Technical Notes#

Status: Draft, 2026-08-03 (all cited sources accessed 2026-08-03; see sources.md).

Scope note. These notes specify a reference liability cash flow projection model for the standardized composite product defined in product-spec.md (same directory). This is not any single insurer’s product. [S#]/[R#] tags refer to the source list in _research/whole-of-life.md via sources.md; [REG-R#] tags refer to the cross-product reference library references/regulatory-and-actuarial-references.md (its own R-numbering; research provenance in _research/regulatory-actuarial.md). std marks standardizations introduced for the reference implementation; unverified marks claims not confirmed against a retrieved document. Parameter values are identical to those in product-spec.md. Two cells share one engine:

  • RefWOL-UW (underwritten guaranteed; the chassis-carrier pattern [S10]) — anchor: male, entry age 40, non-smoker, £150,000 sum assured, level cover, £101.25/month std.

  • RefWOL-O50 (over-50s guaranteed acceptance; the direct-sold pattern [S1] [S4]) — anchor: entry age 70 std, non-smoker, £30/month, £5,000 cash sum (R2 stylised pair).

Neither cell carries any account value, unit fund, or surrender value [S1] [S4] [S7] [S9] [S10]: both are pure-decrement protection models — the projection is premiums in, death benefits and expenses out, weighted by survivorship. This is the deliberate contrast with the US cash-value whole life chassis (no CSV schedule, no dividends, no loans).

Model scope and conventions#

  • Purpose. Project gross best-estimate liability cash flows (premium income, death outgo, expenses; no surrender outgo exists) for single-policy model points of the two cells, on a monthly grid. Reserves, discounting, risk margin and capital are pointed to, not computed (see Valuation and reserve pointers).

  • Projection frequency. Monthly std. Premiums are monthly Direct Debit in the O50 cell [S1] [S4] [S7] [S9] and monthly or annual in the UW cell [S10]; monthly is the natural grid.

  • Timing conventions std. Premiums (and premium-linked commission) at the beginning of the policy month (BOM); deaths during the month resolved at end of month (EOM) against the BOM in-force; lapses at EOM after deaths (death-before-lapse order). Escalation steps (increasing cover, RPI variants) apply at policy anniversaries [S4] [S10]. Annual-grid implementations must preserve the month-13 moratorium boundary.

  • Age basis. Age last birthday (ALB) std. Rationale: the UW chassis defines entry age x as “before the (x+1)th birthday” [S10], which is ALB; the O50 documents price on “age at outset” without stating a basis [S1]. All age lookups in this model are ALB.

  • Currency / units. GBP. Sum assured in £; premiums in £/month; mortality and lapse rates dimensionless per annum, converted to monthly as q_m = 1 − (1 − q)^(1/12) std.

  • Model points. Single-policy expected-value projection: survivorship probabilities multiply per-policy cash flows. No aggregation logic here. Aggregation caps across same-insurer policies (£10,000/£18,000, £100/month [S1] [S4] [S9]) are immaterial to a per-policy model.

  • Claims settlement. Immediate at EOM of the death month std; contractual claims interest (BoE base − 0.5%, floor 0.5% p.a., between death and payment [S1] [S9]) is excluded as a settlement-lag refinement, not a liability driver.

  • Rounding. Full precision carried; cash flows reported to pence std.

Model point attributes#

Attribute

Type

Example (O50 anchor)

Example (UW anchor)

cell

enum {O50, UW}

O50

UW

entry_age

int (ALB)

70

40

sex

enum {M, F}

F std (pick for the anchor cell; attribute carried for basis lookup — O50 pricing itself does not rate by sex in the fetched documents, which state age and smoker status as the rate factors [S1] [S4])

M

smoker

enum {NS, S}

NS

NS (the chassis 3-state definition [S10] collapsed to 2 std)

sum_assured

currency (£)

5,000

150,000

monthly_premium

currency (£/month)

30.00

101.25 std

escalation

enum {level, fixed_5pct, rpi}

level

level (fixed_5pct variant)

cessation_months

int (∞ for UW)

240 (anniversary on/after 90th birthday std)

none — premiums payable for life [S10]

moratorium_months

int

12 [S1] [S4] [S7] [S9]

