5D Library › 跨文化契约、实物期权与东南亚出海博弈论 · English

第六章 委托-代理问题与本地团队的激励

两种时间表

总部希望这个季度的 GMV 翻一番,下个季度看到投资回报。本地合伙人或国家经理希望有一份稳定的收入,并在本地商业圈里维持长期的声誉:今天为了冲量去压分销商、透支网红关系,明年就没人愿意合作了。这不是谁对谁错,而是两种时间表、两种风险承受能力。经济学把这类问题称为委托-代理问题:委托人(总部)把任务交给代理人(本地团队)去做,但看不到代理人的全部努力,双方的目标也不完全一致 (Jensen and Meckling 1976; Holmström 1979)。

一个经典的提醒

克尔在 1975 年发表了一篇标题本身就很说明问题的文章:《论奖励 A 却期望 B 的愚蠢》(Kerr 1975)。组织常常奖励容易测量的东西,却期望得到难以测量的东西。按 GMV 发奖金,却期望本地团队维护品牌和长期客户关系,就是一个典型的例子。

霍姆斯特罗姆与米尔格罗姆把这个直觉写成了严格的模型:多任务委托-代理 (Holmström and Milgrom 1991)。当代理人的工作包含多项任务、各项任务争夺同一份时间与精力、而有些任务难以测量时,对可测量任务的强激励会把精力从不可测量的任务上吸走。他们的结论之一是:在这种情况下,最优的激励往往比单任务模型建议的弱,有时甚至应当用固定工资;只有当难以测量的任务也找到某种(哪怕是有噪声的)测量指标时,可测量任务的激励才可以加强。吉本斯对这一类结论作了通俗的综述 (Gibbons 1998)。

一个数值模型

推导

本地国家经理把精力分给两项任务:短期 GMV(每月测量,噪声方差为 1)与长期的伙伴和品牌建设(只能用 12 个月的代理指标衡量,例如合作伙伴续约率或复购顾客比例,噪声方差为 2;或者根本不测量)。两项任务对总部同样重要且收益递减;参数 ss 表示两项任务争夺时间的程度。国家经理本人也看重长期关系(她在本地商圈的声誉)。她是风险厌恶的,风险厌恶系数为 rr。总部选择两项激励系数,使总剩余最大。

# Holmstrom & Milgrom (1991) multitask agency, numerically. A local country manager splits effort
# between t1 = short-term GMV (measured monthly, noise variance 1) and t2 = long-term partner and
# brand building, measured only by a noisy 12-month proxy such as partner retention or the share of
# repeat customers (noise variance 2), or not measured at all.
#   Value to headquarters:  2*sqrt(t1) + 2*sqrt(t2)   (both tasks matter, with diminishing returns)
#   Effort cost:            (t1^2 + t2^2)/2 + s*t1*t2  (s in [0,1): how much the tasks compete for time)
#   The manager also values t2 for its own sake (her standing with local partners): +0.2 per unit.
# Pay = fixed + b1*GMV + b2*proxy. CARA manager with risk aversion r; headquarters is risk-neutral and
# picks (b1, b2) to maximise total surplus, value - cost + 0.2*t2 - r/2*(b1^2*1 + b2^2*2);
# the fixed part then makes the manager just willing to sign.
import math

INTR = 0.2

def efforts(b1, b2, s):
    """Manager's best response: grad C = (b1, b2 + INTR), efforts non-negative."""
    g1, g2 = b1, b2 + INTR
    t1 = (g1 - s * g2) / (1 - s * s); t2 = (g2 - s * g1) / (1 - s * s)
    if t2 < 0: return g1, 0.0
    if t1 < 0: return 0.0, g2
    return t1, t2

def surplus(b1, b2, s, r):
    t1, t2 = efforts(b1, b2, s)
    cost = (t1 * t1 + t2 * t2) / 2 + s * t1 * t2
    return 2 * math.sqrt(t1) + 2 * math.sqrt(t2) + INTR * t2 - cost - r / 2 * (b1 * b1 + 2 * b2 * b2), t1, t2

def best(s, r, proxy):
    grid = [i / 100 for i in range(201)]
    return max(((*surplus(b1, b2, s, r), b1, b2) for b1 in grid for b2 in (grid if proxy else [0.0])),
               key=lambda x: x[0])

