5D Library › 分拆经济学、防套利身份锁定与多阶授权阶梯 · English
第四章 自购返佣、渠道穿仓与伙伴的模仿
自购返佣:套利成本减半
第三章假定算命师买卡要付全价。许多分销计划并不禁止伙伴通过自己的推广链接购买,于是她买卡时还能拿回 的佣金,每个名额的成本降为 。另一方面,公司交付每一次测算都有成本(算力、人工复核、客服),这里记为 。若公司从每个被转售名额得到的净额 变成负数,渠道就「穿仓」了:卖得越多,亏得越多。
推导
# Chapter 4. [p07] Self-referral and the channel "blow-through".
# As in p05: p2 = 99, card P = 199 for the holder plus k extra slots,
# commission g = 0.5. If the reseller may also buy the card through
# her own affiliate link, she gets g * P back, so a slot costs her
# (1 - g) P / (k + 1). Each reading costs the firm c = 15 to deliver
# (compute, review, support). The channel blows through when the firm's
# net per reading served, (1 - g) P / (k + 1) - c, falls below zero.
P2, CARD, G, C = 99.0, 199.0, 0.5, 15.0
if __name__ == "__main__":
print(" k slot cost slot cost reseller margin firm net/slot")
print(" (no self) (self-ref) at s = p2, self with self-ref")
for k in range(0, 9):
plain = CARD / (k + 1)
self_ = (1 - G) * CARD / (k + 1)
print(" {:2d} {:10.2f} {:11.2f} {:16.2f} {:14.2f}".format(
k, plain, self_, P2 - self_, self_ - C))
kb = next(k for k in range(0, 99) if (1 - G) * CARD / (k + 1) < C)
ka = next(k for k in range(0, 99)
if P2 - (1 - G) * CARD / (k + 1) > G * P2)
print(" with self-referral, resale beats referral from k = {}".format(
ka))
print(" firm net per slot below zero (blow-through) from k = {}".format(
kb))
print(" referral net per reading for comparison: {:.2f}".format(
(1 - G) * P2 - C))
print("\nReading: paying commission on a partner's own purchase halves")
print("the cost of arbitrage: two extra slots already make resale")
print("beat referral, and from six the firm pays to serve a reading.")
# k slot cost slot cost reseller margin firm net/slot
# (no self) (self-ref) at s = p2, self with self-ref
# 0 199.00 99.50 -0.50 84.50
# 1 99.50 49.75 49.25 34.75
# 2 66.33 33.17 65.83 18.17
# 3 49.75 24.88 74.12 9.88
# 4 39.80 19.90 79.10 4.90
# 5 33.17 16.58 82.42 1.58
# 6 28.43 14.21 84.79 -0.79
# 7 24.88 12.44 86.56 -2.56
# 8 22.11 11.06 87.94 -3.94
# with self-referral, resale beats referral from k = 2
# firm net per slot below zero (blow-through) from k = 6
# referral net per reading for comparison: 34.50
#
# Reading: paying commission on a partner's own purchase halves
# the cost of arbitrage: two extra slots already make resale
# beat referral, and from six the firm pays to serve a reading.允许自购返佣时,附赠 2 个名额,转售就胜过推荐(每位客户 65.83 对第三章 p05 中的 49.5);附赠 6 个名额时,公司每服务一个被转售的名额净亏 0.79,而一位正常推荐来的客户能给公司留下 34.50。自购返佣把第三章的门槛从 4 个名额拉低到 2 个,并在 6 个名额处让公司真正亏本。
推导
穿仓条件。 允许自购返佣、附赠 个名额时,公司每个转售名额的净额为 它小于零当且仅当 。本书参数下,从 起穿仓。防穿仓的两条规则彼此独立:佣金不付给本人购买(身份锁的第二个用途),或者附赠名额不可转让。
渠道伙伴会模仿谁
套利一旦比推荐划算,它不会只停留在少数人身上。设有 400 位渠道伙伴,每人每月服务 10 位客户;每个月,每位伙伴随机看一位同行,若对方上个月赚得更多,就以与差距成比例的概率改用对方的做法。这是演化博弈中的成对模仿规则,其平均动态就是复制者动态 (Weibull 1995)。转售还要承担客户投诉的流失(每月 10%)。起点是 2% 的伙伴在转售。
