5D Library › 分拆经济学、防套利身份锁定与多阶授权阶梯 · English
第一章 捆绑何时有利:从两个板块到四个板块
捆绑定理说了什么
多产品定价的经典结论来自 Adams 与 Yellen:当买家对两件商品的估值负相关时,把它们捆在一起出售,可以把买家之间估值的差异「抹平」,从而抽取更多的消费者剩余;同时提供单品与套餐(混合捆绑)往往比只卖套餐更好 (Adams and Yellen 1976)。McAfee、McMillan 与 Whinston 进一步证明,只要估值相互独立,在单品最优价格的基础上加一个略有折扣的套餐,总是有利可图的;捆绑不需要负相关 (McAfee et al. 1989)。Bakos 与 Brynjolfsson 则把这一结论推到大量信息产品:零边际成本、估值独立时,套餐中的产品越多,买家对套餐的总估值越集中,卖家几乎可以抽走全部剩余 (Bakos and Brynjolfsson 1999)。
证据
三条已发表的结论。 (1)负相关估值下,纯捆绑与混合捆绑优于分拆 (Adams and Yellen 1976)。(2)独立估值下,在单品最优价上加一个折扣套餐,收入严格提高 (McAfee et al. 1989)。(3)零边际成本、独立估值、产品数量很大时,捆绑的收入趋于全部消费者剩余 (Bakos and Brynjolfsson 1999)。三条都是关于最优定价的套餐;它们没有说任意价格的套餐都有利。
因此,作者提纲中「分拆击败廉价全包」的命题需要拆成两半:分拆能不能击败最优定价的全包?分拆能不能击败便宜的全通卡?本章先回答前一半。
两个板块:教科书的例子
设五位买家对周期测算板块(C)与声学干预板块(A)的估值负相关。程序在所有可能的价格上精确搜索。
推导
# Chapter 1. [p01] Adams-Yellen on two modules: components, bundle, mixed.
# Five buyers value the cycle module (C) and the acoustic module (A).
# Values are negatively correlated, the textbook case for bundling.
# Marginal cost is zero (digital goods). Every candidate price is a
# buyer's value or a sum of values, so the search below is exact.
from itertools import product
BUYERS = ((100, 10), (80, 40), (60, 60), (40, 80), (10, 100))
def revenue_separate(pc, pa):
return sum((pc if c >= pc else 0) + (pa if a >= pa else 0)
for c, a in BUYERS)
def revenue_bundle(pb):
return sum(pb for c, a in BUYERS if c + a >= pb)
def choice(c, a, pc, pa, pb):
"""the option with the largest surplus; ties go to more product"""
opts = [(0, 0, "none"), (c - pc, pc, "C"), (a - pa, pa, "A"),
(c + a - pb, pb, "C+A")]
best = max(opts, key=lambda o: (o[0], o[1]))
return best if best[0] >= 0 else (0, 0, "none")
def revenue_mixed(pc, pa, pb):
return sum(choice(c, a, pc, pa, pb)[1] for c, a in BUYERS)
if __name__ == "__main__":
vals = sorted({c for c, a in BUYERS} | {a for c, a in BUYERS})
sums = sorted({c + a for c, a in BUYERS})
sep = max((revenue_separate(pc, pa), pc, pa)
for pc, pa in product(vals, vals))
bun = max((revenue_bundle(pb), pb) for pb in sums)
mix = max((revenue_mixed(pc, pa, pb), pc, pa, pb)
for pc, pa, pb in product(vals, vals, sums))
print(" pure components: pC={} pA={} revenue {}".format(
sep[1], sep[2], sep[0]))
print(" pure bundle : pCA={} revenue {}".format(
bun[1], bun[0]))
print(" mixed bundling : pC={} pA={} pCA={} revenue {}".format(
mix[1], mix[2], mix[3], mix[0]))
for c, a in BUYERS:
print(" buyer ({:3d},{:3d}) chooses {}".format(
c, a, choice(c, a, *mix[1:])[2]))
print("\nReading: with negatively correlated values the bundle")
print("beats separate prices and mixed bundling beats both. This is")
print("the case FOR bundling; Chapter 2 asks when it fails.")