print("competition s | risk aversion r | 12-month proxy | best b1 (GMV) | b2 (proxy) | efforts t1, t2 | surplus")
for s in (0.0, 0.5, 0.8):
    for r, proxy in ((1.0, False), (1.0, True), (3.0, True)):
        v, t1, t2, b1, b2 = best(s, r, proxy)
        print(f"{s:13.1f} | {r:15.1f} | {'used' if proxy else 'none':14s} | {b1:13.2f} | {b2:10.2f} |"
              f" {t1:5.2f}, {t2:5.2f}   | {v:7.2f}")
print("a GMV-only bonus when tasks compete strongly (s = 0.8, r = 1, no proxy):")
for b1 in (0.2, 0.5, 1.0, 1.5):
    v, t1, t2 = surplus(b1, 0.0, 0.8, 1.0)
    print(f"   b1 = {b1:.1f}: GMV effort {t1:.2f}, relationship effort {t2:.2f}, surplus {v:.2f}")
# competition s | risk aversion r | 12-month proxy | best b1 (GMV) | b2 (proxy) | efforts t1, t2 | surplus
#           0.0 |             1.0 | none           |          0.63 |       0.00 |  0.63,  0.20   |    2.10
#           0.0 |             1.0 | used           |          0.63 |       0.42 |  0.63,  0.62   |    2.52
#           0.0 |             3.0 | used           |          0.40 |       0.22 |  0.40,  0.42   |    2.09
#           0.5 |             1.0 | none           |          0.27 |       0.00 |  0.23,  0.09   |    1.48
#           0.5 |             1.0 | used           |          0.60 |       0.42 |  0.39,  0.43   |    2.03
#           0.5 |             3.0 | used           |          0.37 |       0.19 |  0.23,  0.27   |    1.66
#           0.8 |             1.0 | none           |          0.21 |       0.00 |  0.14,  0.09   |    1.31
#           0.8 |             1.0 | used           |          0.59 |       0.40 |  0.31,  0.36   |    1.84
#           0.8 |             3.0 | used           |          0.37 |       0.18 |  0.18,  0.23   |    1.49
# a GMV-only bonus when tasks compete strongly (s = 0.8, r = 1, no proxy):
#    b1 = 0.2: GMV effort 0.11, relationship effort 0.11, surplus 1.31
#    b1 = 0.5: GMV effort 0.50, relationship effort 0.00, surplus 1.16
#    b1 = 1.0: GMV effort 1.00, relationship effort 0.00, surplus 1.00
#    b1 = 1.5: GMV effort 1.50, relationship effort 0.00, surplus 0.20

结果与霍姆斯特罗姆–米尔格罗姆的结论一致,几处值得指出:

  • 两项任务互不争夺时间(s=0s = 0),GMV 的最优激励系数是 0.63,与有没有长期指标无关;有了长期指标,关系建设的投入从 0.20 提高到 0.62,总剩余从 2.10 提高到 2.52。
  • 两项任务激烈争夺时间、又没有长期指标时(s=0.8s = 0.8),GMV 的最优激励系数降到 0.21。程序最后几行显示了原因:GMV 奖金一旦达到 0.5,国家经理就会把关系建设的投入降到零,总剩余随之下降。
  • 找到一个长期指标,哪怕噪声很大,GMV 的激励就可以重新加强(s=0.8s = 0.8 时从 0.21 提高到 0.59),两项任务都得到投入。这就是「平衡计分」类做法的经济学依据:问题不在于要不要激励,而在于激励必须同时覆盖相互竞争的任务。
  • 风险厌恶越强(rr 从 1 提高到 3),两项激励都应降低,固定部分相应提高。这与本地合伙人「追求稳定底薪」的偏好在模型中是一致的:对风险厌恶的人来说,高比例的提成是一种昂贵的付酬方式,因为你必须为他承担的风险付钱。

注

模型的边界。 模型假定总部知道各项参数,而现实中总部往往不知道本地团队的风险偏好和两项任务的实际替代程度。模型也没有考虑「激励挤出内在动机」等心理效应;把本地团队对长期关系的重视(程序中的内在回报)当作一个固定的数,是一种简化。

让合伙人自己选:合同菜单

总部往往不知道一个本地合伙人是打算长期合作,还是一两年后就可能离开。一种办法是设计一份合同菜单,让不同类型的合伙人自己选择适合自己的合同,从而在选择中透露自己的类型。这在信息经济学中称为筛选。

推导

两类合伙人:长期型(预计合作三年以上,月折现因子 0.995,风险厌恶 1.0)与短期型(每月有 6% 的可能离开,月折现因子 0.98,风险厌恶 2.0)。两份合同:A「工资型」,固定月薪 8 加 2% 的分成;B「合伙型」,固定月薪 5 加 4% 的分成,再加一笔在第 36 个月发放、离开即失效的递延款 150。

# A contract menu that lets local partners sort themselves (screening). Two kinds of partner apply:
#   long-horizon  - expects to stay 3+ years, monthly discount factor 0.995, risk aversion 1.0
#   short-horizon - may leave within a year (leaves with probability 0.06 a month), discount 0.98,
#                   risk aversion 2.0
# Monthly revenue the partner influences: mean 100, SD 40 (illustrative units, e.g. thousand MYR).
# Menu, 36 months:
#   A 'salary'  : fixed 8 a month + 2 % of revenue
#   B 'partner' : fixed 5 a month + 4 % of revenue + a deferred tranche of 150 paid at month 36
#                 to a partner who is still there (forfeited on leaving)
# Each type's value = expected discounted pay - risk premium (CARA, r/2 * variance of monthly pay).
def value(contract, beta, leave, r):
    fixed, share, tranche = contract
    v, alive = 0.0, 1.0
    for m in range(1, 37):
        alive *= 1 - leave
        monthly = fixed + share * 100 - r / 2 * (share * 40) ** 2
        v += alive * beta ** m * monthly
    return v + alive * beta ** 36 * tranche

