Bayesian Statistical Modeling Of Claim Frequency For Auto Insurance

Auto insurance has become a necessity for Malaysian, making it important to model the claims data so that premiums can be derived with fair and equitable price. This study aims to investigate the best model to fit the data for claims frequency of Malaysia vehicle insurance which composed of 1.21...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Muhammad 'Afif Bin Amir Husin
التنسيق: أطروحة
اللغة:English
الموضوعات:
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
الوصف
الملخص:Auto insurance has become a necessity for Malaysian, making it important to model the claims data so that premiums can be derived with fair and equitable price. This study aims to investigate the best model to fit the data for claims frequency of Malaysia vehicle insurance which composed of 1.21 million policies. These policies comprised of three types of coverages namely Own Damage (OD), Third Party Property Damage (TPPD) and Third Party Bodily Injury (TPBI). First, the frequency data is fit to Bayesian regression models using Poisson, Negative Binomial and Generalized Poisson distribution. After that, all three models are compared with their respective zero-inflated models to analyze the effectiveness of zero-inflated model in handling auto insurance claims data. For the purpose of measurement of good fit, two criteria have been chosen which are Deviance Information Criteria (DIC) and Watanabe-Akaike Information Criteria (WAIC). This research found that Generalized Poisson model outperformed other models evaluated using DIC while Zero-Inflated Negative Binomial have been superior to other models with respect to WAIC.