Scope, conventions & assumptions

This index collects the site’s operating assumptions. Every entry is a one-line statement of a convention and a pointer to the section where it is defined. That section is the authoritative statement: it carries the reasoning and the escape hatch for readers whose organisation chose differently, and it is the one that changes if the convention ever does. A reader using the site as a reference can collect the rules from this page instead of reconstructing them from five chapters.

Where a page or glossary entry adopts one side of a usage the market leaves open, or departs from market usage outright, it says so with the label This site’s convention: (that exact string, so it can be searched for). The label sits at the canonical statement and on the glossary entry; the rows below that open with it are the ones where market usage differs.

Scope

Convention	Rule	Stated in
Scope	Property-catastrophe reinsurance, after the model has run. Every calculation starts from the catastrophe model’s trial-based loss output and works downstream: contracts, portfolios, pricing, capital.	Home
Exclusions	Catastrophe-model internals, exposure and experience rating, reserving and IBNR, and casualty, specialty, life, and health reinsurance are out of scope by design. The boundary is stated, not apologised for.	Home
Loss side only	A handful of operations on trial losses expresses the loss side of the core property-cat structures (quota share, catastrophe excess of loss, aggregate excess of loss) and of programs and portfolios built from them, and does not model the premium side. Other structures add terms to the catalog. Reinstatement premium is the one premium-side term modelled; commissions and profit commissions are named where they arise and not modelled.	Financial Modelling · Quota share · CatXoL
Catalog boundary	The contract catalog covers quota share, CatXoL, and AggXoL with the model event as the occurrence. An hours clause that groups losses differently from the model event, franchise and trigger terms, and count-based nth-event covers need terms the catalog does not have yet. Top & drop is a relationship between layers, named on Programs and not modelled. Reinstatement premiums sit on the premium side, outside the loss-side claim. Surplus share and per-risk excess of loss need losses per policy or location and sit outside post-model scope.	Contract catalog · Programs
Retrocession	Followed one step: the reinsurer’s own portfolio loss is a retro contract’s subject loss and the same terms apply. Designing an outward program is out of scope.	Products and contracts
Applications	API-only: inputs, transformations, outputs. No user interface, deployment, data engineering, or regulatory capital regime.	Applications

Vocabulary

Convention	Rule	Stated in
Trial	Not a year. A trial is one independent simulated realisation of a period (typically a year, not necessarily). The “Y” in YELT and YLT is the industry’s name, inherited rather than endorsed.	The trial worldview · Glossary
Event and occurrence	Two objects. The event is the catalog prototype; the occurrence is its dated instance in a trial, keyed (trial_id, timestamp, event_id). An occurrence may span several YELT rows, one for each cedent, geography, and line of business it touched.	The trial worldview
Scenario	Reserved for a deliberately chosen what-if state. Never a synonym for trial.	Glossary
Contract	The default noun for any reinsurance agreement. Treaty, layer, and program keep their specific senses; a program is the set of protections a cedent buys, never a composition.	Products and contracts · Glossary
Broker market	This site’s convention: the channel IRMI calls the brokerage market. Renamed because brokerage on this site is the fee.	Market participants · Glossary
Subject and covered	Subject is the contract’s input; covered is the filtered, in-period slice. “Subject” is never used for the post-filter stream.	Financial perspectives · Glossary
Retained and retention	This site’s convention: both name the cedent’s side. The reinsurer’s post-recovery position is its net loss, never its “retained loss”.	Products and contracts · Glossary
Composition	The noun and the act. A contract is a composition of financial terms, written $C = T_k \circ \cdots \circ T_1$ and read right to left: $T_1$ applies first. “Pipeline” is the informal name for the data flow.	Contracts as compositions
Inuring and forwarding	Two of the three relationships between a program’s contracts; top & drop is the third. Inuring subtracts one contract’s recoveries from another’s subject: the lower contract inures to the benefit of the higher. Forwarding makes one layer’s gross another’s subject.	Programs · Glossary

