5D Library › 做市商伽马敞口、零日到期期权与波动率收割动力学 · English
第一章 临近到期的伽马:是尖峰,不是墙
伽马是什么,做市商为什么在乎
期权的 Delta 是与它同步变动的股数;伽马(Gamma)是这个股数随价格变化的速度。一个卖出或买入了期权、又不想承担方向风险的做市商,会持有与 Delta 相反的股票头寸,于是价格每动一次,他就必须按伽马交易一次。这笔交易是全书的主题。第二章问它对价格做了什么;本章只问它有多大、落在哪里。
下文全部使用布莱克-斯科尔斯公式 (Black and Scholes 1973),利率取零,价格 100,年波动率 0.20,一年 252 个交易日、每日 6.5 小时。论证不依赖这些数字,它们只是让表格好读。
证据
所用公式。 令 ,认购期权的 Delta 为 ,伽马为 ; 时时间衰减为 ,其中 、 是标准正态的密度与分布函数。这些是教科书结果 (Black and Scholes 1973);程序用 Python 标准库的误差函数计算。
伽马按时间平方根的倒数上升
平值时 ,所以 :伽马按 上升。第一段程序把它列成表。
推导
# Chapter 1. Black-Scholes Greeks from first principles (standard library
# only).
# Question: how does the gamma of an at-the-money option scale as expiry
# approaches, and how wide is the strike range over which that gamma is large?
# Conventions: S = 100, sigma = 0.20 a year, r = 0, 252 trading days of 6.5
# hours.
import math
def phi(x):
return math.exp(-0.5 * x * x) / math.sqrt(2.0 * math.pi)
def Phi(x):
return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))
def d1(S, K, sig, T):
return (math.log(S / K) + 0.5 * sig * sig * T) / (sig * math.sqrt(T))
def gamma(S, K, sig, T):
return phi(d1(S, K, sig, T)) / (S * sig * math.sqrt(T))
def theta_per_day(S, K, sig, T):
# call theta with r = 0, per trading day
return -S * phi(d1(S, K, sig, T)) * sig / (2.0 * math.sqrt(T)) / 252.0
S, SIG = 100.0, 0.20
HOURS = 252 * 6.5
TENORS = [("30 trading days", 30 / 252), ("7 trading days", 7 / 252),
("1 trading day", 1 / 252), ("2 hours", 2 / HOURS),
("30 minutes", 0.5 / HOURS)]
if __name__ == "__main__":
print("A. At-the-money gamma and theta as expiry approaches")
print(" tenor gamma x 30-day theta/day 1-sd move")
g30 = gamma(S, S, SIG, 30 / 252)
for name, T in TENORS:
g = gamma(S, S, SIG, T)
print(
" {:<16}{:8.4f}{:10.2f}{:12.3f}{:10.3f} %".format(
name,
g,
g / g30,
theta_per_day(S, S, SIG, T),
100 * SIG * math.sqrt(T),
)
)
print(
" gamma ratio 30 days / 30 minutes predicted by 1/sqrt(T): {:.2f}"
.format(math.sqrt((30 / 252) / (0.5 / HOURS)))
)
print(
"\nB. Gamma at strikes away from the money, as a share of ATM gamma"
)
print(" strike offset 30 days 1 day 2 hours 30 minutes")
for off in (0.0, 0.5, 1.0, 2.0, 3.0):
K = S * (1 + off / 100)
row = [gamma(S, K, SIG, T) / gamma(S, S, SIG, T) for _, T in
(TENORS[0], TENORS[2], TENORS[3], TENORS[4])]
print(
" {:>5.1f} % {:8.3f}{:8.3f}{:10.3f}{:12.5f}".format(
off, *row
)
)
print("\nC. Strike half-width at which gamma falls to half its ATM value")
for name, T in TENORS:
lo, hi = 0.0, 50.0
for _ in range(80):
mid = 0.5 * (lo + hi)
if gamma(S, S * (1 + mid / 100), SIG, T) > 0.5 * gamma(
S, S, SIG, T
):
lo = mid
else:
hi = mid
print(" {:<16} +{:.3f} %".format(name, lo))
print(
"\nReading: 0DTE gamma is a spike, not a wall. It is larger at the"
" money"
)
print("and narrower in strike by the same factor, 1/sqrt(T).")