推导
# Chapter 4. [p08] Partner dynamics: how fast does a channel turn into
# resellers? 400 channel partners each serve 10 clients a month. Each
# month a partner compares her last-month payoff with that of a random
# other partner and copies the better strategy with probability
# gap / (referral payoff) (pairwise imitation, a discrete replicator
# rule), seed 68. Referral pays g * p2 = 49.5 per client. Resale pays
# p2 - slot cost, slot cost from p05/p07; resale also loses a share of
# clients each month to complaints (10%). Start: 2% resellers.
import random
P2, CARD, G, PARTNERS, CLIENTS = 99.0, 199.0, 0.5, 400, 10
def resale_pay(k, self_ref):
cost = (1 - G) * CARD if self_ref else CARD
return P2 - cost / (k + 1)
def run(k, self_ref, months=24, seed=68):
rng = random.Random(seed)
resell = [rng.random() < 0.02 for _ in range(PARTNERS)]
pay_r = 0.9 * CLIENTS * resale_pay(k, self_ref)
pay_a = CLIENTS * G * P2
scale = pay_a
for _ in range(months):
new = resell[:]
for i in range(PARTNERS):
j = rng.randrange(PARTNERS)
mine = pay_r if resell[i] else pay_a
theirs = pay_r if resell[j] else pay_a
if theirs > mine and rng.random() < (theirs - mine) / scale:
new[i] = resell[j]
resell = new
return sum(resell) / PARTNERS
if __name__ == "__main__":
print(" share of partners reselling after 6, 12 and 24 months")
print(" card 6 mo 12 mo 24 mo")
cases = (("identity lock (k = 0)", 0, False),
("k = 3, no self-referral", 3, False),
("k = 5, no self-referral", 5, False),
("k = 3, self-referral paid", 3, True))
for name, k, sr in cases:
out = [run(k, sr, m) for m in (6, 12, 24)]
print(" {:26s} {:6.3f} {:7.3f} {:7.3f}".format(name, *out))
print("\nReading: once resale pays more than referral, imitation")
print("spreads it, slowly at first and then fast: most of the channel")
print("within two years when self-referral is paid. With the lock, or")
print("whenever resale pays less, the first resellers die out.")
# share of partners reselling after 6, 12 and 24 months
# card 6 mo 12 mo 24 mo
# identity lock (k = 0) 0.000 0.000 0.000
# k = 3, no self-referral 0.010 0.003 0.000
# k = 5, no self-referral 0.020 0.062 0.380
# k = 3, self-referral paid 0.028 0.122 0.877
#
# Reading: once resale pays more than referral, imitation
# spreads it, slowly at first and then fast: most of the channel
# within two years when self-referral is paid. With the lock, or
# whenever resale pays less, the first resellers die out.有身份锁时,起初那 2% 的转售者很快消失;附赠 3 个名额、不许自购返佣时,转售比推荐少赚,转售者同样消失。附赠 5 个名额时,转售略胜推荐,扩散起初很慢,第 12 个月才到 0.062,第 24 个月已到 0.380。附赠 3 个名额而允许自购返佣时,24 个月后 0.877 的伙伴在转售:分销渠道已经变成了一个转售渠道。
假说 · 尚未验证
作者的诠释(尚未验证): 渠道里的转售不是个别人的道德问题,而是价格结构的结果:只要转售比推荐划算,它就会通过同行之间的模仿扩散,而且扩散在一年之内往往看不出来。什么会证明这一判断错:若在保留附赠名额与自购返佣的渠道中,连续两年观察到的转售伙伴比例一直很低(例如低于 5%),说明伙伴之间的模仿远比本模型弱,或者有本模型没有计入的成本(例如名誉)在起作用。
小结
允许自购返佣会让套利成本减半:附赠 2 个名额,转售就胜过推荐;附赠 6 个名额,公司开始为每个转售名额倒贴。渠道伙伴之间的模仿会把有利可图的转售扩散开来:附赠 3 个名额并允许自购返佣时,两年后 0.877 的伙伴在转售。身份锁与「不对本人购买付佣金」两条规则合起来,使转售在一开始就无利可图。
参考文献
Fan, Y. (2026). Unbundling Economics, Anti-Arbitrage Identity Locking and Multi-Tier Licensing Ladders: Orthogonal Demand Partitioning, Resale Cannibalization Bounds, Affiliate Margin Preservation and Five-Figure B2B Upsell Equilibria [分拆经济学、防套利身份锁定与多阶授权阶梯] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090184 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).