THE PREDICTION OF TIME-DEPENDENT CLAIMS FREQUENCY OF HEALTH INSURANCE: RANDOM EFFECT AND ARTIFICIAL MARGINAL MODELS

• Main Author: Vercelli, Celline
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55201
• Institution: Institut Teknologi Bandung

Modeling the claims frequency (number of claims) is an important part in calculating the risk of financial losses. In this Final Project, the expected frequency of claims for the next year is determined based on a panel data of inpatient health insurance claims for underwriting years 2015, 2016, 2017, 2018, and 2019 from a general insurance company. Modeling the claims frequency is carried out using the characteristics of the policyholder, such as: gender; age; and types of diseases; and by using the time-dependent Poisson and Negative Binomial probability distributions which model the interdependence of the data between underwriting years. In this study, the predictions obtained using the Random Effect model, a model using the Bayesian approach, are compared with those obtained using the Artificial Marginal model, which is a model of which parameters are influenced by the characteristics and claims frequency in the previous period. Parameter estimates are obtained using the Maximum Likelihood Estimation method; and Wald’s test is used to determine which parameters are significant. The AIC and BIC values are used to determine the best model from several resulted regression models. For the historical data used, based on the obtained AIC and BIC values, it is found that the Random Effect model is the best model in predicting the claims frequency since its parameters, the characteristic factors which cannot be observed, are always updated in the computing process.