A REFRESHED VISION OF NON-LIFE INSURANCE PRICING - A Generalized Linear Model and Machine Learning Approach

Aug 4, 2026

Upload by Logic

Insurance pricing in non-life insurance must evolve with changing risk landscapes, where some risks are quantifiable while others remain unknown or hard to measure. This project work presents an updated view of pricing methods for a motor portfolio in Portugal by building frequency and severity models with GLM, supported by a machine learning model using Gradient Boosting. The results aim to improve predictive accuracy and produce fairer tariffs for both the insurer and its clients. The Gradient Boosting approach yields the lowest total deviance for frequency, while GLM produces the lowest severity deviance, supporting the integration of machine learning with limited input.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages91

Reading Time3 h 49 min

AccessFree to download

Introduction

Motivation

MEGI Master Degree Program in Statistics and Information Management A REFRESHED VISION OF NON-LIFE INSURANCE PRICING A Generalized Linear Model and Machine Learning Approach Carina de Miranda Clemente Project Work presented as partial requirement for obtaining the Master Degree Program in Statistics and Information Management NOVA Information Management School Instituto Superior de Estatística e Gestão de Informação Universidade Nova de Lisboa NOVA Information Management School Instituto Superior de Estatística e Gestão de Informação Universidade Nova de Lisboa A REFRESHED VISION OF NON-LIFE INSURANCE PRICING By Carina de Miranda Clemente Project Work presented as partial requirement for obtaining the Master’s degree in Statistics and Information Management, with a specialization in Risk Analysis and Management. Supervisor: Gracinda Rita Diogo Guerreiro Co-Supervisor: Jorge Miguel Ventura Bravo November 2022 STATEMENT OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledge the Rules of Conduct and Code of Honor from the NOVA Information Management School. Lisboa, 30/11/2022 DEDICATION This project is dedicated to my grandmother Maria, who had always been one of the strongest supporters of my academic journey. One of her wishes was the successful completion of my studies, which she witnessed with the beginning of the development of this project. Unfortunately, she is not here today to see me deliver it. Avó, I did it. ACKNOWLEDGEMENTS To my coworkers Mariana, Filipa, Rodrigo and Nuno, it would have been impossible to develop this project without your help. You were always there to answer my questions and to guide me through this long path. To my mother Natalina and my boyfriend Bruno, who supported me all along despite my absent times. To my aunt Elisabete and my cousin Catarina, who would always make sure I was being able to hold down the fort. To Catarina and Sofia, who did not get tired of my thousand questions and always listened to my (repetitive) concerns. To my supervisor professor Gracinda and co-supervisor professor Jorge, that despite their tight availability were able to share some of their knowledge with me. ABSTRACT Insurance companies are faced with a constantly changing world, in a daily basis. There are a number of known risks that are quantifiable, alongside with many more that remain unknown or difficult to measure. It is only natural that the pricing of such risks must evolve side by side with the state-of-theart technology that is available. In the late years there has been a rise in the number of studies that conclude on the better fitting of models based on machine learning technology, when it comes to estimate the prices charged by insurance companies in order to hedge the risks they endure. This project work aims to provide a refreshment of the pricing methods applied by a given insurance company operating in Portugal, by developing GLM-based frequency and severity modelling on a subset of the motor portfolio of the company, backed up by a machine learning model: Gradient Boosting. In doing so, there is an expectation of improvement of the accuracy of the model, providing better fitting estimates that could translate into a fairer tariff for both the insurance company and its clients. In fact, it was concluded that the Gradient Boosting approach outputted the lowest total deviance associated with the frequency model. In terms of severity, it was the GLM that produced to the lowest value. With the development of this project, there is now an open path in my company for the inclusion of machine learning methods on the development of insurance tariffs, being here proven that with little required input, this appro

FAQs

What modeling approaches are used for non-life insurance pricing in this project?

The project uses GLM-based frequency and severity modeling and a machine learning approach based on Gradient Boosting.

How does the project evaluate the performance of the models?

Model fit is assessed using deviance metrics, comparing frequency totals and severity values across approaches.

What main conclusions are drawn from the results?

Gradient Boosting provides the lowest total deviance for the frequency model, while GLM achieves the lowest severity deviance.

Aug 4, 2026

Upload by Logic