EARTHQUAKE INSURANCE CLAIM RESERVING BY APPLICATING NEURO-FUZZY INFERENCE SYSTEM: STOCHASTIC GRADIENT DESCENT METHOD

• Main Author: Cleverinto Sondakh, Raynold
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39295
• Institution: Institut Teknologi Bandung

An earthquake insurance business is a long-tail insurance business. In determining the outstanding claims liability of a long-tail insurance business, the Basic Chain Ladder method is commonly used because it is simple and distribution free. In this Final Project, a method to predict the outstanding claims liability using a Neuro Fuzzy Inference System with the Stochastic Gradient Descent optimization method is discussed. The architecture of the Neuro Fuzzy Inference System is modified to determine the development factors of the resulting runoff triangles of the claims data which are based on four characteristics of the claims data. Those characteristics of the underlying earthquakes are: moment magnitude; depth; latitude and longitude. The resulting chain ladder prediction is then compared with the Basic Chain Ladder prediction when the claims characteristics are not taken into account. Initially, it is assumed that the clustering will be based on the earthquakes cresta zones. However, the results of the analysis showed that the clustering is based on the depths of the earthquakes hypocenters.