ROBUST AND SEMI-ROBUST CREDIBILITY MODEL FOR HEAVY-TAILED DATA

• Main Author: Christopher Aryento, Jevan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81486
• Institution: Institut Teknologi Bandung

Credibility model is a model that is used to calculate insurance pure premium by combining industry-standard pure premium with past claim data. The most common credibility model used are Classical Credibility Model, Bayesian Credibility Model, and B¨uhlmann Credibility Model. These credibility models have their advantages and disadvantages, but none of them can provide accurate and simple pure premium calculation for heavy-tailed data which is prone to outlier. Moreover, those models are using sample mean as their statistic to represent past data. Whereas general insurance data is prone to outlier, which will result in over-penalized premium. Hence, this research will provide 2 new credibility model, called Robust Credibility Model, which consist of sample median and sample upper quartile, and Semi-Robust Credibility Model, which consist of sample mean and sample median. Both models is more resistant to outlier. Both models will be tested with B¨uhlmann Credibility Model for 2 types of data, that is heavy-tailed (Pareto-distributed) and light-tailed (exponent-distributed) data. Based on the simulation, Semi-robust Credibility Model is the best model for heavy-tailed data, while B¨uhlmann and Semi-robust Credibility Model has near identical performance for light-tailed data. In conclusion, Robust and Semi-robust Credibility Model is an excellent alternative for heavy-tailed data, but not as good of an alternative for light-tailed data due to its complexity.