Expected recovery against premium, retention and limit options priced on simulated years, limit exhaustion, sublimits and the uninsured scenario that drives the tail.
Cyber insurance is usually bought on price and habit: last year's limit, a retention the broker calls standard, a premium compared with the expiring one. A loss exceedance curve lets a buyer ask better questions. What does the policy return in an average year? What does each extra dollar of limit buy in the tail? Which bad years does the program not touch at all? This guide prices those questions for Halden Marsh Industries, Inc., a fictional manufacturer renewing on January 1, 2027, with a simulated gross ALE of $5,016,300. Every premium and term in the case is a teaching assumption, not market data, but the method applies to any renewal: compare expected recovery with premium, then compare what each option does to the curve.
Worked in full in The Cyber Risk Quantification Handbook by Julian R. Sterling, with every figure reproduced in live Excel workbooks.Get the DRM-free PDF, $19.99 →
The expiring program, which the case uses as the base option, has five features that all matter to the model:
The annual premium is $2,550,000. The model applies each term trial by trial to the 1,000 simulated years described in the guide on how to quantify cyber risk with FAIR, and ignores waiting periods, coinsurance and exclusions other than S5. A real wording deserves the same line-by-line reading before any number is trusted.
Averaged over the 1,000 years, the policy recovers $2,238,423 a year. The premium is $2,550,000, so the expected cost of transfer, premium minus expected recovery, is $311,577. In an average year the company is worse off insured: retained losses of $2,777,876 plus the premium make a total of $5,327,876, against a gross ALE of $5,016,300.
That is normal and not a reason to cancel. Insurance is bought for the tail, not the mean. The net 95th percentile is $12,767,404 against $20,906,669 gross, and the net 99th percentile is $26,253,490 against $45,214,845 gross. Insurance is also what brings the gross breach of the board's one-in-100 tolerance back inside it. The right question is therefore what each option costs in expected terms and what it buys at the percentiles the board cares about.
Three retentions were quoted, all with the $20,000,000 aggregate limit:
| Retention per event | Premium | Expected recovery | Net 95th percentile |
|---|---|---|---|
| $1,000,000 | $2,950,000 | $2,538,851 | $12,542,116 |
| $2,000,000 | $2,550,000 | $2,238,423 | $12,767,404 |
| $5,000,000 | $1,850,000 | $1,620,959 | $14,626,683 |
| Retention per event | Mean retained loss | Mean total cost | 99th percentile total cost |
|---|---|---|---|
| $1,000,000 | $2,477,449 | $5,427,449 | $29,203,490 |
| $2,000,000 | $2,777,876 | $5,327,876 | $28,803,490 |
| $5,000,000 | $3,395,341 | $5,245,341 | $28,103,490 |
Moving from $2,000,000 to $5,000,000 saves $700,000 of premium and gives up $617,465 of expected recovery, so the mean total cost falls by $82,535 and the expected cost of transfer drops to $229,041. The price is paid in the middle of the curve: the net 95th percentile rises by $1,859,279. A company with a healthy balance sheet and a tolerance stated at the one-in-10 and one-in-100 levels can carry that; one whose lenders watch quarterly earnings may not.
A low retention mostly buys back small, frequent losses at a markup. Going down to $1,000,000 raises the premium to $2,950,000 for an expected recovery of $2,538,851, and the net 95th percentile barely moves. Small losses are cheaper to retain and budget for than to insure.
Look again at the net 99th percentile: it is $26,253,490 at every retention. The one-in-100 year is not driven by the deductible. It is driven by losses above the $20,000,000 aggregate limit and by S5, which no option covers. In those years the policy is already exhausted or silent, so who pays the first dollars of each event changes nothing at the top.
The 99th percentile total cost in the second table confirms it: it moves only by the difference in premium between options. If the tail is the concern, the retention is the wrong lever. The limit and the coverage gaps are the right ones.
