5D Library › 订阅制单位经济学、净收入留存与人工智能推理标度律 · English

第七章 「贪贾三之,廉贾五之」:利润率、周转率与「财币欲其行如流水」

为什么在这里读司马迁

《史记·货殖列传》(卷一百二十九)是中国史学中第一篇系统记述商业的文字,而且量化程度少见地高。其中两句话与本书直接相关:一句关于商人赚多少,一句出自计然之策,讲钱为何必须不停流动。本章把它们与第一至第三章并读,并清楚标明何处由阅读变成了诠释。

原文

原典

司马迁《史记·货殖列传》。

節駔會,貪賈三之,廉賈五之,此亦比千乘之家,其大率也。

记计然之策,相传范蠡用之:

論其有餘不足,則知貴賤。貴上極則反賤,賤下極則反貴。貴出如糞土,賤取如珠玉。財幣欲其行如流水。

積著之理,務完物,無息幣。

记范蠡本人:

十九年之中三致千金,再分散與貧交疏昆弟。

(Sima Qian 1959, 1993)

注

原文说了什么、没说什么。 「贪贾三之,廉贾五之」的通行读法是:贪贾获利十分之三,廉贾获利十分之五。原文没有说廉贾为什么赚得更多。本书把差别读作利润率与周转率之间的权衡,这是一种诠释,下面以模型的形式写出,并标为关于文本的假说,而不是文本的本义。

利润率与周转率

第十五段程序把这种诠释写了下来。加价越高卖得越慢;资本的年收益等于加价乘以每年的周转次数,在中间某个加价处达到峰值。

推导

# Chapter 7. "The greedy merchant makes three, the modest merchant five"
# (Shiji, ch. 129), read as margin against turnover. A merchant marks goods up
# by m (as a fraction of cost). A higher markup sells more slowly: the stock
# turns T(m) = T0 exp(-m / s) times a year. The annual return on capital is
# r(m) = m T(m), which peaks at m* = s. We choose s = 0.12 and T0 so that the
# peak is 0.50 a year ("five in ten"), then find the higher markup that
# returns 0.30 ("three in ten"): the greedy merchant's price. Then the same
# arithmetic for a subscription: monthly churn rises with price, c(p) = 0.02
# exp((p - 20) / 15), and undiscounted value per signup is p / c(p).
import math

S = 0.12
T0 = 0.50 * math.e / S

def r(m):
    return m * T0 * math.exp(-m / S)

if __name__ == "__main__":
    lo, hi = S, 2.0
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        if r(mid) > 0.30:
            lo = mid
        else:
            hi = mid
    mg = lo
    print("A. The two merchants")
    print(
        "   modest:  markup {:.2f}, {:.2f} turns a year, return {:.2f} a year"
        .format(S, T0 * math.exp(-1), r(S))
    )
    print(
        "   greedy:  markup {:.2f}, {:.2f} turns a year, return {:.2f} a year"
        .format(mg, T0 * math.exp(-mg / S), r(mg))
    )
    print(
        "   after 10 years of reinvesting: modest x {:.1f}, greedy x {:.1f}"
        .format((1 + r(S)) ** 10, (1 + r(mg)) ** 10)
    )
    print(
        "   markups that return 0.30 a year: {:.3f} (too low) and {:.3f} (too"
        " high)".format(
            next(m / 1000 for m in range(1, 1000) if r(m / 1000) >= 0.30), mg
        )
    )

