Claims Coding Changes

Study guide: Exam 5

Claims Coding Changes are changes in how claims are categorized in the claims system — coverage codes, cause-of-loss codes, claim vs. claimant counting, the treatment of reopened claims, or the line to which a claim is assigned. They move data between cells without changing any underlying cost.

\text{Apparent change} = \text{True change} + \text{Reclassification effect}

• The signature is a sharp, one-time shift concentrated at a single valuation date, offset by a corresponding shift somewhere else. Losses do not appear or disappear in aggregate; they move.
• Coding changes primarily corrupt segment-level analyses. A pure reclassification leaves the total reserve indication unaffected while making the affected lines' triangles unusable — which is dangerous precisely because the aggregate check passes.
• Claim count definitions are the most common casualty: counting per claimant instead of per claim, beginning or ceasing to count claims closed without payment, or treating a reopened claim as new. Any of these makes a count triangle discontinuous and produces a spurious frequency trend.
• Distinguish from Claims Processing Changes, which alter the timing of payments and reserve changes and so change the shape of the development curve. Coding changes alter allocation.
• The remedies, in order of preference: recode the history consistently, map the old and new schemes and restate, or (last resort) select factors from post-change diagonals only and accept the loss of history.
• This is exactly what Friedland's Chapter 4 management interview is for. A coding change is invisible in the triangle until it has already distorted the factors, and it is trivially identifiable by asking.

A Coverage Reclassification An insurer migrates claims systems on 1/1/2023. Slip-and-fall claims at insured premises were coded "GL — premises" in the old system and "GL — operations" in the new one.

What happens to the two sub-line triangles?

Answer The premises triangle shows a sharp drop starting on the 12/31/2023 diagonal; the operations triangle shows a matching rise. The combined GL triangle is unaffected.

An actuary selecting factors from either sub-line's full history blends two definitions. Premises development factors will be understated (the recent diagonal is artificially low relative to its own history) and operations overstated.

Fixes: recode the historical claims to the new scheme, or select factors for the affected sub-lines from post-migration data only. Where neither is possible, the honest answer is to analyze GL in total — the level at which the reclassification nets out — and allocate the result, disclosing that the sub-line split is not supported by the data.

A Claim Count Definition Change Reported claim counts and reported losses at 12 months:

AY	Counts	Losses	Severity
2021	2{,}000	\$9{,}000K	\$4{,}500
2022	2{,}060	\$9{,}500K	\$4{,}612
2023	2{,}120	\$9{,}900K	\$4{,}670
2024	2{,}950	\$10{,}300K	\$3{,}492

Diagnose.

Answer Counts jumped 39\% in 2024 while losses grew only 4\%, so average severity fell 25\%. A genuine 39\% frequency increase with no corresponding loss growth is close to impossible — the extra claims would have to be almost costless.

That is the signature of a counting change: the insurer has started counting something it did not count before, most likely claims closed without payment, or claimants rather than claims.

The consequences run through everything built on counts:

• Frequency-severity projections break — the frequency series has a step change and the severity series has an offsetting one.
• Count development factors are corrupted for any year spanning the change.
• Average case outstanding, the main case adequacy diagnostic, is divided by a redefined denominator and will appear to fall.

The fix is to obtain counts on a consistent definition — usually available by re-extracting with the old filter — before any count-based analysis is run.