5D Library › 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 · English
第八章 实践:现金跑道、历史压力测试与样本太少的问题
现金跑道
一家企业在三十个月的营收下滑中需要多少现金?p14 设定每月成本 1.00,营收从 1.05 逐步降到 0.65,第 7 个月起把成本削减到 0.90。
推导
# Chapter 8. p14: A cash runway through a 30-month slump. A firm's monthly
# costs are 1.00 (say RM 100,000); revenue starts at 1.05 and slides
# evenly to 0.65 by month 30, then recovers 0.01 a month, with noise
# (sd 0.10 a month). From month 7 the firm cuts costs to 0.90. Starting
# cash is 3, 4, 6 or 12 months of costs. Reported over 2,000 paths
# (seed 88): the chance of running out of cash by month 36, and the
# median cash left in month 27, the middle of the forced-sale window,
# for buying from others.
import random
def path(rng, cash):
for m in range(1, 37):
rev = 1.05 - 0.40 * min(m, 30) / 30 + 0.01 * max(0, m - 30)
cost = 1.00 if m <= 6 else 0.90
cash += rev * (1 + rng.gauss(0, 0.10)) - cost
if m == 27:
at27 = cash
if cash < 0:
return True, 0.0 if m <= 27 else at27
return False, at27
if __name__ == "__main__":
rng = random.Random(88)
print(" starting cash out of cash cash in month 27")
print(" (months) by month 36 (median)")
for c in (3, 4, 6, 12):
res = [path(rng, float(c)) for _ in range(2000)]
out = sum(1 for r in res if r[0]) / 2000
mid = sorted(r[1] for r in res)[1000]
print(" {:13d} {:11.3f} {:16.2f}".format(c, out, mid))
# starting cash out of cash cash in month 27
# (months) by month 36 (median)
# 3 0.873 1.40
# 4 0.198 2.43
# 6 0.000 4.40
# 12 0.000 10.41起始现金只有三个月成本时,0.873 的路径在第 36 个月前耗尽现金;四个月时是 0.198;六个月以上则在这些路径中都撑得过去。但能撑过去不等于有余力:六个月储备在第 27 个月,也就是被迫卖出最集中的时候,中位数只剩 4.40 个月的成本;十二个月储备还剩 10.41。要在困境中买入,储备必须大于撑过下跌所需的部分。
用历史下跌做压力测试
p15 把第一章的四次下跌重演为平滑的路径,套在一个六成指数、四成现金、每月再平衡的家族基金上,比较固定每年提取 4 与按过去 36 个月平均值的 1.5% 提取两种规则。
