Athens University Economics and Business (AUEB)
Applications of stochastic analysis in sensitivity analysis and insurance mathematics
Abstract
dc:descriptionThis thesis deals primarily with the use of Malliavin calculus techniques in estimating the sensitivity of functionals of diffusion processes. A brief sketch of the main ideas on which the derivation of these sensitivity estimators are based is presented in the first chapter.The second chapter discusses in detail an issue connected with weight-based sensitivity estimators obtained using the integration-by-parts approach and more generally, via the use of Malliavin calculus techniques. Suppose that X is an R^m–valued random variable whose distribution depends on a real parameter θ and f : R^m→R a measurable function. Assuming that there exists a random variable Hθ such that d/dθ Ef (X)=E[f (X)Hθ ] for all f sufficiently smooth, the objective is to establish sufficient conditions under which the above weight estimator can be extended to non-smooth functions (e.g. measurable f such that f (X) satisfies a moment condition). The analysis presented there addresses this problem in more detail than has been done in the literature on sensitivity analysis using Malliavin techniques thus far.Chapter 3 deals with the problem of obtaining sensitivity estimates for perfor- mance measures associated with diffusion processes in R^m with respect to parameters of the drift and the volatility coefficients, or with respect to the initial conditions. The performance measures we examine are expectations of cylindrical functionals of the diffusion process i.e. expectations of functions depending on the values of the diffusion process at a finite number of points. The approach employs the use of the first variation process in order to express the derivative process of the diffusion process as an Itˆo process and then use the integration by parts formula of the Malliavin calculus to express the derivative of the expectation as an expectation it self. A number of specific examples are discussed in order to address implementation issues. Particular attention is paid to the Ornstein Uhlenbeck process in R^m. For this the intuitive meaning of the technical conditions required for the validity of the results can be understood easily and clearly. Furthermore, in view of the Gaussian nature of this process a detailed study of the variance properties of the obtained sensitivity estimators can be carried out using elementary tools.An application to sensitivity analysis of reinsurance is presented in chapter 4. There we consider a diffusion approximation model of an insurer and a reinsurer in the context of a stop-loss reinsurance contract and examine the problem of the joint ruin probability and its sensitivity with respect to various parameters of the system. The general methodology discussed in chapter 3 is used to this effect.Finally, in chapter 5 we discuss briefly the question of sensitivity analysis of func- tionals of Gaussian processes with respect to perturbations in the mean and the covariance function. In this chapter the approach taken to sensitivity analysis is es- sentially via non-anticipating versions of the Girsanov theorem. A representation of the Gaussian process via a Gaussian measure on a Hilbert space is used together with the Feldman-Hajek theorem on the equivalence of Gaussian measures in order to obtain the desired sensitivity estimators. As expected, a necessary condition for such an approach is that the perturbations lie in the Cameron-Martin space of the covariance operator of the Gaussian process.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Roumelioti, Eleni
- Ρουμελιώτη, Ελένη
Subjects
dc:subject × 10Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/39322