TOTAL ESTIMATION OF CLAIM RESERVES USING DEEP NEURAL NETWORK

• Main Author: Samsir, Rusni
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68346
• Institution: Institut Teknologi Bandung

Claim reserve estimation plays an essential role for insurance companies. Better estimation can affect the profitability of insurance companies, and worse estimates can cause serious consequences. Claims reserves are divided into two: Incurred But Not Reported (IBNR) and Reported But Not Settled (RBNS). Generally, claim reserve estimations are based on run-off triangle data. The run-off triangle data contains an overview of the overall claims and summarizes a data set of individual claims. Chain Ladder, Double Chain Ladder, and Bornhuetter-Ferguson Double Chain Ladder methods are several estimation methods that are pretty well known in estimating claim reserves using aggregate data. Much individual information is lost during the formation of the aggregated data included in calculating claim reserves. Therefore, this study aims to predict the total claims reserve by using the Deep Neural Network approach on individual claim data and determine what components can be controlled in order to obtain an estimated claim reserve with a lower risk. In conclusion, the estimation of the total claims reserve using Deep Neural Network has a lower risk than that of the total claims reserve using the aggregate method. In addition, the influential component in total claims reserve estimation is the recording of claims reporting. Hence it is expected that error in the recording of claims reporting is not more than 10% so that the estimation results of the total claims reserves have a smaller risk.