5D Library › Shareholder Free-Riding Games, Dynamic Vesting Mechanisms and Defection Early-Warning Models · 中文
Chapter 1. Free Riding When Effort Cannot Be Verified
The team problem
When several people produce something together and no one can see how much each contributed, each is tempted to let the others work. Alchian and Demsetz saw this as the reason firms have owners who monitor (Alchian and Demsetz 1972); Holmström proved that in a team whose individual efforts cannot be verified, no rule that shares out exactly the output can give everyone the right incentive (Holmström 1982). A start-up with co-founders on fixed, equal shares is such a team.
p01 writes the simplest version. Each founder chooses effort at cost ; the firm’s value is . A founder who owns share gains from a unit of effort and so chooses , while the effort that maximises the value created is for everyone.
Calculation
# Chapter 1. p01: Free riding when effort cannot be verified. N founders
# each choose effort e_i at cost c e_i^2 / 2; the firm's value is
# theta * (e_1 + ... + e_N). Founder i owns share s_i and so chooses
# e_i = s_i theta / c (Nash), while the value-maximising effort is
# theta / c for everyone. Efficiency = value created net of effort cost,
# relative to the first best. theta = c = 1. Equal splits for N = 1 to
# 5, and some unequal splits for N = 2.
def efficiency(shares):
es = [s for s in shares] # e_i = s_i theta / c
w = sum(es) - sum(e * e for e in es) / 2
fb = len(shares) - len(shares) / 2
return w / fb
if __name__ == "__main__":
print(" founders N equal shares efficiency (2N - 1) / N^2")
for n in range(1, 6):
print(" {:10d} {:12.3f} {:10.3f} {:14.3f}".format(
n, 1 / n, efficiency([1 / n] * n), (2 * n - 1) / n ** 2))
print(" two founders, unequal splits")
for a in (0.5, 0.6, 0.7, 0.9):
print(" {:.1f} / {:.1f} efficiency {:.3f}".format(
a, 1 - a, efficiency([a, 1 - a])))
# founders N equal shares efficiency (2N - 1) / N^2
# 1 1.000 1.000 1.000
# 2 0.500 0.750 0.750
# 3 0.333 0.556 0.556
# 4 0.250 0.438 0.438
# 5 0.200 0.360 0.360
# two founders, unequal splits
# 0.5 / 0.5 efficiency 0.750
# 0.6 / 0.4 efficiency 0.740
# 0.7 / 0.3 efficiency 0.710
# 0.9 / 0.1 efficiency 0.590With equal shares the efficiency is : 0.750 for two founders, 0.556 for three, 0.438 for four and 0.360 for five. Unequal splits do not help when the founders are alike: a 60/40 split gives 0.740, 70/30 gives 0.710 and 90/10 only 0.590, because the minority founder’s effort falls by more than the majority founder’s rises. The loss is not a moral failing of either founder; it is what each does when effort is invisible and the reward is shared.
Holmström’s group target
Holmström’s way out is a rule that pays the team its shares only if output reaches the first-best level, with someone outside the team, typically an investor, keeping the output when the target is missed. In a world without noise, full effort is then an equilibrium: anyone who slackens destroys their own share. p02 adds noise to output.
Calculation
# Chapter 1. p02: A group target with noisy output. Holmstrom showed
# that a rule paying the team its shares only if output reaches a
# target can restore first-best effort, but someone outside the team
# (an investor, the "budget breaker") must keep the output when the
# target is missed. Two founders, theta = c = 1: output
# Y = e_1 + e_2 + noise (sd sigma). Each founder gets Y / 2 if Y >= 2
# (the first-best output), else nothing. Symmetric Nash efforts are the
# fixed points of the best response on a grid (step 0.0025, within one
# step). Reported:
# the highest one, the chance the target is missed although both make
# that effort, and the share of expected output the founders forfeit.
from math import erf, exp, pi, sqrt
def Phi(x):
return 0.5 * (1 + erf(x / sqrt(2)))
def phi(x):
return exp(-x * x / 2) / sqrt(2 * pi)
def paid(mu, sigma, t=2.0):
# E[Y 1{Y >= t}] for Y ~ normal(mu, sigma)
if sigma == 0:
return mu if mu >= t - 1e-9 else 0.0
z = (t - mu) / sigma
return mu * (1 - Phi(z)) + sigma * phi(z)
GRID = [i / 400 for i in range(0, 601)]
def best(other, sigma):
return max(GRID, key=lambda e: 0.5 * paid(e + other, sigma) - e * e / 2)
def equilibria(sigma):
return [e for e in GRID if abs(best(e, sigma) - e) <= 0.0025]
if __name__ == "__main__":
print(" noise sd effort target missed output forfeited"
" efficiency")
for sigma in (0.0, 0.1, 0.3, 0.6):
e = max(equilibria(sigma))
mu = 2 * e
miss = (0.0 if mu >= 2 else 1.0) if sigma == 0 else \
1 - Phi((mu - 2) / sigma)
lost = 1 - paid(mu, sigma) / mu
print(" {:8.1f} {:6.3f} {:13.3f} {:16.3f} {:10.3f}".format(
sigma, e, miss, lost, 2 * e - e * e))
print(" equal shares, no target: effort 0.500, efficiency 0.750")
print(" zero effort is also an equilibrium at sd 0.0: {}".format(
0.0 in equilibria(0.0)))
# noise sd effort target missed output forfeited efficiency
# 0.0 1.000 0.000 0.000 1.000
# 0.1 1.097 0.026 0.023 0.990
# 0.3 1.165 0.136 0.108 0.973
# 0.6 0.005 1.000 0.902 0.010
# equal shares, no target: effort 0.500, efficiency 0.750
# zero effort is also an equilibrium at sd 0.0: TrueWithout noise, the rule restores full effort (efficiency 1.000), though zero effort is an equilibrium too. With small noise (standard deviation 0.1) the founders work harder than the first best, 1.097 each, to make sure of the target, which is still missed 0.026 of the time; efficiency is 0.990. With noise of 0.3 they work 1.165 each, miss 0.136 of the time, forfeit 0.108 of expected output and reach 0.973. With noise of 0.6 the good equilibrium disappears: the target is so uncertain that effort cannot secure it, the founders give up, and efficiency falls to 0.010, far below the 0.750 of plain equal shares. A group penalty is a strong tool for a team that controls its results, and a dangerous one for a team that does not.
Hypothesis · not established
The author’s interpretation (not yet tested): in start-ups whose founders hold fixed, equal shares, the founders’ reported hours and output fall short of those of founders with performance-linked equity, and the gap widens with the number of founders. What would show it wrong: in a pre-registered survey of comparable start-ups, founder effort shows no relation to the share structure once firm stage and funding are held fixed, or the gap does not grow with the number of founders.
Summary
With effort that cannot be verified, equal fixed shares lose a quarter of the value for two founders and more for more founders; tilting the split does not help. A group target can restore effort if results depend mainly on effort, but under heavy noise it can destroy it.
References
Fan, Y. (2026). Shareholder Free-Riding Games, Dynamic Vesting Mechanisms and Defection Early-Warning Models: Moral-Hazard Equity Dilution, Shotgun Buy-Sell Clauses and Behavioral Divergence Signals [股东搭便车博弈、动态股权成熟与背刺早期预警模型] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090243 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).