Published October 2026.
Direct Answer: An exceedance probability curve plots how likely it is that losses will reach or exceed a dollar amount in a defined period—usually one year—not a single “expected” number. Occurrence EP answers the worst single event; aggregate EP answers total loss from all events combined. The 100-year point is a loss level with roughly a 1% annual chance of being equaled or exceeded, not a forecast that you will lose exactly that amount every century. Reading the curve correctly is what turns a modeling report from a black box into a renewal, reinsurance, and capital conversation you can defend.
Your catastrophe modeling report probably arrived as tables, maps, and at least one chart with a gently sloping line and unfamiliar axis labels. Underwriters and brokers may cite a “250-year loss” or “1% EP” as if it were one fixed fact. The curve is the fact: it encodes modeled outcomes, including the rare tail that drives reinsurance and retentions. This article stays on how to read that curve—occurrence versus aggregate, exceedance probability on the axes, return-period points, and why the far right matters. For the wider workflow, see the catastrophe modeling complete guide for 2026.
What your report is showing on the axes
Most vendor and broker EP exhibits use loss on the horizontal axis (sometimes logarithmic) and exceedance probability on the vertical axis, or they swap axes and keep the same relationship. Exceedance probability is the chance that losses in the modeled year will meet or exceed a given threshold. Reports may instead show return period, the approximate inverse of that annual probability. A 1% exceedance probability corresponds to a 100-year return period when the curve is on a per-annual basis.
Read the footnotes before you read the line. The curve is meaningless without peril scope, financial perspective (ground-up, gross, or net of reinsurance), and exposure basis (values, limits, deductibles, business interruption). A ground-up occurrence EP curve for Florida wind is not comparable to a net aggregate EP curve for all perils nationally unless you normalize those choices. Peril context appears in our hurricane, earthquake, and wildfire catastrophe peril analysis.
Probability versus return period
If the vertical axis is labeled “EP” or “exceedance probability,” a point at 2% means there is a 2% modeled chance per year that losses reach at least the loss on the horizontal axis at that height. If the axis is labeled “return period in years,” the same point might read 50 years, because 1 ÷ 0.02 ≈ 50. Brokers often speak in return periods; model output often speaks in EP. They are two labels on the same idea.
Do not treat return period as a countdown timer. A 100-year loss is not “due” after ninety-nine quiet years. It is statistical shorthand for a low-probability, high-severity zone on the curve. Multiple years can exceed that threshold in a row, or you can go decades without touching it, and both outcomes remain consistent with the EP framing.
Occurrence EP versus aggregate EP
Occurrence exceedance probability describes the distribution of the largest single event loss in the year (confirm how the report defines the period). Aggregate exceedance probability describes the distribution of the sum of all event losses in the year. For a schedule with one dominant peak per season, occurrence and aggregate can look similar at high return periods. For a wide footprint with many events, aggregate EP at moderate thresholds often sits to the right of occurrence EP because several smaller events can add up to a total no single occurrence reaches.
Your report says “occurrence EP” when the question is how bad one hurricane, one quake, or one wildfire cluster could be. It says “aggregate EP” when the question is whether annual retained loss could exhaust an aggregate retention, erode an aggregate stop, or trigger an aggregate cat bond index. Treaty language follows the same split: per-occurrence limits respond to occurrence loss; annual aggregates respond to aggregate loss. Our overview of reinsurance treaty structures including quota share, excess of loss, and aggregate maps those mechanics to the curves you see.
A simple numeric contrast
Take illustrative annual event outcomes for one location year, not model output: Event A costs $40M, Event B costs $35M, and no other events apply. Occurrence loss is $40M (the maximum single event). Aggregate loss is $75M ($40M + $35M). On an occurrence EP curve, the point at $40M reflects the probability of any single event reaching $40M. On an aggregate EP curve, the point at $75M reflects the probability the sum of events reaches $75M. Per-occurrence retention tracks the occurrence curve at attachment; an annual aggregate deductible tracks the aggregate curve at that sum.
What the 100-year point really means
When someone says “the model’s 100-year loss is $X,” they mean the loss level on the EP curve where annual exceedance probability is about 1%, or return period is about 100 years, under the report’s peril, perspective, and catalog assumptions. It is a quantile of the modeled distribution, not a budget line item you should expect every year. Average annual loss (AAL) sits far to the left on the same curve—typical years—not in the tail.
Reports rarely print exactly 1.00% EP; you may see 0.95% or 1.05% at the quoted dollar amount. Small changes in uncertainty treatment or loss amplification assumptions can move the 100-year dollar materially. Vendor methodology differs, as summarized in how RMS, AIR, and Verisk approaches quantify catastrophe risk; your job is to read the curve and footnotes in hand.
Why the tail wags the tower
EP curves are flat on the left and steep on the right. Moving from the 60-year to the 30-year point often changes the dollar amount modestly. Moving from the 250-year toward the 100-year region can change it dramatically because you are sampling the sparse tail of the simulated catalog. Reinsurance attachments and aggregate triggers often live in that tail. A modest shift in modeled 250-year occurrence loss can mean a layer clears your program or leaves a gap.
The body of the curve sets AAL and budgeted cat load; the tail sets capacity and solvency stress. Portfolio roll-ups compound tails through correlation and concentration; see catastrophe portfolio management, accumulation, PML, and reinsurance for schedule-level context. On a single EP chart, the tail still dominates whenever someone asks about “worst case” or “1-in-100.”
Occurrence tail versus aggregate tail
For multi-state wind and hail schedules, aggregate tail loss can exceed occurrence tail loss at the same return period because the model sums events across the footprint. For earthquake with one dominant rupture scenario, occurrence and aggregate tails may nearly coincide. When your report shows only one curve, confirm which one it is before you attach it to a treaty clause.
