关于「实证闭环:每月一次的尾部体检、尾部准备金估算器与预注册检验」,《极值尾部与退单准备金》给出的定量结论是什么?

结论:第八章给出每月一次的尾部体检流程、尾部准备金估算器与预注册检验:检验正态千分之一线只需 280 笔订单,检验峰过阈值线需要 9100 笔。书中以标注框区分已发表的证据、可运行的 Python 计算(19 段程序)、原典与作者的假说。

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: 280, 9100

testp0p1nreject if >=sizepower
A0.001000.0109528020.03250.8121
B0.001000.002009100150.04470.8050
C0.041670.125006260.04410.8030
in months of 1,800 orders: A 0.16, B 5.1
查看完整验算程序
# Chapter 8. p82: sample sizes for the pre-registered tests.
# Exact one-sided binomial tests at level 0.05 and power 0.8. For each
# test we find the smallest count n (orders or months) with a critical
# value c such that P(X >= c | p0) <= 0.05 and P(X >= c | p1) >= 0.8.
# Test A: the share of orders above the normal 99.9% line is 0.001
# (p0) against the 0.01095 of the illustrative tail (p11).
# Test B: the share above the GPD 99.9% line is 0.001 (p0) against a
# doubled 0.002 (an under-reserve the shop must catch).
# Test C: months whose chargeback bill beats a 1-in-24-month reserve,
# 1/24 (p0) against 3/24. All rates are illustrative planning values.

import math


def tail_ge(n, c, p):
    # P(X >= c) for X ~ Binomial(n, p), terms in log space, exact fsum
    if c <= 0:
        return 1.0
    lp, lq = math.log(p), math.log1p(-p)
    terms = []
    for k in range(c, n + 1):
        lt = (math.lgamma(n + 1) - math.lgamma(k + 1)
              - math.lgamma(n - k + 1) + k * lp + (n - k) * lq)
        terms.append(math.exp(lt))
        if k > c + 50 and terms[-1] < 1e-18 * terms[0]:
            break
    return math.fsum(terms)


def plan(p0, p1, step, start):
    n = start
    while True:
        c = 0
        while tail_ge(n, c, p0) > 0.05:
            c += 1
        pw = tail_ge(n, c, p1)
        if pw >= 0.8:
            return n, c, tail_ge(n, c, p0), pw
        n += step


print("test  p0        p1        n       reject if >=  size    power")
for name, p0, p1, step, start in (("A", 0.001, 0.01095, 10, 10),
                                  ("B", 0.001, 0.002, 100, 1000),
                                  ("C", 1 / 24.0, 3 / 24.0, 1, 2)):
    n, c, a, pw = plan(p0, p1, step, start)
    print("  %s   %.5f  %.5f  %6d   %6d       %.4f  %.4f"
          % (name, p0, p1, n, c, a, pw))
print("in months of 1,800 orders: A %.2f, B %.1f"
      % (plan(0.001, 0.01095, 10, 10)[0] / 1800.0,
         plan(0.001, 0.002, 100, 1000)[0] / 1800.0))

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 8 章「实证闭环:每月一次的尾部体检、尾部准备金估算器与预注册检验」(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