Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.768149
Title: Robustness and sensitivity of risk evaluations
Author: Pesenti, Silvana Manuela
ISNI:       0000 0004 7652 7386
Awarding Body: City, University of London
Current Institution: City, University of London
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
This thesis is a collection of three contributions to sensitivity analysis of financial and insurance risk evaluations. Sensitivity analysis constitutes an important component of model building, interpretation and validation, particularly for models whose output is at the core of a risk management decision process. We study models comprising a (random) vector of input factors, an aggregation function mapping input factors to a random output, and a risk measure applied to the output. In most typical insurance and financial applications, the model's characteristic - a non-analytical and numerically expensive aggregation function evaluated on numerous input factors - renders most sensitivity analysis methodologies unfeasible. We develop sensitivity analysis procedures applicable specifically for the above model setting. First, we address the estimation of risk measures applied to the model output. The fundamental purpose of a risk measure is to distinguish between different risk profiles. However, strong assumptions on the risk measure's ability to distinguish risk severities lead to non robust estimators. We provide conditions when risk measures exhibit both, robustness and a consistent ranking of risks. Second, we develop a framework termed reverse sensitivity testing, that associates a critical increase in the risk measure to specific input factors. We provide analytical solutions of the stressed distribution of input factors that lead to the required increase in the outputs' risk measure. Third, we introduce a novel sensitivity measure, which quantifies the extent to which the model output is affected by a stress in an individual input factor. Compared to other sensitivity measures in the literature, the proposed measure incorporates the direct impact of the stressed input as well as indirect effects via other input factors that are dependent on the one being stressed. In this way the dependence between inputs is explicitly taken into account.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.768149  DOI: Not available
Keywords: HG Finance
Share: