RESEARCH DOSSIER
Cyber Insurance Sublimits Retentions and Coverage Towers
Abstract. The economic value of a cyber insurance program depends on the contractual conversion of incident losses into recoverable amounts through coverage-specific limits, retained losses, annual aggregates and successive insurance layers. This monograph examines that conversion through comparative analysis of specimens and an issued excess-policy record, selected United States judicial opinions, supervisory evidence, actuarial literature and reproducible hypothetical examples. A sublimit bounds a defined coverage or loss category; a retention determines the portion borne by the insured; a coverage tower distributes eligible loss vertically between contracts according to attachment and exhaustion conditions. These mechanisms interact. Erosion of underlying coverage by a sublimited payment may count toward attachment while the corresponding category remains excluded by the excess agreement. Aggregate depletion, claims expense, time deductibles and reporting conditions change the recoverable amount even when nominal capacity remains unchanged. The research identifies the distinction between potential limit, available limit, recognized underlying payments and realized indemnity as central to measurement. It evaluates incentive effects, risk financing, legal uncertainty and portfolio accumulation, and specifies data requirements for empirical testing. Numerical examples illustrate contractual arithmetic under declared assumptions. The study does not estimate a global cyber loss distribution or derive an optimal real-world program from unavailable claims microdata.
Keywords: Cyber insurance; coverage sublimit; retention; deductible; self-insured retention; excess insurance; attachment point; aggregate exhaustion; follow-form coverage; defense expense; business interruption; coinsurance; risk financing; accumulation; policy wording; actuarial measurement.
Introduction
A cyber event may simultaneously disable operations, expose personal data, generate extortion demands, require forensic services and trigger third-party claims. A headline amount of insurance describes only one part of the financial response. The policy must first recognize the circumstances and categories of loss. A claim must then satisfy the applicable conditions before the contract determines the retained portion, the maximum eligible payment and the capacity remaining after previous losses. A subsequent insurer evaluates its separate promise.
The object of this research is the financial architecture of that sequence. Its three dimensions are the horizontal allocation of losses across differentiated coverage limits, the division of loss between policyholder and insurer through retentions, and the vertical allocation of recoveries through excess contracts. The analysis is anchored in the parent Science Cyber Insurance, whose repository manuscript was consulted for disciplinary scope.
The research asks how contractual language establishes those three dimensions, how they change the distribution and observability of loss, and which data would support comparison across programs and over time. It combines doctrinal comparison of specified instruments, a purposive reading of primary wordings, descriptive market evidence, actuarial deduction and scenario analysis. The wordings are selected for informative differences, not as a probability sample; judicial opinions illustrate interpretation of excess language in their own legal settings. Publication forms and dates identify the examined editions, not the conditions of every subsequent renewal. The research cutoff is November 18, 2026. An extracted document is treated as evidence only for the inspected provisions; no visual inspection of inaccessible PDF pages is asserted.
The sequence from technical incident to financial loss is especially important. Attack activity, vulnerability exploitation, operational disruption, incurred expense, a submitted claim and a paid claim have different observation units. A loss may be genuine without satisfying a particular insurance trigger. Conversely, covered defense costs can consume limits before a liability determination is reached. These distinctions guide the analysis and delimit the interpretation of aggregate market statistics.
Conceptual and Historical Foundations
The contractual unit of financial protection
Cyber insurance is a bundle of promises organized around defined loss and liability categories. First-party cover can address the insured organization's response and recovery costs; third-party cover can address eligible liabilities and defense expenditures. The economic exposure extends beyond the policyholder to customers, employees, suppliers and other persons whose losses may or may not give rise to covered obligations. Define gross economic loss as the quantified financial harm within a declared accounting boundary. Define eligible loss as the portion satisfying a particular insuring agreement, subject to its definitions and exclusions but before financial limitations are applied. This is a useful analytic ordering; actual wording can combine these operations.
The policy limit bounds the insurer's promise at the level specified by the contract. A per-event limit restricts payment for a defined event; an annual aggregate restricts cumulative payment over a policy period. A coverage limit applies to an insuring agreement, and a sublimit imposes a narrower bound on a designated portion. Limits may be nested or additional. The schedule and endorsements determine which relation holds. An advertised aggregate cannot be summed with nested sublimits without double counting.
A retention transfers the initial specified portion of an eligible loss to the insured. A deductible may perform similar economic work while differing in payment and administration. A self-insured retention may require evidence of satisfaction before insurer performance; its treatment of expenses, defense obligations and payments from other insurance must be established by the agreement. A franchise deductible, percentage retention, waiting period and coinsurance provision have different mathematical effects. The term excess can describe the insured's retention in some markets or an insurance layer above underlying insurance. Meaning follows the clause, not the label alone.
