Volatility estimation from observed option prices (Q5944948)

This paper provides a numerical procedure for calibrating the evolution of the underlying to the volatility smile with respect to, simultaneously, time and strike. The market is supposed to calculate prices by means of \[ dS= S(\mu(S,t)\, dt+ \sigma(S,t)\, dW_t). \] The market ``only'' informs us of the prices of options. In other words, it is given \(C(K,T)\), the price of the call with strike \(K\) and maturity \(T\), for a finite set of values of \(K\) and \(T\) and we would like to known \(\sigma(S,t)\). It is well known (Dupire) that \(C(K,T)\) satisfies an equation of the form \[ \frac{1}{2} \sigma^2 \frac{\partial^2C}{\partial K^2}= \frac{\partial C}{\partial T}+ A(T)C+ \biggl(B(T)+ \frac{1}{2} \sigma^2\biggr) \frac{\partial C}{\partial K},\tag{1} \] for some known functions \(A\), \(B\) (depending on dividend yield and interest rate). The numerical procedure goes as follows: Given that \(C\) is known at the corners of a certain grid, one transforms a discrete version of (1) into a nonlinear equation whose variable is the value of \(\sigma\) at a nearby point. This approach of the authors is a variation of a previous algorithm of Andersen and Brotherson-Ratcliffe.