Building and Using Models
Domain 3: Building and Using Models
CB2 tests how to construct, parameterise, run and interpret actuarial models. It expects you to choose between deterministic and stochastic approaches, document assumptions and communicate output distributions rather than a single number.
Model purpose and structure
Start from the decision the model must support. Define the cashflows, time horizon, states and interactions. A deterministic model gives one outcome per scenario; a stochastic model gives a distribution by sampling random drivers. Monte Carlo simulation is common but requires a pseudo-random number generator, a seed and enough simulations for stable percentiles.
Assumptions and parameters
Separate data-driven parameters from expert judgement. Set a base case, then vary one assumption at a time in sensitivity analysis or vary several coherently in scenario analysis. Stress tests examine severe but plausible conditions.
Worked example
A one-year warranty product has premium £100, expected claims £60 and expenses £15. Both cashflows occur on average halfway through the year, and the discount rate is 5% per annum. Discount the expected outgo: (60 + 15) / 1.05^0.5 = 75 / 1.0247 = £73.19. Expected profit is 100 − 73.19 = £26.81. Now increase claim frequency by 10%: claims become £66, discounted outgo becomes (66 + 15)/1.0247 = £79.05, and profit falls to £20.95. The sensitivity shows profit is materially exposed to claim frequency, so the base case alone is insufficient.
Traps
- Mixing nominal and real cashflows or using the wrong discount period.
- Treating a deterministic result as a forecast rather than one scenario.
- Ignoring parameter uncertainty and model error.
- Double counting the same risk in both expenses and claim assumptions.
- Using too few simulations and reporting unstable tail percentiles.