An indication is only as trustworthy as the data feeding it. Before you defend a 12% rate increase, you have to prove the losses, premiums, and exposures behind it are complete, correctly coded, and internally consistent.
KEY: Actuarial Standard of Practice (ASOP) No. 23, Data Quality, has the actuary consider whether data are appropriate for the intended purpose and whether they are reasonable, comprehensive, and consistent, and it calls for reviewing the data for reasonableness and consistency.
Reasonableness. Values should make sense for what they record, so look for entry and coding errors. A claim coded to the wrong accident year, a premium keyed as $1,200 instead of $12,000, or losses recorded gross when the analysis expects net all corrupt the indication.
Comprehensiveness. Every data element and record the analysis needs must be present. Missing months, a data pull that cut off mid extract, or claims that occurred but are not yet reported (a gap in immature years) all understate losses.
Common mistakes
- Trusting an internal total without an external tie. A tidy extract sum of $12,000,000 still hid a dropped territory; only the ledger reconciliation to $12,400,000 caught it.
- Mixing loss and exposure bases. Pairing accident-year losses with written exposures instead of earned produces a ratio that means nothing.
- Averaging away an outlier. Blending a contaminated 1.40 development factor with clean ones buries a real data error inside the selection.
Bottom line
- ASOP 23 has the actuary consider whether data are appropriate for the purpose, reasonable, comprehensive, and consistent, and review them for reasonableness and consistency.
- Reconcile premium and loss data to the company's financial statements (Annual Statement, general ledger) before trusting it.
- Pure premium must equal frequency times severity; if the two paths disagree, a count or dollar field is wrong.
- Loss ratios, frequency, and severity should trend smoothly year to year; explain any jump with a real cause (large loss, law change, coverage change) or treat it as an error, never as noise.
Exam shortcut
When a table has one number that breaks the pattern, that outlier is the answer. Name a real cause (large loss, law change, rate change, mix shift, coding change) or call it an error; never average it in. Always run the frequency-severity identity. If frequency times severity does not equal losses over exposures, stop and hunt the corrupted count or dollar field before anything else.
The full lesson (about 3,247 words, 22 min read) adds 3 worked examples, all 8 common mistakes, a self-check, free in the app.
Learning objectives
- A4
Browse all free Exam 5 lessons or jump into free Exam 5 practice questions.