0 (suicide-only clause instead [S10])

variant_adb_2x

bool (accidental multiplier, one plan [S7])

false

n/a

variant_paid_up

bool (pro-rata paid-up value, one plan [S9])

false

n/a

variant_rpi_increasing

bool (RPI indexation, one plan [S4])

false

n/a

issue_date

date

month 1

month 1

State variables#

Variable

Description

Updated

l(t)

In-force probability at end of month t; l(0) = 1

monthly (deaths, lapses)

CumPrem(t)

Cumulative premiums paid to end of month t (year-1 refund base; crossover tracking)

monthly

N_paid(t)

Count of monthly payments made (pro-rata paid-up numerator [S9])

monthly

SA(t)

Current sum assured / cash sum (escalating variants)

anniversaries

P(t)

Current monthly premium (escalating variants; 0 after cessation)

anniversaries / cessation

paid_up

Pro-rata paid-up state: policy premium-free with reduced payout PU [S9]

on qualifying lapse

PU

Paid-up payout = SA x N_paid / N_expected (pro-rata paid-up variant) [S9]

on paid-up conversion

in_moratorium(t)

Indicator t <= 12 (O50)

monthly

attained_age(t)

entry_age + floor((t−1)/12) (ALB)

monthly

Assumption inputs#

Three classes are distinguished explicitly.

(a) Contractual / guaranteed elements (cited)#

Input

Value

Basis

Premium level at issue

Fixed at outset by age and smoker status (O50 [S1] [S4]) / full underwriting (UW [S10] [S11]); guaranteed never to increase [S1] [S4] [S7] [S9] [S10]

anchors: £30/month for £5,000 at 70 R2; £101.25/month for £150,000 at 40 std

O50 moratorium

12 months; non-accidental death → return of premiums paid (no interest stated); accidental death → full cash sum from day 1

[S1] [S4] [S7] [S9]

O50 premium cessation

Anniversary on/after 90th birthday; cover continues

[S4] [S5] [S9]; pick std

UW terminal illness

Sum assured accelerated on 12-month prognosis; pays once, policy ends

[S10] [S12]

UW suicide clause

Suicide/intentional self-inflicted injury within 12 months of start (or increase) → refund of premiums for that cover

[S10] [S11]

UW escalation (variant)

SA +5%/year, premium +10%/year (2% premium per 1% cover)

[S10]; 5% pick std

O50 RPI variant (one plan)

Cash sum +RPI (floor 0%, cap 10%); premium +RPI x 1.5 (cap 15%); freeze on first declined increase; cash-sum indexation continues post-90

[S4]

Pro-rata paid-up value (one plan’s variant)

If N_paid >= N_expected/2 at premium stop: paid-up payout = SA x N_paid/N_expected; else cancellation with nothing

[S9]

Arrears

60 days to make good; death in window → claim reduced by unpaid amounts; then lapse with no value

[S4] [S9]; pick std

Surrender value

None at any time, either cell

[S1] [S4] [S5] [S7] [S9] [S10]

(b) Insurer-discretionary current elements#

This class is nearly empty — the defining feature of both modern cells. Premiums are guaranteed [S1] [S4] [S7] [S9] [S10]; there are no bonus rates, no reviewable premiums, no unit-linked charges, no asset shares and no MVRs in either cell (those mechanisms belong to with-profits and unit-linked business, out of scope here). The discretionary layer reduces to:

  • New-business rate tables. Insurers do not publish full premium rate tables (research file gap); only quote anchors exist (£20/month at 50 NS → £5,694 [S2]; £25/month NS → £7,643/£6,046/£3,701/£1,893 at 50/60/70/80 [S6]) plus the FCA per-£1,000 averages (£71.73 O50, £8.10 underwritten) R2. The model takes premium as a model-point input; any shipped rate table is a std snapshot calibrated to these anchors.

  • Claims interest rate. Contractual formula, BoE-base-linked (base − 0.5%, floor 0.5%) [S1] [S9]; excluded from the base model std (conventions).

  • Legacy variation only: the unit-linked reviewable design’s review basis (mortality charge scale, review outcomes) is insurer-discretionary [S15]; it is documented as a closed-book variation, not modeled.