A tower is the ordered set of primary and excess promises for a risk. Its nominal capacity is the sum of purchased layer widths where their attachment bands are contiguous. Its actual availability depends on remaining capacity and on the compatibility of the separate promises. A lead insurer can occupy a primary or excess layer; several insurers can share participation in one layer. Horizontal participation within a band and vertical attachment to another band must be distinguished from reinsurance, which transfers risk from an insurer rather than extending a direct policyholder promise.
Historical evidence and specializations
Romanosky, Ablon, Kuehn and Jones (2019) examined U.S. insurance filings in their content analysis of cyber policies. Their comparison connects policy terms, underwriting questions and filed pricing schedules. Variation within the observed forms supports studying financial restrictions at the wording level. The filings are historically informative; they are neither a census of current multinational towers nor a collection of final negotiated prices.
The GAO's 2021 market study recorded an expansion of take-up among clients of a particular broker from 26 percent in 2016 to 47 percent in 2020, alongside industry accounts of reductions in limits in some sectors. These observations illustrate that coverage demand and available capacity can move in different directions. The denominator is the broker's existing clients, not the entire economy. Increased take-up does not establish increased financial adequacy for the same insured risk.
The development of incident-specific coverage has generated a distinct actuarial question. Chong et al. (2025), in Incident-specific cyber insurance, consider the choice of limits and deductibles for different incident types. Their stated scope leaves policy-wide aggregates outside the analysis. This dossier addresses the interaction of those instruments with aggregate and excess arrangements through explicit, restricted illustrations rather than attributing an aggregate model to that work.
The historical record suggests a tension between standardization of quantitative terms and specialization of covered risks. A common currency amount makes policies superficially comparable, while increasingly fine distinctions among incident categories create different payoff functions. A meaningful scientific comparison must therefore preserve the definitions associated with each amount.
Restricted analytical notation
Write x for eligible loss, r for a flat retention, s for an applicable payment sublimit, and a for remaining aggregate capacity. If the contract deducts r first, applies s as a cap on insurer payment and permits no additional recovery, the insurer payment is min(max(x − r, 0), s, a). The insured's retained eligible loss is x minus that payment. Uncovered economic loss must be added separately. All amounts in this relation have the same currency basis and valuation time.
If there is no applicable sublimit, s can be removed from the minimum. If a sublimit instead caps eligible loss before the retention is taken, payment becomes min(max(min(x, s) − r, 0), a). The two expressions can differ. They represent alternative contractual orderings and are not interchangeable definitions of coverage.
These identities use ordinary indemnity arithmetic. They serve as controlled illustrations for interpreting source clauses. They do not constitute a novel theory, a fitted loss model or a normative insurance design.
Primary Documents and Scope of Coverage Sublimits
Nested limits and expense erosion
The AXIS Cyber Insurance Policy specimen (2020 form edition, p. 15) establishes a hierarchy in which coverage limits and sublimits sit within broader policy capacity. Its liability provisions also specify erosion by claim expenses. The contractual significance is that payment for one designated category reduces available capacity for other categories whenever both share the aggregate. The amount attributed to a coverage limit is a bound, not a segregated asset reserved for that category.
The AmTrustCyber sample policy (2021 form edition) likewise states an aggregate structure and identifies coverage-specific maximum amounts within it. Its declarations provide fields for differentiated limits and retentions, alongside time parameters for interruption and supplier-related loss. The sample is a document with placeholders: the presence of a field demonstrates a configurable contract feature, not a priced limit or uniform industry practice. Its extended discovery provision does not increase aggregate capacity.
Compare two hypothetical primary policies, both with a $5 million aggregate and $0.25 million retention. Policy A has a $1 million recovery sublimit nested within the aggregate. Policy B allows eligible recovery expense to access the entire aggregate. For a standalone $2 million eligible recovery loss and otherwise unused limits, A pays $1 million and B pays $1.75 million, assuming the retention is applied before the payment cap. A leaves $1 million of eligible loss with the insured; B leaves $0.25 million. Their aggregate limits are identical but their payable amounts differ by $0.75 million.
Suppose A then covers an independent eligible liability loss of $5 million. Its remaining aggregate is $4 million. After the new $0.25 million retention, the calculated liability payment is limited to $4 million. Total policy payment across the two events is $5 million. If those events occur in the opposite order with equivalent assumptions, the liability payment is $4.75 million, leaving $0.25 million for the recovery claim. Total aggregate payment is again $5 million, but the allocation across expenses changes. Timing and allocation matter even when aggregate totals coincide.
