The IAN Model Annuity Writer

Interest rate risk is not in this table, and that is not an oversight. The interest rate module nets the fall in asset values against the fall in the best estimate liability and the risk margin, and then takes the worse of the up and the down shock. Both of those happen at portfolio level. Splitting the liability offset across seventeen asset lines is an allocation choice, and this model does not make choices it cannot source. The three modules that can be attributed to a line — spread, equity and property — are the three that are here.

“Per year of duration” is a capital cost, not a yield. It answers how much capital a year of exposure costs, and it can invert the ranking: equity release is charged £25.00 per £100 against a BBB corporate’s £17.00, and is the cheaperholding at £1.67 a year against £2.13, because it buys fifteen years of matching rather than eight. Qualifying infrastructure is the other kind of answer — cheaper on both measures, £14.69 and £1.22, which is what that calibration exists to do. Read it no further than that. A charge divided by a duration is notan implied spread — the CLO line would imply 1,660 basis points — and a version of this model that once treated it as one reported that downgrading an asset improved the coverage ratio.

Blank cells are blank for a reason. Gilts, other sovereign and supranational exposures and cash carry nospread charge at all under rule 3D24.2 — that zero is a fact about the rule, not missing data. Equity, property and cash have no duration, so they have no per-year figure; a zero there would sort a 49% equity charge as the cheapest thing on the page.

What filling that column anyway would cost. Capital cost per unit of spread is the number an investment team really wants, and the reason it is not here is above the table. What is worth adding is the consequence: seventeen spreads that had to be estimated to exist would turn a calculation into a house view wearing its clothes, and a reader cannot tell the two apart from the outside. Two of the figures offered to us during that search were fabricated— basis-point numbers manufactured by a search summary and attributed to a page that contains no numbers at all. Both were caught by opening the page. A ranking is only worth as much as its worst input.

“Charged as” is the column that explains the rest. Two lines can sit at the same duration and the same credit quality step and be charged ten times apart: senior RMBS at £8.00 and CLO tranches at £83.00, both 5-year positions at step 2. The whole of that difference is the simple, transparent and standardised treatment — the first qualifies, the second cannot, because a CLO is an actively managed portfolio. Without that column the table reads as broken; with it, the ten-fold gap is the regime doing exactly what it was designed to do.

A longer asset can come out worse here even when it is the cheaper one, and that is the model rather than the market. Move £500m from BBB corporates into qualifying infrastructure — charged 14.69% against 17.00%, plainly the cheaper asset — and the coverage ratio falls by about six tenths of a point. The cause is the matching adjustment. This model computes the widening it gives back as portfolio loss divided by portfolio duration, so a twelve-year asset replacing an eight-year one adds more to the denominator than it takes off the numerator: the implied widening falls, and the relief falls with it by more than the asset loss does. A real annuity writer lengthening its book to match long liabilities would not report a worse position for doing it. Read the module breakdown, not only the headline ratio.

The opening matching adjustment does not respond to the mix. It is an input, 150 basis points, and it stays there whatever is moved. In reality it would change, because different asset classes earn different spreads — and this model has no spread by asset class, for the same reason the table above has no “spread earned” column. What does respond, and is computed properly from whatever mix you build, is the recalculated matching adjustment under a spread stress, the eligible share, and every capital charge.

What the opening balance sheet is calibrated to. The model opens at the median published Solvency II coverage ratio of the four annuity writers in the comparison table below, each read from the same report and the same balance sheet date — 31 December 2025 — that this page already cites for that firm’s asset mix. Standard Life 176%, Legal & General 203%, Rothesay 249%, Pension Insurance Corporation 257%; median 226%. The basis and the page number for each are in that firm’s citation card.

The four are not on the same basis, and that is the weakness of this yardstick. Standard Life’s 176% is its shareholder view, its own narrower measure rather than its group regulatory ratio, and Standard Life is a diversified group rather than an annuity writer. Rothesay’s 249% is a solo entity including transitional relief, 244% without it. Legal & General’s 203% is the reported group figure; the same release gives 210% pro forma after a transaction and a buyback. And all four calculate their requirement on a PRA-approved internal model, while this page is the standard formula. A median of four numbers built four different ways is a yardstick, not a measurement.

It was calibrated to a different figure earlier the same day, and the change is worth explaining. The first calibration used the Bank of England’s published aggregate coverage ratio for UK life insurers, which is one consistent measure across a consistent population — a better statistic in every respect except the one that matters here. That aggregate is every UK life firm, and most of what it contains is unit-linked business where the policyholder carries the investment risk. That is not an annuity book. The population match was traded for basis consistency, deliberately, and both figures are shown on this page so a reader can see the gap rather than wonder which is wrong.

One input was changed, and it was the best estimate liability. £42.0bn on the morning of 18 August, £43.86bn against the sector aggregate, and £43.00bn now, which takes opening own funds to £5.30bn. Raising the liability the assets have to back is the economically natural way to hold less surplus. Other liabilities would have had to carry the whole adjustment on a £50bn balance sheet; the subordinated debt and the transitional measure cannot reach the target at all — removing each of them entirely leaves the model at 230% and 242%. Changing total assets would have rewritten every line of the asset table to move one ratio. Nothing in the four insurers’ own disclosed asset mixes moved, and nothing in the shock calibrations moved.

