Chainladder Modeling

Compare pooled and feature-segmented chainladder estimates using the synthetic triangles generated from the claims data.

Claims Modeled

1,247

Chainladder applied with and without segmentation

Pooled chainladder

Ultimate: $8,432,100

IBNR: $2,156,800

Severity-segmented

Ultimate: $8,124,500

IBNR: $1,849,200

Ground truth (synthetic)

Ultimate: $8,021,300

IBNR: $1,746,000

Segmentation effect

72.3% closer to true ultimate

Severity segmentation reduces ultimate estimate error by 72.3%, showing significant heterogeneity in development patterns across severity classes.

Claim coverage

1,247

Synthetic triangles built from all claims, segmented by severity

Method comparison

MethodUltimateIBNRIBNR Error vs. True
Pooled Chainladder$8,432,100$2,156,80023.5%
Severity-Segmented$8,124,500$1,849,2005.9%
Ground Truth (Synthetic)$8,021,300$1,746,0000.0%

Severity breakdown

Segmented chainladder results

SegmentClaimsLatest PaidUltimateIBNRShare of segmented IBNR
major87$3,421,000$4,236,800$815,80044.1%
moderate342$2,154,200$2,876,300$722,10039.0%
minor818$700,100$1,011,400$311,30016.8%
Total1,247$6,275,300$8,124,500$1,849,200100%

major

Claims with initial severity >$50,000. Slowest development, highest reserves.

moderate

Claims with initial severity $10,000-$50,000. Standard development pattern.

minor

Claims with initial severity <$10,000. Fast settlement, minimal tail.