An expense-inclusive limit also has a different economic meaning from a damages-only limit with separate defense capacity. Assume a $3 million aggregate, $1 million of covered defense expense paid by the insurer, and $3 million additional covered damages after the retention is already satisfied. Only $2 million remains for damages when defense erodes the aggregate. A hypothetical separate defense grant would preserve the damages capacity if its own conditions are met. No conclusion concerning actual liability or settlement value follows from the illustration.
Coverage boundaries, event definitions and time
A sublimit can identify an expense rather than an incident. A single intrusion can generate restoration costs, extortion loss, business interruption and liability; each may have a different coverage grant. An underlying loss may be allocated to more than one accounting category, but duplicate reimbursement must be excluded from financial measurement. Allocation is particularly important when the categories share an aggregate but differ in sublimits.
A dependent business interruption grant can separately define the qualifying service provider, the type of disruption and the restoration period. A waiting period may exclude loss incurred during initial hours, or may operate as a threshold under other wording. A monetary retention may apply in addition. Suppose eligible interruption loss accrues uniformly at $50,000 per hour for 30 hours. A 12-hour time deductible that excludes initial loss leaves $900,000. With an additional $100,000 retention and a $600,000 payment cap, payment is $600,000. A threshold-style waiting period would produce a different eligible loss base; the monetary cap could conceal that difference in this example. One cannot convert hours into a universal money deductible without specifying the loss path and wording.
An incident response timeline captures technical facts alongside financial and legal milestones. Nelson et al. (2025), in NIST SP 800-61 Rev. 3, position incident response within continuing cybersecurity risk management. That guidance supplies a security context; it does not determine policy coverage. In the European Union, GDPR Article 33 requires qualifying supervisory notification without undue delay and, where feasible, within 72 hours of awareness, subject to the stated risk exception. The insurance notice requirement remains a separate time condition. The legal deadline neither establishes a covered claim nor fixes the applicable sublimit.
The evidentiary role of the complete contract
Declarations, base forms, endorsements and schedules must be read together. A base form can say that a sublimit is part of the aggregate, while a specifically negotiated endorsement can modify the relationship. The same financial term may use different currencies, attachment bases or policy periods in international placements. A comparison should record the exact controlling edition, insured entity, effective period, governing law where known, trigger, category definition, cap basis and expense treatment.
The material examined here includes specimens rather than all the executed contracts in a representative market sample. The valid inference is that specified contractual designs exist, not that any design predominates. Verification of a binding program would additionally require its issued declarations, all relevant endorsements and evidence of premium and contractual formation where disputed.
Mechanisms and Evidence for Retention Structures
Risk transfer and loss participation
Under the flat deductible convention with adequate limits, an insured share increases as r rises at fixed loss x. For payment Y equal to min(max(X − r, 0), l), where X is a nonnegative integrable eligible loss and l is a fixed payment limit, expected payment equals the integral of the survival function of X between r and r + l. This follows from writing Y as the length of the interval of loss above r and below r + l. The expression is valid for a single loss, given the financial ordering; it says nothing about legal eligibility.
If the loss distribution is fixed, increasing r cannot increase expected payment. The magnitude depends on the probability mass near and above the affected thresholds. A high retention can eliminate insurance payment for frequent small incidents; a large loss can still exhaust the layer. Changes in the exposure population, prevention, wording or reporting behavior can offset this purely arithmetic comparative static.
In a deliberately simple example, let annual eligible loss consist of one of four mutually exclusive amounts, in millions: 0 with probability 0.65, 0.2 with probability 0.20, 2 with probability 0.10 and 12 with probability 0.05. For a $4 million insurer limit above $0.25 million retention, the expected payment is 0.10 × 1.75 + 0.05 × 4 = $0.375 million. At a $0.75 million retention, it is 0.10 × 1.25 + 0.05 × 4 = $0.325 million. Expected gross loss is $0.84 million; expected residual eligible loss is $0.465 million and $0.515 million respectively. These values exclude premium, tax, expenses, uncovered losses and capital costs. The probabilities are invented to expose arithmetic, not estimated from cyber data.
Premium difference need not equal expected indemnity difference. Acquisition costs, claim administration, underwriting expense, cost of capital and the insurer's assessment of uncertainty also matter. Price comparisons drawn from filed schedules or broker quotes must identify whether the same cover, exposure and retention are being compared.
Multiple retentions and aggregation
An incident-specific deductible can apply separately to every eligible claim or event. A related-events clause can combine incidents or claims into a contractually defined unit. The AXIS specimen and AmTrust sample examined above both contain circumstances in which the largest applicable retention governs instead of repeatedly charging all applicable retentions. The operation is conditional on each form's defined grouping.
