymptotic Normality of the Hill Estimator[J]. Communications in Statistics Stochastic Models, 1998, 14(4): 849-866. doi: 10.1080/15326349808807504

    CrossRef Google Scholar

    [8] CHENG S H, PAN J Z. Asymptotic Expansions of Estimators for the Tail Index with Applications[J]. Scandinavian Journal of Statistics, 1998, 25(4): 717-728. doi: 10.1111/1467-9469.00131

    CrossRef Google Scholar

    [9] CHENG S, DE HAAN L. Penultimate Approximation for Hill's Estimator[J]. Scandinavian Journal of Statistics, 2001, 28(3): 569-575. doi: 10.1111/1467-9469.00255

    CrossRef Google Scholar

    [10] 胡爽, 彭作祥. 基于分块思想的Pickands型估计量[J]. 西南大学学报(自然科学版), 2019, 41(5): 53-58.

    Google Scholar

    [11] 王嫣然, 彭作祥. 基于分块的重尾指数估计量的推广[J]. 西南师范大学学报(自然科学版), 2019, 44(1): 25-28.

    Google Scholar

    [12] 张茜, 彭作祥. 不同观测点重尾参数的估计[J]. 西南师范大学学报(自然科学版), 2019, 44(1): 29-33.

    Google Scholar

    [13] XIONG L, PENG Z X. Heavy Tail Index Estimation Based on Block Order Statistics[J]. Journal of Statistical Computation and Simulation, 2020, 90(12): 2198-2208. doi: 10.1080/00949655.2020.1769622

    CrossRef Google Scholar

    [14] DE HAAN L, FERREIRA A. Extreme Value Theory[M]. New York: Springer-Verlag, 2006.

    Google Scholar

    [15] LEHMANN E L. Elements of Large-Sample Theory[M]. New York: Springer-Verlag, 1999.

    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(823) PDF downloads(140) Cited by(0)

Access History

Other Articles By Authors

On Penultimate Approximation for Distribution of a Class of Hill-Type Estimator

    Corresponding author: PENG Zuoxiang

Abstract: In this paper, an appropriate middle distribution sequence has been defined, which can better approximate the distribution of a class of moment statistics and the induced Hill-type estimator than the normal distribution under the second-order regular variation.

  • 设{Xnn≥1}为独立同分布随机变量序列,{Xin,1≤in}为其升序统计量,公共分布函数F满足1-F$R{V_{ - \frac{1}{\gamma }}}$γ>0被称为重尾指数[1]. 当F未知时,文献[2]提出如下Hill型估计量来估计γ

    其中矩统计量

    Γ(·)表示伽玛函数. 注意Mn(1)(k)=γn(1)(k)为文献[3]提出的Hill估计量. 文献[4-5]研究了其相合性. 文献[6-7]讨论了其渐近正态性. 文献[8]给出Hill估计量分布的渐近展开. 文献[9]在下列二阶正规变化条件下,研究了Hill估计量分布的次渐近逼近:存在辅助函数A(t)→0 (t→∞)在无穷远处符号恒定,使得对x>0

    其中ρ≤0,ARVρ$U = {\left( {\frac{1}{{1 - F}}} \right)^ \leftarrow }$. 近期有关尾指数估计量的研究可参见文献[10-13].

    设{Eii≥1}为独立同标准指数分布的随机变量序列,μ:=E(E1α),σ2:=Var(E1α),Gkα(x):= $P\left\{ {\sum\nolimits_{i = 1}^k {\frac{{E_i^\alpha - \mu }}{{\sigma \sqrt k }}} \le x} \right\}$. 本文将基于Gkα给出Mn(α)(k)和γn(α)(k)分布的次渐近逼近.

    本文主要结论如下:

    定理1  令(2)式成立,$\sqrt k A\left( {\frac{n}{k}} \right) \to 0$$\frac{{\log n}}{k} \to 0$以及${k^{\frac{3}{2}}}A\left( {\frac{n}{k}} \right) \to \infty $. 对x一致地有

    定理2  令(2)式成立,$\sqrt k A\left( {\frac{n}{k}} \right) \to 0$以及存在η∈(0,1)使得${k^{\eta + \frac{1}{2}}}A\left( {\frac{n}{k}} \right) \to \lambda \in [0, \infty ]$. 如果λ=∞,(3)式对x一致成立,否则对x一致地有

    定理3  令(2)式成立,$\sqrt k A\left( {\frac{n}{k}} \right) \to 0$以及存在η∈(0,$\frac{1}{2}$]使得${k^{\eta + \frac{1}{2}}}A\left( {\frac{n}{k}} \right) \to \lambda \in [0, \infty )$. 对x一致地有

    其中

    以及

    其中

    为了证明本文结论,先给出如下3个引理:

    引理1  令(2)式成立. 对x>0且x≠1,有

      由(2)式易得

    再结合(2)式,引理1得证.

    引理2  令(2)式成立. 对任意ε>0,存在一个函数A0~A和充分大的t0使得对tt0x≥1,一致地有

      由(2)式易知log U(x)-γlog xERVρ,结合文献[14]中定理B.2.18可得,对任意ε>0,存在一个函数A0~A以及充分大的t0使得对tt0x≥1,一致地有

    结合三角不等式可得

    由拉格朗日中值定理和Potter界可得

    结合引理1得证.

    引理3  对任意数列fk 0,对x一致地有

      易知E(E1α)3 < ∞,结合文献[15]中定理2.4.3可得

    其中Φϕ分别为标准正态分布函数和概率密度函数,μ3: =E(E1α-μ)3 < ∞. 代入(9)式左边,引理3得证.

    定理1的证明  令{Yi,i≥1}为独立同标准帕累托分布的随机变量序列,{Yin,1≤in}为其升序统计量. 由全概率公式和文献[9]中引理2可得

    其中tn↓0. 注意到$\left\{ {{X_i}} \right\}_{i = 1}^n\mathop = \limits^d \left\{ {U\left( {{Y_i}} \right)} \right\}_{i = 1}^n$. n充分大时可将引理2中的txt分别取为Yn-i+1,nYn-kn,1≤ik. 如果n充分大进一步使得(1-tn)ρ-ε≤1+ε,(1+tn)ρ-ε≥1-ε,(1-tn)ε≥1-ε和(1+tn)ε≤1+ε成立,由Potter界可得${(1 - \varepsilon )^2} \le \frac{{{A_0}\left( {{Y_{n - k, n}}} \right)}}{{{A_0}\left( {\frac{n}{k}} \right)}} \le {(1 + \varepsilon )^2}$. 那么

    结合$\left\{ {\frac{{{Y_{n - i + 1, n}}}}{{{Y_{n - k, n}}}}} \right\}_{i = 1}^k$Yn-kn的独立性,$\sum\nolimits_{i = 1}^k {\frac{{{Y_{n - i + 1, n}}}}{{{Y_{n - k, n}}}}} \mathop = \limits^d \sum\nolimits_{i = 1}^k {\exp } \left( {{E_i}} \right)$以及全概率公式,有

    其中

    以及

    其中Var(V1) < ∞. 由引理3可得

    由切比雪夫不等式可得

    那么

    同理可得下极限的下界. 分别令ε→0,定理1得证.

    定理2的证明  由全概率公式,文献[9]中引理4可得

    由(11)-(13)式,定理2得证.

    定理3的证明  令${x_k} = \sqrt k \left( {{{\left( {1 + \frac{x}{{\alpha \sqrt k }}} \right)}^a} - 1} \right)$,有

    由定理2和引理3知,(5)式成立. 结合(5), (10)式知(6)式成立.

Reference (15)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return