关于「回报水平:多少个月一遇的鲸鱼单有多大」,《极值尾部与退单准备金》给出的定量结论是什么?

结论:第四章把参数换成回报水平:十年一遇的鲸鱼单精确值 27214 元,正态估计只有 1459 元;十年一遇的事件十年内发生的概率是 0.632。

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: 1459, 27214

43200 orders, u 626.7, 2159 excesses, lam 89.96 a month
PWM xi 0.316 sigma 295.8; mean 238.1 sd 275.4
T(months) GPD90% interval exact normal
13569 3276 to 3866 4005 1136
128195 6895 to 9741 10833 1315
6013831 10880 to 17521 20624 1417
12017293 13234 to 22501 27214 1459
bootstrap 90% interval of xi0.269 to 0.362
查看完整验算程序
# Chapter 4. p41: return levels for the whale order, with intervals.
# Model: 24 months of 1,800 orders a month from the illustrative Burr
# XII law (b = 200, c = 2, k = 1.25), fixed seed. Threshold u = top 5%,
# PWM fit, lam = exceedances per month. T-month return level (the size
# exceeded once per T months on average):
#   x_T = u + sigma / xi * ((lam T) ** xi - 1).
# 90% interval by a parametric bootstrap (400 refits, fixed seed),
# set beside the exact Burr value and the normal (mean, sd) value.
# The parent's limiting tail shape is 0.4 (illustrative).

import math
import random
from statistics import NormalDist

B, C, K = 200.0, 2.0, 1.25
PER, MONTHS = 1800, 24
ND = NormalDist()


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


def gpd_pwm(y):
    ys = sorted(y)
    n = len(ys)
    a0 = math.fsum(ys) / n
    a1 = math.fsum(v * (n - 1 - j) / (n - 1) for j, v in enumerate(ys)) / n
    xi = 2.0 - a0 / (a0 - 2.0 * a1)
    return xi, 2.0 * a0 * a1 / (a0 - 2.0 * a1)


def level(u, xi, sig, lam, t):
    return u + sig / xi * ((lam * t) ** xi - 1.0)


def pct(xs, p):
    ys = sorted(xs)
    return ys[min(len(ys) - 1, int(p * len(ys)))]


rng = random.Random(20261119)
n = PER * MONTHS
xs = sorted(burr_q(rng.random()) for _ in range(n))
u = xs[int(0.95 * n)]
y = [x - u for x in xs if x > u]
lam = len(y) / MONTHS
xi, sig = gpd_pwm(y)
m = math.fsum(xs) / n
sd = math.sqrt(math.fsum((x - m) ** 2 for x in xs) / (n - 1))
print("%d orders, u %.1f, %d excesses, lam %.2f a month"
      % (n, u, len(y), lam))
print("PWM xi %.3f sigma %.1f; mean %.1f sd %.1f" % (xi, sig, m, sd))
boot = []
for _ in range(400):
    yb = [sig / xi * ((1.0 - rng.random()) ** (-xi) - 1.0) for _ in y]
    boot.append(gpd_pwm(yb))
print("  T(months)    GPD    90% interval       exact   normal")
for t in (1, 12, 60, 120):
    est = level(u, xi, sig, lam, t)
    bs = [level(u, a, b, lam, t) for a, b in boot]
    exact = burr_q(1.0 - 1.0 / (PER * t))
    normal = m + sd * ND.inv_cdf(1.0 - 1.0 / (PER * t))
    print("  %6d   %7.0f  %7.0f to %7.0f  %7.0f  %7.0f"
          % (t, est, pct(bs, 0.05), pct(bs, 0.95), exact, normal))
lo, hi = pct([a for a, _ in boot], 0.05), pct([a for a, _ in boot], 0.95)
print("bootstrap 90%% interval of xi: %.3f to %.3f" % (lo, hi))

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 4 章「回报水平:多少个月一遇的鲸鱼单有多大」(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