Skip Ribbon Commands
Skip to main content
Sign In

Use Cases

 Content Editor ‭[3]‬

Use Cases

The working groups of the Actuarial Data Science committee are developing various use cases, which are presented below.

 

Neural Networks Meet Mortality Prediction

Working Group *Statistical Methods* of the Actuarial Data Science Committee, 15.09.2020


What added value can neural networks provide for predicting life expectancy in multiple populations? To address this question, we first build a database with mortality rates for Japan, USA, Germany and six other European countries. In further steps, a classical mortality model is computed for each of these populations and the deep artificial neural network presented in Richman and Wüthrich (2018) is trained across populations.

Using these models, mortality predictions for different time periods are generated and compared. It is shown that for the majority of the populations studied, this cross-population neural network has a better forecasting ability compared to a "plain vanilla" Lee-Carter model.

Further information
 

Use (this Solvency II) case! Neural Networks Meet Least Squares Monte Carlo

Working Group *Statistical Methods* of the Actuarial Data Science Committee, 09/15/2020


The topic of artificial intelligence is on everyone's lips right now and the application areas are becoming more and more popular. In this case study, we want to address the risk capital determination of insurance companies in the Solvency II context and we will challenge the classical Least Squares Monte Carlo approach with neural networks. The realistic projection data generated specifically for this purpose from three life insurance and health insurance portfolios prepared in the context of this use case form the central component of this use case.

 w

GLM, Deep Learning and Gradient Boosting in Claims Pricing, Part 1.

Actuarial Data Science Committee in collaboration with the Non-Life Insurance Committee's Pricing Methodology Working Group, 04/23/2020.


What added value can machine learning methods provide for claims pricing? To address this question, loss frequency models are computed for a French auto insurance portfolio using the above methods, cross-validated, and the prediction results are compared. Practical aspects such as tariff organization are also considered. It is shown that "Deep Learning" and "Gradient Boosting" can at least be used to improve classical models. By far the most accurate forecasts are achieved here with "Gradient Boosting".

Further informationther i

Further informationwd

 SpLoginControl - LoginControl

Login


Adressnr.:
Passwort:

 Content Editor ‭[2]‬

GitHub account

This link  will take you to DAV's GitHub account.



 Content Editor ‭[1]‬

DAV
DGVFM
IVS

Contact
Imprint
Data privacy

Become an actuary
Membership information
International involvement

Annual meetings
Autumn meetings
ICA 2018

Practice areas
The European Actuary
European Actuarial Journal