Rémi Bellina

La capacité à accéder à de nouvelles sources de données offre de nombreuses perspectives concernant la gestion des risques des assureurs. Alors que les Insurtechs transforment le marché de l’assurance, tant sur le plan des services et de l’expérience utilisateur que de la création de produits innovants, les assureurs traditionnels misent également sur la donnée pour enrichir leur activité et la connaissance de leurs risques.

D’après l’étude Impact du changement climatique sur l’assurance à l’horizon 2050 produite par FA (France Assureurs), le coût des sinistres climatiques devrait doubler sur la période 2020-2050. Ce document est une synthèse des analyses produites par Milliman en utilisant différentes sources de données (notamment : les arrêtés CatNat de la CCR, les données ERA5 de l’ECMWF, le bilan climatique du printemps 2022 de Météo France, les données RGA de la BRGM et de l’ESDAC et ADMIN-EXPRESS de l’IGN).

Every day, more than 2.5 quintillion bytes of data are created, and that pace is only accelerating. Big data, combined with the increased usage of machine learning algorithms, allows the life insurance industry to model the surrounding world much more effectively than in the past. Insurers can better understand underwriting risks, improve customer service, and lower costs. Better data can also improve the health of customers. In this report, we explore in detail what kind of data can be used to… Voir plus

Every day, more than 2.5 quintillion bytes of data are created, and that pace is only accelerating. Big data, combined with the increased usage of machine learning algorithms, allows the life insurance industry to model the surrounding world much more effectively than in the past. Insurers can better understand underwriting risks, improve customer service, and lower costs. Better data can also improve the health of customers. In this report, we explore in detail what kind of data can be used to achieve these goals, and its limitations. Voir moins

Le marché des insurtechs est en plein essor, en particulier en France. Dans cet article, nous présentons un panorama du secteur, l’apport de ces nouveaux acteurs sur l’offre assurantielle, les enjeux liés à la mise en place d’un dossier d’agrément, et nous concluons sur le rôle des actuaires dans ce nouvel écosystème.

Pour illustrer toute la richesse et les enjeux de ce marché des insurtechs, nous avons eu le plaisir d’échanger avec Samuel Falmagne (CEO et Cofondateur d’Akur8), Philippe… Voir plus

Le marché des insurtechs est en plein essor, en particulier en France. Dans cet article, nous présentons un panorama du secteur, l’apport de ces nouveaux acteurs sur l’offre assurantielle, les enjeux liés à la mise en place d’un dossier d’agrément, et nous concluons sur le rôle des actuaires dans ce nouvel écosystème.

Pour illustrer toute la richesse et les enjeux de ce marché des insurtechs, nous avons eu le plaisir d’échanger avec Samuel Falmagne (CEO et Cofondateur d’Akur8), Philippe Mangematin (Cofondateur de Seyna et de Stoïk, et consultant auprès d’insurtechs) et Tanguy Touffut (CEO et Cofondateur de Descartes Underwriting) dont les interviews sont retranscrites dans cet article. Voir moins

This research paper first presents the technical foundations and key principles of data analytics techniques, then outlines use cases that have been developed by life insurers, reinsurers and insurtechs. All of these cases have proven their ability to generate business value in a broad scope of business applications. We aim to further thinking on how life insurers could leverage data analytics to support their business ambitions. We also present feedback from life insurance industry data… Voir plus

This research paper first presents the technical foundations and key principles of data analytics techniques, then outlines use cases that have been developed by life insurers, reinsurers and insurtechs. All of these cases have proven their ability to generate business value in a broad scope of business applications. We aim to further thinking on how life insurers could leverage data analytics to support their business ambitions. We also present feedback from life insurance industry data leaders, sharing their experience and convictions on artificial intelligence and data analytics. Voir moins

Modelers are trying to anticipate the future of the COVID-19 epidemic based on relevant parameters driving its evolution. This paper explores the potential and the challenge of using BERT, a Natural Language Processing framework, to automate the task of gathering input information and assisting experts for COVID-19 studies.

This paper explores data science applications in reserving from data management through to analysis and all the way to reporting.

French version also available.

The digital revolution has been disrupting industries for some years now. At a time when users increasingly expect trust, responsiveness and transparency, the real challenge is often to provide new services adapted to their emerging needs. The insurance world is not immune to this revolution and telematics is an excellent illustration of this change initiated by insurtech companies. This article focusses on the future of motor insurance and telematics, providing feedback on some of the projects… Voir plus

The digital revolution has been disrupting industries for some years now. At a time when users increasingly expect trust, responsiveness and transparency, the real challenge is often to provide new services adapted to their emerging needs. The insurance world is not immune to this revolution and telematics is an excellent illustration of this change initiated by insurtech companies. This article focusses on the future of motor insurance and telematics, providing feedback on some of the projects led by Milliman's analytics team.

Mandarin version also available Voir moins

Le marché de l’assurance a rarement été marqué par un environnement aussi difficile qu’au cours de ces dernières années. Cette situation met encore davantage en lumière la nécessité au sein des compagnies d’effectuer les diagnostics pouvant aider le management dans sa prise de décisions. L’utilisation de méthodes innovantes et rapides à mettre en oeuvre comme les méthodes d’apprentissage statistique (ou « machine learning ») peut être un avantage compétitif déterminant dans ce contexte.

Insurance pricing is a core business for insurance companies. Our aim here is to tackle the main methods used in non-life pricing and in particular within automobile insurance. The issue at stake is to have a better understanding of the ins and outs of machine learning systems applied to pricing. We compare them to more classical methods, like the very widespread generalized linear model (GLM). To begin with, we highlight a general mathematical framework for estimators in statistics. This… Voir plus

Insurance pricing is a core business for insurance companies. Our aim here is to tackle the main methods used in non-life pricing and in particular within automobile insurance. The issue at stake is to have a better understanding of the ins and outs of machine learning systems applied to pricing. We compare them to more classical methods, like the very widespread generalized linear model (GLM). To begin with, we highlight a general mathematical framework for estimators in statistics. This enables us to compare the different methods. In the first part we present the automobile database used, which gathers half a million of insured. We also underline how to deal with outliers in the data. The second part focuses on the GLM. The aim of the third and fourth parts is to lay down the principles and the full computation of machine learning methods. We specifically go on about the classification and regression trees (CART) and the ensemble methods like bagging, random forests, and boosting. In the last part we eventually draw an analysis between all the results. We conclude with the advantages of the new non-parametric approaches based on trees. They are indeed easy to implement and they offer a synthetic vision of the insurance portfolio. Yet one needs to be cautious with these results. The high performance of the machine learning methods is linked to the database. If the data are the exact realization of a given parametric distribution then a regression model will fit almost perfectly. However, machine learning procedures give an opportunity to better understand the underlying risks of the portfolio, offering a new framework. Voir moins