MODELLING INCURRED BUT NOT REPORTED CLAIM COUNTS USING POISSON INTEGER-VALUED AUTOREGRESSIVE(1) PROCESSES WITH CONDITIONAL LEAST SQUARES ESTIMATION AND ITERATIVE WEIGHTED CONDITIONAL LEAST SQUARES ESTIMATION

• Main Author: Guikajaya, Julianto
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42398
• Institution: Institut Teknologi Bandung

Chain-Ladder and Bornhuetter-Ferguson are methods which are often be used to predict Incurred But Not Reported Claims Reserve. However, because those two methods don’t assume that there is a probability distribution in the claim data which make the prediction result we get isn’t accurate, so it is necessary to consider a parametric model for predicting IBNR reserve. By modifying Kremer’s model(1995), Yang Bai(2016) made a model to predict unclosed claim numbers which is called Integer-Valued Autoregressive(1) Poisson Processes. In this final project, that model will be used to predict unclosed claim numbers. Conditional Least Squares Estimation and Iterative Weighted Conditional Least Squares Estimation Method will be considered to be used in estimating parameter in model. Those two estimation methods will be compared and the best estimation method will be chosen from those two estimation methods by comparing the effectiveness of the two estimation methods from the estimation results we get by using those two methods. After choosing the best estimation method, we will predict the unclosed claim numbers by using the estimation result we get from the best estimation method. The result we get from this final project are Iterative Weighted Conditional Least Square is the best estimation method from two estimation methods and the prediction error the model get is quite small so this model can be considered to be used in predicting unclosed claim numbers. However, it is needed to be watched out that the prediction error we get is underestimated because we ignored parameter estimate error in calculating prediction error.