---
title: the average year is not the year that breaks you
canonical_url: https://ensurance.app/guide/the-average-year-is-not-the-year-that-breaks-you
markdown_url: https://ensurance.app/guide/the-average-year-is-not-the-year-that-breaks-you.md
subtitle: "expected annual loss is an average. the bad year sets the capital, and living systems move some of it, not all"
category: nature-finance
---

# the average year is not the year that breaks you

*expected annual loss is an average. the bad year sets the capital, and living systems move some of it, not all*

Expected annual loss is a number nobody ever pays. Most years come in under it. Then one comes in far over it, and that is the year the reserve, the reinsurance and the plant's insurance were sized for.

**What is tail risk?** Tail risk is the chance of rare, very large losses: the far end of the range of outcomes, where one year costs many times the average. An average such as expected annual loss folds those years in and hides them. The tail is what sets the capital an insurer holds, the reinsurance it buys and whether a balance sheet survives.

Some of what sits under that curve is alive. The forest above a town decides how hard a fire runs, the floodplain upstream decides where the water spreads, and the marsh in front of a street takes energy out of a storm before it arrives. They were doing that work long before anyone modeled it, whether or not anyone pays for them. [**Ensurance**](/manual/ensurance?from=guide) is how the people who depend on that ground can pay for its keeping. It is not the forest or the marsh, and it does not stop a storm.

## expected annual loss and probable maximum loss

A catastrophe model runs thousands of simulated storms, floods or fires against a set of buildings, each event with a yearly chance. Two numbers come out of it.

**Expected annual loss** (EAL, also called average annual loss) is each event's loss times its chance, added up. It is the long-run average cost per year. How a specific action lowers it is set out in [how an intervention lowers the loss](/guide/how-an-intervention-lowers-the-loss?from=guide).

**Probable maximum loss** (PML) is the loss at a chosen rarity: in property catastrophe work, usually the 1-in-100 or 1-in-250 year loss. It is read off the **exceedance curve**, the chart that shows the chance that a year's losses pass any given amount. EAL is the average of that whole curve. PML is one point far out on its right-hand end.

Two portfolios can share an EAL and have very different PMLs. One loses a steady, modest amount most years. The other loses little for a decade and then a great deal in one season. The average treats them as the same business. Nobody who holds capital does.

## the bad year carries the capital

Look at how a catastrophe-exposed premium is built. A 2011 Casualty Actuarial Society ratemaking seminar handout writes it in three parts: expected loss and expense, plus the **net cost of reinsurance**, plus profit, where the profit is meant to pay for the risk the carrier keeps. Only the first term is mostly the average. The other two are mostly about the bad year: the cover bought against it, and the return on the capital held for it. [How insurance premiums are calculated](/guide/how-insurance-premiums-are-calculated?from=guide) walks through each slice.

The same holds outside insurance. A utility, a manufacturer or a lender pays for its bad year through its deductible and retention, its reserves, its borrowing cost and the weeks a site is shut. None of that shows up in an average.

So a cost-benefit case built on the change in EAL alone can undercount adaptation. If work on the ground lowers losses that sit inside a retention or a deductible, the owner keeps that saving directly, and it never shows in an average-based premium comparison. Whether it also lowers a reinsurance or capital charge is the carrier's and the reinsurer's call. In the evidence so far that load mostly stays where it is even when expected loss falls, and one placed policy (Tahoe Donner, below) is the only confirmed case of cover priced on the work. The opposite error is as real: a case that claims the work lowers the 1-in-250 loss, when it mostly moves the frequent years, overcounts. The skill is putting each loss in the band where it actually falls.

One caution: Resources for the Future found in 2025 that the costs of wildfire property retrofits are "orders of magnitude greater than the insurance savings." A premium credit alone rarely pays for the work.

## what living systems move on the loss curve

Split the curve into three bands. The studies were not built by band; each sits where its own wording puts it, and the placement is our reading.