Consider two independent covered events with eligible losses of $400,000 and $600,000, each subject to a $250,000 retention. With adequate limits, separate application produces payments of $150,000 and $350,000, totaling $500,000. If a controlling clause instead aggregates the two as one event with one $250,000 retention, the payment is $750,000. This difference of $250,000 can arise from contract classification while gross eligible loss stays constant. Where another clause applies a lower occurrence limit to the grouped claim, the effect may reverse. Aggregation is both a deductible and a limit issue.
A ransomware campaign and later breach litigation may span several dates and entities. Whether expenses relate to one event or multiple claims is not decided by shared threat-actor attribution alone. The contractual relatedness criteria, event definition, timing and applicable law matter. Statistical labels should therefore distinguish technical incident grouping from insurance aggregation.
Incentive mechanisms and financial resilience
Retention can provide a reason for insureds to invest in preventing losses that remain within their financing layer. It can also discourage notification of minor events when the expected recovery is negligible, subject to notice obligations. The effect on prevention is a competing-mechanism question. A resource-constrained insured may face difficulties funding remediation under a high retention; a well-capitalized insured may deliberately retain frequent losses. Insurance may also improve risk management through service access or underwriting requirements.
The scientific issue is the behavioral margin, rather than an assumption that greater retained loss always improves cybersecurity. For example, observed reductions in insurer claim frequency following retention increases may reflect lower reporting of small losses or changes in the insured population. A causal conclusion about prevention requires incident records independent of claims, security controls measured before renewal, exposure denominators, and a credible comparison.
A first-party incident may require rapid cash outlays before eligibility is settled. An entity with $1 million in liquid resources, a $0.5 million retention and $2 million in urgent expenses may have a material financing gap even if maximum indemnity is substantial. Claims advances, payment timing and independent vendor arrangements affect resilience. The contract's ultimate liability and the timing of cash access should be measured separately.
Statistical observability
A deductible changes the observation process. With perfect reporting above r and loss X, an insurer's positive-payment sample observes X conditional on X exceeding r, with exact distribution dependent on the policy. A payment cap produces right-censoring: observing indemnity at the limit does not identify the full eligible loss. Non-reported incidents below retention are not zeros for incident frequency. An incurred estimate and a final payment are also distinct.
If the underlying positive losses follow distribution F with density f, and X is observed above r but censored at r + l, an uncensored ground-up observation x contributes f(x)/(1 − F(r)) to a likelihood conditional on observation; a capped observation contributes [1 − F(r + l)]/[1 − F(r)]. This expression assumes known thresholds, complete reporting above r, continuous F and a single incident. Discrete observations, reporting delays, risk selection and varying caps require modifications. It prevents treating policy-induced truncation as evidence of naturally smaller loss severity.
Chong et al.'s incident-specific analysis is relevant to the selection of differentiated financial terms, while Zeller and Scherer (2022) use marked point processes to address dynamic and dependent risk. Their illustrated model is not empirically validated against a comprehensive claims dataset. A distributional assumption or theoretical optimization should be assessed against its data and contractual scope before it is used to infer available insurance capacity.
Comparative Measurement of Layer Attachment
Nominal towers and operative contracts
In a simple fully aligned tower with ground-up eligible loss x, flat retention r and layer j of width w_j attaching at accumulated underlying insurer limits b_j, payment y_j is min(max(x − r − b_j, 0), w_j). The primary insurer has b_1 = 0. Each band is measured on a post-retention basis. The formula assumes unused limits, continuous bands, compatible eligibility, recognized exhaustion, no expense or allocation dispute and timely compliance with conditions. If b_j is quoted on a ground-up basis, r must not be subtracted twice.
For a hypothetical tower of $5 million primary, $5 million first excess and $10 million second excess, above $0.5 million retention, the total nominal insurer capacity is $20 million. The first excess begins above $5.5 million ground-up eligible loss; the second above $10.5 million. Payments are as follows, in USD millions.
| Eligible loss | Primary | First excess | Second excess | Insured residual |
|---|---|---|---|---|
| 0.3 | 0 | 0 | 0 | 0.3 |
| 4 | 3.5 | 0 | 0 | 0.5 |
| 8 | 5 | 2.5 | 0 | 0.5 |
| 15 | 5 | 5 | 4.5 | 0.5 |
| 25 | 5 | 5 | 10 | 5 |
The table conserves eligible loss in every row. It describes a payoff under stated assumptions, not a promise by any carrier. If the second layer is shared equally by two insurers and their several obligations are identical, each owes half the calculated second-layer payment. An insurer's insolvency or separate defense cannot be modeled by the arithmetic alone.
