5D Library › 期权隐含波动率曲面、偏度溢价与垂直价差结构的定量风控 · English
第五章 时间止损作为最优停时问题
「7—10 天」能被证明吗
作者的框架中有一条交易规则:熊市看跌价差开仓后,持有 7 到 10 天强制平仓,并希望「证明」这个区间是最优的。作者已确认,这里的「天」指交易日:开仓日记为第 0 个交易日,「第 9 天」指开仓后第 9 个交易日收盘,约为两周的日历时间(没有假日时为 11 到 13 个日历日)。本章与第八章的「第 天」都按交易日计。这在一般意义上不可能成立。「最优」总是相对于一个目标与一组假设而言的:交易者要最大化什么(期望收益、某种效用、回撤约束下的收益),标的怎样运动,期权怎样被定价,平仓要付多少成本。换一组假设,最优的平仓时间就会改变。能做的是:把这些假设写清楚,把「何时平仓」写成一个最优停时问题,求出在这些假设下的最优规则,再检查在哪些参数下它落在 7—10 天附近、在哪些参数下不是。
最优停时与斯内尔包络
设 是在时刻 平仓能拿到的价值(市场价减去平仓成本), 是任何不预知未来的平仓时间(停时)。问题是求
在离散时间中,答案由倒推给出:到期时 ,之前每一步
即「现在平仓」与「继续持有的期望价值」取大者。 称为 的斯内尔包络 (Snell 1952),最优规则是在 的第一刻平仓。连续时间的版本是一个自由边界问题,佩斯基尔与希里亚耶夫的专著给出了系统的理论 (Peskir and Shiryaev 2006)。美式期权的定价(第三章)就是这类问题的一个例子。
作者的模型
注
以下是作者的模型,参数都是示意值。 它的目的是展示最优平仓时间依赖于哪些量,不是对任何真实市场或交易策略的估计;模型的数学结论不是对现实的预测证据。
- 时钟: 以交易日计时,一年 252 个交易日;周末与假日既不计波动,也不计时间损耗(这是一种简化:实际市场中期权在周末也会损失一部分时间价值)。
- 价差: 买入行权价 100、卖出行权价 95 的看跌期权,开仓时剩余 21 个交易日(约 30 个日历日),标的价格 100,按卖价开仓(借方 1.581;第三章按日历日计时,中间价为 1.518,加上同样的 0.06 为 1.578,差别来自两种计时方式)。
- 市场定价: 每条腿按固定的隐含波动率(100 为 0.22,95 为 0.245,「黏性行权价」)用布莱克—斯科尔斯公式定价,利率 3%。
- 标的的真实运动: 波动率 (低于隐含波动率,即存在方差风险溢价);漂移 。(年化)是交易者开仓时的看空信号,它以时间常数 个交易日衰减;信号消失后,标的回到带有股权风险溢价 的正常漂移。
- 成本: 提前平仓付两个半价差 ;到期时若价差为实值,也必须在到期前平仓,付 (实物交割的期权);若价差已作废,则任其失效,不付成本。
- 目标: 最大化期望贴现价值,没有风险厌恶。
模型用每个交易日 12 步的二叉树表示标的的真实运动(每一步的漂移取该步内的平均值),在每个节点比较「现在平仓」与「继续持有」。程序同时计算每一种固定的时间止损(第 个交易日收盘无条件平仓)的期望收益,以及依状态而定的最优规则;最后在 12 种情形中每次只改变一个参数,看最优的固定平仓日怎样移动。