| band | what it looks like | what living systems move | evidence nearest this band |
|---|---|---|---|
| Frequent | The losses most years see: the small flood, the grass fire, the hailstorm | Often a real share where the peril is set on the ground (flood, wildfire). Nothing for hail, which is weather | Florida's best-protected coastal-habitat areas had about half the flood claim frequency in frequent, less severe storms (Swiss Re Institute, 2025, observed National Flood Insurance Program claims). Tahoe Donner's 2025 wildfire policy, priced on forest treatment, carried an 89% lower deductible than the stated counterfactual (sponsors' figures, placed policy; a wildfire deductible, placement is ours) |
| Mid-size | The bad year in a decade or two | Some, specific to the mechanism and the place | Coastal wetlands cut modeled annual flood loss in Ocean County, New Jersey by 16% on average and up to 70% at low elevation (Narayan et al., 2017, built on industry catastrophe models). Ecological forestry in the Northern Sierra cut modeled average annual loss by 40–60% (The Nature Conservancy and WTW, 2021) |
| Extreme | The year the top of the tower is bought for: a major hurricane's wind at landfall, a large earthquake, a record fire | Little for peak perils. Coastal flood is the partial exception (see reefs, below). A record event can exceed any buffer | Our published position: [what reinsurance cannot cede](/guide/what-reinsurance-cannot-cede?from=guide) and [the model learns from a past that is leaving](/guide/the-model-learns-from-a-past-that-is-leaving?from=guide) |

The mid-size studies report only a change in the average; they do not say which band the saving came from. When you see a percentage from a nature study, ask which part of the curve moved.

The line between bands is not sharp, and it differs by peril and by whose loss you count. Here the extreme band means a peak-peril loss at the reinsured top of the tower, not every 100-year flood on a coast. Beck and colleagues (2018) modeled economic flood damage behind the world's coral reefs, not insured loss, and found that without them, annual expected flood damage would double, frequent-storm costs would triple and 100-year flood damage would rise 91%. On some coasts the effect reaches well up the curve. That still does not make a reef a substitute for tail cover: at the top of the tower the ground does not replace reinsurance, retrocession or catastrophe bonds, and anyone who says it does is overclaiming.

The middle bands also carry the money. Secondary perils, the frequent and mid-size hazards such as flood, wildfire and severe convective storm, caused a record 92% of 2025's $107 billion in global insured natural catastrophe losses (Swiss Re Institute, 2026). Not all of that is small, and not all of it is set on the ground: the Los Angeles wildfires were about $40 billion, and hail is weather. [Secondary perils are not secondary](/guide/secondary-perils-are-not-secondary?from=guide) sorts them peril by peril.

:::johnson
**the average year does not break anyone.**

The bad year sets the capital, the reinsurance and the survival question. Where the peril is set on the ground, living systems can move a share of the frequent and mid-size losses, and little of the extreme tail, so the case should claim the first and keep cover for the second.

[read next: the bet that isn't in your scenarios →](/guide/the-bet-that-isnt-in-your-scenarios?from=guide)
:::

## what this means for the case

**Put the right loss in the case.** Ask for the exceedance curve with and without the work, on the same events and the same buildings, not just the two averages. Then report the change in EAL and the change at the points where your retention and deductible sit, and at the lower reinsurance layers if your carrier will show them. That is where living systems earn their keep. The full method, worked on a hypothetical utility with averages only, as a floor, is the last post in this series: [the cost-benefit case for the ground upstream](/guide/the-cost-benefit-case-for-the-ground-upstream?from=guide).

**Keep tail cover for the top of the tower.** Raising, hardening and reinsurance do real work, and the extreme band still needs them; the ground replaces none of them. It is the part of the loss nobody else is paying to lower.

**Read your average by peril.** For a US place, the Federal Emergency Management Agency (FEMA) [splits expected annual loss by peril](/place?from=guide); see which hazards carry the average and which of those are set partly on the ground.

## where ensurance fits

**Ensurance** funds the condition of a named piece of ground before the season. A certificate ties the funding to the forest, floodplain or marsh it pays for, and proceeds go to the people doing the keeping. Over time a policy certificate can move from ensured toward entrust, permanent protection; ensured alone does not finish that path.

The limits, plainly. We do not compute a tail, run catastrophe models or sell a risk model; a modeler's output is an input. Ensurance does not hold tail risk, reinsure anyone or pay claims. No carrier or company has booked a loss reduction funded through ensurance; the evidence above is the field's, not ours.

If you write or buy property cover, the services behind this, delivered with technical partners (a climate and nature risk assessment that reads suppliers' models, wildfire and flood resilience work, monitoring of the work), are on [ensurance for insurers](/solutions/insurers?from=guide&topic=after-the-score).

## frequently asked questions

### what is tail risk?

The chance of rare, very large losses at the far end of the range of outcomes. An average hides it; the tail sets the capital, the reinsurance and whether a balance sheet survives.

