Incurred but not reported (IBNR) claims estimation using machine learning techniques
- Author
- Manatsa, Emmanuel
- Title
- Incurred but not reported (IBNR) claims estimation using machine learning techniques
- Abstract
- This study investigates the estimation of Incurred But Not Reported (IBNR) claims in Zimbabwe’s non-life insurance sector by comparing traditional actuarial methods with modern machine learning (ML) techniques. While the Chain-Ladder and Bornhuetter-Ferguson methods have long been used for IBNR forecasting, they often fall short in adapting to non-linear, dynamic claim behavior. This research employs Random Forest, Gradient Boosting Machine (GBM), and Long Short-Term Memory (LSTM) models to evaluate their predictive accuracy against conventional approaches. Using historical claims data from selected insurers, the models were assessed using MAE, RMSE, and MAPE performance metrics. Results show that ML models, particularly GBM, outperform traditional methods in predictive accuracy, although concerns about interpretability and regulatory acceptance remain. The study concludes that while traditional models provide transparency and simplicity, ML methods offer superior adaptability and forecasting power. It recommends a hybrid approach, combining actuarial insights with ML innovation, as a pathway to improved reserving accuracy, financial solvency, and regulatory compliance in emerging insurance markets like Zimbabwe
- Date
- June 2025
- Publisher
- BUSE
- Keywords
- Machine Learning Techniques
- IBNR Claims Estimation
- Supervisor
- Dr. Magodora
- Media
-
Manatsa, Emmanuel.pdf
Part of Incurred but not reported (IBNR) claims estimation using machine learning techniques