推导
# Chapter 8. p15: Stress tests on the declines of p01. Each decline is
# replayed as a smooth path, falling by the same total over the same
# number of months (the real paths had rallies, as in p02). A family
# fund of 100 holds 60% in the index and 40% in cash at 0%, rebalanced
# monthly. Two payout rules, paid monthly: a fixed 4 a year, or 1.5% a
# year of the fund's average over the last 36 months. Reported at the
# trough: the fund, and the payout relative to the first month's.
import importlib.util
def load(name):
spec = importlib.util.spec_from_file_location(name, name + ".py")
m = importlib.util.module_from_spec(spec)
spec.loader.exec_module(m)
return m
p01 = load("p01_history")
def replay(fall, months, rule):
step = (1 + fall) ** (1 / months)
v, hist, first = 100.0, [100.0] * 36, None
for _ in range(months):
v = 0.6 * v * step + 0.4 * v
pay = 4.0 / 12 if rule == "fixed" else 0.015 / 12 * sum(hist) / 36
v -= pay
hist = hist[1:] + [v]
first = first or pay
return v, pay / first
if __name__ == "__main__":
print(" episode months fund at trough"
" payout at trough")
print(" fixed 1.5% fixed 1.5%")
for name, d0, p0, d1, p1 in p01.EPISODES:
n = round(p01.months(d0, d1))
a, pa = replay(p1 / p0 - 1, n, "fixed")
b, pb = replay(p1 / p0 - 1, n, "rule")
print(" {:<17} {:6d} {:5.1f} {:5.1f} {:5.2f} {:5.2f}"
.format(name, n, a, b, pa, pb))
# episode months fund at trough payout at trough
# fixed 1.5% fixed 1.5%
# Dow 1929-32 34 20.4 24.9 1.00 0.58
# Nasdaq 2000-02 31 33.8 38.5 1.00 0.70
# S&P 500 2000-02 31 58.2 63.8 1.00 0.84
# S&P 500 2007-09 17 56.2 59.1 1.00 0.90在 1929 至 1932 年的路径上,基金在低点只剩 20.4(固定提取)或 24.9(按比例提取):每月再平衡意味着用现金不断买入下跌中的指数,在跌幅近九成的下跌里,现金也被拖下去了。按比例提取的规则在低点把提取额自动降到原来的 0.58,固定提取则维持 1.00。2000 至 2002 年的标普 500 路径上,基金低点是 58.2 与 63.8。机械的再平衡在慢崩盘里本身就是一种「越跌越买」,应当与第四章的投入纪律一起设计。
四次下跌能教我们多少
推导
# Chapter 8. p16: How much can four declines teach? From the four
# peak-to-trough durations of p01, a 90% interval for the typical
# duration is built on the log scale with Student's t (one-sided 95%
# points for 1, 3 and 7 degrees of freedom: 6.314, 2.353, 1.895). Then
# the same spread is assumed with 2, 4 or 8 episodes, to show how slowly
# the interval narrows.
import importlib.util
from math import exp, log, sqrt
def load(name):
spec = importlib.util.spec_from_file_location(name, name + ".py")
m = importlib.util.module_from_spec(spec)
spec.loader.exec_module(m)
return m
p01 = load("p01_history")
T95 = {2: 6.314, 4: 2.353, 8: 1.895}
if __name__ == "__main__":
d = [p01.months(a, c) for _, a, _, c, _ in p01.EPISODES]
lg = [log(x) for x in d]
m = sum(lg) / len(lg)
s = sqrt(sum((x - m) ** 2 for x in lg) / (len(lg) - 1))
print(" durations (months): " + ", ".join(
"{:.1f}".format(x) for x in d))
print(" typical duration {:.1f} months, log sd {:.3f}".format(
exp(m), s))
print(" episodes 90% interval for the typical duration (months)")
for n in (2, 4, 8):
h = T95[n] * s / sqrt(n)
print(" {:8d} {:.1f} to {:.1f}".format(n, exp(m - h), exp(m + h)))
# durations (months): 34.1, 31.0, 30.5, 17.0
# typical duration 27.2 months, log sd 0.318
# episodes 90% interval for the typical duration (months)
# 2 6.6 to 112.6
# 4 18.7 to 39.6
# 8 22.0 to 33.7四次下跌的持续月数是 34.1、31.0、30.5 与 17.0,在对数尺度上的典型值是 27.2 个月。只凭这四次,典型持续时间的 90% 区间是 18.7 至 39.6 个月;若只有两次,是 6.6 至 112.6 个月;即使有八次同样分散的样本,区间仍有 22.0 至 33.7 个月。「三十个月」不能当作定律使用,第一章的模型也只是其中一种机制的示意。
利益披露
投资声明: 本书讨论的是流动性、实物期权与长期财富保全的数学,不是投资建议。书中的模型参数是示意,历史指数数据只用于说明下跌的形态;任何投资、借贷或家族信托的安排,请咨询持牌的专业顾问。
注
预先登记的检验: 第二章关于被迫卖出高峰时间的假说,以及第三章关于困境协议收购回报的假说,应在收集数据前登记数据来源、事件定义与分析方法,结果不论正负都发表。
小结
现金储备要分成两部分:撑过下跌所需的部分,以及留给第 24 至 30 个月困境买入的部分。机械再平衡在慢崩盘中会把现金也拖下去;按比例的提取规则会自动降低提取额。几次历史下跌不足以确定「三十个月」这样的数字,所有计划都应在下跌更短或更长的情况下也能成立。
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).