Olivier Lopez, UPMC Paris VI

 

 

In this short course, we will present how the use of survival analysis techniques may help to predict the amount of RBNS claims (Reported But Not Settled). By taking account for the censoring and truncation present in the databases, we will see how to model the dependence between the amount of a claim and the time before its settlement, and how to use it to estimate the final amount. We will also consider closely related problems (evaluation of the customer value in a simple framework).
 
I - A simple censoring model for RBNS claims
               1) Description of the model
               2) Specificities compared to a classical survival analysis problem
 
II - Copula modeling
               1) Copula model and estimation
               2) Prediction method
 
III - General censoring and truncation
               1) General bivariate censoring/truncation model
               2) Dependent censoring
               3) Another example: customer value in online subscription
 
IV - Machine learning techniques
               1) Regression trees
               2) Random forests
 

 

 

Thursday 17/05/2018 : 14:00 - 18:15

Friday 18/05/2018 : 14:00 - 18:15