Definition
Estimated proportion of current loss reserves that may be subject to emerging liability classes for which the ML IBNR model has no historical reporting pattern to learn from.
The Measurement Gap
ML IBNR models learn loss development patterns from historical data. They identify how claims emerge, develop, and settle for each liability class based on decades of experience. This works well for established liability classes where the development pattern is stable and well-understood.
The problem is that new liability classes emerge for which no historical development pattern exists. PFAS environmental liability, AI-generated content liability, social media defamation under homeowners policies, cannabis-related claims in states with recent legalization. These liabilities have no historical development triangle for the model to learn from. The model cannot estimate what it has never seen, and it does not know what it does not know.
Legacy reserve adequacy metrics test whether reserves are sufficient for known liability classes. They do not quantify the exposure sitting in liability territory where no model, ML or traditional, has a basis for estimation. An actuary reviewing reserve adequacy may conclude that reserves are adequate based on model output while an entire category of emerging liability goes unestimated.
ELBSI closes this gap by identifying active liability classes that have emerged since the ML model's last training date, estimating the exposure for each, and expressing the total as a percentage of carried reserves. This makes the unmodeled territory visible to the reserving committee and the board.
Decision Supported:
How much of the reserve book sits in liability territory where the ML model is structurally incapable of estimating IBNR, and should the actuarial team maintain a separate emerging-liability provision outside the model's output?
Formula
Plain Language: Identify active liability classes that have emerged since the ML model's last training date. Estimate the exposure (insured limit multiplied by probability of claim emergence) for each identified class. Express as a percentage of total reserves.
Formula:
ELBSI = Sum of (Exposure_e x P_emergence_e) / Total_reserves x 100
Variables:
- e: identified emerging liability class (e.g., PFAS environmental liability, social media defamation under homeowners, cannabis-related claims, AI liability)
- Exposure_e: total insured limits potentially exposed to liability class e (dollars, from policy in-force data)
- P_emergence_e: actuarial judgment estimate of probability that liability class e generates material claims within 3 years (0 to 1, necessarily subjective)
- Total_reserves: total carried reserves (dollars)
Units:
Percentage of total reserves potentially exposed to emerging liabilities the model cannot estimate.
Data Requirements:
| Data Element | Source System | Classification |
|---|---|---|
| Emerging liability class inventory | Legal/regulatory monitoring, industry publications (ISO, NAIC bulletins) | MANUAL |
| Insured limits by policy and coverage type | Policy administration system | NATIVE |
| Probability of emergence estimates | Actuarial judgment panel | MANUAL |
| ML model training date and liability classes included | Model documentation | MANUAL |
| Total carried reserves | Financial reporting / actuarial department | NATIVE |
Proposed formula, pending validation pilots.