Optimal importance sampling for Lévy processes (Q2289777)

Lévy processes are stochastic processes with stationary independent increments. They are used as models for asset price when jump risk is relevant, either directly or as building blocks for models. The present paper develops sampling estimators for Monte Carlo pricing of European and path-dependent options in models driven by Lévy processes. Using results from the theory of large deviations for processes with independent increments, an explicit asymptotic approximation for the variance of the pay-off under a time-dependent Esscher-style change of measure is computed. Minimizing this asymptotic variance, an importance sampling estimator of the option price is obtained. It is shown that the proposed estimator is logarithmically optimal among all importance sampling estimators. Numerical test in the variance gamma model shows consistent variance reduction with a small computational overhead. In the Introduction of the paper a short review of the Lévy processes to model a financial market is presented. To model a financial market with a Lévy process it is assumed that the market consists of a risk-free asset \(S_{t}^{0} = 1\) and \(n\) risky assets \(S^{1}, S^{2}, \ldots, S^{n}\), where \[ S^{i}_{t} = S^{i}_{0} . e^{x^{i}_{t}}, \] and \((X^{1}, X^{2}, \ldots, X^{n})\) is a Lévy process, such that \(S^{i}\) is a martingale for each \(i\), under the risk-neural probability. Also, some methods for computing of the expectation of this process are reminded. The Fourier or Carr and Madan method and the American options are discussed. For high dimensional problems, or in the case of strong path dependence, the Monte Carlo method is considered. The convergence of this method is discussed. Finally, the main principles of the importance sampling method are reminded. In Section 2 the notations and results from the theory of large deviations are recalled. First, the large deviations principle on abstract spaces is formulated. Also, the large deviation principle for Lévy processes is developed. Section 3 provides a representation for the proxy of the variance, a simplified representation in the case of concave log-payoffs and easy to verify criterion for concavity. Here, the used importance sampling estimator is based on the path-dependent Esscher transform. The approach of the considerations is instead to minimize a proxy of the variance. Proposition 5 provides an expression for such a proxy. The notion of an asymptotically optimal is introduced. In Theorem 8 it is shown the solution of the important sampling problem. Section 4 presents computations for European basket and Asian options. The results of the previous section are applied to several options pay-offs encountered in practice. The general European pay-offs and the arithmetic asian put option are considered. In Section 5 the results of the paper are illustrated with numerical computations in the multivariate variance gamma model. In the Appendix of the paper two technical results are proved.