关于「峰过阈值:皮坎兹—巴尔凯马—德哈恩定理与超额分布的收敛」,《极值尾部与退单准备金》给出的定量结论是什么?

结论:第二章讲皮坎兹与巴尔凯马—德哈恩的峰过阈值定理,说明对数正态在有限阈值上看起来像形状 0.2148 的幂律尾。

数据口径:书中的订单、退单、请求量与参数均为示意;全部计算由书中程序生成;准备金与服务器冗余的讨论属于非个人化经营与财务方法,不构成理财、会计、税务或投资建议,实际金额与生产变更须由经营者人工决定。 第一章显示,在形状约 0.4 的示意订单尾部上,真实千分之一分位是正态估计的 2.57 倍,万分之一分位是 5.56 倍;样本上的正态估计中位数只有真实值的 0.363,最大的 20 笔订单占去离差平方和的 0.353。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.2148

local shape r'(u) at threshold quantile p (limit in last column)
parent/ p0.90.990.9990.9999 0.999999limit
Burr XII0.32470.38970.39840.39970.40000.40
lognormal s=10.30000.24810.21480.19210.16270.00
exponential0.00000.00000.00000.00000.00000.00
normal-0.2698-0.1271-0.0822-0.0605-0.03940.00
Burr XII at p=0.95: u = 632.0 yuan, GPD scale r(u) = 278.1
查看完整验算程序
# Chapter 2. p21: the local tail shape seen from finite thresholds.
# Model: for a threshold u the excess law is close to a GPD whose
# scale is the reciprocal hazard r(u) = S(u) / f(u) and whose shape is
# the slope r'(u) (Pickands 1975; Balkema and de Haan 1974). We compute
# r'(u) by a central difference at the 0.9 to 0.999999 quantiles of
# four parents: the illustrative Burr XII order law (limit 0.4), a
# lognormal with log-sd 1 (limit 0), an exponential (limit 0) and a
# normal (limit 0). Parameters are illustrative.

import math
from statistics import NormalDist

ND = NormalDist()
B, C, K = 200.0, 2.0, 1.25


def burr(u):
    s = (1.0 + (u / B) ** C) ** (-K)
    f = K * C * u ** (C - 1.0) / B ** C * (1.0 + (u / B) ** C) ** (-K - 1)
    return s, f


def burr_q(p):
    return B * ((1.0 - p) ** (-1.0 / K) - 1.0) ** (1.0 / C)


def lnorm(u):
    z = math.log(u) / 1.0
    return 0.5 * math.erfc(z / math.sqrt(2.0)), ND.pdf(z) / u


def lnorm_q(p):
    return math.exp(ND.inv_cdf(p))


def expo(u):
    return math.exp(-u), math.exp(-u)


def expo_q(p):
    return -math.log(1.0 - p)


def norm(u):
    return 0.5 * math.erfc(u / math.sqrt(2.0)), ND.pdf(u)


def norm_q(p):
    return ND.inv_cdf(p)


def local_xi(law, u):
    h = 1e-4 * max(1.0, abs(u))
    s1, f1 = law(u + h)
    s0, f0 = law(u - h)
    return (s1 / f1 - s0 / f0) / (2.0 * h)


levels = (0.9, 0.99, 0.999, 0.9999, 0.999999)
print("local shape r'(u) at threshold quantile p (limit in last column)")
print("  parent  / p  " + "".join("%9s" % ("%g" % p) for p in levels)
      + "   limit")
for name, law, q, lim in (("Burr XII", burr, burr_q, 0.4),
                          ("lognormal s=1", lnorm, lnorm_q, 0.0),
                          ("exponential", expo, expo_q, 0.0),
                          ("normal", norm, norm_q, 0.0)):
    row = "".join("%9.4f" % local_xi(law, q(p)) for p in levels)
    print("  %-14s%s  %6.2f" % (name, row, lim))
u95 = burr_q(0.95)
s, f = burr(u95)
print("Burr XII at p=0.95: u = %.1f yuan, GPD scale r(u) = %.1f"
      % (u95, s / f))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《极值理论广义帕累托尾部、超级鲸鱼客单上限与极端退单准备金母典》第 2 章「峰过阈值:皮坎兹—巴尔凯马—德哈恩定理与超额分布的收敛」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-244 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026244,
  author    = {Fan, Yafu},
  title     = {Extreme-Value Theory and Generalised Pareto Tails: Whale-Order Ceilings and Reserves for Extreme Chargebacks},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论