Data, modelling and communication of results
Domain 4 tests data quality, model design and validation, and communication of results to non-actuaries.
Data
Check accuracy, completeness, consistency, timeliness, granularity and relevance. Reconcile to accounts, investigate outliers, and document missing-data treatments and proxies. Poor data is a limitation to report, not a problem to bury.
Modelling
Choose deterministic versus stochastic, asset versus liability, and the granularity the decision needs. Document assumptions and their basis, test sensitivity, validate against experience and independent review, and record model limitations. TAS 100 requires reporting of data, assumptions, methods, results, uncertainty and limitations.
Communication
Write for the audience: decision-makers need the conclusion, the uncertainty and the implications, not a model dump. State material judgements and their effect.
Worked example
A claims extract holds 1,000 claims totalling £2,500,000, so the naive mean is £2,500. Reconciliation identifies 40 duplicate records totalling £200,000. Corrected: £2,300,000 over 960 claims = £2,395.83, a 4.2% reduction. Reasoning: the duplicates are not random - they cluster in one handler's cases - so the corrected mean is more credible, but the process failure must also be reported to the data owner.
Traps
Reporting a point estimate without uncertainty; assuming a more complex model is more accurate; failing to document judgements; over-fitting a short history; answering the actuary's question rather than the client's.