Búsqueda

Comparative performance analysis between Gradient Boosting models and GLMs for non-life pricing

Comparative performance analysis between Gradient Boosting models and GLMs for non-life pricing
Recurso electrónico / Electronic resource
Registro MARC
Tag12Valor
LDR  00000cam a22000004b 4500
001  MAP20210035753
003  MAP
005  20220911185740.0
008  211217s2021 esp|||| ||| ||eng d
040  ‎$a‎MAP‎$b‎spa‎$d‎MAP
084  ‎$a‎6
1001 ‎$0‎MAPA20210037092‎$a‎Martínez de Lizarduy Kostornichenko, Viktor
24510‎$a‎Comparative performance analysis between Gradient Boosting models and GLMs for non-life pricing‎$c‎Viktor Martínez de Lizarduy Kostornichenko
260  ‎$a‎Madrid‎$b‎Universidad Carlos III de Madrid‎$c‎2021
300  ‎$a‎118 p.
5050 ‎$a‎Trabajo Fin de Master del Master en Ciencias Actuariales y Financieras de la Escuela de Postgrado de la Universidad Carlos III de Madrid. Tutores: José Miguel Rodríguez-Pardo del Castillo, Jesús Simón del Potro Curso 2020-2021
520  ‎$a‎Modelling the behavior of risks is one of the most fundamental pillars in the insurance business throughout all its branches. Actuarial practitioners have always been interested in finding the best statistical tools to capture the nature of the risks they undertake from their clients, and in the last decades these techniques have thrived through the implementation and expansion of Machine Learning, both to process and handle large amounts of data, as well as to carry out advanced computations. Specifically, and as the purpose of this document, we will be focusing on the Gradient Boosting algorithms from the sub-family of ensemble methods used for regression to predict and model basic pricing variables such as frequency and claim severities, and compare their predictive and pricing capabilities with classical Generalized Linear Models. In our study case of a French insurance motor portfolio, we found that Gradient Boosting models have a stronger predictive performance and a higher pricing ability to adjust the premiums to both high risk and low risk profiles. And finally, we conclude that these models can be used to support and improve GLMs and their pricing results as Machine Learning continues to settle in the actuarial modeling paradigm.
650 4‎$0‎MAPA20080592011‎$a‎Modelos actuariales
650 4‎$0‎MAPA20080588953‎$a‎Análisis de riesgos
650 4‎$0‎MAPA20080592059‎$a‎Modelos predictivos
650 4‎$0‎MAPA20080604127‎$a‎Tarificación a priori
650 4‎$0‎MAPA20170005476‎$a‎Machine learning
650 4‎$0‎MAPA20080573935‎$a‎Seguros no vida
650 4‎$0‎MAPA20080594589‎$a‎Análisis comparativo
650 4‎$0‎MAPA20160001693‎$a‎Modelos GLM
650 4‎$0‎MAPA20210037177‎$a‎Modelos GBM
650 4‎$0‎MAPA20080664510‎$a‎Trabajos de investigación
7001 ‎$0‎MAPA20140014897‎$a‎Rodríguez-Pardo del Castillo, José Miguel
7001 ‎$0‎MAPA20160001525‎$a‎Simón del Potro, Jesús Ramón
7102 ‎$0‎MAPA20080455026‎$a‎Universidad Carlos III de Madrid
830 0‎$0‎MAPA20160014013‎$a‎Trabajos Fin de Master
856  ‎$q‎application/pdf‎$w‎1113083‎$y‎Recurso electrónico / Electronic resource