Modeling the frequency of traffic accident compensation claims by jasa raharja using the GSTARX model

Abstract

Fluctuations in traffic accident frequency increase PT Jasa Raharja's exposure to claims and financial risks, while total compensation claims are stochastic because they are influenced by incident frequency and accident severity. Although the Generalized Space-Time Autoregressive with Exogenous Variable (GSTARX) model has been widely used to model spatial-temporal data, its application remains primarily in macroeconomics, simulation studies, and non-insurance phenomena. Hence, its application to model the frequency of traffic accident compensation claims, influenced by spatio-temporal relationships and external factors such as calendar variations, remains relatively limited. Therefore, this study aims to model and forecast the frequency of traffic accident compensation claims in four police districts, namely Batu Police District, Malang City Police District, Malang Police District, and Pasuruan Police District, for the period 2020-2024 using the GSTARX model. The data used are monthly secondary data from PT Jasa Raharja Malang Branch, with exogenous variables in the form of calendar variables. The analysis was conducted through data characteristic identification, stationarity testing, spatial-temporal correlation analysis using the Cross Correlation Function (CCF), model identification, location-weight determination, parameter estimation using Seemingly Unrelated Regression (SUR), and forecasting and model accuracy evaluation. The results of the study show that the GSTARX(2,1)(0,0,0) model with CCF(0)-based neighborhood location weights and cross-correlation normalization location weights produces relatively identical forecasting performance.