Fourier Transform Option Pricing

Definition

Fourier transform option pricing values European options whenever the characteristic function of the log-price is known analytically, by writing the option price (or its Fourier transform) in closed form in terms of that characteristic function and inverting numerically with the Fast Fourier Transform (FFT) (Carr-Madan 1999). It sidesteps the need for a tractable density and delivers prices across a whole grid of strikes at once.

Key Ideas

• Work with the characteristic function, not the density. Many models (Heston, jump-diffusion, variance gamma) have a simple characteristic function $\phi_T(u)=E[e^{iu s_T}]$ of the log price even when the density $q_T$ is complicated or unknown in closed form.
• The damping trick. The call price $C_T(k)$ as a function of log-strike $k$ is not square-integrable ($C_T(k)\to S_0$ as $k\to-\infty$). Multiply by an exponential factor, $c_T(k)=e^{\alpha k}C_T(k)$ with a damping parameter $\alpha>0$, to make it integrable.
• Closed-form transform. The Fourier transform of the damped call has the explicit form $$\psi_T(v)=\frac{e^{-rT}\,\phi_T\!\big(v-(\alpha+1)i\big)}{\alpha^2+\alpha-v^2+i(2\alpha+1)v},$$ purely in terms of the characteristic function.
• Invert with the FFT. $C_T(k)=\frac{e^{-\alpha k}}{\pi}\int_0^\infty e^{-ivk}\psi_T(v)\,dv$ is a direct Fourier integral; discretizing on a log-strike grid (with Simpson's-rule weights) lets the FFT return option prices at many strikes simultaneously — enabling real-time pricing, marking, and hedging of books with thousands of options.
• The role of $\alpha$. $\alpha>0$ is required (at $\alpha=0$ the denominator vanishes at $v=0$, a singularity); $\alpha$ must keep $\phi_T(v-(\alpha+1)i)$ finite, so it is bounded by the moments the model admits. A separate (hyperbolic-sine damped) transform handles near-maturity/at-the-money time value.

How It Works

Given a model's characteristic function, plug it into $\psi_T(v)$, choose $\alpha$, set up a log-strike grid matched to the FFT frequency grid, and run one FFT to obtain a strip of call prices. Because the transform is analytic, the only error sources are the numerical truncation and discretization of the integral, both controllable.

Why It Matters

• Speed and scale. One FFT prices an entire strike strip, making calibration and large-book revaluation fast — the practical enabler for using realistic characteristic-function models in production.
• Unlocks characteristic-function models. It is the standard pricing engine for models defined through their characteristic function — Heston stochastic volatility, jump-diffusion, and the variance gamma model (Madan) used to illustrate the method.
• Reorients the problem. Prices the option value directly rather than decomposing into risk-neutral probabilities with discontinuous payoffs — numerically cleaner.

Open Questions

• Choosing $\alpha$ and the FFT grid (spacing, truncation) to balance accuracy against aliasing/truncation error; poor choices degrade deep-in/out-of-the-money prices.
• Extensions and refinements (COS method, fractional FFT, saddlepoint approximations) improve accuracy/efficiency over the basic scheme.
• Robustness for very short maturities and heavy-tailed models where the damped integrand decays slowly.

Related

• Heston Model
• Jump-Diffusion Model
• Carr-Madan (1999) Option Valuation Using the Fast Fourier Transform
• Peter Carr
• Dilip B. Madan
• GARCH Option Pricing