Worked reading: $500M total insured value
Assume a simplified commercial property schedule with $500M total insured value across coastal wind-exposed locations, ground-up perspective, hurricane peril only, occurrence EP, one-year horizon. The report’s curve (illustrative, not a vendor quote) might read approximately as follows:
At 10-year (10% EP), modeled occurrence loss might sit near $25M—enough to matter for deductibles and lower tower layers. At 50-year (2% EP), perhaps $95M. At 100-year (1% EP), perhaps $140M. At 250-year (0.4% EP), perhaps $220M. These numbers are invented for pedagogy; your report will differ. The exercise is how to use them.
Suppose your physical damage deductible is $5M per occurrence and you carry $100M excess of $5M per occurrence. At 50-year occurrence loss of $95M, retained loss is $5M and the layer pays $90M. At 100-year loss of $140M, the layer pays $100M and $35M remains uninsured or flows to a higher layer if you bought one. The EP curve gives a ladder of thresholds so you can ask whether each rung matches a contract.
Switch to aggregate EP on the same schedule. If the 100-year aggregate point is $180M while the 100-year occurrence point is $140M, a treaty with a $150M annual aggregate retention could be penetrated by aggregate loss even when no single occurrence exceeds $140M—multiple events in one season. Phrase it for your file: 100-year occurrence $140M, 100-year aggregate $180M means single-event capital peaks lower than annual summed stress.
Check whether the curve is before or after reinsurance. A gross occurrence EP curve at $140M at 100-year does not tell you net retention until you apply the program. Net EP curves shift left when reinsurance is efficient and can kink where layers attach. If the report only shows ground-up, reconstruct net with your layer chart or ask the broker for a net view.
The “your report says X” decoder
“AAL is $8M.” Expected value of annual loss across simulated years—center of mass of the EP curve, not the tail. Useful for pricing; weak for limit adequacy.
“1% occurrence EP loss is $140M.” Same as 100-year occurrence in annual framing; one-event severity at that quantile.
“250-year aggregate is $310M.” Sum-of-events view at 0.4% annual exceedance; use for aggregate covers and some cat bond triggers.
“EP curve shifted higher at 100-year after the update.” Tail repricing, not necessarily proportional change to AAL or every layer premium. Ask what changed: exposure, vulnerability, or event set. Broader update economics sit in climate risk pricing and catastrophe model updates in 2026; your curve still demands line-by-line reading.
“PML” on a broker summary. Often a return-period loss with internal definitions. Match the acronym to the EP point on the chart; do not assume PML equals a number from a different curve type. When tables and charts disagree, trust the documented export and reconcile before bind.
Using EP curves in renewal dialogue
Bring the curve, not only the headline 100-year figure. Mark deductibles, layer attachments, and co-participation on the loss axis for the relevant occurrence or aggregate EP. Ask where the net curve crosses those marks at 50-year, 100-year, and 250-year. That aligns risk management, broking, and finance on what is priced versus retained.
For buying decisions, compare marginal tail: incremental loss between 100-year and 250-year points relative to premium for the covering layer beats debating AAL alone. Modeling fees are a separate procurement topic—see catastrophe modeling costs, fees, and who pays.
Common chart variants and axis traps
Some exhibits plot EP on a logarithmic scale to compress the tail. “Non-exceedance” flips the vertical axis: that curve rises left to right while an exceedance curve falls. Reading the wrong orientation inverts tail intuition.
Multi-year EP curves exist for bond tenors. A five-year EP is not the same as a one-year EP at the same return-period label. Align the horizon with the contract before you cite a point.
FAQ
What does an EP curve actually show?
An EP curve shows the modeled probability that losses in a defined period will meet or exceed each loss amount on the chart, given the perils, exposure, and financial perspective in the report. It is a full distribution summary, not a single forecast. Each point pairs a dollar threshold with an exceedance probability or return period. You use it to see typical years on the left and rare tail outcomes on the right.
What is the difference between occurrence and aggregate exceedance probability?
Occurrence EP describes the largest single event loss in the period, while aggregate EP describes the sum of all event losses in the period. A year with two major events can produce a high aggregate loss with a lower occurrence loss. Treaty structures mirror the distinction: per-occurrence limits respond to occurrence EP, and annual aggregates respond to aggregate EP.
What does the 100-year loss point on an EP curve mean?
The 100-year point is the loss level at roughly 1% annual exceedance probability under the report’s definitions. It means the model assigns about a one-in-one-hundred chance per year that losses reach or exceed that amount, not that the loss repeats on a calendar schedule. It is a tail quantile for planning and reinsurance, not expected annual spend.
Why does the EP curve matter more than average annual loss?
AAL summarizes the average of many simulated years and sits on the low-probability side of the curve, which aligns with pricing load but not with limit adequacy or insolvency stress. Reinsurance attachments, retentions, and board tail questions live at 100-year, 250-year, and beyond. The curve shows those regions; AAL alone does not.
How do EP curves set reinsurance attachment points?
Teams overlay occurrence or aggregate EP points with proposed attachments and limits to see where net loss remains at selected return periods. If 100-year occurrence loss pierces a proposed retention, the tower or terms adjust until net EP at key points fit appetite. Aggregate EP supports annual aggregate deductibles and aggregate stop-loss structures.
What are the most common misreads of EP output?
People treat return period as a due date, confuse occurrence with aggregate, compare curves with different peril or net-versus-gross definitions, and cite AAL when the question is tail adequacy. Another frequent error is ignoring footnotes on catalog, epistemic uncertainty, and exposure date. Always match the curve type to the contract clause before you bind.