关于「运程报告的后验校准:低谷购买、巴纳姆陈述与布赖尔分数」,《均值回归与封面诅咒》给出的定量结论是什么?

结论:随便挑一个月,下个月更好的比例是 0.500,这正是对称性给出的值。可是在 \(\rho\) 为 0.3、只在最差的 0.05 份额月份购买时,下个月更好的比例是 0.924,平均回升 1.426 个标准差;在 \(\rho\) 为 0.6、最差 0.15 份额时是 0.770。即使购买门槛放宽到最差的 0.30 份额,比例也还有 0.716 到 0.786。一份只写「会好起来」的报告,在这些购买者身上的命中率就是这么高,它不需要任何知识。

数据口径:书中门店、达人、顾客与报告的参数均为示意;全部计算由书中程序生成;本书是非个人化的经营分析,不构成经营、法律或理财建议,涉及运程报告与原典之处只分析结构与概率,不作任何个人占断。 书中以标注框区分已发表的证据、可运行的 Python 计算(22 段程序)、原典与作者的假说。

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: 0.05, 0.15, 0.3, 0.30, 0.500, 0.6

rhobottom qbuyers: next bettermean gainany month
0.30.050.924+1.4260.500
0.30.150.861+1.0810.500
0.30.300.786+0.8130.499
0.60.050.850+0.8440.500
0.60.150.770+0.6170.500
0.60.300.716+0.4740.500
a report that only says 'it will get better' is right that
often with no knowledge at all
查看完整验算程序
# Chapter 6. p61: People consult a report in their worst month.
# 10,000 shop owners, 24 months each. A month's index is a permanent level
# plus fresh noise, scaled so neighbouring months correlate at rho. An
# owner buys a report in any month that falls below a cutoff (the bottom
# q of all months). The report has no effect. We count how often the next
# month is better, for buyers and for a random month. Fixed seed;
# parameters illustrative.
# 低谷求签:人们在最差的月份购买报告,报告本身毫无作用。
import math
import random

rng = random.Random(6)
OWNERS, MONTHS = 10000, 24


def cdf(z):
    return 0.5 * (1.0 + math.erf(z / math.sqrt(2.0)))


def quantile(p):
    lo, hi = -10.0, 10.0
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        lo, hi = (mid, hi) if cdf(mid) < p else (lo, mid)
    return 0.5 * (lo + hi)


def run(rho, q):
    a, b = math.sqrt(rho), math.sqrt(1.0 - rho)
    cut = quantile(q)
    buy_n = buy_up = all_n = all_up = 0
    gains = []
    for _ in range(OWNERS):
        s = rng.gauss(0.0, 1.0)
        m = [a * s + b * rng.gauss(0.0, 1.0) for _ in range(MONTHS)]
        for t in range(MONTHS - 1):
            up = m[t + 1] > m[t]
            all_n += 1
            all_up += up
            if m[t] < cut:
                buy_n += 1
                buy_up += up
                gains.append(m[t + 1] - m[t])
    return buy_up / buy_n, all_up / all_n, math.fsum(gains) / len(gains)


if __name__ == "__main__":
    print("rho   bottom q   buyers: next better   mean gain   any month")
    for rho in (0.3, 0.6):
        for q in (0.05, 0.15, 0.30):
            r_buy, r_all, g = run(rho, q)
            print("{:.1f}    {:.2f}          {:.3f}            {:+.3f}"
                  "       {:.3f}".format(rho, q, r_buy, g, r_all))
    print("a report that only says 'it will get better' is right that")
    print("often with no knowledge at all")

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《均值回归与封面诅咒、运程报告后验校准与塞翁失马序列独立性母典》第 6 章「运程报告的后验校准:低谷购买、巴纳姆陈述与布赖尔分数」(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-202 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026202,
  author    = {Fan, Yafu},
  title     = {Regression to the Mean and the Cover Jinx: Posterior Calibration of Fortune Reports and Serial Independence in the Tale of the Lost Horse},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论