推导
# When should a 100/95 bear put spread with 21 TRADING days to expiry (about 30
# calendar days) be closed? An optimal stopping problem (Snell envelope) on a binomial
# tree, 12 steps per trading day. Time is counted in TRADING days throughout: a year is
# 252 trading days and "day d" means the close of the d-th trading day after entry
# (entry = day 0).
# THE AUTHOR'S MODEL, all numbers ILLUSTRATIVE: the market marks each leg at a fixed
# implied vol (sticky strike); the underlying moves with vol sP and drift
# mu(t) = r + a e + b (1 - e), e = exp(-t / tau): a bearish signal a that fades with time
# constant tau (trading days), then drift b (an equity premium). Closing costs c (two
# half-spreads); ce at expiry (ce = c: an in-the-money spread must be closed before
# expiry; ce = 0: cash-settled). The trader maximises expected discounted value (no risk
# aversion). Weekends and holidays are treated as carrying no variance and no time decay.
import math
TD_YEAR, DAYS, M = 252, 21, 12 # trading days a year, to expiry, steps per day
def ncdf(x):
return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))
def put(S, K, T, r, s):
if T <= 0:
return max(K - S, 0.0)
d1 = (math.log(S / K) + (r + 0.5 * s * s) * T) / (s * math.sqrt(T))
return K * math.exp(-r * T) * ncdf(-d1 + s * math.sqrt(T)) - S * ncdf(-d1)
BASE = dict(a=-1.0, tau=2.5, b=0.06, sP=0.18, v1=0.22, v2=0.245, c=0.06, ce=0.06, r=0.03)
def solve(a, tau, b, sP, v1, v2, c, ce, r, days=DAYS, m=M, S0=100.0):
n, dt = days * m, 1 / (TD_YEAR * m)
u = math.exp(sP * math.sqrt(dt)); dn = 1 / u; disc = math.exp(-r * dt)
mark = lambda i, j: (put(S0 * u ** (2 * j - i), 100, (n - i) * dt, r, v1)
- put(S0 * u ** (2 * j - i), 95, (n - i) * dt, r, v2))
def prob(i): # drift averaged over the step
e = tau * m * (math.exp(-i / m / tau) - math.exp(-(i + 1) / m / tau))
mu = r + a * e + b * (1 - e)
return (math.exp(mu * dt) - dn) / (u - dn)
end = lambda j: mark(n, j) - (ce if mark(n, j) > 0 else 0)
w = [end(j) for j in range(n + 1)]; stop = {}
for i in range(n - 1, 0, -1):
p = prob(i); new = []
for j in range(i + 1):
cont = disc * (p * w[j + 1] + (1 - p) * w[j]); now = mark(i, j) - c
stop[i, j] = now > cont; new.append(max(now, cont))
w = new
p = prob(0); v_opt = disc * (p * w[1] + (1 - p) * w[0])
mass, fixed, exit_day = [1.0], {}, [0.0] * (days + 1)
for i in range(1, n + 1): # forward pass: probabilities
p = prob(i - 1)
mass = [(mass[j] * (1 - p) if j < i else 0) + (mass[j - 1] * p if j else 0)
for j in range(i + 1)]
if i % m == 0:
k = i // m
