-
设{Xn,n≥1}为独立同分布随机变量序列,公共分布函数F(x)为重尾分布,即
$\bar{F} \in R V_{-\frac{1}{\gamma}}$ ,其中重尾指数γ>0,$\bar{F}(x)=1-F(x)$ .分布函数F(x)未知时,文献[1]提出了极值指数的Hill估计量. 在此基础上文献[2-3]进一步证明了Hill估计量$\sqrt{k}$ 阶渐近正态性. 作为Hill估计量的一个推广,文献[4-5]提出了如下定义的半参数估计量其中{Xi,n,1≤i≤n}为其升序统计量. 有关极值指数估计量及其应用的更多研究见文献[6-14].
本文受文献[7]构造伪估计量的启发,将半参数估计量
$\hat{\gamma}_{n}(k)$ 的Xn-k,n替换为非随机的分位数$U\left(\frac{n}{k}\right)$ ,得到如下伪估计量其中
${\mathbb{1}}_{\{\cdot\}}$ 为示性函数. 本文将利用Lyapunov中心极限定理[2]和Cramér-Wold设计[15],建立估计量$\hat{\gamma}_{n}(k)$ 与伪估计量$\tilde{\gamma}_{n}(k)$ 的渐近关系. 本文假定$U(t) = \left( {\frac{1}{{1 - F}}} \right)\overleftarrow {} \left( t \right)$ 为二阶正规变化函数,即存在辅助函数A(t),使得对所有x>0成立. 显然,A(t)∈RVρ,ρ≤0.
对连续可微单增函数f(x),定义
其中
$\bar{F}_{n}(t)=\frac{1}{n} \sum_{i=1}^{n} {\mathbb{1}}_{\left\{X_{i}>t\right\}}$ . 特别地,当f(x)=log x时,有$\hat{A E}_{\log }\left(X_{n-k, n}\right)=\hat{\gamma}_{n}(k)$ 和$\hat{A E}_{\log }\left(U\left(\frac{n}{k}\right)\right)=\tilde{\gamma}_{n}(k)$ . 下面给出本文的主要结果,即$\hat{A E}_{f}\left(X_{n-k, n}\right)$ 和$\hat{AE}_{f}\left(U\left(\frac{n}{k}\right)\right)$ 之间的渐近关系.定理1 在条件(1)下,假设序列k满足
$k=k(n) \rightarrow \infty, \frac{k}{n} \rightarrow 0$ 和$\sqrt{k} A\left(\frac{n}{k}\right)=O(1)(n \rightarrow \infty)$ . 函数f满足f′∈RVa-1,且0≤4aγ < 1. 则特别地,令f(x)= log x,可以得到如下推论.
推论1 在条件(1)下,假设序列k满足
$k=k(n) \rightarrow \infty, \frac{k}{n} \rightarrow 0$ 和$\sqrt{k} A\left(\frac{n}{k}\right)=O(1)(n \rightarrow \infty)$ ,则有引理1 假设条件(1)成立. 函数f满足f′∈RVa-1,且0≤4aγ < 1. 当
$t \rightarrow \infty$ 时,有证 使用分部积分法,可得
再根据文献[2]的命题B.1.10,当t充分大时,对任意的δ>0有
加之
$\bar{F} \in R V_{-\frac{1}{\gamma}}$ 和f∈RVa,得同理,
引理1得证.
引理2 在条件(1)下,假设序列k满足
$k=k(n) \rightarrow \infty, \frac{k}{n} \rightarrow 0$ 和$\sqrt{k} A\left(\frac{n}{k}\right)=O(1)(n \rightarrow \infty)$ . 函数f满足f′∈RVa-1,且0≤4aγ < 1. 令则当
$n \rightarrow \infty$ 时,有其中
$h(a, \gamma):=-\frac{6(1-a \gamma)(1-2 a \gamma)}{(4 a \gamma-1)(3 a \gamma-1)}-1, \delta>0$ . 特别地,$\sum_{i=1}^{n} Z_{i, n}^{(f)} \stackrel{d}{\longrightarrow} N(0, h(a, \gamma))$ .证 根据
$\left(A E_{f}\left(U\left(\frac{n}{k}\right)\right)\right)^{2}=\frac{E\left(\left(f(X)-f\left(U\left(\frac{n}{k}\right)\right)\right)^{2} \mathbb{1}_{\left\{x>U\left(\frac{n}{k}\right)\right\}}\right)}{2 \bar{F}\left(U\left(\frac{n}{k}\right)\right)}$ 和$\bar{F}\left(U\left(\frac{n}{k}\right)\right)=\frac{k}{n}$ ,可得使用分部积分法,可得
加之
$\bar{F} \in R V_{-\frac{1}{\gamma}}, f \in R V_{a}$ 和(4)式,当$n \rightarrow \infty$ 时,可得再结合引理1,可得
此外,当n充分大时,对任意的δ>0有
最后,根据Lyapunov中心极限定理可知
$\sum_{i=1}^{n} Z_{i, n}^{(f)}$ 收敛. 引理2得证.定理1的证明 使用分部积分法,可得
将(6)式与(7)式分别平方后再相减,可得
首先考虑等式(8)右边的第一项与第三项. 令
$v=x U\left(\frac{n}{k}\right)$ 再结合$\frac{1}{\bar{F}_{n}\left(X_{n-k, n}\right)}=\frac{n}{k}$ ,可得由文献[2]定理2.4.1知,
$\sqrt{k}\left(\frac{X_{n-k, n}}{U\left(\frac{n}{k}\right)}-1\right) \stackrel{d}{\longrightarrow} N(0, 1)$ . 根据f′∈RVa-1,且$\bar{F}_{n}\left(X_{n-k, n}\right)=\frac{k}{n}$ ,当n充分大时,可得再结合引理1和文献[7]的引理1,当n充分大时,可得
接下来考虑等式(8)右边的第二项. 注意到
再根据引理2,当n充分大时,可得
从而有
最后考虑等式(8)右边的第四项. 根据
$\sqrt{k}\left(\frac{X_{n-k, n}}{U\left(\frac{n}{k}\right)}-1\right) \stackrel{d}{\longrightarrow} N(0, 1)$ ,可得综上所述,当n充分大时,可得
定理1得证.