Data

Convention	Rule	Stated in
Trials	Equiprobable and independent: each of $N$ trials carries probability $1/N$, and one trial’s outcome says nothing about another’s. Every cedent’s YELT shares one trial index, so losses add within a trial across cedents.	The trial worldview · Portfolio analytics
Trial identity	A property of one simulation and its trial numbering. trial_id identifies a trial within one simulation and numbering, and event_id an event within one event catalog, so the occurrence key (trial_id, timestamp, event_id) matches only within one simulation and numbering. Equal trial ids in loss sets from two simulations (another vendor’s model, and usually another model version or climate view) name unrelated trials, and no check on the ids detects the difference. Every Caldera Re YELT comes from one simulation, which is what lets the roll-up sum per trial; the 20- and 1,000-trial tiers are numbered separately.	The trial worldview · Portfolios · Portfolio analytics
YELT schema	A four-column industry core (trial_id, timestamp, event_id, loss) and a five-column enrichment specific to this site (peril, geography_id, lob_id, cedent_id, tiv): one workable design, not a standard.	The trial worldview · Glossary
Primary key	(trial_id, timestamp, event_id, cedent_id, geography_id, lob_id). event_id names a prototype, so timestamp is part of the key, and one occurrence can span several rows.	The trial worldview
Sign convention	This site’s convention: loss-positive. loss and tiv are strictly positive in a base YELT and non-negative in a transformed one; a negative loss anywhere is a validation error, and other financial record types are a schema extension, never a sign flip.	The trial worldview · escape hatch: If your losses are negative · Glossary
Units	Millions of US dollars, one currency. Losses add directly across cedents; currency conversion is treated as outside the analytics.	The trial worldview · At production scale
YELT in, YELT out	Every financial term consumes and produces a YELT at row resolution: one row per occurrence, cedent, geography, and line of business. The excess formulas are written on occurrence and trial totals, and the two excess terms back-allocate their gross to the rows, by erosion across a trial’s occurrences and by share within one occurrence. A YLT is a chosen, lossy reduction: no term outputs one.	The trial worldview · Occurrence excess · Aggregate excess · Portfolios · Glossary
Rounding	Computed unrounded, displayed rounded: one decimal for losses (two in the Trial 9 split tables), two for capital and premium. Components and totals are rounded separately, so displayed parts can miss a displayed total in the last place.	Caldera Re
Demo tier	Every worked example runs on the 20-trial demo tier: the mechanics are exact, the amounts are not. The phrase “20-trial demo tier” is the standing marker.	The trial worldview · Caldera Re

Perspectives

Convention	Rule	Stated in
Point of view	The reinsurer’s, unless stated otherwise. Subject is what the cedent’s book presents; gross is what the reinsurer assumes; recoveries are what it retrocedes; net is gross minus recoveries.	Products and contracts · Glossary
Loss perspectives	Five perspectives on one set of dollars: subject, covered, gross, cedent retained, net. Expected loss splits exactly across them; tail risk does not. A residual tail is a Co-TVaR, not a difference of TVaRs.	Financial perspectives
Net	Gross is net until an outward protection exists. Net needs a second contract; a single contract’s four perspectives are the whole story.	Financial perspectives
Coverage basis	Losses occurring, throughout. An occurrence belongs to a contract when its event date falls inside the period; risks attaching and underwriting year need a policy inception date no YELT row carries.	Contract period · Glossary

