关于「单笔凯利上限:偏斜收益、估计误差与分数凯利」,《低波动窗口的看跌期权》给出的定量结论是什么?

结论:第六章用凯利准则给单笔仓位定上限,基准世界的凯利比例只有 0.0291,估计误差逼人取四分之一。

数据口径:书中的「做空」只指损失有上限的看跌敞口,即买入看跌期权与看跌借方价差;全部价格以合成指数的指数点表示,不使用任何历史行情。 书中参数为示意;全部计算由书中程序生成。

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: 0.0291

p(fall)mean RP(R<-0.9)max Rf*g(f*)g(f*/2)g(f*/4)
0.005-0.2300.559524.30.0000+0.00000+0.00000+0.00000
0.015+0.1530.469732.10.0291+0.00203+0.00157+0.00094
0.030+0.7360.357729.80.1645+0.04477+0.03596+0.02289
查看完整验算程序
# Chapter 6. p61: Kelly fraction of capital spent on one put premium.
# Outcome R = day-30 mark / premium - 1 (R >= -1) of a 45-day put
# struck at 3,800, bought in a quiet rally, in the world of p22-p23,
# for three daily rally-to-fall chances (0.005, 0.015, 0.030) and a
# 0.02 variance-risk premium; 10,000 synthetic paths each. Kelly (1956)
# growth g(f) = mean ln(1 + f R); f* solves mean R / (1 + f R) = 0 by
# bisection on [0, 0.99]; f* = 0 when the mean of R is not positive.
import math
import random

S0, R_, DT = 4000.0, 0.03, 1.0 / 365.0
KAP, XI, RHO, SKEW, VRP = 5.0, 0.55, -0.7, 0.40, 0.02
MU = (0.08, 0.15, -0.90)
TH = (0.020, 0.030, 0.090)
N, HOLD, TENOR, K, V0 = 10000, 30, 45, 3800.0, 0.0121


def ncdf(x):
    return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))


def put_price(s, k, t, r, sig):
    sq = sig * math.sqrt(t)
    d1 = (math.log(s / k) + (r + 0.5 * sig * sig) * t) / sq
    return k * math.exp(-r * t) * ncdf(-(d1 - sq)) - s * ncdf(-d1)


def implied(s, k, v, t, reg):
    a = KAP * t
    m = TH[reg] + (v - TH[reg]) * (1.0 - math.exp(-a)) / a
    return max(math.sqrt(max(m, 1e-8)) + VRP - SKEW * math.log(k / s),
               0.05)


def outcomes(rng, pf):
    trans = [[0.990, 0.006, 0.004],
             [0.025 - pf, 0.975, pf],
             [0.010, 0.030, 0.960]]
    t0, t1 = TENOR / 365.0, (TENOR - HOLD) / 365.0
    prem = put_price(S0, K, t0, R_, implied(S0, K, V0, t0, 1))
    c = math.sqrt(1.0 - RHO * RHO)
    out = []
    for _ in range(N):
        s, v, reg = S0, V0, 1
        for _ in range(HOLD):
            u, row = rng.random(), trans[reg]
            reg = 0 if u < row[0] else (1 if u < row[0] + row[1] else 2)
            vp = max(v, 0.0)
            z1, z2 = rng.gauss(0.0, 1.0), rng.gauss(0.0, 1.0)
            s *= math.exp((MU[reg] - 0.5 * vp) * DT
                          + math.sqrt(vp * DT) * z1)
            v += (KAP * (TH[reg] - vp) * DT
                  + XI * math.sqrt(vp * DT) * (RHO * z1 + c * z2))
        mark = put_price(s, K, t1, R_, implied(s, K, max(v, 0.0), t1, reg))
        out.append(mark / prem - 1.0)
    return out


def growth(rs, f):
    return math.fsum(math.log(1.0 + f * r) for r in rs) / len(rs)


def kelly(rs):
    if math.fsum(rs) <= 0.0:
        return 0.0
    lo, hi = 0.0, 0.99
    for _ in range(60):
        mid = 0.5 * (lo + hi)
        d = math.fsum(r / (1.0 + mid * r) for r in rs)
        lo, hi = (mid, hi) if d > 0.0 else (lo, mid)
    return 0.5 * (lo + hi)


rng = random.Random(42561)
print("p(fall)  mean R  P(R<-0.9)  max R   f*      g(f*)    g(f*/2)"
      "   g(f*/4)")
for pf in (0.005, 0.015, 0.030):
    rs = outcomes(rng, pf)
    f = kelly(rs)
    tail = len([r for r in rs if r < -0.9]) / N
    print("{:.3f}   {:+.3f}   {:.4f}    {:>5.1f}  {:.4f}  {:+.5f}  {:+.5f}"
          "  {:+.5f}".format(pf, math.fsum(rs) / N, tail, max(rs), f,
                             growth(rs, f), growth(rs, f / 2.0),
                             growth(rs, f / 4.0)))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《宏观下行期低隐含波动率做空窗口、双希腊值凸性共振与三十日胜率边界母典》第 6 章「单笔凯利上限:偏斜收益、估计误差与分数凯利」(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-425 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026425,
  author    = {Fan, Yafu},
  title     = {Shorting Low-Implied-Volatility Windows in Macro Downturns: Delta-Vega Dual Convexity, Bear-Rally Compression and Thirty-Day Hit-Rate Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论