Second order multiscale stochastic volatility asymptotics: stochastic terminal layer analysis and calibration (Q309158)

This paper deals with the class of multiscale stochastic volatility models. Let \(X\) denote the price of a non-dividend-paying asset whose dynamics under the historical probability measure \(P\) is defined by \(dX_{t}=\mu X_{t}dt + f(Y_{t},Z_{t}) X_{t} dW_{t}^{(0)}\), \(dY_{t}=\varepsilon^{-1}\alpha(Y_{t})dt+\varepsilon^{-1/2}\beta(Y_{t})dW_{t}^{(1)}\), \(dZ_{t}=\delta c(Z_{t})dt+\delta^{1/2} g(Z_{t})dW_{t}^{(2)}\). Here, \((W_{t}^{(0)},W_{t}^{(1)},W_{t}^{(2)})\) are \(P\)-Brownian motions with correlation structure \(d\langle W^{(0)},W^{(1)}\rangle_{t}=\rho_1 dt\), \(d\langle W^{(0)},W^{(2)}\rangle_{t}=\rho_2 dt\), \(d\langle W^{(1)},W^{(2)}\rangle_{t}=\rho_{12} dt\), where \(|\rho_1|, |\rho_2|, |\rho_{12}|<1\), \(1+2\rho_{1}\rho_{2}\rho_{12}-\rho_{1}^2-\rho_{2}^2-\rho_{12}^2>0\). It is assumed that \(0<\varepsilon\ll 1\) and \(0<\delta\ll 1\) so that \(Y\) and \(Z\) represent fast and slowly varying factors of volatility, respectively. Furthermore, it is assumed that the fast factor is mean-reverting. The authors derive a pricing approximation which is valid for any European-style option and establish the accuracy of the proposed pricing approximation. An explicit formula for the implied volatility surface induced by the proposed option pricing approximation is presented. A procedure for calibrating the class of multiscale stochastic volatility models to the empirically observed implied volatility surface of liquid calls and puts is proposed. This calibrating procedure is applied to S\&P 500 index call and put options data.