val = sum(q * (mark(i, j) - c if i < n else end(j))
for j, q in enumerate(mass))
fixed[k] = val * math.exp(-r * k / TD_YEAR)
live = [1.0] # mass still open under the optimal rule
for i in range(1, n + 1):
p = prob(i - 1)
live = [(live[j] * (1 - p) if j < i else 0) + (live[j - 1] * p if j else 0)
for j in range(i + 1)]
if i < n:
for j in range(i + 1):
if stop[i, j]:
exit_day[math.ceil(i / m)] += live[j]; live[j] = 0.0
exit_day[days] += sum(live)
return v_opt, fixed, exit_day, stop
def debit(v1, v2, c, r=0.03, **_):
T = DAYS / TD_YEAR
return put(100, 100, T, r, v1) - put(100, 95, T, r, v2) + c
if __name__ == "__main__":
v, fixed, ex, stop = solve(**BASE); d0 = debit(**BASE)
print(f"base: a {BASE['a']}, tau {BASE['tau']} td, b {BASE['b']}, sP {BASE['sP']},"
f" implied {BASE['v1']}/{BASE['v2']}, c = ce = {BASE['c']}")
days = (1, 3, 5, 7, 8, 9, 10, 12, 15, 21)
print(f"debit at the ask {d0:.3f}; expected profit (value - debit)"
f" by exit trading day (TD):")
print(" TD " + "".join(f"{d:7d}" for d in days))
print(" profit" + "".join(f"{fixed[d] - d0:+7.3f}" for d in days))
cum = [sum(ex[: d + 1]) for d in range(DAYS + 1)]
mean = sum(d * q for d, q in enumerate(ex))
print(f"optimal state-dependent rule: profit {v - d0:+.3f};"
f" closed by trading day 7 {cum[7]:.0%},")
print(f" by day 10 {cum[10]:.0%}, at expiry {ex[DAYS]:.0%};"
f" mean exit trading day {mean:.1f}")
u = math.exp(BASE["sP"] * math.sqrt(1 / (TD_YEAR * M)))
reg = []
for d in (3, 6, 9, 12, 18):
s = [100 * u ** (2 * j - M * d) for j in range(M * d + 1) if stop[M * d, j]]
reg.append(f"day {d}: " + (f"{min(s):.1f}-{max(s):.1f}" if s else "none"))
print(" close when the spot is in " + ", ".join(reg[:3]))
print(" " + ", ".join(reg[3:]))
print("scenario (one change from base) best TD profit hold optimal within 0.01")
for lab, kw in (("base", {}), ("slower signal tau 5 td", dict(tau=5.0)),
("faster signal tau 1.5 td", dict(tau=1.5)),
("stronger signal a -1.5", dict(a=-1.5)),
("no equity premium b 0", dict(b=0.0)),
("equity premium b 0.12", dict(b=0.12)),
("no variance premium sP .22", dict(sP=0.22)),
("larger variance premium sP .15", dict(sP=0.15)),
("no skew v2 .22", dict(v2=0.22)),
("wider spreads c = ce = .12", dict(c=0.12, ce=0.12)),
("cash-settled ce 0", dict(ce=0.0)), ("no signal a 0", dict(a=0.0))):
P = dict(BASE, **kw); v, fixed, ex, stop = solve(**P); d0 = debit(**P)
best = max(fixed, key=fixed.get)
band = [d for d in fixed if fixed[d] >= fixed[best] - 0.01]
print(f" {lab:30s} {best:4d} {fixed[best] - d0:+.3f} {fixed[DAYS] - d0:+.3f}"
f" {v - d0:+.3f} days {min(band)}-{max(band)}")
# base: a -1.0, tau 2.5 td, b 0.06, sP 0.18, implied 0.22/0.245, c = ce = 0.06
# debit at the ask 1.581; expected profit (value - debit) by exit trading day (TD):
# TD 1 3 5 7 8 9 10 12 15 21
# profit -0.032 +0.068 +0.113 +0.131 +0.134 +0.136 +0.136 +0.134 +0.125 +0.125
# optimal state-dependent rule: profit +0.240; closed by trading day 7 20%,
# by day 10 40%, at expiry 26%; mean exit trading day 13.5
# close when the spot is in day 3: none, day 6: 103.3-106.8, day 9: 102.0-106.1
# day 12: 100.7-105.4, day 18: 99.3-102.0
# scenario (one change from base) best TD profit hold optimal within 0.01
# base 10 +0.136 +0.125 +0.240 days 7-14
# slower signal tau 5 td 21 +0.479 +0.479 +0.556 days 21-21
# faster signal tau 1.5 td 6 +0.027 -0.014 +0.113 days 4-9
# stronger signal a -1.5 21 +0.294 +0.294 +0.402 days 10-21
# no equity premium b 0 21 +0.275 +0.275 +0.346 days 21-21
# equity premium b 0.12 7 +0.098 -0.021 +0.157 days 5-9
# no variance premium sP .22 21 +0.267 +0.267 +0.270 days 21-21
# larger variance premium sP .15 7 +0.102 -0.019 +0.228 days 5-10
# no skew v2 .22 6 +0.084 -0.078 +0.114 days 5-8
# wider spreads c = ce = .12 21 +0.034 +0.034 +0.120 days 21-21
# cash-settled ce 0 21 +0.155 +0.155 +0.255 days 21-21
# no signal a 0 1 -0.118 -0.195 -0.058 days 1-5结果
基准情形。 第 1 个交易日平仓的期望收益为 −0.032(来回付了两次买卖价差,信号还没来得及兑现);随后逐步上升,在第 9 到 10 个交易日达到最高(都是 +0.136,程序取第 10 天,二者相差不到 0.001),之后缓慢下降,第 15 个交易日为 +0.125,持有到期也是 +0.125。所以在这组参数下,最优的固定时间止损落在 7—10 天的区间里。但这个最优点非常平坦:第 7 到 14 个交易日的期望收益与最优值相差都不到 0.01,约为最大亏损的 0.6%;持有到期只差 0.011。
依状态而定的规则更好。 最优的平仓规则不是一个日期,而是一个区域:开仓后头几个交易日什么情况下都不平仓(信号还在);之后,当标的处在多头行权价 100 以上、价差偏虚值的区域时平仓(第 9 个交易日约为 102.0 到 106.1),其他时候继续持有。这个区域随时间向下移动,到第 18 个交易日时为 99.3 到 102.0。原因正是第四章的净希腊值:标的在 100 以上时,价差的净伽马为正、净 vega 为正,在实际波动率低于隐含波动率时,每天损失方差风险溢价,又因为负 Delta 承受股权风险溢价;标的跌向 95 时,这两项都转为有利。价格高于约 106 到 107 时,价差几乎一文不值,平仓成本超过剩余价值,于是任其持有。按这个规则,期望收益为 +0.240,约为最佳固定时间止损(+0.136)的 1.8 倍;到第 7 个交易日收盘为止只有 20% 的路径已经平仓,26% 持有到期,平均平仓日为第 13.5 个交易日。图三右图画出了这个区域。
改变一个参数,最优日就移动。 12 种情形中,最优的固定平仓日从第 1 个交易日到持有到期(第 21 个交易日)都有:
- 信号衰减变慢( 个交易日),最优是持有到期;信号衰减变快(1.5 个交易日),最优是第 6 个交易日(第 4 到 9 天都接近最优);信号变强(),最优也是持有到期,但第 10 到 21 天都接近最优。
- 没有股权风险溢价(),或者没有方差风险溢价(实际波动率等于隐含波动率 0.22),最优都是持有到期:信号消失后继续持有不再有系统性的损失,提前平仓只是白付成本。
- 股权风险溢价加倍()、方差风险溢价变大(实际波动率 0.15)或没有偏度(两腿都按 0.22 定价),最优日提前到第 6 或 7 个交易日。
- 买卖价差加倍时,提前平仓要多付成本,最优变为持有到期(+0.034,作废的价差不付平仓成本);现金交割(到期不付成本)时最优也是持有到期。
- 完全没有信号()时,所有规则的期望收益都为负,最好的选择是不开仓。
独立检查与单笔交易的噪声
下面的程序用另写的蒙特卡洛代码(40 万条按交易日模拟的路径)独立检查基准情形,并看单笔交易的结果有多分散;最后不用路径,直接对第 9 个交易日与到期时标的价格的对数正态分布做数值积分,得到两种规则的精确期望值。 是以最大亏损(借方)为单位的收益。
