COS Method For Option Pricing

Many advanced asset-price models (jump-diffusions, stochastic volatility) don't have a simple closed-form option price like Black-Scholes, but they do have a known characteristic function — a Fourier transform of the asset's return distribution that fully describes it, even when the density itself has no simple formula. Monte Carlo simulation can price options under these models but is slow if you need many prices quickly. The COS method offers a much faster alternative: since the characteristic function is essentially the Fourier coefficients of the distribution, it can be used to reconstruct the density as a Fourier cosine series on a bounded interval, and that series can then be integrated term by term against the option's payoff, giving the option price as a fast, finite sum.

$$v(x) \approx \sum_{k=0}^{N-1} {}' A_k \cos\!\left(k\pi \frac{x - a}{b-a}\right)$$

In words: the option's value is approximated by summing a modest number of cosine terms $A_k$ (computed directly from the model's characteristic function), rather than simulating thousands of random asset paths — the prime on the sum just means the first term is weighted by one-half.

The method is dramatically faster than Monte Carlo for models where the characteristic function is known analytically (like the Heston stochastic-volatility model), often computing prices in milliseconds where simulation would take seconds — the trade-off is that it requires the characteristic function to be available in closed form, and the truncated cosine series introduces a small, controllable approximation error rather than the sampling error of Monte Carlo.