Insurance Loss Estimation Engine
A multi-peril loss estimation platform that calibrates and blends commercial CAT model outputs with internal claims experience for sharper portfolio loss estimates.
Hurricane · Earthquake · Flood
Perils
XGBoost calibration
Method
Post-event response packs
Output
Problem & use case
Single-vendor CAT model reliance led to systemic bias and large deviations vs realized losses.
Solution & engineering approach
An XGBoost-based calibration layer on top of RMS RiskLink outputs, trained on historical event-vs-actual loss data.
Architecture & data flow
- 01
Pull CAT model outputs across hurricane / earthquake / flood
- 02
Join with internal historical loss experience
- 03
Train XGBoost calibration model per peril
- 04
Blend calibrated outputs into portfolio view
- 05
Publish event response packs after live events
Technology stack
Implementation status
Implemented — XGBoost calibration layer on RMS model outputs with portfolio blending