ネットポジションは計算するもので、表計算するものではありません。

Programs, treaties, layers and participationslink

The structure a cession is computed from — versioned, ordered and shared across markets.

A program holds treaties; a treaty holds layers; a layer holds participations. Participations carry the share each counterparty takes of that layer, so a placement spread across five markets is represented as five rows against one layer rather than as a note about who is on the risk.

Layers are explicitly ordered and reorderable, because the sequence in which cover attaches is part of the contract rather than a display preference. Removing a layer or a participation is a delete against the structure, so the resulting cession changes and can be re-derived rather than silently disagreeing with a stored total.

A cession engine that distinguishes treaty formslink

Quota share, per-claim excess of loss and aggregate stop loss are separate forms, because they respond differently.

Cession runs as a pure calculation over the program and the loss list. Quota share cedes proportionally, excess of loss attaches per claim with reinstatements capping the annual charge, and aggregate stop loss attaches on the retained aggregate for the period and can never pay more than its limit however the losses arrive.

That last distinction is the reason stop loss is its own treaty form rather than a flavour of excess of loss. Charging an aggregate cover once per claim lets a twelve-event year charge it twelve times: the same $15m cover pays $15.00m read correctly and $30.48m read per claim, on the same losses. Neither figure looks wrong in a summary — both report the treaty as within programme — which is exactly why the modelling has to be right rather than reviewed.

Stop-loss recovery is allocated back across claims pro-rata on their retained share, so per-claim nets still sum to the reported net and the two views cannot disagree.

Risk transfer testing, before the rest of it means anythinglink

SSAP 62R 10-10 and ERD, reported independently — because a contract that fails both is a deposit, not reinsurance.

Every other capability on this page assumes an answer this test supplies. Cession, recoverables, credit exposure, Schedule F credit and the IFRS 17 presentation all presume the contracts in the book are reinsurance. A contract that does not transfer significant insurance risk is a financing arrangement and must be deposit-accounted — its premium and its recoveries never touch the underwriting result at all.

Two measures are computed and BOTH are returned. The 10-10 rule asks how often the reinsurer loses: at least a 10% probability of a 10% or greater present-valued loss. It is simple, widely cited, and known to reject genuine low-frequency high-severity layers — a catastrophe layer may carry a 4% chance of a total loss and still be unmistakably reinsurance. Expected reinsurer deficit asks how badly it loses when it does, as a probability-weighted average severity against a threshold. The engine reports each against the bar it was tested on and does not silently pick a winner, because which measure governs is an accounting-policy decision and burying it in a boolean is how a rejected contract quietly becomes an accepted one.

The engine refuses rather than approximates. A contract on which the reinsurer keeps nothing after ceding commission has no denominator for a loss ratio, and returning infinity or substituting one would produce a confident verdict from nothing — so it asks for review as a financing arrangement instead. An empty scenario set is not a contract that failed the test, it is a contract that has not been tested, and it is rejected on those terms. Probabilities are rounded to significant figures rather than to fixed decimal places, so a remote catastrophe outcome at 1e-7 keeps its meaning instead of being flattened to zero in the flattering direction. The "substantially all" exemption is a separate call rather than an automatic branch, because an exemption applied automatically is an exemption nobody notices being claimed.

Assessments are persisted against the treaty with an exceptions view, so a contract that failed or was never tested is findable rather than resting on whoever ran it last.

Recoveries from case to collection, with ageinglink

What is recoverable, what has aged, and what has actually been collected — kept apart.

A recovery is raised against the claim that earned it and moves through its own lifecycle to collection. Ageing is reported on the outstanding balance, so a recoverable that has sat for two hundred days is visible as that rather than as part of a single recoverable total.

Collection is an explicit event rather than a status edit. That separation is what keeps "we are owed this" and "we have received this" from collapsing into one number at the point where the difference matters most, which is a quarter end.

Counterparty credit exposure, net of collateral heldlink

Recoverable balance less collateral actually held — with concentration, rating and a provision.

Counterparties carry AM Best, S&P and Moody's ratings with the date each was taken, an authorisation flag and a required collateral percentage. Exposure is snapshotted as at a date with recoverable balance, collateral held, net exposure, ceded premium year to date, the rating at that date, a default probability, a provision amount and a concentration percentage.

Recording the rating as at the snapshot date rather than reading today's rating is what makes a historic exposure report reproducible. A credit position recomputed with current ratings answers a different question from the one it was asked.

Collateral is held as its own record with a release event and an expiry view, so cover that is about to lapse is visible before it does rather than after a recovery is refused.

Bordereaux with a lifecycle, not an attachmentlink

Generated, submitted, queried, accepted — each a state, with the query recorded.

Bordereaux are generated for a treaty and period with a row count and a total, then move through submission to acceptance. A query carries its note on the record, so the reason a cedant and a reinsurer disagree is part of the bordereau rather than a thread in somebody's inbox.

Row count and total sit on the bordereau itself, which is what makes a reconciliation a comparison rather than a re-derivation.

Accounts, settlements and disputes that do not overwritelink

Settlements clear or reverse; disputes escalate, resolve or withdraw. Every transition is a record.

Technical accounts are generated for a period and settled. A settlement clears or reverses as an explicit transition rather than being edited back to a previous state, so a reversed settlement leaves the reversal visible rather than leaving no trace of the original.

Disputes carry a category, an amount in dispute and an assignee, and move through escalation, resolution or withdrawal with the settled amount recorded on resolution. A dispute that resolves for less than the amount claimed records both figures, because the difference is the number a reinsurance function is actually managing.