推论1的证明 令f(x)=log x时有
$\hat{A E_{\log }}\left(X_{n-k, n}\right)=\hat{\gamma}_{n}(k)$ 和$\hat{A E_{\log }}\left(U\left(\frac{n}{k}\right)\right)=\tilde{\gamma}_{n}(k)$ . 再结合定理1,可得推论1得证.
On Asymptotic Relationship Between Heavy-Tailed Index Estimator and Its Pseudo-Estimator
-
Abstract: As an important research object of extreme value theory, extreme value index is often used to describe the thick tailed degree of distribution. Therefore, the estimation of extreme value index has attracted the attention of many statisticians, and is widely used in insurance, actuarial, finance and other fields. In this paper, the pseudo-estimator has been obtained by replacing the random threshold of the heavy tailed index estimator with the non-random threshold, and then the asymptotic relationship between the original estimator and the pseudo-estimator been established.
-
-
[1] 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. [2] DE HAAN L, FERREIRA A. Extreme Value Theory [M]. New York: Springer, 2006. [3] RESNICK S I. Heavy-Tail Phenomena[M]. New York: Springer, 2007: 39-69. [4] CAEIRO F, GOMES M I. A Class of Asymptotically Unbiased Semi-Parametric Estimators of the Tail Index [J]. Test, 2002, 11(2): 345-364. doi: 10.1007/BF02595711 [5] GOMES M I, MARTINS M J, NEVES M. Alternatives to a Semi-parametric Estimator of Parameters of Rare Events-the Jackknife Methodology[J]. Extremes, 2000, 3(3): 207-229. doi: 10.1023/A:1011470010228 [6] KINSVATER P, FRIED R, LILIENTHAL J. Regional Extreme Value Index Estimation and a Test of Tail Homogeneity [J]. Environmetrics, 2016, 27(2): 103-115. doi: 10.1002/env.2376 [7] STUPFLER G. On a Relationship Between Randomly and Non-Randomly Thresholded Empirical Average Excesses for Heavy Tails [J]. Extremes, 2019, 22(4): 749-769. doi: 10.1007/s10687-019-00351-5 [8] 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. doi: https://www.cnki.com.cn/Article/CJFDTOTAL-XNND201905010.htm [9] AHMED H, EINMAHL J H J. Improved Estimation of the Extreme Value Index Using Related Variables [J]. Extremes, 2019, 22(4): 553-569. doi: 10.1007/s10687-019-00358-y [10] PICKANDS J. Statistical Inference Using Extreme Order Statistics [J]. The Annals of Statistics, 1975, 3(1): 119-131. [11] DEKKERS A L M, EINMAHL J H J, DE HAAN L. A Moment Estimator for the Index of an Extreme-Value Distribution [J]. The Annals of Statistics, 1989, 17(4): 1833-1855. [12] SCHMIDT R, STADTMVLLER U. Non-Parametric Estimation of Tail Dependence [J]. Scandinavian Journal of Statistics, 2006, 33(2): 307-335. doi: 10.1111/j.1467-9469.2005.00483.x [13] DREES H, HUANG X. Best Attainable Rates of Convergence for Estimators of the Stable Tail Dependence Function [J]. Journal of Multivariate Analysis, 1998, 64(1): 25-46. doi: 10.1006/jmva.1997.1708 [14] HOGA Y. Detecting Tail Risk Differences in Multivariate Time Series [J]. Journal of Time Series Analysis, 2018, 39(5): 665-689. doi: 10.1111/jtsa.12292 [15] BILLINGSLEY P. Convergence of Probability Measures [M]. Hoboken: John Wiley & Sons, 1999. -
计量
- 文章访问数: 507
- HTML全文浏览数: 507
- PDF下载数: 71
- 施引文献: 0