5D Library › 分销网络幂律激活、网红矩阵组合优化与激励稳定边界 · English

第五章 多层分佣的几何与金字塔的饱和

多层分佣需要多少人在下面

多层直销计划把佣金分摊到上线的好几层。p09 设定一个总计 50% 的七层分佣(20%、10%、8%、5%、4%、2%、1%),每位成员每月流水 300 令吉,其中 200 令吉是维持资格的必要购买。

推导

# Chapter 5. p09: The geometry of multi-level pay. A plan pays 50% of
# volume in all, spread over seven levels: 20, 10, 8, 5, 4, 2 and 1
# percent. Each member moves RM 300 a month, RM 200 of it the purchase
# needed to stay qualified. If every member brings in b recruits, a
# member with l full levels below has b + b^2 + ... + b^l people under
# her. How many does she need for RM 3,000 a month of override, and in
# a finished network (a full b-ary tree, depth 12) what share of members
# earn more in override than their own required purchase?
RATES = (0.20, 0.10, 0.08, 0.05, 0.04, 0.02, 0.01)
VOLUME, BUY = 300.0, 200.0

def override(b, levels):
    return sum(r * VOLUME * b ** (i + 1)
               for i, r in enumerate(RATES[:levels]))

if __name__ == "__main__":
    print("   b   levels for RM 3,000   people below   override (RM)")
    for b in (2, 3, 5):
        for lv in range(1, 8):
            if override(b, lv) >= 3000 or lv == 7:
                people = sum(b ** k for k in range(1, lv + 1))
                note = "" if override(b, lv) >= 3000 else "  not reached"
                print("   {}   {:19d}   {:12,d}   {:13,.0f}{}".format(
                    b, lv, people, override(b, lv), note))
                break
    print("   full tree of depth 12")
    print("   b   members        share above purchase   share with no"
          " recruit")
    for b in (2, 3, 5):
        depth = 12
        members = sum(b ** d for d in range(depth + 1))
        above = 0
        for d in range(depth + 1):
            if override(b, min(7, depth - d)) > BUY:
                above += b ** d
        leaves = b ** depth
        print("   {}   {:13,d}  {:21.3f}   {:20.3f}".format(
            b, members, above / members, leaves / members))
#    b   levels for RM 3,000   people below   override (RM)
#    2                     7            254           1,824  not reached
#    3                     5            363           5,229
#    5                     3            155           4,050
#    full tree of depth 12
#    b   members        share above purchase   share with no recruit
#    2           8,191                  0.250                  0.500
#    3         797,161                  0.111                  0.667
#    5     305,175,781                  0.200                  0.800

若每人只招两人,即使七层全满、下面有 254 人,每月的层级佣金也只有 1,824 令吉,达不到 3,000 令吉;每人招三人,需要五层、363 人;每人招五人,需要三层、155 人。再看一个「完成」的网络(深度十二层的满树):每人招五人时成员总数是 305,175,781,远超任何一个国家的可及人口;即使在这样的满树里,层级佣金超过自己必要购买的成员也只有 0.200,而 0.800 的成员下面一个人也没有。几何结构决定了:多数人必然在最底层。

有限市场中的饱和

p10 把同一个计划放进一个只有五万人可以招募的市场,按招募速度随市场饱和而放慢的方式运行三十六个月,并累计每位成员在网络中的净得失(必要购买按半价计为自用价值)。

推导

# Chapter 5. p10: A multi-level network in a finite market. 50,000
# people can be reached. The network starts with 10 members; each month
# every member recruits one more person with probability
# 0.3 (1 - members / 50,000). Members use the plan of p09: RM 300 of
# volume a month, of which RM 200 is a required purchase; the purchase
# is counted at half its price as worth it to the member, RM 30 is
# margin on outside retail, and overrides go up seven levels. After 36
# months, each member's net gain over her own time in the network is
# summed. For comparison, a single-tier affiliate pays nothing to join
# and keeps 50% of her outside retail. Seed 87.
import random

