Evaluated predictive models forecasting health insurance case reserves

Evaluated predictive models forecasting health insurance case reserves

Our client wanted to accurately forecast the incurred but not reported (IBNR) reserves to have reliable estimates of monthly costs. 

We developed a framework where we compared several actuarial and advanced statistical methods and delivered a detailed report and predictions by completing these tasks: 

  • Evaluated existing completion factor methods used to forecast IBNR reserves

  • Compared and applied several deterministic (chain ladder, paid PMPM, separation methods, etc.) and stochastic (Mack, Munich CL, etc.) actuarial methods

  • Provided working prototypes of selected actuarial methods and machine learning techniques, such as neural networks

  • Created reusable and flexible software in statistical programming tools (R, Python) that implemented advanced statistical methods, such as generalized linear models

The client was very pleased with our recommended statistical methods and working prototypes, which predicted forecasting accuracy increase by 30% and better confidence intervals.