# pure components: pC=60 pA=60 revenue 360
# pure bundle : pCA=110 revenue 550
# mixed bundling : pC=100 pA=100 pCA=120 revenue 560
# buyer (100, 10) chooses C
# buyer ( 80, 40) chooses C+A
# buyer ( 60, 60) chooses C+A
# buyer ( 40, 80) chooses C+A
# buyer ( 10,100) chooses A
#
# Reading: with negatively correlated values the bundle
# beats separate prices and mixed bundling beats both. This is
# the case FOR bundling; Chapter 2 asks when it fails.分拆时两个板块各卖 60,收入 360;套餐卖 110,收入 550;混合捆绑把单品价提到 100、套餐卖 120,收入 560。两端的买家只买自己偏爱的单品,中间三位买套餐。这正是 Adams 与 Yellen 所说的机制:套餐把「一个高、一个低」的估值加成一个稳定的总数。
四个板块、四种需求结构
把四个板块记为 C(周期测算)、A(声学干预)、M(代谢配餐)与 B(商业脚本),用 8,000 位模拟买家在四种需求结构下比较四种方案:
- indep:四个板块的估值独立,均匀分布在 0 到 100 之间;
- ortho:每位买家只想要一个板块,其余板块对他几乎没有价值(单一尺度的正交需求);
- segm:92% 是普通消费者,只对 C、A、M 有 0 到 100 的估值;8% 是工作室,对商业脚本 B 的估值在 1,000 到 5,000 之间(正交,并且相差三个数量级);
