An Example in Reserving

Below is the modeled reserve distribution for 57 workers compensation companies as of year-end 1997. Drag the vertical blue line to choose a desired reserve. For demo purposes, the reserve set by the industry ($4.7B) is added alongside the actual reserve requirement based on future 10th year incurred values.

During the tested period, many companies experienced a quick decline in loss ratio. The Proxima model captured this with only 45 data points per triangle. The industry, on the other hand, was over-reserved by over a billion dollars, lying far outside the Proxima predicted range.

Review a Business Segment

The Proxima model captured unique development patterns and trends across the industry for both paid and incurred loss ratios using only public information.

Validation

All models are tested on point and distributional accuracy metrics for both paid and incurred loss projections. The 57 demo results are compared to the Mack Chain Ladder method on weighted root mean squared error (WRMSE), negative log predictive density (NLPD), 90% coverage probability (COV 90), and the Kolmogorov-Smirnov distributional test (KS). Proxima shows a clear elevation in performance across all metrics. On the right, a P-P plot shows the observations are well-distributed throughout the predicted distributions.