Primary evidence of unequal follow form
The publicly filed Starr excess policy (2021–2022) identifies an underlying cyber liability policy with a $5 million limit above a $500,000 SIR. Section V of its excess wording treats underlying sublimits in two distinct ways: excess coverage for a sublimited category requires specific endorsement, while payments under that underlying sublimit can still count in determining erosion. Its insuring clause also specifies acceptable sources of covered underlying payments. This is evidence of the structure of one issued public record; the example tower above is independent of that contract.
The Great American excess form (2020 form edition) exhibits a comparable separation between recognition of payments under underlying sublimits and the absence of excess cover for the associated claims. Its own conditions address exhaustion and separate notice. A follow-form description therefore requires attention to the excess policy's departures from underlying grants. Actual insurance payments can reduce the vertical threshold for future eligible losses without making an excluded category itself recoverable.
A controlled illustration isolates the distinction. Let the primary layer have $5 million annual capacity, with a $1 million nested sublimit for category A, and let an excess layer of $5 million recognize primary sublimit payments for erosion but exclude A. Ignore retention for this calculation. A category A loss of $4 million yields $1 million primary and no excess. Remaining primary capacity is $4 million. A later independent $7 million eligible category B loss yields $4 million primary and $3 million excess if step-down mechanics recognize that earlier payment. Across the two events, total payment is $8 million and residual is $3 million. An alternative excess policy that did not recognize that erosion and maintained its original $5 million attachment for B would pay $2 million for B, leaving an additional $1 million gap, if no other provision filled it. The alternative is a declared counterfactual, not attributed to the cited forms.
The existence of this difference means that claim-specific coverage and recognized underlying exhaustion require separate fields in any claims database. A single variable recording remaining tower capacity cannot represent both.
Exhaustion by payment and contractual gaps
A qualifier on the source of underlying payment can materially affect attachment. In Qualcomm v. Certain Underwriters (California Court of Appeal, 2008), a D&O dispute involved a $16 million settlement with an underlying insurer against a $20 million underlying limit. The court interpreted the particular excess language as requiring satisfaction of the specified underlying payment condition. The opinion is a non-cyber comparison of exhaustion doctrine, not a national cyber insurance rule.
The New York appellate court's JPMorgan Chase v. Indian Harbor decision (2012) also rejected the proposed method of treating an insured's absorption of the settlement shortfall as sufficient under the disputed excess wording. These opinions demonstrate the importance of text and governing law when an insured seeks to bridge a compromised lower layer. They cannot determine the outcome of a contract explicitly permitting different sources of qualifying payment.
Suppose a $5 million primary insurer settles for $4 million, the insured pays $1 million to bridge the difference, and qualifying loss above that point amounts to $3 million. If the excess policy permits the insured's covered-loss payment to count toward underlying exhaustion, the $3 million may fall within the excess layer, subject to remaining conditions. If a different policy demands actual payment of the full $5 million by the named underlying insurer and the condition is enforceable, the excess may remain unattached. The $1 million bridge can thus alter both the immediate uninsured amount and eligibility for the next layer. The comparison is conditional, not a legal opinion on a live dispute.
Depletion and insolvency
The Australian-market Emergence Cyber Excess Protection wording (March 2021, general conditions) permits the layer to follow reductions of underlying cover through qualifying payments, and specifies how it functions after exhaustion. Payments outside scope and the inability of an underlying insurer to pay are treated differently from qualifying erosion. Its terms also address changes to primary insurance. This document establishes a contractual comparison in an Australian-market form; it does not establish all Australian law.
If an underlying layer has $2 million uncollectible because of insolvency, an excess agreement can preserve its attachment rather than substituting for the missing insurer. That exposure differs from exhaustion by $2 million in covered payments. In a probabilistic study of recovery, insurer counterparty condition is therefore distinct from cyber incident severity.
Aggregate exhaustion after multiple events can alter attachments during the year. In the preceding examples, earlier recognized payments reduce later effective underlying capacity only under an assumed contractual step-down. A reinstatement, a fresh retention or a separate aggregate can produce alternative results. A change in one layer's form at renewal need not propagate to all others.
Comparative market evidence and its denominator
The NAIC 2026 market report for the 2025 reporting year distinguishes primary, excess and endorsement business and reports an increase in the total loss ratio from 48.7 percent in 2024 to 53.0 percent in 2025. These are market report figures, not layer-level severity probabilities. Policy counts, premium volumes and loss ratios have different numerators and denominators; the number of policy records does not reveal the number of unique insured organizations or distinct complete towers.