(c) Behavioral / experience assumptions (modeler’s view)#

CMI access is honestly restricted: current tables and the Projections Model are limited to Authorised Users/Subscribers; older publications are free — so a reference basis must be a std proxy shaped like the named tables, and cannot redistribute current qx values REG-R22 R7.

Input

Recommended public basis

Basis tags

UW mortality

Assured-lives shape: CMI “00” series permanent assurances AMC00/AMS00/AMN00, AFC00/AFS00/AFN00 (publicly downloadable; the latest published assured-lives whole of life base tables) x A/E factor 100% std; AM92/AF92 as the teaching-table alternative shape

R6 R7 REG-R24; factor std

O50 mortality

Population-level: ONS national life tables qx (single year of age, sex; freely downloadable under OGL) x anti-selection loading 120%, level across durations std

REG-R32; loading std (see below)

Mortality improvement

None in base std; sensitivity: “CMI_20xx with long-term rate p% std” is the market-standard expression, but the model is subscriber-restricted — a flat 1% p.a. improvement is the std sensitivity proxy

REG-R30 REG-R22

O50 accidental-death share of year-1 deaths

3% std — accidental deaths are a small minority at 70+; no public split was found (research gap)

std

UW suicide share of year-1 deaths

1% std — refund instead of sum assured; immaterial, carried for completeness

std

Terminal illness acceleration (UW)

Model TI claims as deaths accelerated by 6 months on average; base model ignores the acceleration (pays at death) std

timing std; benefit [S10]

Lapse

std tables below; no public UK WoL lapse study was retrieved (research gap); the FCA documents the lapse-supported dependence qualitatively

R2; tables std

Expenses

O50: acquisition £150/policy + commission 25% of year-1 premiums std; maintenance £30/policy/year inflating 3% p.a. std. UW: acquisition £300/policy + initial commission std; maintenance £50/policy/year inflating 3% std. Commission existence per one plan’s disclosure (intermediary “paid by commission as a percentage of total annual premium” [S1]); all levels std

[S1]; levels std

Why the O50 basis is population-plus-loading, not assured lives. Guaranteed acceptance removes underwriting, so the pool cannot be better than population and self-selects worse: the CMI is analysing non-underwritten whole of life experience separately from underwritten — direct recognition of the anti-selection distinction R7 — and the FCA’s price differential (£71.73 vs £8.10 per £1,000) reflects guaranteed-acceptance anti-selection, older entry ages and shorter durations R2. No insurer discloses its guaranteed-acceptance pricing basis (expected — proprietary; research file gap), so the 120% loading on ONS population rates is a std placeholder to be calibrated; population mortality is itself heavier than insured experience REG-R32, so the loading is deliberately modest. The UW cell uses an assured-lives shape (“00” series R6) because full underwriting restores select experience.

Reference base lapse tables std (annual rates; shapes are drafting constructions — no public product-specific study; replace with experience):

Policy year

1

2

3–5

6+

after premium cessation

O50 w_base

8%

6%

4%

4%

0% (no premiums due — no lapse)

UW w_base

6%

5%

3%

2%

n/a (premiums for life)

Cash flow components and recursions#

Notation (defined once, used throughout; shared with product-spec.md)#

Symbol

Meaning

t

policy month, t = 1, 2, …; y = policy year = floor((t−1)/12) + 1

a(t)

attained age (ALB) = entry_age + floor((t−1)/12)

P(t)

monthly premium due at BOM of month t (0 after cessation / in paid-up state)

SA(t)

sum assured / cash sum in month t (constant unless escalating variant)

T_cess

months from start to premium cessation (O50: to anniversary on/after 90th birthday; anchor 240); ∞ for UW

CumPrem(t)

Σ_{s<=t} P(s)

q(y)

annual mortality rate for policy year y (basis per cell, class (c))

q_m(y)

monthly mortality = 1 − (1 − q(y))^(1/12) std

w(y), w_m(y)

annual / monthly lapse rates, w_m = 1 − (1 − w)^(1/12) std

δ_acc

accidental share of year-1 deaths (O50), 0.03 std

δ_su

suicide share of year-1 deaths (UW), 0.01 std

l(t)

in-force probability at end of month t; l(0) = 1

DB_na(t), DB_ac(t)

death benefit for non-accidental / accidental death in month t (O50)

k_adb

accidental multiplier after year 1: 1 (base) or 2 (one plan’s variant [S7])

E[·]

expectation over decrements (survivorship weighting)

Dimensional check: premiums and benefits are £; q_m, w_m, δ are dimensionless; every expected cash flow below is £ per month per policy issued.