RATES = (0.20, 0.10, 0.08, 0.05, 0.04, 0.02, 0.01)
M, MONTHS = 50000, 36

if __name__ == "__main__":
    rng = random.Random(87)
    up = [None] * 10
    net = [0.0] * 10
    for month in range(MONTHS):
        n = len(up)
        for i in range(n):
            net[i] += 30.0 - 100.0
            j, lvl = up[i], 0
            while j is not None and lvl < 7:
                net[j] += RATES[lvl] * 300.0
                j, lvl = up[j], lvl + 1
        p = 0.3 * (1 - n / M)
        for i in range(n):
            if rng.random() < p:
                up.append(i)
                net.append(0.0)
        if month in (11, 23, 35):
            losers = sum(1 for x in net if x < 0) / len(net)
            print("   month {:2d}: members {:6,d}, share with a net loss"
                  " {:.3f}".format(month + 1, len(net), losers))
    s = sorted(net, reverse=True)
    top = sum(s[:len(s) // 100])
    gains = sum(x for x in s if x > 0)
    print("   top 1% of members take {:.3f} of all net gains".format(
        top / gains))
    print("   median member net (RM): {:,.0f}".format(s[len(s) // 2]))
    print("   single tier (no fee, RM 50 a month on retail): share with"
          " a loss 0.000")
#    month 12: members    349, share with a net loss 0.607
#    month 24: members  7,314, share with a net loss 0.631
#    month 36: members 43,165, share with a net loss 0.724
#    top 1% of members take 0.524 of all net gains
#    median member net (RM): -160
#    single tier (no fee, RM 50 a month on retail): share with a loss 0.000

第十二个月成员只有 349 人,第二十四个月 7,314 人,第三十六个月 43,165 人,接近市场上限;亏损成员的比例从 0.607 升到 0.724。前 1% 的成员拿走全部净收益的 0.524。相比之下,单层分销员不缴费、不必压货,只赚自己对外零售的佣金,没有人会亏钱;他们的收入同样重尾,但重尾来自销售,不来自招募。

法律界线

马来西亚《直销与反金字塔计划法令 1993》(第 500 号法令)第 VA 部分(第 27A、27B 条)禁止金字塔计划,其附表所列的特征之一,是参与者的奖金或利益「完全或主要」来自招募或介绍他人,而不是来自销售 (Malaysia 1993)。美国联邦贸易委员会 2018 年的多层营销指引也指出,判断的关键在于报酬是否以对真实顾客的零售为基础,而不是参与者自己的购买或招募 (Federal Trade Commission 2018)。

假说 · 尚未验证

作者的诠释(尚未验证): 单层直推 50% 佣金加上持牌主理人的批发差价,可以在不依赖招募收入的前提下,让顶尖分销员获得与多层计划相当的收入。什么会证明它错: 在实际运行的计划中,持牌主理人收入的相当部分(例如超过三成)仍来自与招募挂钩的款项,或顶尖分销员的收入显著低于同类多层计划的顶层。

本书不是法律意见;任何分佣方案在实施前都须由当地执业律师审查。

小结

多层分佣的数学结构决定了多数成员在最底层,市场饱和后多数人亏损,收益集中在最早加入的少数人手中。单层分佣的收入同样重尾,但来自销售而非招募,这正是法律界线所在。

参考文献

Federal Trade Commission. 2018. Business Guidance Concerning Multi-Level Marketing. Federal Trade Commission, Washington, DC.
Malaysia. 1993. Direct Sales and Anti-Pyramid Scheme Act 1993 (Act 500). Laws of Malaysia.

Fan, Y. (2026). Affiliate Network Power Laws, Influencer Portfolio Optimization and Incentive Stability Bounds: Heavy-Tailed Productivity, Markowitz Traffic Allocation and Single-Tier Commission Equilibria [分销网络幂律激活、网红矩阵组合优化与激励稳定边界] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090229 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).