关于「最优平仓边界不是一个固定比例:格点上的斯奈尔包络」,《期权提前止盈与最优停时》给出的定量结论是什么?

结论:第四章在三叉格点上解出最优平仓边界,它随状态与结构变化,不是固定比例;单独的 50% 止盈在两种报价假设下都排在末位。

数据口径:书中价格一律以指数点表示,路径全部合成,参数为示意;全部计算由书中程序生成。 书中以标注框区分已发表的证据、可运行的 Python 计算(21 段程序)、原典与作者的假说。

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: 50

sticky marking: debit 1.704, iv at entry 0.1600
ruleP&L cost 0.05P&L cost 0.15
hold to expiry-0.1120-0.1120
optimal (Snell)-0.0414-0.0987
take 50%, stop 21-0.1237-0.2237
take 65%, stop 21-0.1159-0.2159
take 50%, no stop-0.1645-0.2023
stop 21 only-0.1074-0.2074
rational marking: debit 1.773, iv at entry 0.1746
ruleP&L cost 0.05P&L cost 0.15
hold to expiry-0.1805-0.1805
optimal (Snell)-0.0500-0.1487
take 50%, stop 21-0.1427-0.2427
take 65%, stop 21-0.1417-0.2417
take 50%, no stop-0.2204-0.2579
stop 21 only-0.1387-0.2387
查看完整验算程序
# Chapter 4. p43: exact values of closing rules for the put debit spread
# (100/95) on the lattice and vol chain of p41 (index points).
# Holder's forecast (illustrative): realized vol 0.14 calm, 0.32 panic;
# calm to panic 0.005 a day with an index jump of three grid steps down,
# panic to calm 0.10 a day; zero drift overall (the index is a
# martingale). Two marking hypotheses: "sticky" (0.16 calm, 0.34 panic)
# and "rational": the square root of 1.10 times the chain's expected
# average variance over the remaining life (a variance premium of ten
# percent, jumps included). Each rule is valued by backward induction,
# so the numbers are exact expectations on the lattice, not samples.
# P&L = value minus the entry debit; closing costs 0.05 or 0.15.
import math


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


def bs_put(s, k, t, vol):
    if t <= 0.0:
        return max(k - s, 0.0)
    sd = vol * math.sqrt(t)
    d1 = (math.log(s / k) + 0.5 * sd * sd) / sd
    return k * ncdf(-(d1 - sd)) - s * ncdf(-d1)


N, DT = 45, 1 / 365
SIG, IV = (0.14, 0.32), (0.16, 0.34)
P_UP, P_DN, JUMP = 0.005, 0.10, -3
DX = 0.34 * math.sqrt(DT)
LO, HI = -4 * N - 4, N + 4


def probs(sig, c):
    m = math.log(c) - 0.5 * sig * sig * DT
    a = (sig * sig * DT + m * m) / (DX * DX)
    pu, pd = 0.5 * a + 0.5 * m / DX, 0.5 * a - 0.5 * m / DX
    return pu, 1.0 - pu - pd, pd


PR = (probs(SIG[0], 1.0 / (1 - P_UP + P_UP * math.exp(JUMP * DX))),
      probs(SIG[1], 1.0))


def spot(x):
    return 100.0 * math.exp(x * DX)


def nxt(F, x, s):
    def mix(y, t):
        pu, pm, pd = PR[s]
        return pu * F[t][y + 1] + pm * F[t][y] + pd * F[t][y - 1]
    if s == 0:
        return (1 - P_UP) * mix(x, 0) + P_UP * mix(x + JUMP, 1)
    return (1 - P_DN) * mix(x, 1) + P_DN * mix(x, 0)


def avg_var():
    # A[s][d]: expected summed variance over d days starting in state s
    jv = P_UP * (JUMP * DX) ** 2
    var = (SIG[0] ** 2 * DT + jv, SIG[1] ** 2 * DT)
    A = [[0.0] * (N + 1) for _ in (0, 1)]
    for d in range(1, N + 1):
        A[0][d] = var[0] + (1 - P_UP) * A[0][d - 1] + P_UP * A[1][d - 1]
        A[1][d] = var[1] + (1 - P_DN) * A[1][d - 1] + P_DN * A[0][d - 1]
    return [[0.0] + [math.sqrt(1.10 * A[s][d] / (d * DT))
                     for d in range(1, N + 1)] for s in (0, 1)]


RAT = avg_var()


def make_mark(kind):
    def mark(S, d, s):
        v = IV[s] if kind == "sticky" else RAT[s][d]
        t = d / 365
        return bs_put(S, 100.0, t, v) - bs_put(S, 95.0, t, v)
    return mark


def value(mark, cost, tp, stop, optimal=False):
    size = HI - LO + 1
    debit = mark(100.0, N, 0)
    mx = 5.0 - debit
    W = [[min(max(100.0 - spot(i + LO), 0.0), 5.0) for i in range(size)]
         for _ in (0, 1)]
    for d in range(1, N + 1):
        nW = [[0.0] * size for _ in (0, 1)]
        for s in (0, 1):
            for i in range(size):
                m = mark(spot(i + LO), d, s)
                if i < 4 or i > size - 2:
                    nW[s][i] = m - cost
                    continue
                c = nxt(W, i, s)
                if optimal:
                    nW[s][i] = max(m - cost, c)
                elif (m - debit) / mx >= tp or d <= stop:
                    nW[s][i] = m - cost
                else:
                    nW[s][i] = c
        W = nW
    return W[0][-LO] - debit, debit


RULES = (("hold to expiry", 9.0, 0, False),
         ("optimal (Snell)", 9.0, 0, True),
         ("take 50%, stop 21", 0.50, 21, False),
         ("take 65%, stop 21", 0.65, 21, False),
         ("take 50%, no stop", 0.50, 0, False),
         ("stop 21 only", 9.0, 21, False))
for kind in ("sticky", "rational"):
    mk = make_mark(kind)
    print("%s marking: debit %.3f, iv at entry %.4f"
          % (kind, mk(100.0, N, 0),
             IV[0] if kind == "sticky" else RAT[0][N]))
    print("  rule                P&L cost 0.05   P&L cost 0.15")
    for name, tp, stop, opt in RULES:
        a = value(mk, 0.05, tp, stop, opt)[0]
        b = value(mk, 0.15, tp, stop, opt)[0]
        print("  %-18s    %+.4f          %+.4f" % (name, a, b))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《期权提前止盈最优停时、恐慌脉冲波动率收割与持有至到期回吐陷阱母典》第 4 章「最优平仓边界不是一个固定比例:格点上的斯奈尔包络」(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-426 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026426,
  author    = {Fan, Yafu},
  title     = {Optimal Early Profit-Taking in Options: Harvesting Volatility Spikes, Twenty-One-Day Decay Cliffs and the Expiration Hold Trap},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论