Monthly processing order (both cells) std#

At month t while in force and not paid-up:

  1. BOM: premium P(t) received if t <= T_cess (O50) or always (UW); commission/premium expense deducted as an expense flow, not from any fund (there is no fund).

  2. BOM: maintenance expense for the month.

  3. Anniversary (t ≡ 1 mod 12, t > 12): apply escalation to SA and P (variants only) [S4] [S10].

  4. EOM: deaths at rate q_m(y) applied to l(t−1); benefit per the rules below.

  5. EOM: lapses at rate w_m(y) applied to survivors of step 4; death-before-lapse std. In the pro-rata paid-up variant a “lapse” with N_paid >= N_expected/2 converts to paid-up (state change, no cash flow) instead of termination [S9].

  6. Update l(t) = l(t−1) x (1 − q_m(y)) x (1 − w_m(y)).

Paid-up policies (the pro-rata paid-up variant) and post-cessation O50 policies skip steps 1 and 5 (no premiums due, so no lapse decrement std) and continue steps 2, 4, 6 with w_m = 0. Step 3 is also skipped, with one exception: in the RPI-increasing variant the cash sum continues to index at anniversaries after premiums cease at 90 [S4] (the premium step, being zero, stops).

RefWOL-O50 recursions#

Premiums (level base design):

P(t) = P x 1{t <= T_cess},        CumPrem(t) = P x min(t, T_cess)

Death benefit split during the 12-month moratorium [S1] [S4] [S7] [S9]:

DB_na(t) = CumPrem(t)   if t <= 12          (return of premiums paid, no interest)
         = SA           if t >  12
DB_ac(t) = SA           if t <= 12          (full cash sum from day 1)
         = k_adb x SA   if t >  12          (k_adb = 2: one plan's variant [S7])

Expected cash flows in month t (per policy issued):

E[premium](t) = l(t−1) x P(t)
E[death outgo](t) = l(t−1) x q_m(y) x [ (1−δ_acc) x DB_na(t) + δ_acc x DB_ac(t) ]   if t <= 12
                  = l(t−1) x q_m(y) x [ (1−δ_acc) + δ_acc x k_adb ] x SA           if t >  12
E[expenses](t) = l(t−1) x [maintenance(t)] + commission/acquisition at their BOM timing

(with k_adb = 1 the post-moratorium death outgo is simply l(t−1) q_m SA). Lapse generates no cash flow: there is no surrender value [S1] [S4] [S5] [S7] [S9] — its entire effect is through l(t). That is the arithmetic meaning of “lapse-supported”: every lapse extinguishes a paid-up-style liability for nothing, and the FCA records that without the continuing-payer cross-subsidy “insurers would need to rely on lapses to remain profitable” R2.

Crossover (tipping point): cumulative premiums first exceed the cash sum at

t* = floor(SA / P) + 1    (months, level premiums, t* <= T_cess)

Anchor: floor(5000/30) + 1 = 167 months = 13 years 11 months, reproducing the FCA’s stylised example exactly R2. Total premiums payable are capped at P x T_cess (anchor: £7,200 vs £5,000 cash sum). A crossover exists iff SA < P x T_cess; the FCA notes entrants at 79–80 are most exposed and that the majority of policies still pay out more than premiums paid R2.

Pro-rata paid-up variant [S9]: on premium stop at month t with N_paid(t) >= N_expected/2 (N_expected = T_cess):

PU = SA x N_paid(t) / N_expected        (worked example: 180/240 x £3,500 = £2,625 [S9])

thereafter DB_na = DB_ac = PU (the moratorium is long past), premiums 0, lapse 0 std.