Three aggregate limits were quoted at the $2,000,000 retention:
| Aggregate limit | Premium | Expected recovery | Probability the limit is exhausted |
|---|---|---|---|
| $10,000,000 | $1,950,000 | $1,758,727 | 7.6 percent |
| $20,000,000 | $2,550,000 | $2,238,423 | 2.7 percent |
| $35,000,000 | $3,050,000 | $2,498,589 | 1.1 percent |
| Aggregate limit | Net 95th percentile | Net 99th percentile | Mean total cost |
|---|---|---|---|
| $10,000,000 | $15,009,765 | $35,220,372 | $5,207,573 |
| $20,000,000 | $12,767,404 | $26,253,490 | $5,327,876 |
| $35,000,000 | $10,900,481 | $24,854,924 | $5,567,711 |
Limit exhaustion is the number to read first. At $20,000,000 the policy runs out in 2.7 percent of simulated years; at $10,000,000 it runs out in 7.6 percent. Each of those years is a year when the company carries everything above the limit alone.
Raising the limit to $35,000,000 costs $500,000 more premium for $260,166 more expected recovery, so the expected cost of transfer climbs to $551,411. What it buys is a net 99th percentile lower by $1,398,566. Cutting the limit to $10,000,000 saves $600,000 but raises the net 99th percentile by $8,966,882. The asymmetry is the finding: the first $20,000,000 of limit does most of the tail work, and the extra limit up to $35,000,000 adds little because the remaining tail comes from S5.
A headline limit overstates protection when sublimits and exclusions cap specific scenarios. Three features shape the Halden Marsh curve:
For S5, the levers are controls and, where available, a separate cover. The controls analysis prices them with the same trials; all three proposed controls together cost $2,180,000 and cut the gross ALE by $1,868,525, something no insurance option does.
Price is only half the comparison. The board's tolerance points, at most 10 percent above $15 million and at most 1 percent above $40 million, are a pass or fail screen that each option must clear before its cost is discussed. On the current program the net curve reads 4.1 percent and 0.4 percent, both inside. An option that saves premium but pushes either point back over its limit is not a saving; it is a decision to accept more risk than the board approved, and it needs the board's explicit agreement. Run the screen on the net curve of every quote, and record the result next to the premium so the trade-off is visible on one line.
The free Cyber Risk Monte Carlo Model applies a retention and a limit to S1 alone. On the current terms it returns an expected recovery of $1,164,843, a net ALE of $474,966 and a net 95th percentile of $2,000,000, exactly the retention, because a single covered event in a one-in-20 year is absorbed by the policy above the deductible. Change the retention or the limit and the figures move at once, which makes the trade-off concrete before the full portfolio is rerun.
With that page in hand, the conversation moves from "what did we pay last year" to "what does each dollar of premium buy at the percentile we care about". The full insurance chapter, including how the net figures feed the board's appetite test and incident disclosure, is in The Cyber Risk Quantification Handbook.
Usually yes, because insurance is bought for the tail. Halden Marsh pays $2,550,000 for an expected recovery of $2,238,423, but its net 99th percentile falls to $26,253,490 from $45,214,845 gross, and the board's one-in-100 tolerance is met.
Compare the premium saved with the expected recovery given up and the change in the net 95th percentile. Moving from $2,000,000 to $5,000,000 saves $700,000 of premium, gives up $617,465 of recovery and raises the net 95th percentile by $1,859,279.
Read the probability that the limit is exhausted and the net 99th percentile for each option. At $20,000,000 the limit runs out in 2.7 percent of simulated years; a $35,000,000 limit lowers the net 99th percentile by only $1,398,566 because uninsured S5 drives the remaining tail.
It caps recovery for one scenario whatever the headline limit. The cloud outage is capped at $5,000,000 per event and payment fraud at $1,000,000 after a $250,000 retention, so their large years stay mostly retained. Model each sublimit trial by trial, as the free Cyber Risk Monte Carlo Model does for retention and limit.
This guide is one arc of The Cyber Risk Quantification Handbook. The book follows one company from its heat map to five quantified scenarios: calibrated estimates, the FAIR model, Monte Carlo simulation, the portfolio loss exceedance curve against board tolerance, the return on controls, the cyber insurance renewal and an incident materiality decision.
Paperback PDF edition, $19.99 Book page and companion workbooks
Free download: Cyber Risk Monte Carlo Model (Excel).
Illustrative figures computed by the book model; the company, its scenarios and every estimate are fictional. Rules described in principle as read in November 2026. Education only, not insurance, legal or security advice.