SPATIAL MODELLING OF CLAIM FREQUENCY IN NON-LIFE INSURANCE

• Main Author: (NIM: 10113092), DEVINA
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/21674
• Institution: Institut Teknologi Bandung

Premium rating in non-life insurance need two important components which are <br /> <br /> <br /> claim frequency and claim size. The usage of Generalized Linear Mixed Model <br /> <br /> <br /> (GLMM) is to model a linear relationship between characteristics of a policy <br /> <br /> <br /> holder and its claim frequency by log link. GLMM model includes spatial random <br /> <br /> <br /> effect to be able to detect the spatial dependency pattern with random spatial <br /> <br /> <br /> effect is assumed to have Conditional Autoregressive (CAR) distribution. Bayesian <br /> <br /> <br /> approach is used to estimate parameters with Markov Chain Monte Carlo (MCMC) <br /> <br /> <br /> algorithm because of the inclusion of spatial random effect in model. Deviance <br /> <br /> <br /> Information Criterion (DIC) and Predictive Model Choice Criterion (PMCC) are <br /> <br /> <br /> quantification to choose better model between model with and without spatial <br /> <br /> <br /> random effect. Model is applied to a dataset by assuming the observed spatial <br /> <br /> <br /> structure because it’s not given. PMCC says model with random spatial effect is <br /> <br /> <br /> a better model but not by DIC, one of the cause is the assumptions that are used but <br /> <br /> <br /> model with spatial random effect could predict claim frequency in each area better.