RPI-increasing variant (one plan [S4]), r_y = RPI inflation for year y:

SA(y+1) = SA(y) x (1 + min(max(r_y, 0), 0.10))
P(y+1)  = P(y)  x (1 + min(max(1.5 x r_y, 0), 0.15))       while y < cessation
SA continues to index after premiums cease at 90; first declined increase freezes both
(premium step floored at 0 **[std]** — [S4] defines an increase only, no decrease).

RefWOL-UW recursions#

Premiums guaranteed level (base): P(t) = P for all t; no cessation age [S10]. Escalating variant (5% std), applied at anniversaries [S10]:

SA(y) = SA_0 x 1.05^(y−1),      P(y) = P_0 x 1.10^(y−1)

Death/terminal-illness benefit: the sum assured is paid once on death or earlier terminal illness diagnosis (12-month prognosis), and the policy ends [S10] [S12]. The base model pays SA(y) at death (TI acceleration ignored std; a TI module would move a fraction of claims ~6 months earlier std with no change in amount). Suicide within 12 months refunds premiums [S10]:

E[death outgo](t) = l(t−1) x q_m(y) x [ (1−δ_su) x SA(y) + δ_su x CumPrem(t) ]   if t <= 12
                  = l(t−1) x q_m(y) x SA(y)                                      if t >  12

Lapse (2 months’ unpaid premiums, no reinstatement [S10]) again generates no cash flow — no cash-in value at any time [S10] — and only reduces l(t). Milestone-benefit exercises and requested increases are out of scope (they would step SA and P; anti-selection flagged in model risks) [S10] [S12].

Cash flow outputs (per policy issued, month t)#

Cash flow

Formula

Sign

Premium income

l(t−1) x P(t)

+

Death outgo

per cell formulas above

−

Acquisition expense + initial commission

at t = 1 (and commission % x premiums in year 1, O50) std

−

Maintenance expense

l(t−1) x (annual maintenance / 12) x (1.03)^(y−1) std

−

Surrender outgo

none — identically zero in both cells [S1] [S4] [S7] [S9] [S10]

—

Claims interest

excluded std (contractual BoE−0.5% floor 0.5% between death and payment [S1] [S9])

—

Policyholder behavior modeling#

All dynamic formulas are std reference constructions; no public UK whole of life lapse/persistency study was retrieved (research gap), so shapes are drafting assumptions with the qualitative anchors cited.

  • Base lapse std. Duration-declining tables above; converted monthly. Rationale for the declining shape: sunk premiums with zero surrender value and (O50) the approaching paid-out-in-full status discourage late lapse.

  • Moratorium-completion effect (O50) std. No extra lapse spike at month 13: the moratorium gives no incentive to lapse (lapsing returns nothing at any time). The year-1 rate is set highest instead (affordability/buyer’s-remorse attrition; the 30-day cooling-off with full refund [S1] [S4] is modeled as never-issued business, out of scope).

  • Crossover-aware lapse (O50) std. Sensitivity module, off in base: w(t) = w_base(y) x (1 + β x 1{CumPrem(t) > SA}), β = 0.5. Rationale: Consumer Duty communications must enable informed choice about the over-payment risk R2, which could raise post-tipping-point lapses; the FCA has seen no evidence that a significant proportion of customers reach the premium caps R2. β is a pure stress dial.

  • Pro-rata paid-up selection (one plan’s variant) std. Once N_paid >= N_expected/2, all would-be lapses convert to paid-up (rational: forfeiture is strictly dominated; mechanics per [S9]); before the halfway point, lapse means total loss, so the base w applies. This converts lapse profit into a retained pro-rata liability — the variant exists precisely to remove the forfeiture cliff, and materially weakens lapse support (sensitivity mandatory).

  • Premium reduction options std. One-off reductions (three of the O50 plans [S1] [S4] [S9]) are not modeled; they are economically a partial lapse with proportionate SA reduction.

  • Escalation opt-out (UW variant) std. Increasing-cover holders decline an increase with probability 10% per anniversary; three declines remove the option [S10]; base model assumes full take-up.

  • Payment holidays (one plan [S9]) are ignored std (≤ 12 months’ premiums deferred or netted; second-order).

