关于「原典并读:壹赏、定分与信赏必罚」,《VCG 拍卖与社群激励》给出的定量结论是什么?
结论:《商君书·赏刑》的原话是「壹赏,壹刑,壹教」:赏只出于战功一途,刑不分等级、功过不相抵,「明赏不费,明刑不戮」。《定分》与《吕氏春秋·慎势》所引慎子的「一兔走,百人逐之」字句不同,意思相同:归属未定的有价物引起争夺,作者的模型(图洛克竞赛)显示 100 人争夺时 99% 的价值被耗散,界定归属则不耗散。「信赏必罚」不见于《商君书》,较早的出处是《韩非子·外储说右上》狐偃之语,又见《汉书·宣帝纪》赞。法家的赏罚服务于君主、参与不可退出、约束的是臣民而不是君主;机制设计的激励相容以参与者自身的最优选择与自愿参与为前提。二者同名不同义,相似只在结构层面。
数据口径:书中以标注框区分已发表的证据、可运行的 Python 计算(16 段程序)、原典与作者的模型和假说;书中所有数据都是虚构或模拟的。
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。 与结论吻合的打印值: 100
| (A) one hare, n chasers: equilibrium effort each, total effort (= share of | |
| the hare's value dissipated) | |
| n | 2: iterated 0.2500, formula 0.2500; total 0.500; deviation gain 0.0000 |
| n | 3: iterated 0.2222, formula 0.2222; total 0.667; deviation gain 0.0000 |
| n | 10: iterated 0.0900, formula 0.0900; total 0.900; deviation gain 0.0000 |
| n | 100: iterated 0.0099, formula 0.0099; total 0.990; deviation gain 0.0000 |
| names and shares settled (owner known, theft punished): effort 0 | |
| (B) deterrence: q F > | 1 needed |
| detection 0.9, penalty | 1.5: expected penalty 1.35 -> deterred |
| detection 0.5, penalty | 1.5: expected penalty 0.75 -> not deterred |
| detection 0.5, penalty | 3.0: expected penalty 1.50 -> deterred |
| detection 0.1, penalty | 3.0: expected penalty 0.30 -> not deterred |
| detection 0.1, penalty | 12.0: expected penalty 1.20 -> deterred |
| (C) to get 60 % of agents to work, announced reward needed | 0.6 / rho |
| paid with probability 1.00: announce 0.60 | |
| paid with probability 0.80: announce 0.75 | |
| paid with probability 0.50: announce 1.20 | |
| paid with probability 0.25: announce 2.40 |
查看完整验算程序
# Chapter 7. The author's toy models beside "one hare runs, a hundred chase it"
# (Shangjunshu, Dingfen; Lushi chunqiu, Shenshi). An analogy, not a reconstruction.
# (A) Unassigned prize: n chasers spend effort x_i, win the hare (value 1) with
# probability x_i / sum x (Tullock lottery). Symmetric equilibrium x* = (n-1)/n^2;
# total effort (n-1)/n of the prize is dissipated. Checked by damped iteration of
# the best reply sqrt(O) - O to the others' total O, then by a grid search for a
# profitable deviation. Assigned prize ("names and shares settled"): no one chases.
# (B) Enforcement ("bi fa"): a thief takes a hare worth 1 unless deterred; expected
# penalty q F (detection probability q, fine F). Deterred iff q F >= 1 (Becker 1968).
# (C) Credibility ("xin shang"): a reward R promised for a task, paid with probability
# rho; agents with costs ~ U[0,1] work iff rho R >= cost. Share working = rho R.
def best_response(others, grid=4000):
best = (0.0, 0.0)
for k in range(grid + 1):
x = 0.5 * k / grid
u = (x / (x + others) if x + others > 0 else 0.0) - x
if u > best[0]:
best = (u, x)
return best[1]
if __name__ == "__main__":
print("(A) one hare, n chasers: equilibrium effort each, total effort (= share of")
print(" the hare's value dissipated)")
for n in (2, 3, 10, 100):
x = 0.1
for _ in range(5000): # damped best-reply iteration
O = (n - 1) * x; x = 0.995 * x + 0.005 * max(0.0, O ** 0.5 - O)
O = (n - 1) * x; dev = best_response(O)
gain = (dev / (dev + O) - dev) - (x / (x + O) - x)
print(f" n = {n:3d}: iterated {x:.4f}, formula {(n - 1) / n ** 2:.4f};"
f" total {n * x:.3f}; deviation gain {max(gain, 0.0):.4f}")
print(" names and shares settled (owner known, theft punished): effort 0")
print("(B) deterrence: q F >= 1 needed")
for q, F in ((0.9, 1.5), (0.5, 1.5), (0.5, 3.0), (0.1, 3.0), (0.1, 12.0)):
print(f" detection {q:.1f}, penalty {F:5.1f}: expected penalty {q * F:5.2f}"
f" -> {'deterred' if q * F >= 1 else 'not deterred'}")
print("(C) to get 60 % of agents to work, announced reward needed = 0.6 / rho")
for rho in (1.0, 0.8, 0.5, 0.25):
print(f" paid with probability {rho:.2f}: announce {0.6 / rho:.2f}")唯一数据源与引用
范雅福(Fan, Y.). (2026). 《机制设计、维克里—克拉克—格罗夫斯(VCG)拍卖与数字社群激励相容工程》第 7 章「原典并读:壹赏、定分与信赏必罚」(版本 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23836274
DOI: 10.5281/zenodo.23836274 · 阅读本章所在专著 · 5DL-B-050 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026050,
author = {Fan, Yafu},
title = {Mechanism Design, Vickrey–Clarke–Groves Auctions and Incentive-Compatible Community Engineering},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
doi = {10.5281/zenodo.23836274},
url = {https://doi.org/10.5281/zenodo.23836274}
}