5D Library › Slow-Crash Liquidity Reserves, Real-Asset Option Valuation and Intergenerational Wealth Preservation · 中文
Chapter 6. The Family Fund: Merton’s Rule and a Dynamic Cap on Withdrawals
The spending rate from Merton’s rule
Merton solved in continuous time for the best portfolio and spending rate of an investor with constant relative risk aversion (Merton 1969). p11 computes the results for different degrees of risk aversion with illustrative market parameters.
Calculation
# Chapter 6. p11: Merton's rule for a family fund. With constant relative
# risk aversion g, market excess return mu - r = 0.04, sd 0.18, cash
# r = 0.02 and impatience rho = 0.03, the best share in the market is
# pi = (mu - r) / (g sd^2) and the best spending rate (of current
# wealth, infinite horizon) is
# c = (rho - (1 - g) (r + (mu - r)^2 / (2 g sd^2))) / g.
# Wealth then grows at the time-average rate
# r + pi (mu - r) - pi^2 sd^2 / 2 - c.
MU_R, SD, R, RHO = 0.04, 0.18, 0.02, 0.03
def merton(g):
pi = MU_R / (g * SD ** 2)
c = (RHO - (1 - g) * (R + MU_R ** 2 / (2 * g * SD ** 2))) / g
growth = R + pi * MU_R - pi ** 2 * SD ** 2 / 2 - c
return pi, c, growth
if __name__ == "__main__":
print(" risk aversion market share spending rate wealth growth")
for g in (1, 2, 3, 5, 8):
print(" {:13d} {:12.3f} {:13.4f} {:13.4f}".format(
g, *merton(g)))
# risk aversion market share spending rate wealth growth
# 1 1.235 0.0300 0.0147
# 2 0.617 0.0312 0.0073
# 3 0.412 0.0288 0.0049
# 5 0.247 0.0260 0.0029
# 8 0.154 0.0240 0.0018With a risk aversion of 2, the best market share is 0.617 and the best spending rate about 0.0312 a year, leaving the wealth itself a time-average growth of only 0.0073. Higher risk aversion means a lower market share and a lower spending rate, but also lower growth of wealth: under these parameters, no combination lets a family take 4% a year and still see its wealth grow. Merton’s rule spends a share of current wealth, not a fixed amount.
A family fund over a hundred years
The popular “4% rule” comes from historical back-tests of thirty years of withdrawals by American retirees (Bengen 1994); it was not designed for a family fund meant to last a hundred years with a growing number of beneficiaries. p12 holds the fund 60% in the market and 40% in cash, gives the market’s real yearly return the crash years of Chapter 5, lets the beneficiaries grow 1.5% a year, and compares five withdrawal rules.
Calculation
# Chapter 6. p12: A family fund over 100 years. The fund starts at 100
# (real terms) and holds 60% in the market and 40% in cash at a real
# rate of 0. The market's real yearly return has mean 4% and sd 18%,
# with crashes as in p10 (chance 0.03 a year of a crash year, mean -42%,
# sd 10%; otherwise mean 5.42%, sd 16.23%). The number of heirs grows
# 1.5% a year. Five payout rules: a fixed 4 a year; a fixed 5 a year;
# 4% of the fund's current value; 3.5% or 1.5% of its average value
# over the last three years. A fund below 10 counts as spent. Per-heir
# payouts are shown relative to year 1. 5,000 paths, seed 88.
import random
def market(rng):
if rng.random() < 0.03:
return rng.gauss(-0.42, 0.10)
return rng.gauss(0.0542, 0.1623)
def run(rng, rule, years=100):
v, hist, heirs = 100.0, [100.0] * 3, 1.0
first, at50, spent = None, None, False
for t in range(1, years + 1):
v *= 1 + 0.6 * market(rng)
if rule == "fixed 4":
pay = 4.0
elif rule == "fixed 5":
pay = 5.0
elif rule == "4% of value":
pay = 0.04 * v
elif rule == "3.5% of 3-yr mean":
pay = 0.035 * sum(hist) / 3
else:
pay = 0.015 * sum(hist) / 3
pay = min(pay, max(v, 0.0))
v -= pay
hist = hist[1:] + [v]
heirs *= 1.015
if first is None:
first = pay
if t == 50:
at50 = pay / heirs / first
if v < 10:
spent = True
return spent, v, pay / heirs / first, at50
if __name__ == "__main__":
rng = random.Random(88)
q = lambda v, a: sorted(v)[int(a * (len(v) - 1))]
print(" rule spent median fund per heir,"
" per heir,")
print(" by 100 at year 100 year 100 year 50,")
print(" (median) 10th pct")
for rule in ("fixed 4", "fixed 5", "4% of value", "3.5% of 3-yr mean",
"1.5% of 3-yr mean"):
res = [run(rng, rule) for _ in range(5000)]
spent = sum(1 for r in res if r[0]) / 5000
print(" {:<17} {:6.3f} {:11.0f} {:9.3f} {:9.3f}".format(
rule, spent, q([r[1] for r in res], 0.5),
q([r[2] for r in res], 0.5), q([r[3] for r in res], 0.1)))
# rule spent median fund per heir, per heir,
# by 100 at year 100 year 100 year 50,
# (median) 10th pct
# fixed 4 0.926 0 0.000 0.000
# fixed 5 0.978 0 0.000 0.000
# 4% of value 0.554 10 0.024 0.058
# 3.5% of 3-yr mean 0.405 16 0.039 0.069
# 1.5% of 3-yr mean 0.024 135 0.308 0.203Taking a fixed 4 a year, the chance of running the fund out within a hundred years is 0.926; a fixed 5, 0.978. Taking 4% of current value lowers the chance to 0.554; 3.5% of the average over the past three years, 0.405. Only when the rate is cut to 1.5% of the three-year average, close to the portfolio’s time-average growth, does the chance fall to 0.024, with a median fund of 135 in year 100. Even so, because beneficiaries multiply, each receives in year 100 only 0.308 of what each received in year 1.
Note
A dynamic cap on withdrawals: each year the family fund takes no more than its three-year average value times an estimate of the portfolio’s time-average growth; the estimate is revised every ten years against actual growth; distributions are made per household rather than per person, or the rules state plainly that each person’s share will fall as beneficiaries multiply.
Summary
Merton’s rule and ergodicity reach the same conclusion: a fund meant to last a hundred years must spend close to the portfolio’s time-average growth, far below 4% under the illustrative settings. Fixed-amount withdrawals almost surely exhaust the fund; spending a low share of a smoothed value lets the fund last, while the growing number of beneficiaries still dilutes each person’s share.
References
Fan, Y. (2026). Slow-Crash Liquidity Reserves, Real-Asset Option Valuation and Intergenerational Wealth Preservation: Thirty-Month Drawdown Kinetics, Distressed Resource Acquisition and Ergodic Dynasty Trusts [慢崩盘流动性储备、实物资产期权估值与跨代财富保全动力学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090231 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).