Contracts

Convention	Rule	Stated in
Contracts at 100%	A composition never carries the participation share. Participation is a scaling applied at portfolio roll-up, and standalone contract metrics are reported at 100%.	Contracts as compositions
Occurrence definition	The YELT’s event_id grouping. The treaty’s hours clause is assumed to coincide with the model’s event boundaries; the occurrence definition as a term of its own is out of scope.	Occurrence excess · Glossary
Reinstatements	Encoded in the composition as the aggregate limit $(1+n)\,\ell$, a claims-side encoding that reproduces the recoveries. The reinstatement premium is a separate stream computed from the CatXoL’s claims-order output, a table keyed by occurrence, not a negative loss. In pricing that includes it, capital is measured on the gross net of reinstatement premium; the standalone-pricing and portfolio ratios are stated with and without it, and the verdicts use the view with it.	CatXoL · Reinstatement premium · Standalone program pricing
Erosion attribution	Claims order. The aggregate excess books each occurrence’s contribution to the aggregate gross chronologically; proportional attribution across a trial’s occurrences is a recognised variant the site does not use; inside one occurrence, whose geography and line-of-business rows have no claims order, the erosion is split in proportion to the rows’ losses.	Aggregate excess · Glossary
Event limit	This site’s convention: the cap sits on the subject occurrence. Filed wordings cap the ceded side; a ceded-side limit $\ell$ is the subject-side $\ell_e = \ell / q$.	Quota share · Glossary
Annual limit (quota share)	This site’s convention: the cap sits on the subject aggregate, before the cession, as the event limit’s does. A ceded-side annual limit $\ell$ is the subject-side $\ell_a = \ell / q$.	Quota share · Glossary
Contract period	Stays inside the trial year. Windows that cross a year boundary, multi-year covers, and anniversary resets are out of scope.	Contract period
Inuring	Acts between two contracts’ outputs, not inside a composition. It takes the inuring contract’s gross loss out of the beneficiary’s subject loss before the beneficiary’s own terms run.	Contracts as compositions · Programs

Metrics, capital, and pricing

Convention	Rule	Stated in
EP curve	$P(L \geq x)$, with $\geq$, so the $r$-th largest of $N$ equiprobable trials has exceedance probability $r/N$; the return period is its reciprocal.	Exceedance probability curves
VaR and TVaR	This site’s convention: VaR is always an observed trial loss, the $k$-th largest with $k = \lfloor (1-\alpha)\,N \rfloor$; interpolating quantile definitions exist and are not used. Two further rules are plain choices, not departures from market usage: when $k = 0$ the quote does not exist and the code returns NaN, and TVaR averages the worst $k$ ranks.	Value at Risk · Tail Value at Risk · Glossary
Attachment and exhaustion	This site’s convention: attachment is strict ($L > A$), where the EP curve counts with $\geq$. Exhaustion needs no label: $L \geq A + \ell$, because a loss exactly at the top of the layer pays the full limit. Both conditions describe the recovery, not the EP curve’s rank rule. The basis (trial total or occurrence) follows the contract.	Attachment probability · Exhaustion probability · Glossary
Capital	This site’s convention: the unexpected loss, $K = \text{TVaR}_\alpha - \text{EL}$ at $\alpha = 90\%$ (one economic-capital convention, not a regulatory one).	Standalone program pricing · Glossary
Technical premium	$\text{EL} + \epsilon\,\text{EL} + K\,r_c$ with $\epsilon = 5\%$ and $r_c = 10\%$, an illustrative decomposition chosen for this site. With paid reinstatements, premium plus expected reinstatement premium covers these costs and $K$ is measured on the outcome net of it. Brokerage, taxes, uncertainty loads, and the market cycle are left out.	Standalone program pricing · Glossary
Loss ratio	This site’s convention: modelled expected loss over premium (plus expected reinstatement premium where a contract has paid reinstatements), an ex ante figure, where market practice reserves the term for realised incurred over earned.	Glossary
Rate on line and RAROC	Rate on line is premium over limit; RAROC is $(\text{premium} - \text{EL} - \epsilon\,\text{EL}) / K$, with expected reinstatement premium added to the premium, and $K$ measured net of it, where a contract has paid reinstatements. Every premium-derived ratio on the site is a demonstration of method at demo scale, not a market benchmark.	Glossary · Glossary · Portfolio analytics · Caldera Re
Premium figures	100% layer figures for the six modelled contracts: $68M against a company-wide ~$250M. Caldera Re books its participation share of each.	Caldera Re

If an entry here and its source section ever disagree, the source section is right and this page is out of date. The glossary carries the definitions the conventions rest on; the Caldera Re appendix carries the data and contracts every number traces to.