Message Board

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

2023 Volume 48 Issue 5
Article Contents

YANG Xi, PENG Zuoxiang. The Asymptotic Properties of Generalized Quantiles[J]. Journal of Southwest China Normal University(Natural Science Edition), 2023, 48(5): 37-41. doi: 10.13718/j.cnki.xsxb.2023.05.005
Citation: YANG Xi, PENG Zuoxiang. The Asymptotic Properties of Generalized Quantiles[J]. Journal of Southwest China Normal University(Natural Science Edition), 2023, 48(5): 37-41. doi: 10.13718/j.cnki.xsxb.2023.05.005

The Asymptotic Properties of Generalized Quantiles

More Information
  • Corresponding author: PENG Zuoxiang
  • Received Date: 04/09/2022
    Available Online: 20/05/2023
  • MSC: O211.4

  • Bellini et al. (2014) proposed the generalized quantile and studied its basic properties. In particular, for heavy-tailed distributions, the paper also analyzed the asymptotic relationship between expectiles and quantiles. Based on the assumption that the generalized quantile is a regularly varying function, this paper studies its asymptotic relationship with quantiles.
  • 加载中
  • [1] BELLINI F, KLAR B, MVLLER A, et al. Generalized Quantiles as Risk Measures [J]. Insurance: Mathematics and Economics, 2014, 54: 41-48. doi: 10.1016/j.insmatheco.2013.10.015

    CrossRef Google Scholar

    [2] BRECKLING J, CHAMBERS R. M-Quantiles [J]. Biometrika, 1988, 75: 761-772. doi: 10.1093/biomet/75.4.761

    CrossRef Google Scholar

    [3] KOENKER R, BASSETT G. Regression Quantiles [J]. Econometrica, 1978, 46(1): 33. doi: 10.2307/1913643

    CrossRef Google Scholar

    [4] NEWEY W K, POWELL J L. Asymmetric Least Squares Estimation and Testing [J]. Econometrica, 1987, 55(4): 819. doi: 10.2307/1911031

    CrossRef Google Scholar

    [5] CHEN Z H. Conditional Lp-Quantiles and Their Application to the Testing of Symmetry in Non-Parametric Regression [J]. Statistics & Probability Letters, 1996, 29(2): 107-115.

    Google Scholar

    [6] ABDOUS B, REMILLARD B. Relating Quantiles and Expectiles under Weighted-Symmetry [J]. Annals of the Institute of Statistical Mathematics, 1995, 47(2): 371-384. doi: 10.1007/BF00773468

    CrossRef Google Scholar

    [7] EFRON B. Regression Percentiles Using Asymmetric Squared Error Loss [J]. Statistica Sinica, 1991, 1(1): 93-125.

    Google Scholar

    [8] YAO Q, TONG H. Asymmetric Least Squares Regression Estimation: a Nonparametric Approach [J]. Journal of Nonparametric Statistics, 1996, 6(2-3): 273-292. doi: 10.1080/10485259608832675

    CrossRef Google Scholar

    [9] SCHNABEL S K, EILERS P H C. A Location-Scale Model for Non-Crossing Expectile Curves [J]. Stat, 2013, 2(1): 171-183. doi: 10.1002/sta4.27

    CrossRef Google Scholar

    [10] DAOUIA A, GIRARD S, STUPFLER G. Estimation of Tail Risk Based on Extreme Expectiles [J]. Journal of the Royal Statistical Society Series B: Statistical Methodology, 2018, 80(2): 263-292. doi: 10.1111/rssb.12254

    CrossRef Google Scholar

    [11] DAOUIA A, GIRARD S, STUPFLER G. Extreme M-Quantiles as Risk Measures: From L1 to Lp Optimization [J]. Bernoulli, 2019, 25(1): 264-209.

    Google Scholar

    [12] POLLARD D. Asymptotics for Least Absolute Deviation Regression Estimators [J]. Econometric Theory, 1991, 7(2): 186-199. doi: 10.1017/S0266466600004394

    CrossRef Google Scholar

    [13] PORTNOY S, KOENKER R. Adaptive L-Estimation for Linear Models [J]. The Annals of Statistics, 1989, 17(1): 362-381.

