Scoring a Risk With No Loss History and No Agreed Scale
A specialty insurer is assessing a new risk: mispricing of a parametric weather product launched eight months ago. There is no internal loss history. The 5x5 matrix defines impact by financial bands, with the top band at 'over USD 20m'. The likelihood scale defines 4 as 'likely within 12 months' and 5 as 'almost certain'. The product has USD 400m of notional exposure and the modeling team estimates a 1-in-50-year loss of USD 60m, but the model is new and unvalidated. The business head wants a score of 8, arguing the product is small relative to the group. The risk function notes that two peers have withdrawn similar products after losses. The assessment must be finalized for the quarterly ORM committee.
1. Recommend inherent likelihood and impact and explain how you handle the absence of internal loss history and the unvalidated model.
Read the worked answer
Impact should be 5. The modeled 1-in-50-year loss of USD 60m exceeds the top band of 'over USD 20m' by a wide margin, and the scale is defined by exposure, not by the size of the product relative to the group. The business head's argument that the product is small relative to the group is a portfolio-relative argument that the impact scale does not recognize; using it would silently redefine the scale. The unvalidated model is a reason to treat the USD 60m as uncertain, not a reason to discount it - model risk pushes the estimate up, not down.
Likelihood is harder. With eight months of operation and no internal losses, internal frequency data is uninformative. Use external evidence: peer withdrawals after losses on comparable products indicate the event is plausible within a year. A likelihood of 3 (possible) is defensible; 4 requires evidence that the drivers are present in this portfolio. Document the basis: external events, model uncertainty, and product novelty. Record the model validation gap as a separate issue with an owner.
2. The business head proposes 8 (likelihood 2 x impact 4). Explain the specific errors in that combination and how you would document the disagreement.
Read the worked answer
Two errors. First, impact 4 caps the impact below the modeled tail: a USD 60m loss is in band 5 on the firm's own scale. Choosing 4 appears to be reverse-engineering the score to reach a comfortable total. Second, likelihood 2 (unlikely) is asserted without support when the only available evidence - peer experience and an unvalidated model on a novel product - points the other way. A 2 x 4 combination on this evidence is not a judgment call; it is a score selected for its total.
The documentation should record the second line's score of 5 x 3 = 15, the evidence for each dimension, the business head's proposed score and stated rationale, and the reason it was not adopted. The disagreement should be escalated to the ORM committee rather than settled bilaterally. The record should also note that the model validation gap is tracked as an issue, because if the model is later validated downward the score can be revisited on evidence - which is the legitimate route to a lower score, not re-labeling the impact band.