Carr-Madan (1999) Option Valuation Using the Fast Fourier Transform

Summary

Carr and Madan show how to value European options with the Fast Fourier Transform whenever the characteristic function of the log return is known analytically. Rather than numerically recovering risk-neutral probabilities (which involve discontinuous payoffs), they derive a simple analytic expression for the Fourier transform of the (exponentially damped) option price in terms of the characteristic function, then invert it by FFT. The method prices an entire strip of strikes at once, enabling real-time pricing and hedging of large option books, and is illustrated on the variance gamma model.

Key Claims

• Characteristic-function pricing. For any model whose risk-neutral log-price characteristic function $\phi_T(u)$ is known in closed form, option values follow without needing the density.
• Damping for integrability. The call price $C_T(k)$ in log-strike $k$ is not square-integrable (it tends to $S_0$ as $k\to-\infty$); the modified price $c_T(k)=e^{\alpha k}C_T(k)$, $\alpha>0$, is, and can be Fourier-transformed.
• Analytic transform (eq. 6). $\displaystyle \psi_T(v)=\frac{e^{-rT}\phi_T(v-(\alpha+1)i)}{\alpha^2+\alpha-v^2+i(2\alpha+1)v}$, from which $C_T(k)=\frac{e^{-\alpha k}}{\pi}\int_0^\infty e^{-ivk}\psi_T(v)\,dv$.
• FFT inversion. The inversion integral is a direct Fourier transform, discretized on a log-strike grid (Simpson weights) so a single FFT yields prices at many strikes — real-time pricing, marking, and hedging even for books with thousands of options.
• Damping parameter $\alpha$. $\alpha=0$ makes the denominator vanish at $v=0$ (a singularity), so $\alpha>0$ is needed; it is bounded by the model's finite moments (keeping $\phi_T(v-(\alpha+1)i)$ finite). A separate hyperbolic-sine-damped transform handles near-maturity at-the-money time value.
• Illustration. The variance gamma model — simple characteristic function, complicated density — demonstrates the method against a case with known prices.

Concepts Introduced or Extended

• Fourier Transform Option Pricing

Entities Mentioned

• Peter Carr
• Dilip B. Madan

Quotes

"This paper shows how the fast Fourier Transform may be used to value options when the characteristic function of the return is known analytically."

"Our use of the FFT in the inversion stage permits real time pricing, marking, and hedging using realistic models, even for books with thousands of options."

My Take

One of the most cited "engineering" papers in derivatives: it turned the abstract fact that many models have tractable characteristic functions into a fast, production-grade pricing routine. The two ideas that make it work — exponential damping to force integrability, and pricing the option value directly instead of decomposing into discontinuous probability elements — are simple but decisive, and the FFT's strike-strip output is exactly what calibration needs. The practical soft spot is the damping parameter $\alpha$ and the grid: they must be tuned to the model's moments and to the strike range, and the basic scheme loses accuracy in the wings and at very short maturities, motivating the later COS/fractional-FFT refinements. It is the natural pricing engine behind the Heston and jump-diffusion models catalogued elsewhere in the wiki.