Development and application of statistical learning methods in insurance and finance

Makariou, Despoina (2022) Development and application of statistical learning methods in insurance and finance. PhD thesis, London School of Economics and Political Science.

Text - Submitted Version Download (3MB)

Abstract

This thesis deals with the development and application of statistical learning methods in insurance and finance. Firstly, we focus on an insurance-linked financial instrument type namely catastrophe bond. Given the intricacies, and the over-the-counter nature of the market where these instruments are traded, we introduce a flexible statistical learning model called random forest. We use real data in order to predict the spread of a new catastrophe bond at issuance and identify the importance of various variables in their ability to predict the spread in a purely predictive framework. Finally, we develop and implement a series of robustness checks to ensure repeatability of prediction performance and predictors’ importance results. Secondly, we explore a decision-making problem which is faced in an abundance of interdisciplinary settings referring to the combination of different experts’ opinions on a given topic. Focusing on the case where opinions are expressed in a probabilistic manner, we suggest employing a finite mixture modelling methodology to capture various sources of heterogeneity in experts’ opinions, and assist the decision maker to test their very own judgement on opinions weights allocation too. An application in an actuarial context is presented where different actuaries report their opinions about a quantile-based risk measure to decide on the level of reserves they need to hold for regulatory purposes. Finally, we focus on the problem of regression analysis for multivariate count data in order to capture the dependence structures between multiple count response variables based on explanatory variables, which is encountered across several disciplines. In particular, we introduce a multivariate Poisson-Generalized Inverse Gaussian regression model with varying dispersion and shape for modelling different types of insurance claims and their associated counts and we provide a real-data application in non-life insurance.

Item Type:	Thesis (PhD)
Additional Information:	© 2022 Despoina Makariou
Library of Congress subject classification:	H Social Sciences > HG Finance Q Science > QA Mathematics
Sets:	Departments > Statistics
Supervisor:	Barrieu, Pauline and Chen, Yining
URI:	http://etheses.lse.ac.uk/id/eprint/4391

Actions (login required)

Record administration - authorised staff only