推导
# Independent Monte Carlo check of the base case of p09 (same ILLUSTRATIVE parameters,
# same TRADING-day clock: 252 a year, 21 to expiry): simulate one step per trading
# day, close at the close of trading day 9 or hold to expiry, and look at the spread of
# single-trade outcomes; then compute the two expectations exactly by numerical
# integration. Returns on risk R = profit / debit (the debit is the max loss).
import math, random
def ncdf(x):
return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))
def put(S, K, T, r, s):
if T <= 0:
return max(K - S, 0.0)
d1 = (math.log(S / K) + (r + 0.5 * s * s) * T) / (s * math.sqrt(T))
return K * math.exp(-r * T) * ncdf(-d1 + s * math.sqrt(T)) - S * ncdf(-d1)
a, tau, b, sP, v1, v2, c, ce, r = -1.0, 2.5, 0.06, 0.18, 0.22, 0.245, 0.06, 0.06, 0.03
Y, DAYS = 252, 21 # trading days a year; trading days to expiry
def spread(S, d): # d = trading days since entry
return put(S, 100, (DAYS - d) / Y, r, v1) - put(S, 95, (DAYS - d) / Y, r, v2)
def drift(t0, t1): # integral of mu(t) dt over trading days t0..t1, in years
e = tau * (math.exp(-t0 / tau) - math.exp(-t1 / tau))
return (r * (t1 - t0) + a * e + b * ((t1 - t0) - e)) / Y
if __name__ == "__main__":
rng = random.Random(38); n, stop_day = 400000, 9
debit = spread(100.0, 0) + c
step = [drift(d, d + 1) - 0.5 * sP * sP / Y for d in range(DAYS)]
vol = sP * math.sqrt(1 / Y)
rs, rh = [], []
for _ in range(n):
S = 100.0
for d in range(DAYS):
S *= math.exp(step[d] + vol * rng.gauss(0, 1))
if d + 1 == stop_day:
ps = math.exp(-r * stop_day / Y) * (spread(S, stop_day) - c) - debit
pay = spread(S, DAYS); pay -= ce if pay > 0 else 0 # worthless: let it lapse
ph = math.exp(-r * DAYS / Y) * pay - debit
rs.append(ps / debit); rh.append(ph / debit)
mean = lambda x: sum(x) / len(x)
def sd(x):
m = mean(x)
return math.sqrt(sum((y - m) ** 2 for y in x) / (len(x) - 1))
for lab, x in (("close on trading day 9", rs), ("hold to expiry", rh)):
q = sorted(x)
se = sd(x) * debit / math.sqrt(n)
print(f"{lab}: mean profit {mean(x) * debit:+.3f} (SE {se:.3f}),"
f" R mean {mean(x):+.3f}, SD of R {sd(x):.2f}")
print(f" P(loss) {sum(y < 0 for y in x) / n:.2f}, P(R <= -0.95)"
f" {sum(y <= -0.95 for y in x) / n:.2f}; R at 5th/50th/95th pct"
f" {q[n // 20]:+.2f} / {q[n // 2]:+.2f} / {q[19 * n // 20]:+.2f}")
dif = [x - y for x, y in zip(rs, rh)]
cov = mean([x * y for x, y in zip(rs, rh)]) - mean(rs) * mean(rh)
rho = cov / (sd(rs) * sd(rh))
print(f"paired D = R(TD 9) - R(hold): mean {mean(dif):+.3f}, SD {sd(dif):.2f},"
f" correlation of the two R {rho:.2f}")
# exact expectations by numerical integration over the lognormal spot (no paths)
def expect(day, f, k=4000):
m0 = math.log(100.0) + drift(0, day) - 0.5 * sP * sP * day / Y
s0 = sP * math.sqrt(day / Y); tot = wt = 0.0
for i in range(-k, k + 1):
z = 8.0 * i / k; w = math.exp(-0.5 * z * z)