MENU = {"A salary": (8, 0.02, 0), "B partner": (5, 0.04, 150)}
TYPES = {"long-horizon": (0.995, 0.0, 1.0), "short-horizon": (0.98, 0.06, 2.0)}
for t, (beta, leave, r) in TYPES.items():
    vals = {c: value(k, beta, leave, r) for c, k in MENU.items()}
    pick = max(vals, key=vals.get)
    print(f"{t:13s}: " + ", ".join(f"{c} {v:6.1f}" for c, v in vals.items()) + f"  -> picks {pick}")

# Expected cost to headquarters per hire (undiscounted, with each type's chance of staying):
for c, (fixed, share, tranche) in MENU.items():
    for t, (beta, leave, r) in TYPES.items():
        stay = sum((1 - leave) ** m for m in range(1, 37))
        cost = stay * (fixed + share * 100) + (1 - leave) ** 36 * tranche
        print(f"cost of {c} for a {t} partner: {cost:6.1f} over 36 months "
              f"(still there at month 36: {(1 - leave) ** 36:.0%})")
# long-horizon : A salary  318.0, B partner  378.9  -> picks B partner
# short-horizon: A salary  103.7, B partner   79.2  -> picks A salary
# cost of A salary for a long-horizon partner:  360.0 over 36 months (still there at month 36: 100%)
# cost of A salary for a short-horizon partner:  139.8 over 36 months (still there at month 36: 11%)
# cost of B partner for a long-horizon partner:  474.0 over 36 months (still there at month 36: 100%)
# cost of B partner for a short-horizon partner:  142.0 over 36 months (still there at month 36: 11%)

两类合伙人各自选择了为他们设计的合同:长期型选择合伙型合同(价值 378.9 对 318.0),短期型选择工资型合同(103.7 对 79.2)。递延款与较高的分成对长期型更有吸引力,因为他们更可能拿到递延款,也更不在意每月收入的波动。代价是:对长期型合伙人,合伙型合同的总成本(474.0)高于工资型(360.0)。这部分额外成本是总部为「区分类型、留住长期伙伴」付出的价格;它是否划算,取决于一个长期伙伴能为企业多创造多少价值,而这在模型中没有计算。

假说 · 尚未验证

合同菜单筛选假说(尚未确立)。在招募东南亚本地合伙人或国家经理时,同时提供「较高固定收入」与「较低固定收入 + 较高分成 + 递延款」两种合同,选择后者的人在三年内的留任率与合作伙伴续约率会明显高于选择前者的人,而且这种差别不能完全由事前可观察的特征(年龄、经验、家庭状况)解释。如果两组的留任率无差别,或者差别完全由事前特征解释,这一假说就被削弱。

期权与股权的作用

许多出海企业给本地核心合伙人股权或期权。从本章的模型看,它们的作用是把长期任务的价值「写进」报酬:期权的价值取决于企业几年后的价值,而企业价值包含了品牌、渠道和关系这些难以按月测量的东西。但期权也有代价。它把企业的风险转嫁给了往往比总部更厌恶风险的本地合伙人;在许多东南亚国家,外国母公司的期权对本地员工的法律、税务和兑现安排都很复杂;而且如果企业的估值主要由总部的决策决定,本地合伙人会把期权看作「总部的彩票」,而不是对自己努力的回报。第八章的核查表因此要求把激励安排与剩余控制权矩阵放在一起审查。

本章小结

总部与本地团队的目标不一致,大部分不是因为谁的品德问题,而是因为时间表、风险承受力与可测量性不同。多任务模型说明:只奖励可测量的短期指标,会把精力从长期任务上吸走;为长期任务找到哪怕有噪声的指标,才能让短期激励重新加强。合同菜单可以让不同类型的合伙人自我选择。下一章暂时离开模型,回到东方兵学与西方盟约传统的原典。

参考文献

Gibbons, Robert. 1998. “Incentives in Organizations.” Journal of Economic Perspectives 12 (4): 115–32.
Holmström, Bengt. 1979. “Moral Hazard and Observability.” Bell Journal of Economics 10 (1): 74–91.
Holmström, Bengt, and Paul Milgrom. 1991. “Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design.” Journal of Law, Economics, and Organization 7: 24–52.
Jensen, Michael C., and William H. Meckling. 1976. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.” Journal of Financial Economics 3 (4): 305–60.
Kerr, Steven. 1975. “On the Folly of Rewarding a, While Hoping for b.” Academy of Management Journal 18 (4): 769–83.

Fan, Y. (2026). Cross-Cultural Contracts, Real Options and Game Theory of Southeast Asian Market Entry: Relational Contracting, Optimal Stopping Thresholds and Institutional Distance, Read Beside the Art of War [跨文化契约、实物期权与东南亚出海博弈论] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23753261 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).