- pareto:四个板块的估值独立,服从尾指数 1.5 的 Lomax 分布(重尾,方差无穷),以 2,000 的预算封顶。
为了不让样本中某一位「巨鲸」决定价格,程序要求每个价格至少卖给 1% 的买家。
推导
# Chapter 1. [p02] Four modules, four demand regimes, four price plans.
# Modules: C cycle, A acoustic, M metabolic, B business scripts.
# 8,000 simulated buyers per regime (seeds 68..71):
# indep : values iid uniform on 0..100 (light tails, independent)
# ortho : each buyer wants ONE module (uniform 0..100), the others
# are worth 0..5 to her (orthogonal demand, one scale)
# segm : 92% consumers value C, A, M uniform 0..100 and B 0..10;
# 8% studios value B uniform 1,000..5,000 and C, A, M 0..100
# (orthogonal AND three orders of magnitude apart)
# pareto: values iid Lomax, scale 20, tail index 1.5 (heavy tails),
# capped at a budget of 2,000
# Plans: SEP optimal separate prices; BUN optimal pure bundle;
# PASS separate prices plus a cheap all-access pass priced at
# 1.5 x the mean separate price; MIX best of separate prices
# scaled by s (1, 1.25, 1.5, 2) plus a pass on a price grid.
# A price must sell to at least 1% of buyers (no pricing for a single
# whale in the sample).
import random
N, MODS = 8000, 4
def draw(regime, rng):
if regime == "indep":
return [rng.uniform(0, 100) for _ in range(MODS)]
if regime == "ortho":
v = [rng.uniform(0, 5) for _ in range(MODS)]
v[rng.randrange(MODS)] = rng.uniform(0, 100)
return v
if regime == "segm":
v = [rng.uniform(0, 100) for _ in range(3)]
if rng.random() < 0.08:
return v + [rng.uniform(1000, 5000)]
return v + [rng.uniform(0, 10)]
return [min(2000.0, 20 * ((1 - rng.random()) ** (-1 / 1.5) - 1))
for _ in range(MODS)]
def best_price(values):
s = sorted(values)
n = len(s)
top = n - n // 100
return max((s[i] * (n - i), s[i]) for i in range(top))[1]
def revenue(pop, prices, pass_price=None):
tot = 0.0
for v in pop:
sep_s = sum(max(0.0, x - p) for x, p in zip(v, prices))
sep_r = sum(p for x, p in zip(v, prices) if x >= p)
t = sum(v)
if pass_price is not None and t >= pass_price and \
t - pass_price >= sep_s:
tot += pass_price
else:
tot += sep_r
return tot / len(pop)
def plans(pop):
prices = [best_price([v[j] for v in pop]) for j in range(MODS)]
pb = best_price([sum(v) for v in pop])
sep = revenue(pop, prices)
bun = revenue(pop, [1e18] * MODS, pb)
cheap = 1.5 * sum(prices) / MODS
pas = revenue(pop, prices, cheap)
mix = max(revenue(pop, [s * p for p in prices], pb * k / 10)
for s in (1, 1.25, 1.5, 2) for k in range(6, 17))
return prices, pb, sep, bun, cheap, pas, mix
if __name__ == "__main__":
print(" regime SEP BUN PASS 1.5x MIX SEP/BUN")
for k, regime in enumerate(("indep", "ortho", "segm", "pareto")):
rng = random.Random(68 + k)
pop = [draw(regime, rng) for _ in range(N)]
pr, pb, sep, bun, cheap, pas, mix = plans(pop)
print(" {:6s} {:8.2f} {:8.2f} {:11.2f} {:8.2f} {:9.3f}".format(
regime, sep, bun, pas, mix, sep / bun))
print(" prices C,A,M,B = {}".format(
", ".join("{:.1f}".format(p) for p in pr)))
print(" bundle {:.1f}; cheap pass {:.1f}".format(
pb, cheap))
print("\n (revenue per buyer; cost zero)")
# regime SEP BUN PASS 1.5x MIX SEP/BUN
# indep 99.82 121.05 73.37 122.31 0.825
# prices C,A,M,B = 52.3, 50.5, 48.3, 47.4
# bundle 156.8; cheap pass 74.4
# ortho 25.29 28.95 25.29 28.95 0.873
# prices C,A,M,B = 52.5, 53.8, 54.3, 48.4
# bundle 50.8; cheap pass 78.4
# segm 206.54 139.65 151.75 207.16 1.479
# prices C,A,M,B = 48.7, 50.8, 51.1, 2558.5
# bundle 2647.3; cheap pass 1015.9
# pareto 31.05 42.51 38.58 41.95 0.730
# prices C,A,M,B = 46.0, 39.5, 30.0, 47.9
# bundle 92.7; cheap pass 61.3
#
# (revenue per buyer; cost zero)结果与提纲的直觉相反:在 indep、ortho 与 pareto 三种结构下,最优纯捆绑都胜过分拆,分拆只得到捆绑收入的 0.825、0.873 与 0.730。正交需求本身不足以让分拆获胜:当每位买家只想要一个板块时,套餐价格可以设在单品价附近,额外的零星估值反而成了白得的收入。重尾估值也不足以让分拆获胜(第二章会把尾指数一路调低)。
只有在 segm 结构下,分拆是最优捆绑的 1.479 倍。原因在于价格的数量级:B 板块的最优单价为 2558.5,消费者板块约为 50;一个统一的套餐价格要么设在 2647.3,把所有消费者挡在门外,要么设在消费者能接受的水平,把工作室的剩余白白送掉。混合捆绑(MIX)几乎只能追平分拆。
假说 · 尚未验证
作者的诠释(尚未验证): 作者业务的真实需求接近 segm:商业脚本板块的买家是少数工作室与从业者,他们的估值比个人用户高两到三个数量级,而个人用户对四个板块的兴趣大体彼此独立。在这种结构下,应当分拆出售,并把 B 板块单独定价。什么会证明这一判断错:若用真实的盲测意愿支付数据(而非本书的参数)估计出 B 板块与消费者板块的估值同处一个数量级,或者个人用户对 B 板块也有可观的估值,那么 indep 或 pareto 的结论适用,最优捆绑将胜过分拆。
小结
捆绑定理说的是最优定价的套餐。在独立、单一尺度正交与重尾三种结构下,最优纯捆绑都胜过分拆;分拆只有在「正交」与「数量级分离」同时成立时才获胜,在本书参数下收入高出 1.479 倍。至于便宜的全通卡,那是另一个问题,留给第二章。
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).