tot += w * f(math.exp(m0 + s0 * z)); wt += w
return tot / wt
def at_expiry(S):
x = spread(S, DAYS)
return x - ce if x > 0 else 0.0
e9 = math.exp(-r * stop_day / Y) * expect(stop_day,
lambda S: spread(S, stop_day) - c)
eh = math.exp(-r * DAYS / Y) * expect(DAYS, at_expiry)
d_exact = (e9 - eh) / debit
print(f"exact (numerical integration): TD 9 {e9 - debit:+.3f},"
f" hold {eh - debit:+.3f}, mean D {d_exact:+.4f}")
z9 = (mean(rs) * debit - (e9 - debit)) / (sd(rs) * debit / math.sqrt(n))
zh = (mean(rh) * debit - (eh - debit)) / (sd(rh) * debit / math.sqrt(n))
print(f" MC minus exact, in standard errors: TD 9 {z9:+.1f}, hold {zh:+.1f}"
f" (same paths: correlated)")
k = (2.8 * sd(dif) / d_exact) ** 2
print(f"independent trades for 80 % power to detect the exact mean D"
f" (two-sided 5 %): {k:,.0f}")
print(f" = {k / 50:,.0f} years at 50 independent trades a year")
# close on trading day 9: mean profit +0.138 (SE 0.002), R mean +0.087, SD of R 0.65
# P(loss) 0.50, P(R <= -0.95) 0.01; R at 5th/50th/95th pct -0.84 / +0.01 / +1.28
# hold to expiry: mean profit +0.130 (SE 0.003), R mean +0.082, SD of R 1.28
# P(loss) 0.59, P(R <= -0.95) 0.48; R at 5th/50th/95th pct -1.00 / -0.75 / +2.12
# paired D = R(TD 9) - R(hold): mean +0.005, SD 1.05, correlation of the two R 0.58
# exact (numerical integration): TD 9 +0.136, hold +0.125, mean D +0.0067
# MC minus exact, in standard errors: TD 9 +1.4, hold +1.4 (same paths: correlated)
# independent trades for 80 % power to detect the exact mean D (two-sided 5 %): 192,009
# = 3,840 years at 50 independent trades a year三种方法一致:第 9 个交易日平仓的期望收益,数值积分为 +0.136,二叉树为 +0.136,蒙特卡洛为 +0.138(标准误 0.002);持有到期分别为 +0.125、+0.125 与 +0.130(标准误 0.003)。蒙特卡洛的两个偏差都约为 1.4 个标准误,而且同号,因为两种规则用的是同一批路径。
但单笔交易的结果极其分散。第 9 个交易日平仓时 的标准差为 0.65,一半的交易亏损;持有到期时标准差为 1.28,近一半的交易(48%)几乎亏掉全部借方(),另有约 5% 的交易赚到借方的两倍以上。两种规则在同一路径上的差 的精确平均值只有 +0.0067,标准差却是 1.05。要以 80% 的把握度在 5% 的显著性水平上分辨这个差别,需要约 19 万笔相互独立的交易,即每年 50 笔、持续约 3,800 年。这说明了一件重要的事:个人交易经验不可能「证明」某个平仓日是最优的;即使模型完全正确,几百笔交易的结果也几乎全是噪声。第八章的检验设计正是为此而来。
假说 · 尚未验证
依价格状态而定的平仓规则优于单纯的时间止损(尚未确立)。 作者的假说是:在真实的期权数据中,「持仓超过五个交易日且标的高于多头行权价时平仓」这类依状态而定的规则,比固定的 7—10 个交易日时间止损有更高的平均收益(以最大亏损为单位)。如果在第八章预先锁定的样本外数据中,这个规则的平均收益不高于时间止损,或者差别的置信区间完全落在 ±0.15 之内,这个假说就是错的。
本章小结
「7—10 天是最优的」不能在一般意义上证明。本书按作者的确认,把「天」统一为交易日。把平仓写成最优停时问题后,在作者的模型中,最优的固定平仓日取决于信号衰减速度、股权风险溢价、方差风险溢价、偏度与成本,在 12 种情形中从第 1 个交易日到持有到期不等;基准情形的最优日为第 9 到 10 个交易日,但第 7 到 14 个交易日几乎一样好,持有到期也只差 0.011,而依价格状态而定的规则明显更好。即使模型正确,单笔交易的噪声也大到几百笔交易无法分辨不同的平仓规则。「7—10 天」应当被看作需要检验的经验规则。
参考文献
Fan, Y. (2026). Implied Volatility Surfaces, Skew Premia and Quantitative Risk Control of Vertical Spreads: Microstructure Pricing and Time-Stop Dynamics under Non-Gaussian Tails [期权隐含波动率曲面、偏度溢价与垂直价差结构的定量风控] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23796765 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).