Four decisions, all downstream of Mark's question on July 16. We measured the tenure gradient,
built it into all six models, and found something that inverts the obvious conclusion —
so these are genuine choices, not ratifications. The green banner shows
what you asked → what we built; then pick what we carry into the paper and the
toolkit. Q1 and Q3 are the ones that matter most; if we run out of time, answer those by
email and we will proceed on the recommendations for the rest.
★ = researcher's recommendation.
Q1
Tenure-varying loss — what becomes the default?M. MONDELLO
✓You asked (Jul 16): should losses
“differ by how long you have had them renew” rather than a flat 66% every year?
✓We built it — CLV is now
Σ n_t × margin(t), so loss can vary by year. Measured gradient
0.625 → 0.416 across tenure; shipped as a smoothed 4.22%/year credit.
loss_trend=0.0 is currently the default and an exact no-op.
Which basis should the published headline use?
A
Keep flat as the headline;
publish the tenure gradient as a sensitivity dimension until the NAIC
cross-carrier test confirms it★ rec
B
Make the filing basis
(revenue-neutral relativity) the headline now
C
Make the planning basis
(level effect, +$2.62M on the book) the headline now
D
Publish flat and tenure side-by-side
everywhere, as we do for the two CLV variants
Q2
Where should the curve come from?
✓You flagged (Jul 16) that the model should
use “modeled predictions for loss” rather than one assumption.
✓We built a curve measured by
retention_adjusted_loss_ratio() then smoothed by a premium-weighted log-linear
fit (R² 0.887, floored at 0.70). Raw yearly ratios are too noisy to price off —
tenure 6 bounces back to 0.517 on thin premium.
What should the published default curve be, given users will apply this to their own books?
A
Fit on the user's own book,
with ours as a documented illustration only★ rec
B
Ship our fitted curve as a
general-purpose default
C
Credibility-blend the user's
own experience toward an industry curve
D
Judgmentally selected curve,
actuary-supplied, no default at all
Q3
May a tenure relativity enter a filed exhibit?
✓You asked whether long-tenured customers
are “more profitable over time.” ✓We found the
opposite of the obvious answer. On a revenue-neutral basis, recognising that new
business improves with tenure means long-tenured business carries relatively more
loss: the 10+ cohort's mean CLV falls −7.0% while every younger cohort rises
1–3%.
Heads-up: “loss ratio improves with tenure” and
“give loyal policyholders a discount” are not the same proposition. A
tenure relativity built naively from a loss-ratio-by-tenure exhibit is likely backwards.
How should the paper position tenure in rate indications?
A
Planning use only. Publish
the filing-basis result as a cautionary finding, recommend against a tenure
relativity without further work★ rec
B
Permit it in filings on the
unconstrained variant, with the four-fifths screen extended to age bands
C
Silent — out of scope for this paper;
present the mechanics without a recommendation
D
Escalate to a dedicated section with
a state-by-state review (CA / NY / CO)
Q4
Case studies — which grain goes in the paper?M. MONDELLO
✓You asked (Jul 16) for “a very small
cohort of policyholders — two people renew, five people renew for ten years,” because
“the more complex the subject, the simpler the example.”
✓We built exactly that as
cas_clv.cohorts.mark_cohort() — seven named policyholders, every cell
hand-checkable. Two of seven destroy value; the cohort still earns +$3,127.
What carries into the paper's exposition?
A
Lead with the 7-policyholder
cohort, then scale to tenure bands and the full book★ rec
B
Tenure bands only — the 4.1× CLV
gradient is the cleaner headline
C
CLV quintiles only — matches the
ratemaking relativity exhibits
D
All of it as an appendix; keep the
main text purely narrative