    print(
        "\nB. A subscription priced p a month, churn c(p) = 0.02 exp((p - 20)"
        " / 15)"
    )
    print("   price   monthly churn   value per signup   first-year revenue")
    print("                           p / c              per signup")
    for p in (10, 15, 20, 25, 30, 40):
        c = 0.02 * math.exp((p - 20) / 15)
        first = sum(p * (1 - c) ** t for t in range(12))
        print(
            "   {:>5}     {:6.3f}          {:8.0f}           {:8.1f}".format(
                p, c, p / c, first
            )
        )
    print("   value per signup peaks at p = 15, where"
          " churn's price sensitivity")
    print("   equals 1/p")
    print("\nReading: the 'greedy' price is not the price"
          " that earns least per sale but")
    print("the one that turns capital, or customers,"
          " too slowly. In both cases the")
    print("best price is set by how fast volume falls"
          " as price rises, not by the")
    print("margin alone.")
# A. The two merchants
#    modest:  markup 0.12, 4.17 turns a year, return 0.50 a year
#    greedy:  markup 0.29, 1.05 turns a year, return 0.30 a year
#    after 10 years of reinvesting: modest x 57.7, greedy x 13.8
#    markups that return 0.30 a year: 0.036 (too low) and 0.285 (too high)
#
# B. A subscription priced p a month, churn c(p) = 0.02 exp((p - 20) / 15)
#    price   monthly churn   value per signup   first-year revenue
#                            p / c              per signup
#       10      0.010               974              113.4
#       15      0.014              1047              166.5
#       20      0.020              1000              215.3
#       25      0.028               896              258.0
#       30      0.039               770              292.1
#       40      0.076               527              322.7
#    value per signup peaks at p = 15, where churn's price sensitivity
#    equals 1/p
#
# Reading: the 'greedy' price is not the price that earns least per sale but
# the one that turns capital, or customers, too slowly. In both cases the
# best price is set by how fast volume falls as price rises, not by the
# margin alone.

A 段把模型校准为廉贾的年收益为 0.50(「五之」)。他加价 0.12,存货一年周转 4.17 次。年收益 0.30(「三之」)的贪贾加价 0.29——每笔多赚一倍以上——存货一年却只周转 1.05 次。十年复投后,廉贾的资本增长 57.7 倍,贪贾 13.8 倍。还有一个过低的加价也只得 0.30,即 0.036:曲线有两侧,「贪」的是越过峰值的那一侧。

「无息币」与「财币欲其行如流水」是把同一算术说成了一条规则:收益等于利润率乘以流速,闲置的钱流速为零。第三章的现金低谷是它的现代形式——获客成本回收慢的企业,有钱闲置在它的客户身上。

B 段把这一算术用于流失率随价格上升的订阅产品。每位注册用户的价值,即价格除以流失率,在价格 15 处达到峰值(每位 1047);价格 40 时为 527。但每位注册用户的首年收入随价格一直上升,到 40 时为 322.7。一家只求首年收入最大的企业,选的就是贪贾的价格。

假说 · 尚未验证

作者关于文本的假说(尚未验证): 《史记》此句的「三」与「五」反映的是周转速度的差别,而不只是加价的差别;早期注释凡解释此句者,会把廉贾的优势归于买卖得时、薄利多销,而不是归于更高的价格。什么结果会证明它错:注释把差别解释为诚信、风险或社会地位的问题,完全不涉及贸易的速度。

范蠡的两次分散

范蠡三致千金,再分散之。与第二章的天花板并读,这个故事有一层量化的意味:以高收益复投的商人,只有在还有增长空间时才能无限复利;分散钱财的人,是选择不把资本硬推进一个已趋饱和的市场。这同样只是类比,原文给出的理由是德行,而不是算术。

小结

《史记》记载贪贾三之、廉贾五之,并说财币欲其行如流水、无息币。若读作利润率与周转率之间的权衡——这是诠释,不是文本明说的意思——一个简单模型就能重现「三」与「五」:贪贾的加价是廉贾的两倍多,存货周转却只有其四分之一;十年复利把差距拉大到 57.7 倍对 13.8 倍。同样的算术把订阅产品的最佳价格定在流失率的价格敏感度与利润率相平衡之处,低于首年收入最大的价格。

参考文献

Sima Qian. 1959. Shiji [Records of the Grand Historian]. In Chinese; Punctuated Edition. Zhonghua Book Company.
Sima Qian. 1993. Records of the Grand Historian: Han Dynasty II. Revised. Translated by Burton Watson. Columbia University Press.

Fan, Y. (2026). Subscription Unit Economics, Net Revenue Retention and AI Inference Scaling Laws: Cohort Survival Mixtures, Negative-Churn Differential Equations, Token Compute Margins and Metcalfe Congestion Limits [订阅制单位经济学、净收入留存与人工智能推理标度律] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090164 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).