### what is the difference between expected annual loss and probable maximum loss?

Expected annual loss is the long-run average loss per year across every modeled event. Probable maximum loss is the loss at one chosen rarity, usually the 1-in-100 or 1-in-250 year event.

### can nature-based solutions reduce tail risk?

Some of the loss curve, mostly in the frequent and mid-size bands, where studies show real, mostly modeled reductions for flood and wildfire. Some coastal studies reach further up the curve. Peak-peril tail capital is largely untouched, and a record event can exceed any buffer.

### why does a cost-benefit analysis on average loss undercount adaptation?

Because the premium and the balance sheet also pay for the bad year. A saving inside a retention or deductible is the owner's, and an average misses it; whether a reinsurance or capital charge falls is the carrier's call.

## sources

[Casualty Actuarial Society, Ratemaking seminar handout (2011)](https://casact.org/sites/default/files/presentation/rpm_2011_handouts_ws1-jabben.pdf) — Catastrophe-exposed premium written as expected loss and expense, plus net cost of reinsurance, plus profit for the risk retained

[Resources for the Future, Working Paper 25-30, *From Risk to Reward* (2025)](https://www.rff.org/publications/working-papers/from-risk-to-reward-insurance-discounts-for-wildfire-mitigation/) — Wildfire property retrofit costs are "orders of magnitude greater than the insurance savings"

[Swiss Re Institute, "Natural habitats can reduce flood losses" (2025)](https://www.swissre.com/institute/research/topics-and-risk-dialogues/climate-and-natural-catastrophe-risk/coastal-and-flood-protection-natural-habitats.html) — Florida NFIP claims, 2009–2022: best-protected areas had about half the claim frequency in frequent, less severe storms; frequency only

[The Nature Conservancy, Tahoe Donner wildfire resilience insurance (April 2025)](https://www.nature.org/en-us/newsroom/ca-wildfire-resilience-insurance0/) — $2.5 million cover over 1,345 acres; premium 39% lower and deductible 89% lower than without the forest work; sponsors' figures

[Narayan, S. et al. (2017), "The value of coastal wetlands for flood damage reduction in the northeastern USA," *Scientific Reports* 7: 9463](https://doi.org/10.1038/s41598-017-09269-z) — Ocean County, New Jersey: 16% lower annual flood loss on average, up to 70% at low elevation; modeled with industry catastrophe models

[The Nature Conservancy and WTW, *Wildfire Resilience Insurance* (2021)](https://wtwco.com/en-US/Insights/2021/07/wildfire-resilience-insurance-quantifying-the-risk-reduction-of-ecological-forestry-with-insurance) — Northern Sierra ecological forestry: modeled average annual loss 40–60% lower

[Beck, M. W. et al. (2018), "The global flood protection savings provided by coral reefs," *Nature Communications* 9: 2186](https://doi.org/10.1038/s41467-018-04568-z) — Without reefs, annual expected flood damage would double, frequent-storm costs triple and 100-year damage rise 91%; modeled

[Swiss Re Institute, press release, 19 March 2026](https://www.swissre.com/press-release/Wildfires-storms-floods-contribute-to-record-92-of-global-insured-losses-in-2025-says-Swiss-Re-Institute/7b39b1a5-b878-4a55-a5ff-bf5aa561a675) — $107 billion global insured natural catastrophe losses in 2025; secondary perils a record 92%; Los Angeles wildfires about $40 billion

## the series

1. [the score is the easy half](/guide/the-score-is-the-easy-half?from=guide) — **Read first:** What a climate risk score leaves undecided, and the five questions after it
2. [the average year is not the year that breaks you](/guide/the-average-year-is-not-the-year-that-breaks-you?from=guide) — Expected annual loss is an average; the bad year sets the capital, and living systems move some of it, not all
3. [the bet that isn't in your scenarios](/guide/the-bet-that-isnt-in-your-scenarios?from=guide) — A stress test reports the risk you modeled; the source your site depends on is usually the scenario nobody ran
4. [the watershed isn't in the asset register](/guide/the-watershed-isnt-in-the-asset-register?from=guide) — A climate risk model needs hazards, assets and damage curves; the ground that sets the hazard has no row
5. [waiting is a position](/guide/waiting-is-a-position?from=guide) — When the bad years get worse, the protection is worth more and each year of waiting costs more
6. [the cost-benefit case for the ground upstream](/guide/the-cost-benefit-case-for-the-ground-upstream?from=guide) — A worked example on a hypothetical water utility, from the loss to a payment a budget can book