The original report's segment distinctions are useful for observing market structure but do not identify how often individual excess contracts attach. Estimating the probability of first-excess attachment requires the number of covered exposure periods with comparable towers and contract-recognized exhaustion. It also requires tracking prior erosion. Excess premiums per insured cannot be inferred by dividing aggregate excess premium by all primary and endorsement policy records.
A longitudinal design would use policy-period and layer-level observations with consistent inflation and currency treatment. Policy years, accident years, report years and payment years should be distinguished. A large claim paid in a later calendar year may be associated with an earlier policy period. Comparisons across jurisdictions must account for reporting and regulatory differences and the distribution of firm sizes and industries.
An attachment frequency estimate can use the number of exposed layer-years with at least one qualifying payment, divided by all observed layer-years of the same design and maturity. It captures the chance of positive payment in those layer-years, not the conditional probability of an excess insurer paying after every qualifying incident. Both metrics require explicit denominators and censoring treatment.
Financial, Legal, and Institutional Consequences
Claims coordination and accounting evidence
Claims management converts uncertain financial exposure into documented obligations. A structured loss record should identify incident and claim dates, covered entity, cause and affected service, loss components, coverage classification, retention and expense treatment, payments, reserves and remaining limits. A separate attachment record should identify the source and status of underlying payments and whether the relevant excess insurer recognizes their erosion.
The cited Starr and Great American forms contain separate excess notice conditions and procedures for changes to underlying wording. Those provisions illustrate why a primary insurer's acceptance of a claim cannot automatically stand for a decision by the entire tower. Documentary consistency across contracts is a measurable component of risk transfer. A claim can exhaust a lower layer numerically while higher coverage remains contested.
Reserving requires attention to expenses and allocation. If an insured's estimated ultimate eligible loss is $12 million in a continuous $5 million primary plus $10 million excess program above a $0.5 million retention, a simple gross allocation yields $5 million primary and $6.5 million excess. If $2 million of the loss falls in an excess-excluded category, the allocation may change materially depending on primary sublimits and recognized erosion. A reserve based on headline loss without that classification risks overstating recovery. The example assumes no previous payments or other complications.
The insurer's accounting liability and the insured's insurance recoverable are distinct reporting perspectives. The same unresolved event can generate estimates that differ because the parties assess legal eligibility, timing and uncertainty differently. A scientific dataset should record the reporting basis of each estimate rather than treating differences as errors by default.
Capital, accumulation and reinsurance
The use of sublimits and higher attachments can reduce exposure to specific high-severity categories under a fixed loss distribution. But transforming individual risks does not remove dependence. A common service provider outage can generate simultaneous claims under many programs, including portfolios with high retentions if the event is sufficiently severe. An insurer's share of a layer is a contingent obligation across its participating risks, not a fixed amount independently accessible under every imaginable event.
For portfolio contributions Y_i with finite second moments, variance of their sum equals the sum of individual variances plus twice the sum of pairwise covariances. This standard identity explains why layer losses may remain concentrated despite lower marginal payments. It is incomplete as a measure of catastrophe risk: equal covariance can accompany very different joint tails. The Zeller and Scherer study provides a research precedent for representing different cyber event types and accumulation through dynamic stochastic processes; its simulation illustrates model behavior rather than a calibrated global distribution.
Reinsurance changes an insurer's net risk and capital exposure. The insured's direct entitlement remains determined by its direct contracts. A facultative placement may itself follow form; an excess-of-loss treaty attaches to defined ceded loss under a separate aggregate or occurrence definition. Neither should be counted as additional direct insured capacity unless a legally applicable direct promise actually exists.
The OECD catastrophic risk assessment (2026, section 3.6.1) identifies constraints on higher cyber limits and on protection for potentially catastrophic incidents. Its discussion supports studying capacity and dependency together. Public support, where considered, involves separate questions of public risk-bearing, incentive design and implementation. This dossier does not estimate the expected fiscal cost or recommend a universal government backstop.
Coverage clarity and international institutions
The Lloyd's Market Association's follow-form clarification (2020) addresses when underlying cyber wording can satisfy market clarity requirements for specified follow-form excess and facultative arrangements, recommending that relevant wording be identified and recorded. This is an institutional precedent for retaining the exact underlying wording in a layered underwriting record. Its application is limited to the described market context.
