Sensitivities in Option Pricing Models

The inverse problem in finance consists of determining the unknown parameters of the pricing equation from the values quoted from the market. We formulate the inverse problem as a minimization problem for an appropriate cost function to minimize the difference between the solution of the model and the market observations. Efficient gradient based optimization requires accurate gradient estimation of the cost function. In this thesis we highlight the adjoint method for computing gradients of the cost function in the context of gradient based optimization and show its importance. We derive the continuous adjoint equations with appropriate boundary conditions for three main option pricing models: the Black-Scholes model, the Heston's model and the jump diffusion model, for European type options. These adjoint equations can be used to compute the gradient of the cost function accurately for parameter estimation problems.

The adjoint method allows efficient evaluation of the gradient of a cost function F(σ) with respect to parameters σ where F depends on σ indirectly, via an intermediate variable. Compared to the finite difference method and the sensitivity equation method, the adjoint equation method is very efficient in computing the gradient of the cost function. The sensitivity equations method requires solving a PDE corresponding to each parameter in the model to estimate the gradient of the cost function. The adjoint method requires solving a single adjoint equation once. Hence, for a large number of parameters in the model, the adjoint equation method is very efficient.

Due to its nature, the adjoint equation has to be solved backward in time. The adjoint equation derived from the jump diffusion model is harder to solve due to its non local integral term. But algorithms that can be used to solve the Partial Integro-Differential Equation (PIDE) derived from jump diffusion model can be modified to solve the adjoint equation derived from the PIDE.