Problem Framing, Assumption Setting and Modelling
Domain 2 examines how you behave when the question is deliberately under-specified: a business problem, partial data, a limited word count. Marks reward choosing a method that fits the decision, setting assumptions defensibly and expressing uncertainty, not producing the longest calculation.
Framing the problem
- State the decision being supported, the time horizon, the stakeholder and the materiality threshold before choosing a technique.
- Match the model to the decision: deterministic for a single known scenario, stochastic or simulation where the shape of the distribution drives the answer, proxy or curve-fitting models where full runs are impractical.
- Test data for volume, homogeneity, credibility and relevance; historical experience may not survive a change in distribution channel or regulation.
Setting assumptions
- Separate best-estimate, prudent and pricing bases; name which one the question requires.
- Apply credibility weighting where the portfolio is thin, blending own experience with a benchmark.
- Document expert judgement, then evidence it with sensitivities and a reasoned central estimate.
Worked example
A block of 2,000 policies in force for one year shows 180 lapses, a crude rate of 9%, against an industry benchmark of 6%. With a credibility factor of k = 1,000 policies, Z = 2,000 / (2,000 + 1,000) = 0.67, so the assumption is 0.67 × 9% + 0.33 × 6% ≈ 8.0%. The reasoning continues: on a unit-linked book, higher lapses cut future fund-charge income and release reserves, so the sign of the impact depends on whether reserves exceed the value of future charges. Because the block is new, persistence is highly uncertain, so the answer should present 8% as the central assumption with a sensitivity at 12% and explain which way surplus moves.
Traps that cost marks
- Using a crude count of events without adjusting for exposure.
- Assuming the direction of a lapse or mortality change without tracing the cash flows.
- Giving a single number with no sensitivity or range.
- Building an elaborate model when a simple closed-form estimate answers the question.
- Failing to record the judgement, so a reviewer cannot follow the reasoning.