EIOPA's 2022 supervisory statements announcement highlights target-market clarity, systemic exclusions and management of non-affirmative cyber exposures. The relevance here is both consumer-facing and prudential: the coverage description must correspond to the risk actually assumed. A headline aggregate alone is insufficient evidence of value when essential categories have narrow bounds. The announcement is supervisory material, not an international uniform rule for interpreting every cyber contract.
The comparative cases and documents have different institutional status: U.S. state appellate interpretations of specific disputes, a public procurement record, specimens, an Australian-market wording, European supervisory material and international policy analysis. The GDPR is a statutory instrument within its applicable legal scope. The NIST publication is technical guidance. None can substitute for the controlling wording and governing legal system of a specific placement. International scientific analysis requires classifying sources by their legal and empirical force.
Access, concentration and the protection gap
The maximum marketed capacity is not equivalent to the risk financing available to an individual enterprise. Coverage-specific gaps, self-financing requirements, excluded categories and contractual incompatibility can coexist with apparently high aggregate limits. Those gaps affect smaller entities through liquidity and affordability, while larger enterprises can face constraints on total purchased high-severity capacity.
Measuring a protection gap requires specifying a financial harm denominator. The gap against all economic loss differs from the gap against defined eligible loss or the gap against a chosen financing target. A company's own compensation of affected customers also differs from total social harm. These distinctions matter when public discussions combine market premium growth with estimates of cyber damage.
Integrated Scientific Synthesis and Research Agenda
Results of the contractual comparison
The evidence supports three connected conclusions. First, the sublimit is a distributional restriction over a defined category of eligible loss, often nested within a broader aggregate in the examined primary forms. Consequently, the aggregate and coverage sublimits cannot be treated as additive unless the exact contract grants separate capacity.
Second, retained loss is determined by amount, grouping and timing. The same monetary deductible can produce different total recovery when claims are combined or separated; a time deductible requires an interruption loss path. The appropriate unit of actuarial observation is therefore contract-specific.
Third, excess attachment contains both arithmetic and contractual components. The examined excess wordings show that recognized erosion of the underlying layer can diverge from the scope of excess coverage. The judicial examples demonstrate that payment-source conditions can govern attachment even when financial loss surpasses the nominal threshold. A coverage tower is a collection of conditional promises whose practical value depends on compatibility and availability.
These conclusions synthesize existing legal and actuarial concepts with direct comparison and calculated examples. They do not establish new universal empirical laws. Their strength is the identification of which quantities must remain distinct in loss measurement.
A reproducible comparative exercise
A reproducible study should specify the loss vector, relevant financial parameters and event order before computing total payments. The following illustration assumes one retention per event of $0.2 million, an annual primary aggregate of $3 million, a nested annual category A payment sublimit of $0.8 million, and a $4 million annual excess layer that excludes A but recognizes all qualifying primary payments for aggregate depletion. Other categories are eligible in both layers; expense treatment is included in the eligible loss values. There is no reinstatement.
Event A1 has $1.5 million eligible loss. Primary pays $0.8 million and excess pays zero. Category A sublimit is exhausted; $2.2 million primary aggregate remains. Event B1 has $2.5 million eligible loss. After retention, $2.3 million remains; primary pays $2.2 million and excess pays $0.1 million. Event A2 has $0.9 million eligible loss. Its sublimit is exhausted and excess excludes it, so it pays zero. Event B2 has $3 million eligible loss. Primary is exhausted; applying the retention again produces $2.8 million excess payment.
Total gross eligible loss is $7.9 million; primary payment is $3 million; excess payment is $2.9 million; residual is $2 million. Remaining excess aggregate is $1.1 million. The residual decomposes into $0.2 + $0.5 million for A1, $0.2 million for B1, $0.9 million for A2 and $0.2 million for B2. The arithmetic verifies total conservation. Under a counterfactual excess endorsement covering A after underlying sublimit exhaustion, results depend on the new attachment and coverage basis; simply adding the excess limit to the sublimit would be unsupported.
This exercise shows why a simple loss vector must be coupled to a contract state at each event. An empirical reconstruction should retain both the order and identity of payments. The calculation can be reproduced by subtracting event retention once, limiting primary payment to remaining category and policy caps, updating each affected cap, then applying the qualifying residual to excess only if its coverage and exhaustion rules allow it.
Empirical evaluation priorities
A longitudinal study of actual insurance programs would collect a protected panel of anonymized insured-year records and link these to layer-level wordings, revisions, independent cybersecurity indicators, incident records and paid and incurred financial data. It should distinguish observed companies without insurance from companies with insurance but no submitted claim; otherwise selection and reporting effects cannot be separately assessed.
