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

第一章 慢崩盘的动力学:三十个月从哪里来

大跌用了多久

「大崩盘往往要走三十个月左右」是一个流行的说法。先看几次有名的下跌从高点到低点用了多久。

推导

# Chapter 1. p01: How long the great declines took. Index closing values
# at peak and trough (public index histories): Dow Jones Industrial
# Average 1929 to 1932, with its 1930 bear-market rally; Nasdaq
# Composite and S&P 500 2000 to 2002; S&P 500 2007 to 2009 for contrast.
from datetime import date

EPISODES = (
    ("Dow 1929-32", date(1929, 9, 3), 381.17, date(1932, 7, 8), 41.22),
    ("Nasdaq 2000-02", date(2000, 3, 10), 5048.62, date(2002, 10, 9),
     1114.11),
    ("S&P 500 2000-02", date(2000, 3, 24), 1527.46, date(2002, 10, 9),
     776.76),
    ("S&P 500 2007-09", date(2007, 10, 9), 1565.15, date(2009, 3, 9),
     676.53),
)

def months(a, b):
    return (b - a).days / 30.4375

if __name__ == "__main__":
    print("   episode            fall     months to trough")
    for name, d0, p0, d1, p1 in EPISODES:
        print("   {:<17} {:6.1%}   {:16.1f}".format(
            name, p1 / p0 - 1, months(d0, d1)))
    low, high = 198.60, 294.07
    print("   Dow rally 13 Nov 1929 to 17 Apr 1930: {:+.1%} in {:.1f} months"
          .format(high / low - 1, months(date(1929, 11, 13),
                                           date(1930, 4, 17))))
    print("   fall still to come after the rally top: {:.1%}"
          .format(41.22 / high - 1))
    print("   share of the whole fall left after the first low: {:.3f}"
          .format((198.60 - 41.22) / (381.17 - 41.22)))
#    episode            fall     months to trough
#    Dow 1929-32       -89.2%               34.1
#    Nasdaq 2000-02    -77.9%               31.0
#    S&P 500 2000-02   -49.1%               30.5
#    S&P 500 2007-09   -56.8%               17.0
#    Dow rally 13 Nov 1929 to 17 Apr 1930: +48.1% in 5.1 months
#    fall still to come after the rally top: -86.0%
#    share of the whole fall left after the first low: 0.463

道琼斯工业平均指数从 1929 年 9 月的 381.17 跌到 1932 年 7 月的 41.22,跌幅 89.2%,用了 34.1 个月;纳斯达克综合指数 2000 至 2002 年跌 77.9%,用了 31.0 个月;标普 500 同期跌 49.1%,30.5 个月。但 2007 至 2009 年的标普 500 跌 56.8% 只用了 17.0 个月。三十个月是几次著名下跌的共同特征,不是定律。

1929 年的例子还说明了另一件事:第一波暴跌在 1929 年 11 月 13 日见到 198.60 后,指数在五个月内反弹 48.1%;但从反弹的高点算起,后面还要再跌 86.0%。第一个低点之后,全部跌幅还剩 0.463 没有走完。

一个带阻尼的模型

为什么下跌会拖这么久?一种解释来自债务:债务不是一天到期,而是分散在几年里陆续到期;每一笔到期的债务都要按新的、更低的抵押品价值重新定价,于是基本面价值随着到期阶梯一步步下移 (Fisher 1933)。价格则以一定速度追随基本面(阻尼),中央银行不定期注入流动性,造成反弹。p02 把这三者写进一个月度模型。

推导

# Chapter 1. p02: A damped model of a slow crash. Log price x chases a
# log "fundamental" f with speed k = 0.25 a month (damping), plus noise
# (sd 0.06 a month). f falls as debt rolls over: maturities are spread
# evenly over 30 months, and each month's maturities reprice, so f
# falls by 0.70 / 30 a month for 30 months, then recovers 0.015 a month.
# Central-bank injections arrive at random (chance 0.12 a month) and
# push x up by 0.10 at once. 2,000 paths of 60 months, seed 88.
import random
from math import exp

K, SD, DROP, T, UP, P_INJ, JUMP = 0.25, 0.06, 0.70, 30, 0.015, 0.12, 0.10

def path(rng, months=60):
    x = f = 0.0
    out = [1.0]
    for t in range(months):
        f += -DROP / T if t < T else UP
        x += K * (f - x) + rng.gauss(0, SD)
        if rng.random() < P_INJ:
            x += JUMP
        out.append(exp(x))
    return out

def rallies(p, size=0.20):
    # a rally: a rise of `size` from the lowest point since the last one
    n, low, armed = 0, p[0], True
    for v in p[1:]:
        if v < low:
            low, armed = v, True
        elif armed and v >= low * (1 + size):
            n, low, armed = n + 1, v, False
    return n

if __name__ == "__main__":
    rng = random.Random(88)
    trough, fall, rall, r10, early = [], [], [], [], 0
    for _ in range(2000):
        p = path(rng)
        k = min(range(61), key=lambda i: p[i])
        trough.append(k)
        fall.append(p[k] - 1)
        rall.append(rallies(p[:k + 1]))
        r10.append(rallies(p[:k + 1], 0.10))
        early += k < 18
    q = lambda v, a: sorted(v)[int(a * (len(v) - 1))]
    print("   months to trough: median {}, 10% to 90% {} to {}".format(
        q(trough, 0.5), q(trough, 0.1), q(trough, 0.9)))
    print("   fall at trough: median {:.1%}, 10% to 90% {:.1%} to {:.1%}"
          .format(q(fall, 0.5), q(fall, 0.1), q(fall, 0.9)))
    for name, v in (("10%", r10), ("20%", rall)):
        print("   rallies of {} or more before the trough: median {},"
              " 90th pct {}".format(name, q(v, 0.5), q(v, 0.9)))
    print("   share of paths with the trough before month 18: {:.3f}"
          .format(early / 2000))
#    months to trough: median 34, 10% to 90% 29 to 41
#    fall at trough: median -50.7%, 10% to 90% -55.1% to -46.2%
#    rallies of 10% or more before the trough: median 3, 90th pct 5
#    rallies of 20% or more before the trough: median 1, 90th pct 2
#    share of paths with the trough before month 18: 0.000

在示意参数下,基本面在三十个月里下移约一半,价格的低点出现在第 34 个月(中位数),十分之八的路径落在第 29 至第 41 个月之间,没有一条路径在第 18 个月之前见底。见底前通常有三次 10% 以上的反弹,有一次 20% 以上。慢崩盘的长度在这个模型里由债务到期的阶梯决定,而反弹由注资决定:两者都不是价格本身的「规律」。

p02 的四条示意路径、四百条路径的中位数与基本面(p02)。

假说 · 尚未验证

作者的诠释(尚未验证): 由债务驱动的大型下跌,其持续时间主要取决于债务到期结构的长度,而不是估值偏离的程度。什么会证明它错: 在跨国的历史下跌样本中,下跌持续月数与下跌前企业或家庭债务的加权平均剩余期限没有正相关,或与下跌前估值偏离的相关性更强。

小结

几次著名的大跌从高点到低点都用了约三十个月,但也有十七个月的反例;1929 年的第一个低点之后还剩近一半的跌幅。把债务到期阶梯、阻尼与注资写进模型,可以得到持续约三十个月、中间夹着几次反弹的下跌。这是一个机制的示意,不是预测工具。

参考文献

Fisher, Irving. 1933. “The Debt-Deflation Theory of Great Depressions.” Econometrica 1 (4): 337 to 357.

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).