Worked example#

RefWOL-O50 anchor cell: entry age 70 (ALB), non-smoker, P = £30/month, SA = £5,000, T_cess = 240 months (anniversary on/after 90th birthday std), base design (k_adb = 1, no pro-rata paid-up value). Illustrative walk-through basis std (placeholder, not attributable to any table): q(y) = 0.024 x 1.10^(y−1) — i.e. a 0.020 population-style rate at 70 x the 120% anti-selection loading, with 10% p.a. age progression; lapse 8%/6%/4%/4% (years 1/2/3–5/6+), 0 after cessation; δ_acc = 3%. Monthly rates: q_m(1) = 1 − (1−0.024)^(1/12) = 0.0020223; w_m(1) = 1 − (1−0.08)^(1/12) = 0.0069244. Expenses omitted from the table for clarity. E[death outgo](t) = l(t−1) x q_m x (0.97 x DB_na + 0.03 x DB_ac) for t <= 12, and l(t−1) x q_m x 5,000 thereafter. All £, full precision carried, displayed rounded.

t

y

CumPrem

DB non-acc

DB acc

l(t−1)

E[premium]

E[death outgo]

1

1

30.00

30.00

5,000

1.00000

30.00

0.36

6

1

180.00

180.00

5,000

0.95613

28.68

0.63

12

1

360.00

360.00

5,000

0.90601

27.18

0.91

13

2

390.00

5,000.00

5,000

0.89792

26.94

10.00

24

2

720.00

5,000.00

5,000

0.82785

24.84

9.22

60

5

1,800.00

5,000.00

5,000

0.66359

19.91

9.88

120

10

3,600.00

5,000.00

5,000

0.42564

12.77

10.31

166

14

4,980.00

5,000.00

5,000

0.27420

8.23

9.85

167

14

5,010.00

5,000.00

5,000

0.27131

8.14

9.74

240

20

7,200.00

5,000.00

5,000

0.09992

3.00

6.57

241

21

7,200.00

5,000.00

5,000

0.09828

0.00

7.16

Trace, month 1: E[death] = 1.0 x 0.0020223 x (0.97 x 30 + 0.03 x 5,000) = 0.0020223 x 179.10 = £0.36 — the year-1 death outgo is dominated by the small accidental tail paying the full cash sum, not the premium refund. Trace, month 13: the moratorium ends and the full £5,000 becomes payable for any death: E[death] = 0.89792 x 0.0022271 x 5,000 = £10.00 (q(2) = 0.0264 → q_m = 0.0022271) — a ~11x jump in expected death outgo at the month-12/13 boundary, the signature discontinuity of this product. Month 167 is the crossover: CumPrem = £5,010 first exceeds the £5,000 cash sum (13 years 11 months, reproducing R2). Month 241: premiums have ceased (E[premium] = 0) but death outgo continues — and rises, because lapses stop std and mortality steps up at the year-21 anniversary; the post-cessation period is pure outgo, funded by the pre-cessation premium margins and lapse releases.

Valuation and reserve pointers#

This library projects gross best-estimate liability cash flows; valuation layers consume them and are cited, not reproduced:

  • Solvency UK best estimate. The best estimate is the probability-weighted average of future cash flows discounted at the relevant risk-free term structure, on realistic assumptions, gross of reinsurance (PRA Rulebook Technical Provisions 3.1) REG-R1 — exactly what this model’s expected cash flows feed. Technical provisions = best estimate + risk margin, market-consistent (2.3, 2.4) R3 REG-R1.

  • Risk margin. Cost-of-capital method at 4% (Solvency UK rate, effective 31 December 2024 definitions) R3, with the life-business risk-tapering factor lambda = 0.9 (floor 0.25) from SI 2023/1346 REG-R4. Requires an SCR runoff — cited-not-specified.

  • Solvency UK frame. Assimilated Solvency II law was revoked 31 December 2024 and restated into PRA rules effective the same date; the PRA Rulebook, not EU text, is the operative source R4. Legacy back-books (the unit-linked variation [S15]) may carry TMTP, which adjusts technical provisions, not projected cash flows REG-R3.

