Building a portfolio and reading EP curves

A portfolio is a named set of locations you want to analyse together. It is the unit of work for anything that spans more than one address: aggregate hazard exposure, accumulation checks, and — where available — loss output.

Creating one and loading locations

Create a portfolio from Portfolio, then add locations. Each location carries coordinates and, optionally, the attributes that make downstream analysis meaningful: property type, insured value, construction type, year built, storeys, square footage.

Those optional fields are worth populating. Hazard scoring works from coordinates alone, but anything that turns intensity into a loss estimate needs to know what is standing there. A portfolio of bare coordinates gives you accumulation analysis; a portfolio with exposure attributes gives you considerably more.

How many locations a portfolio holds depends on your plan — see pricing.

Aggregate hazard exposure

With locations loaded, the portfolio view scores each one and aggregates. This answers the accumulation question: how much of this portfolio sits in high-hazard geography for a given peril, and where is it concentrated?

The output that usually matters is the concentration, not the average. A portfolio with a moderate mean hurricane score can still be unacceptable if the exposure is clustered on one stretch of coast, and the per-location breakdown is where you see that. Sort by score, look at the top of the list, and check whether those locations are geographically independent of each other.

When a live event is active, the portfolio view also shows which locations fall inside its consequence zones — the fastest way to answer "what of ours is in this" during an event.

What an EP curve is

An exceedance probability curve is the standard way catastrophe losses are expressed. It plots loss on one axis against the probability of exceeding that loss in a year on the other. Read it in either direction: pick a probability and read off the loss, or pick a loss and read off how often you would expect to exceed it.

Two curves are conventionally shown together:

• OEP, occurrence exceedance probability — the distribution of the largest single event loss in a year. This is the curve that speaks to per-occurrence reinsurance.
• AEP, aggregate exceedance probability — the distribution of total annual loss across all events. This is the curve that speaks to aggregate covers and to your own annual result.

AEP sits above OEP, because a year's total is at least as large as its worst single event. If you ever see them cross, something is wrong with the run.

From the curve come the numbers people quote: AAL, the annual average loss, which is the mean of the annual loss distribution and the right number for pricing; and PML at a return period, which is a point read off the curve and the right number for capital and reinsurance structuring. A 250-year PML is the loss level exceeded with probability 1 in 250 in any given year. It is emphatically not a loss that arrives on a 250-year timetable.

Reading one honestly

The tail of an EP curve is the least certain part of it and the part everyone quotes. Return periods far beyond the length of the historical record are extrapolations from a fitted distribution, and reasonable modelling choices produce materially different answers out there. Treat a far-tail PML as an estimate with wide uncertainty, not a measurement.

A curve is also only as good as the exposure behind it. If insured values in the portfolio are stale, every number derived from them is stale by the same proportion — and a curve looks equally authoritative either way.

Availability

EP curve output depends on validated stochastic event sets for the peril in question, and is an Enterprise capability — see pricing. Where a peril's event set has not completed validation, the platform does not produce a curve for it rather than producing one you should not rely on. Validation documentation is published alongside the model.

Aggregate hazard exposure and accumulation analysis do not depend on stochastic modelling and are available wherever portfolios are.

Where to go next

Reading a hazard score covers the per-location input. Exports covers getting portfolio output into your own systems.