关于「幂律还是对数正态:先检验尾巴,再谈鲸鱼」,《鲸鱼客群与尊享档》给出的定量结论是什么?
结论:程序 p12 在 0.50 到 0.98 分位数之间取 60 个候选起点,每本账簿造 200 本合成账簿。结果出人意料。账簿 LN 选出的起点是 134.4,尾部 427 位客户,指数 2.574,距离 0.0300,p 值 0.430,远高于门槛,幂律没有被拒绝。账簿 PT 选出的起点是 125.4,离真实起点 120 不远,尾部 329 位客户,指数 2.659,与真值 1 加 1.6 很接近,p 值却只有 0.100,恰好压在门槛上。
数据口径:核对原始文献之后做了蓝图修正:幂律必须先检验,一本纯对数正态的合成账簿通过了幂律拟合优度检验,p 值 0.430,只有似然比检验给出 z 为 -1.271、无法裁决;前 2 % 的利润份额随尾部形状从 0.121 变到 0.768,只能按区间报;「自动显影」用 BG/NBD 模型认出下一年前 2 % 客户的命中率是 0.625,不是 1;诱饵效应在自然呈现下可能减弱或反转;「隐式探针」改写为对所有访客公开、可自选的入口。 书中参数为示意,全部计算由书中程序生成。
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: 0.0300, 0.100, 0.430, 125.4, 134.4, 2.574
| book | x_min | n_tail | alpha | KS D | GOF p (200 reps) |
| LN | 134.4 | 427 | 2.574 | 0.0300 | 0.430 |
| PT | 125.4 | 329 | 2.659 | 0.0453 | 0.100 |
| rule: p < 0.1 rejects the power law (Clauset et al. 2009) |
查看完整验算程序
# Chapter 1. p12: Clauset-Shalizi-Newman fit with a bootstrap p-value.
# Model: for each candidate x_min (60 quantiles from 0.50 to 0.98) fit
# the continuous power law alpha = 1 + n / sum ln(x / x_min) and keep
# the x_min with the smallest Kolmogorov-Smirnov distance D. The
# goodness-of-fit p is the share of 200 semi-parametric bootstrap
# books whose own best D is at least the observed D (Clauset et al.
# 2009). Books LN and PT are the synthetic books of p11.
import math
import random
def make_books(n):
rng = random.Random(39901)
ln = [math.exp(rng.gauss(3.6, 1.25)) for _ in range(n)]
pt = []
for _ in range(n):
if rng.random() < 0.12:
pt.append(120.0 * (1.0 - rng.random()) ** (-1.0 / 1.6))
else:
pt.append(math.exp(rng.gauss(3.0, 0.6)))
return ln, pt
def fit_tail(s, start):
tail = s[start:]
m = len(tail)
xm = tail[0]
a = 1.0 + m / math.fsum(math.log(v / xm) for v in tail)
d = 0.0
for i, v in enumerate(tail):
model = 1.0 - (v / xm) ** (1.0 - a)
d = max(d, abs((i + 1) / m - model), abs(i / m - model))
return xm, a, d, m
def best_fit(xs):
s = sorted(xs)
n = len(s)
best = None
for j in range(60):
q = 0.50 + 0.48 * j / 59
r = fit_tail(s, int(q * n))
if best is None or r[2] < best[2]:
best = r
return best
def gof_p(xs, fit, reps, rng):
xm, a, d0, m = fit
body = [v for v in xs if v < xm]
n = len(xs)
hits = 0
for _ in range(reps):
syn = []
for _ in range(n):
if rng.random() < m / n or not body:
u = 1.0 - rng.random()
syn.append(xm * u ** (-1.0 / (a - 1.0)))
else:
syn.append(body[rng.randrange(len(body))])
if best_fit(syn)[2] >= d0:
hits += 1
return hits / reps
if __name__ == "__main__":
ln, pt = make_books(3000)
rng = random.Random(39902)
print("book x_min n_tail alpha KS D GOF p (200 reps)")
for name, xs in (("LN", ln), ("PT", pt)):
f = best_fit(xs)
p = gof_p(xs, f, 200, rng)
print("{:<5} {:>7.1f} {:>6,} {:>5.3f} {:>7.4f} {:>6.3f}".format(
name, f[0], f[3], f[1], f[2], p))
print("rule: p < 0.1 rejects the power law (Clauset et al. 2009)")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《帕累托鲸鱼客群自动显影、高净值信号自筛选与十倍客单尊享层跃迁母典》第 1 章「幂律还是对数正态:先检验尾巴,再谈鲸鱼」(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-399 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026399,
author = {Fan, Yafu},
title = {Pareto Whale Auto-Detection, High-Net-Worth Self-Selection Signals and Ten-X VIP Tier Escalation},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}