关于「分位数对冲:预算有限时,先保住尾巴的哪一段」,《尾部保护与凯利边界》给出的定量结论是什么?

结论:第四章说明以成功概率为目标的分位数对冲会放弃深尾,同一预算下资本低于 0.6 的概率反而从 0.00415 升到 0.00480。

数据口径:用默顿跳扩散模型描述营收指数,并让保费报价对跳跃风险收取示意的溢价,行权 0.80 的固定保费保护,保费是平均赔付的 1.669 倍,买方平均收益为 -0.401;越只保极端情形,保费相对平均赔付越贵。 书中参数为示意;全部计算由书中程序生成;本书是非个人化法律、税务或理财建议,不推荐任何具体金融产品、保险产品、机构或交易策略,不承诺收益,不鼓励杠杆。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.00415, 0.00480, 0.6

budget 0.00470 = 40 % of full cover 0.01176
planP(M<TAU)E shortfallES 0.975P(M'<0.6)E log M'
no cover0.05950.006320.33720.004150.14024
quantile0.01430.003310.28690.004800.13985
linear0.01430.003310.28690.004800.13985
proportional0.03570.003790.26700.000220.13970
log growth0.01700.003390.28770.002680.13994
frontierquantile: P(M<TAU)ES 0.975 | log: P(M<TAU)ES 0.975
0.20.02520.3373|0.02750.3256
0.40.01430.2869|0.01700.2877
0.60.00730.2377|0.00880.2409
0.80.00250.1932|0.00290.1951
1.00.00000.1618|0.00000.1618
查看完整验算程序
# Chapter 4. p42: one budget, four objectives, different tails.
# Same grid, operator and budget as p41. Covers c (0 <= c <= D) bought
# at exp(-R) * q per unit in each bin; the premium B is paid by every
# strategy, so M' = M - B + c. Strategies: quantile hedge (max success
# probability), linear shortfall (max p / q first), proportional cover
# (a plain protection struck where M = TAU), and log growth (maximise
# E log M' by water-filling). Then a budget frontier. Illustrative.
import math

MU, SIG = 0.05, 0.18
LAM, JM, JS = 0.10, -0.35, 0.15
R, LAMQ = 0.03, 0.20
A, L, TAU = 0.12, 1.0, 0.85
DF = math.exp(-R)


def Phi(x):
    return 0.5 * math.erfc(-x / math.sqrt(2.0))


def comps(mu, lam):
    kap = math.exp(JM + 0.5 * JS * JS) - 1.0
    base = mu - 0.5 * SIG * SIG - lam * kap
    return [(math.exp(-lam) * lam ** n / math.factorial(n), base + n * JM,
             math.sqrt(SIG * SIG + n * JS * JS)) for n in range(13)]


def cdf(cs, x):
    return math.fsum(w * Phi((x - m) / s) for w, m, s in cs)


CP, CQ = comps(MU, LAM), comps(R, LAMQ)
E = [-4.0 + 0.02 * i for i in range(261)]
FP = [0.0] + [cdf(CP, e) for e in E[1:260]] + [1.0]
FQ = [0.0] + [cdf(CQ, e) for e in E[1:260]] + [1.0]
S = [math.exp(0.5 * (E[i] + E[i + 1])) for i in range(260)]
P = [FP[i + 1] - FP[i] for i in range(260)]
Q = [FQ[i + 1] - FQ[i] for i in range(260)]
M = [1.0 + A + L * (s - 1.0) for s in S]
D = [max(TAU - m, 0.0) for m in M]
FULL = DF * math.fsum(q * d for q, d in zip(Q, D))


def greedy(key, budget):
    c, left = [0.0] * 260, budget
    for i in sorted((i for i in range(260) if D[i] > 0), key=key):
        cost = DF * Q[i] * D[i]
        f = min(1.0, left / cost)
        c[i] = f * D[i]
        left = max(0.0, left - f * cost)
        if left <= 0.0:
            break
    return c


def log_fill(budget):
    def cov(nu):
        return [min(D[i], max(0.0, P[i] / (nu * DF * Q[i]) - (M[i] - budget)))
                if Q[i] > 0 else 0.0 for i in range(260)]
    lo, hi = -20.0, 20.0
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        cost = DF * math.fsum(q * x for q, x in zip(Q, cov(math.exp(mid))))
        lo, hi = (mid, hi) if cost > budget else (lo, mid)
    return cov(math.exp(hi))


def report(c, budget):
    mp = [m - budget + x for m, x in zip(M, c)]
    fail = math.fsum(P[i] * (1.0 - c[i] / D[i]) for i in range(260)
                     if D[i] > 0)
    short = math.fsum(p * (d - x) for p, d, x in zip(P, D, c))
    tail, need, acc = sorted(zip(mp, P)), 0.025, []
    for m, p in tail:
        use = min(p, need)
        acc.append(use * (1.0 - m))
        need -= use                  # step
        if need <= 0.0:
            break
    deep = math.fsum(p for m, p in zip(mp, P) if m < 0.6)
    g = math.fsum(p * math.log(m) for m, p in zip(mp, P))
    return fail, short, math.fsum(acc) / 0.025, deep, g


B = 0.40 * FULL
plans = [("no cover", [0.0] * 260, 0.0),
         ("quantile", greedy(lambda i: -P[i] / (Q[i] * D[i]), B), B),
         ("linear", greedy(lambda i: -P[i] / Q[i], B), B),
         ("proportional", [0.40 * d for d in D], B),
         ("log growth", log_fill(B), B)]
print("budget {:.5f} = 40 % of full cover {:.5f}".format(B, FULL))
print("plan          P(M<TAU)  E shortfall  ES 0.975  P(M'<0.6)  E log M'")
for name, c, b in plans:
    f, sh, es, dp, g = report(c, b)
    print("{:12s}  {:.4f}    {:.5f}     {:.4f}    {:.5f}    {:.5f}".format(
        name, f, sh, es, dp, g))
print("frontier   quantile: P(M<TAU)  ES 0.975 | log: P(M<TAU)  ES 0.975")
for share in (0.2, 0.4, 0.6, 0.8, 1.0):
    b = share * FULL
    f1, _, e1, _, _ = report(greedy(lambda i: -P[i] / (Q[i] * D[i]), b), b)
    f2, _, e2, _, _ = report(log_fill(b), b)
    print("  {:.1f}               {:.4f}     {:.4f}  |      {:.4f}     {:.4f}"
          .format(share, f1, e1, f2, e2))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《尾部黑天鹅期权护城河与分位数对冲、跳扩散模型与凯利边界下的固定保费现金流保护母典》第 4 章「分位数对冲:预算有限时,先保住尾巴的哪一段」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-290 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026290,
  author    = {Fan, Yafu},
  title     = {Tail-Risk Option Moats and Quantile Hedging: Jump-Diffusion Models and Fixed-Premium Protection of Core Cash Flows under Kelly Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论