# A. At-the-money gamma and theta as expiry approaches
# tenor gamma x 30-day theta/day 1-sd move
# 30 trading days 0.0578 1.00 -0.046 6.901 %
# 7 trading days 0.1197 2.07 -0.095 3.333 %
# 1 trading day 0.3166 5.48 -0.251 1.260 %
# 2 hours 0.5708 9.88 -0.453 0.699 %
# 30 minutes 1.1417 19.76 -0.906 0.349 %
# gamma ratio 30 days / 30 minutes predicted by 1/sqrt(T): 19.75
#
# B. Gamma at strikes away from the money, as a share of ATM gamma
# strike offset 30 days 1 day 2 hours 30 minutes
# 0.0 % 1.000 1.000 1.000 1.00000
# 0.5 % 1.000 0.927 0.777 0.36198
# 1.0 % 0.995 0.736 0.365 0.01743
# 2.0 % 0.969 0.294 0.018 0.00000
# 3.0 % 0.926 0.065 0.000 0.00000
#
# C. Strike half-width at which gamma falls to half its ATM value
# 30 trading days +8.726 %
# 7 trading days +4.061 %
# 1 trading day +1.503 %
# 2 hours +0.829 %
# 30 minutes +0.413 %
#
# Reading: 0DTE gamma is a spike, not a wall. It is larger at the money
# and narrower in strike by the same factor, 1/sqrt(T).A 段是大家常引用的部分。一日期权的伽马是 30 日期权的 5.48 倍,只剩三十分钟的期权是 19.76 倍; 规则预测 19.75。时间衰减同步上升——那是为这份伽马付的账单。
B、C 两段是通常被略去的部分:伽马较大的行权价区间宽度以同一个倍数收窄。伽马降到平值一半时的半宽,30 日为 +8.726 %,一日为 +1.503 %,三十分钟为 +0.413 %。偏离平值 2% 处,一日期权保留峰值伽马的 0.294,两小时期权只保留 0.018。
所以把零日到期(0DTE)期权说成「伽马墙」只对了一半。伽马确实极大,但到了尾盘只集中在不到百分之一宽的行权价带里。它对价格是否重要,完全取决于价格是否落在这条带里,以及期权在谁手里——那是第三章的主题。
Charm:什么都没动,对冲也在动
对冲流的第二个、更安静的来源是时钟。价格不变时,期权的 Delta 仍随到期临近而漂移:虚值认购的 Delta 衰减向零,实值认购的衰减向一。这个变化率叫 charm。第二段程序对单张认购期权,把它与同一时间窗内一次典型价格波动带来的伽马部分作比较。
