Application of actuarial judgement to complex, unstructured general insurance problems
Unstructured problems are those where the data does not answer the question directly - a new liability, a novel product, a one-off capital decision. The exam tests method and defensible judgement, not a textbook formula.
Method
- Frame the decision: who decides, what materiality, what horizon, what precision changes the answer. A range that changes no decision is wasted work.
- Build from exposure, not only experience: insured values, limits, attachment points, penetration and wordings.
- Triangulate with analogous classes, calibrated expert elicitation with documented rationale, scenarios and stress tests, and market benchmarks.
- Show sensitivity, identify the dominant assumption, recommend a point estimate inside a range, and say why.
Worked example
You must estimate silent cyber exposure in a commercial property book with no usable loss data. Steps: identify policies where cyber damage could trigger cover and review exclusion wording; estimate the proportion of sums insured exposed (say 15%); apply a damage ratio from cyber incidents (say 20% of exposed value); attach through policy excesses; aggregate assuming some correlation. With 400m of exposed sums insured, expected gross loss is about 400 x 0.15 x 0.20 = 12m before excesses and non-modelled events. Present the range, flag correlation and wording uncertainty, and recommend a reserving or exclusion action.
Traps
Fitting a chain ladder to two data points; hiding judgement behind unexplained expert opinion; giving a single number with no range; ignoring that wordings can be changed.