    Google Scholar

    [14] RESNICK S I. Point Processes[M]// Extreme Values, Regular Variation and Point Processes. New York: Springer, 1987: 123-161.

    Google Scholar

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

    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(1336) PDF downloads(157) Cited by(0)

Access History

Other Articles By Authors

The Asymptotic Properties of Generalized Quantiles

    Corresponding author: PENG Zuoxiang

Abstract: Bellini et al. (2014) proposed the generalized quantile and studied its basic properties. In particular, for heavy-tailed distributions, the paper also analyzed the asymptotic relationship between expectiles and quantiles. Based on the assumption that the generalized quantile is a regularly varying function, this paper studies its asymptotic relationship with quantiles.

  • X的分布函数为F,其生存函数为$\bar{F}=1-F$. 假定X服从重尾分布,即$\bar{F}$满足如下的正规变化条件:

    记为$\bar{F} \in R V_{-\frac{1}{\gamma}}$.

    $H(x):[0, \infty) \longrightarrow[0, \infty)$为严格单调递增、可导的凸函数,满足

    文献[1]定义X的广义分位数

    $q_\tau^* \in \arg \min _q \mathrm{E}\left[\eta_\tau(X-q)-\eta_\tau(X)\right]$,其中$\eta_\tau(x)=|\tau-1(x \leqslant 0)| H(|x|)$. 它进一步研究了M分位数[2]的性质. 当$H(x)=x, x^2, x^p, p>1$时,(1)式分别为一般分位数$q_\tau \in \arg \min _q \tau \mathrm{E}\left[(X-q)^{+}\right]+$ $(1-\tau) \mathrm{E}\left[(X-q)^{-}\right]$ [3]、期望分位数[4]$L_p$分位数[5].

    文献[6-9]说明了$q_\tau$与期望分位数之间的联系. 文献[10-11]则得到了$q_\tau$$L_p$分位数的渐近关系. 有关$q_\tau$,期望分位数和$L_p$分位数应用的更多研究见文献[12-13]. 本文受文献[11]的启发,在假定$H \in R V_\alpha$$\alpha>1$的基础上,讨论极端情况下$q_\tau^*$$q_\tau$的渐近关系.

    下面给出本文的主要结果,即$q_\tau^*$$q_\tau$的联系. 首先建立$\bar{F}\left(q_\tau^*\right)$$1-\tau$的联系,结论如下:

    定理1    假定$\bar{F} \in R V_{-\frac{1}{\gamma}}, H \in R V_\alpha$. 对任意的$\alpha>1, c>0$,有$\mathrm{E}[H(c|X|)] <\infty$$0 <\gamma <\frac{1}{\alpha-1}$. 则

    其中$\mathrm{B}(a, b)=\int_0^1 x^{a-1}(1-x)^{b-1} \mathrm{~d} x$是贝塔函数.

    由于$\bar{F}\left(q_\tau\right)=1-\tau$,因此基于定理1我们易得到$q_\tau^*$$q_\tau$的联系(定理2).

    定理2    在定理1的条件下,有

    在证明定理1之前,先引入3个引理.

    引理1    若对任意的$c>0$,有$\mathrm{E}[H(c|X|)] <\infty$,则$q_\tau^* \in \arg \min _q \mathrm{E}\left[\eta_\tau(X-q)-\eta_\tau(X)\right]$当且仅当$(1-\tau) \mathrm{E}\left[H^{\prime}\left(\left(X-q_\tau^*\right)^{-}\right)\right]=\tau \mathrm{E}\left[H^{\prime}\left(\left(X-q_\tau^*\right)^{+}\right)\right]$.

       参见文献[1]引理3.

    引理2    若对任意的c>0,有$\mathrm{E}[H(c|X|)] <\infty$,则$q_\tau^*$是(0,1)上的单调递增函数且当$\tau \rightarrow 1$时,$q_\tau^* \rightarrow+\infty$.

