Penerapan Model Regression Tree Credibility (RTC) dalam Estimasi Loss Ratio sebagai Dasar Penetapan Premi Asuransi Kendaraan Bermotor

Anifa Hashilah, Dr. Adhitya Ronnie Effendie, S.Si, M.Si., M.Sc.

2026 | Skripsi | S1 ILMU AKTUARIA

Loss Ratio merupakan salah satu indikator utama dalam mengevaluasi kinerja underwriting dan kecukupan premi pada asuransi kendaraan bermotor karena mencerminkan proporsi klaim terhadap premi yang diterima. Namun, prediksi loss ratio seringkali sulit dilakukan akibat heterogenitas portofolio dan keterbatasan volume data pada masing-masing kelompok risiko. Penelitian ini menerapkan model Regression Tree Credibility (RTC) untuk memprediksi loss ratio dengan mengelompokkan risiko ke dalam rating cell homogen berdasarkan variabel yang berlandaskan regulasi OJK (SEOJK No.6/SEOJK.05/2017) untuk menghasilkan estimasi loss ratio yang menyeimbangkan pengalaman klaim lokal dan estimasi kolektif. Menggunakan data klaim kendaraan bermotor periode 2019–2024, loss function dari model Regression Tree Credibility (RTC) dibandingkan berdasarkan Relative Prediction Error (RPE). Hasil menunjukkan model Regression Tree Credibility (RTC) memberikan prediksi loss ratio dengan RPE terendah dibandingkan model Bühlmann–Straub tanpa partisi regression tree biasa. Estimasi loss ratio yang dihasilkan selanjutnya dapat menjadi jembatan untuk indikasi penyesuaian premi melalui Loss ratio Method sehingga penelitian ini memberikan kontribusi praktis dalam mendukung evaluasi risiko dan penetapan premi asuransi kendaraan bermotor berbasis data.

Loss ratio is a key indicator for evaluating underwriting performance and premium adequacy in motor vehicle insurance, as it reflects the proportion of claims relative to premium received. However, predicting loss ratio is often challenging due to portfolio heterogeneity and limited data volume within individual risk groups. This study applies the Regression Tree Credibility (RTC) model to predict loss ratio by partitioning risk into homogeneous rating cells based on variables grounded in OJK regulation (SEOJK No.6/SEOJK.05/2017), then applying the Bühlmann–Straub credibility model within each rating cell to produce loss ratio estimates that balance local claims experience against the collective estimate. Using motor vehicle claims data from 2019–2024, the four loss functions of the RTC model were compared based on Relative Prediction Error (RPE). The results show that the RTC model produced the lowest RPE outperforming both the unpartitioned Bühlmann–Straub model and the conventional regression tree. The resulting loss ratio estimates were then used as a bridge to premium adjustment indications through the Loss Ratio Method, providing a practical contribution toward data-driven risk evaluation and premium setting in motor vehicle insurance.

Kata Kunci : Loss Ratio, Premium, Regression Tree, Kredibilitas Bühlmann–Straub