关于「孵化池的幸存者偏差:乘法增长、淘汰下限与被看见的头部」,《达人经纪的幂律、抽查与买断》给出的定量结论是什么?

结论:第一章用带淘汰下限的乘法增长模型说明,只看仍活跃的达人会把头部率放大 1.74 倍,只看被外界看见的达人会放大 10.70 倍,而下限本身会把对数正态变成幂律。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是个人化的法律或理财建议。 第六章把阶梯分成、买断金与抽查率合成三联定价试算器,买断金取剩余佣金的 1.712 倍。

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: 1.74, 10.70

pool 2000, 24 months, floor 3000, visible >100000, head >= 500000
drift vol alive vis heads true |alive |visible drift seen
0.00 0.351152 187 28 0.0140 0.0243 0.1497 +0.1249
-0.02 0.35948 110 16 0.0080 0.0169 0.1455 +0.1215
0.00 0.301240 107 14 0.0070 0.0113 0.1308 +0.1163
0.02 0.251584 171 12 0.0060 0.0076 0.0702 +0.1064
reference row: the head rate looks 1.74 times larger among survivors
and 10.70 times larger among visible creators, whose drift reads +0.1249
a month although the true drift is zero
查看完整验算程序
# Chapter 1. p11: survivorship bias in a creator incubation pool.
# Model (illustrative parameters, not data): 2000 newly signed
# creators start near 10000 followers (lognormal spread 0.5).  Each
# month the log follower count takes a normal step with drift MU and
# volatility SIG.  A creator who falls below 3000 followers is dropped
# for good.  After 24 months a "head" has at least 500000 followers;
# outsiders can "see" a creator at 100000 or more.  The program
# compares the true head rate with the rates read off survivors and
# off visible creators, and the drift an outsider would infer.
import math
import random

N, MONTHS = 2000, 24
F0, FLOOR, VISIBLE, HEAD = 10000, 3000, 100000, 500000


def run(mu, sig, rng):
    alive = vis = head = 0
    g_vis = []
    lf, lv, lh = math.log(FLOOR), math.log(VISIBLE), math.log(HEAD)
    for _ in range(N):
        x0 = math.log(F0) + rng.gauss(0.0, 0.5)
        x, ok = x0, True
        for _ in range(MONTHS):
            x += rng.gauss(mu, sig)
            if x < lf:
                ok = False
                break
        if not ok:
            continue
        alive += 1
        if x >= lv:
            vis += 1
            g_vis.append((x - x0) / MONTHS)
        if x >= lh:
            head += 1
    g = math.fsum(g_vis) / len(g_vis)
    return alive, vis, head, g


rng = random.Random(20261112)
print("pool {}, {} months, floor {}, visible >= {}, head >= {}".format(
    N, MONTHS, FLOOR, VISIBLE, HEAD))
print(" drift  vol  alive  vis  heads  true   |alive  |visible  "
      "drift seen")
rows = []
for mu, sig in ((0.0, 0.35), (-0.02, 0.35), (0.0, 0.30),
                (0.02, 0.25)):
    a, v, h, g = run(mu, sig, rng)
    rows.append((a, v, h, g))
    print("{:6.2f} {:4.2f} {:6d} {:4d} {:6d}  {:.4f} {:.4f}  {:.4f}"
          "   {:+.4f}".format(mu, sig, a, v, h, h / N, h / a, h / v, g))
a, v, h, g = rows[0]
print("reference row: the head rate looks {:.2f} times larger among"
      " survivors".format((h / a) / (h / N)))
print("and {:.2f} times larger among visible creators, whose drift"
      " reads {:+.4f}".format((h / v) / (h / N), g))
print("a month although the true drift is zero")

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《网红经纪机构孵化幸存者幂律、达人飞单抽样审计博弈与违约买断期权定价》第 1 章「孵化池的幸存者偏差:乘法增长、淘汰下限与被看见的头部」(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-172 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026172,
  author    = {Fan, Yafu},
  title     = {Survivor Power Laws in Influencer-Agency Incubation, Sampling-Audit Games against Off-Platform Deals and Option Pricing of Contract Buyouts},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论