关于「出价上限:把鉴别、挑窗与诅咒折进一个数,以及仓位与出局」,《翡翠鉴别与赌石赢家诅咒》给出的定量结论是什么?
结论:第六章把三项修正折成出价上限:挑窗、六人竞价、证书可靠时为窗口估值的 0.309,只做门店检验时为 0.274;在天真买家占多数的公盘里应当离场;凯利比例在示意优势下为 0.090。
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,也不是任何一件石头的鉴定意见。 第六章把三项修正折成出价上限:挑窗、六人竞价、证书可靠时为窗口估值的 0.309,只做门店检验时为 0.274;在天真买家占多数的公盘里应当离场;凯利比例在示意优势下为 0.090。
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: 0.274, 0.309
| bid cap as a share of the naive window value W, pi | 0.99 |
| N k | 1 k = 4 |
| 1 | 0.743 0.512 |
| 3 | 0.530 0.365 |
| 6 | 0.448 0.309 |
| 12 | 0.387 0.267 |
| 24 | 0.341 0.235 |
| same, pi | 0.8622 (no lab test), k = 4, N = 6: 0.274 |
| check in the p12 world, pi | 1: |
| k N cap/W rivals my win rate my return | |
| 1 | 3 0.535 capped 0.3307 +0.246 |
| 1 | 3 0.535 naive 0.1115 +0.028 |
| 4 | 6 0.311 capped 0.1666 +0.229 |
| 4 | 6 0.311 naive 0.0014 -0.277 |
| 4 | 12 0.269 capped 0.0813 +0.254 |
| 4 | 12 0.269 naive 0.0001 -0.512 |
核心公式
\[\frac{\mathrm{cap}}{W} = \Big[\pi\cdot\frac{1}{F_k}\cdot e^{-s_r e_N} + (1-\pi)\,r_{\mathrm{fake}}\Big]\cdot\frac{1-c}{1+g}\]
查看完整验算程序
# Chapter 6. p61: A bid cap that undoes identification, window and curse.
# 出价上限(参数示意):W 为买手把窗口当整块时的估值。
# cap / W = [pi / F_k * exp(-s_r e_N) + (1 - pi) * r_fake]
# * (1 - c) / (1 + g)
# pi:p31 的翡翠后验;F_k:p41 的挑窗高估倍数(s = 0.4);e_N:p11 的
# 最大估计偏差,s_r = 0.4 为买手读窗误差(与 p12 相同);r_fake = 0.05
# 为非翡翠料的残值份额;c = 0.10 为切磨成本;g = 0.20 为目标回报。
# 最后在 p12 的世界里核对两种场:其余 N - 1 人也按上限出价(对称),
# 或其余人都是天真买手(只有你一人按上限出价)。
import math
import random
from statistics import NormalDist
ND = NormalDist()
S_SPOT, S_READ, R_FAKE, COST, GAIN = 0.4, 0.4, 0.05, 0.10, 0.20
MARGIN = 0.2
def order_moments(k, s, lo=-9.0, hi=9.0, steps=18000):
h = (hi - lo) / steps
m1, m2 = [], []
for i in range(steps + 1):
x = lo + i * h
f = k * ND.pdf(x) * ND.cdf(x) ** (k - 1)
w = 1 if i in (0, steps) else (4 if i % 2 else 2)
m1.append(w * x * f)
m2.append(w * math.exp(s * x - s * s / 2) * f)
return math.fsum(m1) * h / 3, math.fsum(m2) * h / 3
def cap_ratio(pi, k, n):
f_k = order_moments(k, S_SPOT)[1]
e_n = order_moments(n, S_SPOT)[0]
core = pi / f_k * math.exp(-S_READ * e_n) + (1 - pi) * R_FAKE
return core * (1 - COST) / (1 + GAIN)
def check(k, n, ratio, seed, alone, stones=20000):
rng = random.Random(20261112 + seed)
nets, paid = [], []
for _ in range(stones):
v = math.exp(rng.gauss(0.0, 1.0))
best = max(rng.gauss(-S_SPOT ** 2 / 2, S_SPOT) for _ in range(k))
window = v * math.exp(best)
reads = [window * math.exp(rng.gauss(-S_READ ** 2 / 2, S_READ))
for _ in range(n)]
mine = reads[0] * ratio
rmul = (1 - MARGIN) if alone else ratio
rival = max([r * rmul for r in reads[1:]] + [0.0])
if mine > rival:
nets.append(v * (1 - COST))
paid.append(mine)
return len(paid) / stones, math.fsum(nets) / math.fsum(paid) - 1
def main():
print("bid cap as a share of the naive window value W, pi = 0.99")
print(" N k = 1 k = 4")
for n in (1, 3, 6, 12, 24):
print("%4d %.3f %.3f"
% (n, cap_ratio(0.99, 1, n), cap_ratio(0.99, 4, n)))
print("same, pi = 0.8622 (no lab test), k = 4, N = 6: %.3f"
% cap_ratio(0.8622, 4, 6))
print("check in the p12 world, pi = 1:")
print(" k N cap/W rivals my win rate my return")
for k, n in ((1, 3), (4, 6), (4, 12)):
r = cap_ratio(1.0, k, n)
for alone in (False, True):
win, ret = check(k, n, r, 10 * k + n, alone)
print(" %d %2d %.3f %-6s %.4f %+.3f"
% (k, n, r, "naive" if alone else "capped", win, ret))
if __name__ == "__main__":
main()唯一数据源与引用
范雅福(Fan, Y.). (2026). 《翡翠静水比重贝叶斯鉴别、赌石开窗信号博弈与原石拍卖赢家诅咒母典》第 6 章「出价上限:把鉴别、挑窗与诅咒折进一个数,以及仓位与出局」(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-206 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026206,
author = {Fan, Yafu},
title = {Hydrostatic Specific Gravity and Bayesian Jadeite Identification, Window-Cut Signalling Games and the Winner's Curse in Rough-Jade Tenders},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}