       $q_\tau^*$的单调性见文献[1]的命题5(f).若当$\tau \rightarrow 1$时,$q_\tau^* \not \rightarrow+\infty$. 则由单调收敛定理可得$q_\tau^*$收敛到一个有限数值,记为$q^*$. 函数$\mathrm{E}\left[H^{\prime}\left(\left(X-q_\tau^*\right)^{-}\right)\right]$$\mathrm{E}\left[H^{\prime}\left(\left(X-q_\tau^*\right)^{+}\right)\right]$是连续的,利用控制收敛定理及引理1,当$\tau \rightarrow 1$时,$\mathrm{E}\left[H^{\prime}\left(\left(X-q_\tau^*\right)^{+}\right)\right]=0$,即$\mathrm{P}\left(X \leqslant q^*\right)=1$. 这与X服从重尾分布F,其上端点为+∞[15]矛盾.

    引理3    假定$\bar{F} \in R V_{-\frac{1}{\gamma}}, H \in R V_\alpha$. 对任意的$\alpha>1, b \geqslant 1$c>0,有$\mathrm{E}[H(c|X|)] <\infty, 0 < \gamma <\frac{1}{b(\alpha-1)}$. 则

       记

    其中

    由文献[14]的命题0.7(b)可知$H^{\prime} \in R V_{a-1}$. 根据文献[14]的命题0.5,控制收敛定理及正规变化函数的局部一致收敛性,有

    由文献[15]的定理B.2.18,对任意$\epsilon >0, 0 <\delta <\gamma^{-1}-b(\alpha-1)$,存在$t_0=t_0(\epsilon, \delta)$使得对$t \geqslant t_0, x>1$,有

    因此,对$t>2, q_\tau^* \geqslant t_0$,有

    由于$\gamma^{-1}-b(\alpha-1)>0, \gamma^{-1}-b(\alpha-1)-\delta>0$. 因而$(t-1)^{b(a-1)} t^{-\frac{1}{\gamma}-1}$$(t-1)^{b(\alpha-1)+\hat{\delta}} t^{-\frac{1}{\gamma}-1}$在(2,+∞)上关于t可积. 同样利用控制收敛定理,可得

    因此,由(3)-(5)式可得

    $\varDelta_2(\tau)$,同样由文献[15]的定理B.2.18可知,对任意$\epsilon>0, 0 <\delta <\gamma^{-1}-b(\alpha-1)$,存在$t_0^*=t_0^*(\epsilon, \delta )$使得对$q_\tau^* \geqslant t_0^*, s>1$,有

    因此,

    注意到

    $\epsilon \rightarrow 0$可得

    由(2),(6)与(7)式可知结论成立,引理证毕.

    定理1的证明    由引理1可知

    等价于

    其中$I_1\left(q_\tau^*\right)=\mathrm{E}\left[H^{\prime}\left(X-q_\tau^*\right) 1_{\left(X>q_*^*\right)}\right], I_2\left(q_\tau^*\right)=\mathrm{E}\left[H^{\prime}\left(\left|X-q_\tau^*\right|\right)\right]$. (8)式两边同除以$H^{\prime}\left(q_\tau^*\right)$,有

    由引理3,令b=1有

    $q_\tau^* \rightarrow+\infty$时,则$\bar{F}\left(q_\tau^*\right) \rightarrow 0$,由(9)式知$A_1(\tau)=\frac{I_1\left(q_\tau^*\right)}{H^{\prime}\left(q_\tau^*\right)} \rightarrow 0$.

    注意到,文献[14]的命题0.7(a)表明,由$\mathrm{E}[H(c|X|)]$的存在性可得$\mathrm{E}\left[H^{\prime}(c|X|)\right]$的存在性. 因此对第二项$A_2(\tau)$,利用$\mathrm{E}\left[H^{\prime}(c|X|)\right]$的存在性及文献[14]的命题0.8(i),当$q_\tau^* \rightarrow+\infty$时,有

    对第三项A3(τ),有

    由文献[14]的命题0.7(b)可得$H^{\prime} \in R V_{\alpha-1}$,类似引理3的证明有

    注意到,当$\tau \rightarrow 1$时,有$q_\tau^* \rightarrow+\infty$,则(10)式第二项中的$\left(1-\frac{X}{q_\tau^*}\right)^{\alpha-1} 1_{\left(|X| <q_\tau^*\right)}$几乎处处收敛到1且小于等于$2^{a-1}$,由控制收敛定理可得

    由(10),(11)与(12)式可得A3(τ)收敛到1. 因此,

    联合(8),(9)与(13)式可得定理得证.

Reference (15)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return