  • Realistic-lapse warning. A best estimate on realistic assumptions REG-R1 embeds the lapse-support profits: raising assumed lapses lowers the BEL of the O50 cell. The FCA’s articulation of the reliance on lapses R2 makes lapse the assumption to govern hardest (TAS 100 justified-assumptions discipline R8).

  • IFRS 17. UK-adopted IFRS 17 (adopted 16 May 2022, effective 1 January 2023, replacing IFRS 4) applies to IFRS reporters REG-R38; the fulfilment-cash-flow engine consumes the same projections with different discounting/aggregation layers.

  • Professional standards. TAS 100 v2.0 (effective 1 July 2023) applies to all UK technical actuarial work including this modeling R8; TAS 200: Insurance v2.0 (effective 1 January 2025) applies additionally to insurance technical actuarial work REG-R34.

Key sensitivities and model risks#

Dominant assumptions, in order, for a guaranteed-acceptance (O50) block:

  1. Lapse — sensitivity analysis mandatory. With no surrender value, every lapse is a pure profit release; the FCA itself records the reliance on lapses for profitability R2. BEL is monotonically decreasing in lapse rates; run at 0.5x / 1x / 2x base lapse and at zero lapse (the conduct-stress floor). The pro-rata paid-up variant [S9] converts post-halfway lapses into paid-up liabilities and collapses most of the lapse sensitivity — model it as a separate variant, never as a small adjustment.

  2. Guaranteed-acceptance mortality and anti-selection. The 120% x ONS loading is a std placeholder; the true basis is proprietary and the CMI’s non-underwritten whole of life analysis was pending as of the fetched announcement R7 REG-R32. Year-one anti-selection interacts with the moratorium: the refund design exists precisely because year-1 non-accidental mortality is anti-selected.

  3. Longevity past the crossover and past cessation. Post-90 the policy is pure outgo; improvement assumptions (CMI_20xx-style, subscriber-restricted REG-R30) directly lengthen it. For the UW cell, whole-of-life duration makes the liability improvements- and discount-dominated.

  4. Expense inflation vs fixed premiums. Premiums are small (£30/month anchor) and level; maintenance expenses inflate. The expense margin erodes mechanically — a per-policy expense assumption error compounds over 20+ year horizons.

  5. Escalation take-up (UW variant). Premium escalates at 2x the benefit rate [S10]; the variant is premium-margin-accretive but lapse-sensitive (escalating premiums into fixed incomes); opt-out behavior (three declines end the option [S10]) is unobserved std.

Known modeling pitfalls:

  • Moratorium boundary. The month-12/13 discontinuity (~11x jump in expected death outgo in the worked example) must not be smoothed by annual-grid interpolation; if projecting annually, split year 1 explicitly.

  • Refund base. The year-1 non-accidental benefit is cumulative premiums paid, not the cash sum and not an annualized premium; with the arrears rule, claims in the 60-day window are further reduced by unpaid amounts [S9].

  • Lapse after cessation. There are no premiums to stop paying after T_cess; applying a lapse decrement there silently destroys liability. Set w = 0 post-cessation std (and in paid-up states).

  • Accidental-multiplier double-count. The 2x applies to accidental death on/after the first anniversary only [S7]; applying it in year 1 (where accidental already pays 1x SA in the base plans, and that variant’s own year-1 accidental benefit is 1x the cash sum [S7]) or to all deaths overstates outgo.

  • Anti-selective options (UW). Milestone-benefit increases without underwriting [S10] and smoker-status reviews [S10] are exercised against the office; excluding them is a std scope choice that understates tail risk on large-sum business.

  • Terminal illness timing (UW). TI pays the same amount earlier; ignoring acceleration understates the present value slightly. Do not model TI as an additional decrement — it accelerates the death benefit, it does not add one [S10] [S12].

  • Basis mixing. The O50 cell uses a population-plus-loading basis, the UW cell an assured-lives shape R6 REG-R32; feeding either cell the other’s basis produces plausible-looking but wrong margins (the FCA’s £71.73 vs £8.10 differential R2 is the scale of the error).

  • Claims interest. Excluded std; if added, it is a settlement-lag uplift at BoE − 0.5% (floor 0.5%) on death claims [S1] [S9], not a discounting change.