Q1
§3.5 — does the filing/planning split earn its place, and what do we call it?
✓ You askedMondello why the filing basis is revenue-neutral, and said you could not follow the section. ✓ We answered it live on 21 August — hold the overall rate level neutral and redistribute on CLV — and Robinson then put the harder question: “I still question the filing basis versus planning basis here, and whether that’s necessary.” The researcher’s own view on the call was that filing and planning are confusing words for what they name.
Now pick what §3.5 becomes in the delivered paper.
What the two bases actually do, at v3.3. Same tenure curve, two normalisations. On the revenue-neutral basis the book total is unchanged to $0.09 of rounding, so the effect is pure redistribution: the 1 yr cohort +2.85% and the 10+ yrs cohort -7.43% — the tenure credit runs backwards. On the level-effect basis every cohort gains (1 yr +17.28%, 10+ yrs +4.84%) and the book rises by $2,615,287. The missing sentence, which Robinson reasoned out on the call: the flat-loss-ratio comparator is the tenure-average loss ratio, not the first term's — so a ten-year customer has already banked the improvement and a first-year one has it all ahead. That is the whole reason the gradient is steepest at low tenure. It appeared nowhere in the manuscript through v3.2; it is now §3.5's “What the comparison is against” paragraph.
A
Keep both bases, rename them — revenue-neutral relativity basis and level-effect basis. The words then say what the arithmetic does, and the section keeps the paper's main tenure caution.recommended
B
Keep both, keep the names — add the missing context and the baseline sentence only. Least churn across the toolkit, the planner and the deck.
C
Keep both, different words — e.g. rate-neutral and portfolio-value. Name the pair you would use.
D
Publish one basis only — the revenue-neutral one, since it is the toolkit default, and move the other to the framework doc. This is the option Robinson's question points at.
Q2
The two CLV variants were named backwards. Minute the fix.
✓ You askedRobinson, Mondello whether the naming convention should be flipped — you read constrained as the regulatorily constrained, i.e. filing-safe, variant. ✓ We built the neutral pair clv_base / clv_with_growth, keeping the old column names as deprecated aliases for one release so nothing that already runs breaks. ✓ You ratified it verbally on 21 August — this question exists only to put it in the minutes, because it amends the binding two-variant decision of 2 July.
Confirm the pair we carry into the published toolkit.
A
Neutral pair — clv_base / clv_with_growth. Removes the ambiguity instead of pointing it the other way.recommended
B
Flip the original — clv_constrained becomes the filing base. Matches the instinct, but the same word still carries the ambiguity.
C
Name the use, not the constraint — clv_filing / clv_planning.
D
Revert to the v2.0 names and add a clarifying paragraph.
Q3
Cross-sell was credited every renewal term. What default do we ship?
✓ You askedRobinson how cross-sell value is calculated, and whether $150 per term was plausible — noting that once it is done, it will not be done again. It was not plausible, and it was a defect. ✓ We built a one-time event with an annual hazard: the sale is credited at most once, and cross_sell_value is now documented as the CLV of the secondary product.
Now pick the shipped default for the hazard p.
For scale: the v2.0 per-term treatment valued the same assumptions at $693.40 per customer — an annuity of cross-sales. The corrected figure is $376.21, of which $98.85 is the single sale and $277.36 the recurring upsell uplift.
A
p = 1.0 — the sale lands at the first renewal. Most favourable timing, so the figure is a ceiling on growth-programme value: $376.21 per customer here.recommended
B
p = 0.25 — a four-year expected time to first cross-sale. More realistic, but it is another fitted assumption we cannot support from this book.
C
No default — require the user to state p explicitly, so the assumption can never be implicit.
D
Drop cross-sell from the shipped illustration entirely and document the mechanism only.
Q4
The tenure curve did not reproduce on real data. How do we ship it?
✓ You askedMondello on 16 July whether the loss ratio varies by tenure. ✓ We built it into all six families — and then tested it against three real public datasets. The designed five-year cross-carrier test turns out not to be executable on public data at all: no public source carries carrier identity and policyholder tenure together.
Now pick how the curve ships.
What the three datasets said. Wisconsin LGPIF: +19.93%/yr over tenures 0–3 (R² 0.949) — until the final cell, one fund-year of catastrophe, turns it into -7.93%. One year of data flips the sign. Spanish households: +1.80% and not monotone. Shipped synthetic curve: +4.22%.
A
Mechanism, not parameter — keep loss_trend = 0.0 as an exact no-op, ship the curve as an illustration, and accept a user-supplied curve. The published finding stays the direction, which holds for any magnitude.recommended
B
Remove the shipped curve — expose the machinery with no default curve at all, forcing every user to supply one.
C
Ask CAS for a data call — loss ratio by policyholder tenure across carriers. Schedule P sets the bar: between-carrier dispersion is 24.4% against a ten-year curve effect of 35.0%.
D
Ship the LGPIF-fitted +19.93% as an alternative curve alongside the synthetic one.
Q5
Which segment should carry the retention-adjusted loss ratio?
✓ You askedRobinson whether the retention-adjusted loss ratio should be shown by segment rather than by tenure — because the decision it supports is charging less for business expected to persist longer at the same first-term loss ratio, and a by-tenure exhibit cannot show that. ✓ We built it, and the result is sharper than the question assumed.
Now pick the segment the published exhibit leads with.
The result. The 3+ products segment has the worst first-term loss ratio of the three — 0.6652 — and still indicates a credit of 0.9425, on 7.92 expected years against 2.72 for mono-line. A single-term exhibit ranks it last of three; the lifetime exhibit ranks it second.
A
Product count — the closest analogue to a credit insurers actually file (the multi-policy discount), and the segment with a real persistence gradient.recommended
B
Region — a conventional rating variable, but this book's retention barely varies across it, so the exhibit says little.
C
A rating variable the POG names — tell us which and we will fit it.
D
Revert to the by-tenure view only.