推导
# Chapter 1. Where expiry-day hedge flow comes from: price (gamma) or clock
# (charm)? A dealer holding one call hedges with -delta shares. Over one hour
# the hedge must change by d(delta). We split that change into the part caused
# by a typical price move (gamma x one-hour 1-sd move) and the part caused
# purely by the passage of time with the price unchanged (charm). S = 100,
# sigma = 0.20, r = 0, 6.5-hour session ends 16:00.
import math
def Phi(x):
return 0.5 * (1.0 + math.erf(x / math.sqrt(2.0)))
def phi(x):
return math.exp(-0.5 * x * x) / math.sqrt(2.0 * math.pi)
def delta(S, K, sig, T):
if T <= 0:
return 1.0 if S > K else 0.0
d1 = (math.log(S / K) + 0.5 * sig * sig * T) / (sig * math.sqrt(T))
return Phi(d1)
def gamma(S, K, sig, T):
d1 = (math.log(S / K) + 0.5 * sig * sig * T) / (sig * math.sqrt(T))
return phi(d1) / (S * sig * math.sqrt(T))
S, SIG = 100.0, 0.20
YEAR_H = 252 * 6.5
def hm(h):
return "{:02d}:{:02d}".format(int(h), int(round((h - int(h)) * 60)))
# clock in hours, e.g. 10.0 -> 6 hours to the 16:00 close
def T_at(clock):
return (16.0 - clock) / YEAR_H
if __name__ == "__main__":
move = S * SIG * math.sqrt(1.0 / YEAR_H)
print("One-hour 1-sd price move: {:.3f} (= {:.3f} %)".format(move, move))
print("\nChange in the hedge of ONE call over each window")
print(
" strike window delta |gamma part| charm part "
" charm/gamma"
)
for K in (100.25, 100.5, 101.0):
for a, b in ((10.0, 11.0), (14.0, 15.0), (15.0, 15.75), (15.5, 15.9)):
Ta, Tb = T_at(a), T_at(b)
d_a = delta(S, K, SIG, Ta)
charm = delta(S, K, SIG, Tb) - d_a
gpart = gamma(S, K, SIG, Ta) * move * (b - a) ** 0.5
print(
" {:6.2f} {}-{} {:7.4f} {:11.4f} {:+10.4f} {:9.2f}"
.format(
K, hm(a), hm(b), d_a, gpart, charm, abs(charm) / gpart
)
)
print(
"\nReading: at 10:00 the clock term is 5-19 % of the price term;"
" in the"
)
print(
"last hour it is of the same order. An out-of-the-money call loses"
" delta on"
)
print(
"the clock alone, so a dealer long calls buys back hedge shares with"
" no price"
)
print(
"move at all. In this grid the clock term never exceeds the price"
" term."
)
# One-hour 1-sd price move: 0.494 (= 0.494 %)
#
# Change in the hedge of ONE call over each window
# strike window delta |gamma part| charm part charm/gamma
# 100.25 10:00-11:00 0.4207 0.1596 -0.0079 0.05
# 100.25 14:00-15:00 0.3617 0.2650 -0.0542 0.20
# 100.25 15:00-15:45 0.3076 0.3045 -0.1511 0.50
# 100.25 15:30-15:54 0.2380 0.2768 -0.1829 0.66
# 100.50 10:00-11:00 0.3424 0.1500 -0.0145 0.10
# 100.50 14:00-15:00 0.2388 0.2192 -0.0818 0.37
# 100.50 15:00-15:45 0.1570 0.2081 -0.1352 0.65
# 100.50 15:30-15:54 0.0770 0.1292 -0.0763 0.59
# 101.00 10:00-11:00 0.2073 0.1167 -0.0219 0.19
# 101.00 14:00-15:00 0.0778 0.1029 -0.0556 0.54
# 101.00 15:00-15:45 0.0222 0.0457 -0.0221 0.48
# 101.00 15:30-15:54 0.0022 0.0062 -0.0022 0.36
#
# Reading: at 10:00 the clock term is 5-19 % of the price term; in the
# last hour it is of the same order. An out-of-the-money call loses delta on
# the clock alone, so a dealer long calls buys back hedge shares with no price
# move at all. In this grid the clock term never exceeds the price term.10:00 时时钟项只是价格项的一小部分:三个行权价上为 0.05 到 0.19。到最后一小时,两者变成同一量级:对行权价 100.25(略虚值)的认购,15:00-15:45 的比值为 0.50,15:30-15:54 为 0.66。在这张表的范围内,时钟项从未超过价格项。
方向对第二章很重要。持有虚值认购的做市商持有空头对冲;Delta 随时钟衰减时,他在价格完全不动的情况下回补股票。下一章行权价钉死的机制,部分就来自这股流量。
注
这两张表没有说的。 它们描述的是一张期权的对冲。市场看到的流量是所有做市商账簿的加总,而符号取决于做市商对每张期权是多头还是空头。单张期权的伽马表对这个符号一无所知,而这个符号正是第三章的全部问题。
小结
临近到期的伽马按 上升——剩三十分钟时约为剩三十天时的二十倍——而伽马较大的行权价带以同一倍数收窄到不足百分之一。到期日的伽马是尖峰,不是墙。最后一小时里,光是时钟(charm)带来的对冲变化,就与一次典型价格波动带来的同一量级。这些都没有说伽马在谁手里;那是后面两章的事。
参考文献
Fan, Y. (2026). Dealer Gamma Exposure, Zero-Day Options and Volatility Harvesting Dynamics: Delta-Hedging Feedback Loops, Strike Pinning, Iron Condor Tail Ruin and Variance Swap Replication [做市商伽马敞口、零日到期期权与波动率收割动力学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090160 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).