Skip to content

Latest commit

 

History

History
81 lines (52 loc) · 2.37 KB

gop.md

File metadata and controls

81 lines (52 loc) · 2.37 KB

title: General Option Pricing author: Keith A. Lewis institute: KALX, LLC classoption: fleqn fleqn: true abstract: European option pricing and greeks thanks: Thank you Peter Carr and Bill Goff for your valuable feedback. ...

\newcommand{\Var}{\operatorname{Var}} \newcommand{\RR}{𝑹}

European option valuation involves calculating the expected value of the option payoff using the underlying at expiration. Greeks are derivatives of the option value with respect to model parameters. This short note derives formulas for these that can be used for any underlying.

Let $F$ have mean $f$ and variance $f^2s^2$. We can write $F = f(1 + s X)$ where $X$ has mean 0 and variance 1. If $X$ is standard normal we have the Bachelier model. Let $\Phi$ be the cdf of $X$.

Note $F\le k$ iff $X \le (k - f)/fs = x(k)$. $k(x) = f(1 + s x)$.

$dx(k)/df = 1/fs$.

$v = E[\nu(F)]$.

$dv/df = E[\nu'(F)(1 + sX)]$.

$d^nv/df^n = E[\nu''(F)(1 + sX)^n]$.

$dv/ds = E[\nu'(F)fX]$.

Let $\Phi(x, n) = E[X^n 1(X\le x)]$ be the partial moments.

By Carr-Madan $f(x) = f(a) + f'(a)x + \int_{-\infty}^a f''(k)(k - x)^+,dk + \int_a^{\infty} f''(k)(x - k)^+,dk$.

$E[\delta_a(g(X))] = E[\delta_a(Y)]$, $Y = g(X)$.

$\Psi(y) = P(Y\le y) = P(g(X)\le y) = P(X\le g^{-1}(y)) = \Phi(g^{-1}(y))$

$\psi(y) = \Phi'(y) = \phi(g^{-1}(y))dg^{-1}(y)/dy = \phi(g^{-1}(y))/g'(g^{-1}(y))$.

Bachelier

$F = f(1 + sX)$, $g(x) = f(1 + sx)$.

Let $f(x) = (x - a)^2/2 1(x \le a)$

$f'(x) = -(x - a)^2/2 \delta_a(x) + (x - a) 1(x \le a)$

$f''(x) = -(x - a)^2/2 \delta'_a(x) - 2(x - a) \delta_a(x) + 1(x \le a)$.

$E[(k - F)^+] = E[(k - F)1(F \le k)] = k\Phi(x(k)) - E[F1(X \le x(k))]$.

$E[F1(X \le x)] = E[f(1 + s X)1(X\le x)] fP(X\le x) + fs E[X 1(X\le x)]$

$p = E[(k - F)^+] = (k - f)\Phi(x(k)) + fs\Phi(x(k), 1)$

$dp/df = E[-(F/f)1(F\le k)] = E[-(1 + s X)1(F\le k)] = -\Phi(x(k)) - s \Phi(x(k), 1)$.

$d^2p/df^2 = $

$d/df E[1(F\le k)] =

$dp/ds = E[-X1(F\le k)] = - \Phi(x(k), 1)$.

$1 + s X = e^{sZ - s^2/2}$, $X = (e^{sZ - s^2/2} - 1)/s$

$s^2 = (e^{s^2} - 1)$, $s^2 = \log(s^2 + 1)$.