The stresses are those of Commission Delegated Regulation (EU) 2015/35 as amended: interest rate risk at Articles 166 and 167, equity risk and the symmetric adjustment at Articles 169 and 172, property risk at Article 174, spread risk at Articles 176 to 178a, the exemptions and the qualifying-infrastructure treatment at Article 180, the market risk correlations at Annex IV, and the loss-absorbing capacity of technical provisions at Article 206.

Where the rules now live. The article numbers are the origin of each shock and remain how practitioners refer to them, but since 31 December 2024 the operative rules have been the PRA Rulebook’s, in the Solvency Capital Requirement — Standard Formula Part. Interest rates sit at rules 3D5 (up) and 3D6 (down); the two equity types at 3D7.2 and 3D7.3, their 39% and 49% shocks at 3D9 and their 0.75 correlation at 3D7.6; the symmetric adjustment at 3D12, bounded at ±10% by 3D12.4; property at 3D15; spread risk across 3D16 to 3D25, with senior simple, transparent and standardised securitisations at 3D21.3 and 3D21.5 and non-STS at 3D21.8; the sovereign and multilateral exemption at 3D24.2 to 3D24.4 and the qualifying-infrastructure factors at 3D24.16 to 3D24.21, with what counts as infrastructure at 3D2 and 3D3; the market risk correlation matrix at rule 3.11A(2), which is no longer a standalone Annex; and the loss-absorbing capacity of technical provisions at rules 6.1 to 6.3, with deferred tax at 6.4 and 6.5.

The two corridors part company in January 2027. The UK’s symmetric adjustment corridor is ±10 percentage points and no change to it has been proposed. The EU’s widens to ±13 percentage points from 30 January 2027, under Article 1(49) of Directive (EU) 2025/2, which replaces Article 106(3) of the Solvency II Directive. From that date two insurers holding the same equity face different capital depending on which side of the Channel they sit. This model applies the UK figure.

The spread tables are the rulebook’s own, not a scaling of one another. Four separate calibrations are carried here, each transcribed from the table printed in the rule that this page cites for it: rated bonds and loans from rule 3D17.3, unrated ones from 3D17.4, senior STS securitisations from 3D21.3, and qualifying infrastructure from 3D24.16. Three of those four are BANDED — the charge is b × duration up to duration 5 and then a + b × (duration minus the band start) in steps of five years to 20 — so a table read only at its first row is wrong everywhere else. Qualifying infrastructure in particular has its own a and b figures at every credit quality step and is not a discount applied to the corporate bond row: at credit quality step 3 and duration 12 the infrastructure charge is 14.69% against 22.0% for the same exposure charged as ordinary corporate credit, and no single scaling factor reproduces the table at more than one duration. Non-STS securitisation, at 3D21.8, is the one that is not banded: it is min(b × duration, 1) throughout, with b running from 12.5% a year at credit quality step 0 to 100% at step 5.

Type 1 equity is an OECD test, not an EEA one. Rule 3D7.2 makes an equity type 1 if it is listed on a regulated market in a country that is a member of the OECD, so US, Japanese and Swiss listed equities are type 1. Type 2 is what is left: equity that is not listed on such a market — unlisted and private equity — together with commodities and other alternative investments, and anything the interest rate, property and spread sub-modules do not pick up. The EEA limb was struck out of the UK rule at onshoring in 2019, years before the restatement carried the narrowed test forward.

UK divergence. Solvency UK reformed the matching adjustment — eligibility widened, the fundamental spread recalibrated — and cut the risk margin. The restatement carried the standard formula market risk modules across materially unchanged. So the shocks here are on solid ground; the matching adjustment parameters are the part a UK reader should treat as indicative.

Moving several sliders at once is not a combined scenario. The standard formula does not add its market modules together. It aggregates them on the correlation matrix at Annex IV, now rule 3.11A(2), precisely because interest rates, spreads, equity and property are not assumed to move to their one-in-200 points simultaneously. The SCR figure on this page does that aggregation. The BALANCE SHEET does not — it applies every shock set above, in full, at the same time. That is deliberate, because watching one shock at a time is how the mechanism becomes visible, and the page says so on screen the moment a second slider moves. A combination of full shocks is a sensitivity, not a capital requirement, and nobody is required to hold capital against it.

Approximations, stated. Every asset and the best estimate liability carries a single modified duration and moves first-order; convexity is ignored. The interest rate stress is a level shift read off the Article 166/167 table at each item’s duration, not a re-discounting of a stressed term structure. Currency and concentration risk are set to zero — a sterling annuity book hedges currency to immaterial, and concentration risk needs single-name exposures this model does not carry.

What is missing, and it matters. The loss-absorbing capacity of deferred tax (Article 207) is not modelled at all. It is a real and material offset for these firms, so the modelled SCR here is conservative. Saying so is better than guessing a rate.