To assess retention changes, a within-insured comparison before and after renewal could control for exposure, security investment, industry and reporting changes. Even with fixed effects, endogenous selection of retentions and contemporaneous changes in risk remain potential confounders. A credible natural experiment would require evidence that the change in retention is plausibly independent of anticipated loss, conditional on measured factors. This study neither possesses such a panel nor claims causal identification.
To assess attachment, a study can reconstruct recognized depletion using paid claim histories and observed wording changes. It should classify omitted claims, unsettled claims, adverse legal decisions and declined payments separately. Sensitivity analysis should vary unresolved allocation, settlement recognition and exhaustion rules, instead of imposing the same uncensored distribution on all layers.
For empirical market comparison, sampling should stratify direct primary, excess and endorsements and identify the location of the insured risk and governing jurisdiction. Currency conversion should use declared dates; exposure adjustment and inflation methodology should be transparent. The source records reviewed here do not provide a linked international panel. This limitation bounds all descriptive and comparative conclusions.
The scientific agenda extends to service-provider dependence, loss timing, contract versioning and insurer counterparty risk. Each can alter financial protection without altering the headline limit. These priorities are practical research directions derived from the examined materials and remain open to new evidence.
Conclusion
Cyber insurance sublimits, retentions and coverage towers determine the distribution of recoveries from a specified loss across coverage categories, the insured's retained exposure and successive insurance promises. Their interaction establishes the financial meaning of purchased coverage.
Primary documents reveal nested limits and deductible provisions whose operation depends on defined events, expense categories and prior payments. Excess wordings illustrate the distinction between coverage for a loss category and recognition of payments toward underlying erosion. Selected appellate decisions show how an exhaustion condition can turn on actual qualifying payment rather than nominal loss. Actuarial research and public market reports provide a basis for examining incident-specific choices, dynamic dependence and capacity, within their documented empirical limits.
The principal analytical contribution is a contract-aware interpretation of financial protection: a nominal amount must be accompanied by the coverage scope, retention convention, available aggregate, recognized underlying exhaustion, excess eligibility and payment timing. The examples demonstrate why these quantities cannot be collapsed into a single headline figure. Reliable empirical comparison requires linked wording, claims and exposure records; the lack of a representative international linked dataset precludes universal estimates of adequacy or causal effects.
Scientific understanding of cyber insurance thus advances through precise contract interpretation and disciplined measurement of the actual risk transfer. A coverage tower has a financial meaning determined by the promises that remain effective for the event, the available capacities under those promises, and the conditions under which each successive promise becomes payable.
References
AXIS. (2020). AXIS Cyber Insurance Policy. Specimen AXIS 1012561 0120, 22 pp. AXIS specimen
Technology Insurance Company, Inc. (2021 form edition). AmTrustCyber sample policy. CYDEC 0321, distributed by Victor Insurance. AmTrustCyber sample
Starr Surplus Lines Insurance Company. (2021). Starr Secure Excess Liability Policy, No. 1000634720211, July 1, 2021–July 1, 2022. County of San Bernardino public records. SS FL 001 XS and SS FL 002 XS (10/12). Starr excess policy
Great American Insurance Group. (2020 form edition). Excess Follow Form Insurance Policy. CYX 4000. Great American excess form
Emergence Insurance Pty Ltd. (2021, March). Cyber Excess Protection: Important Information & Policy Wording. CXP-001.1. Emergence policy wording
Qualcomm, Inc. v. Certain Underwriters at Lloyd's, London, 161 Cal. App. 4th 184 (Cal. Ct. App. 2008). California appellate opinion
JPMorgan Chase & Co. v. Indian Harbor Insurance Co., 98 A.D.3d 18 (N.Y. App. Div. 2012). New York appellate opinion
Romanosky, S., Ablon, L., Kuehn, A., & Jones, T. (2019). Content analysis of cyber insurance policies: how do carriers price cyber risk? Journal of Cybersecurity, 5(1), tyz002. DOI: 10.1093/cybsec/tyz002. Journal of Cybersecurity article
Chong, W. F., Linders, D., Quan, Z., & Zhang, L. (2025). Incident-specific cyber insurance. ASTIN Bulletin, 55(2), 395–425. DOI: 10.1017/asb.2025.9. ASTIN Bulletin article
Zeller, G., & Scherer, M. (2022). A comprehensive model for cyber risk based on marked point processes and its application to insurance. European Actuarial Journal, 12, 33–85. DOI: 10.1007/s13385-021-00290-1. European Actuarial Journal article
U.S. Government Accountability Office. (2021). Cyber Insurance: Insurers and Policyholders Face Challenges in an Evolving Market. GAO-21-477. GAO report
National Association of Insurance Commissioners. (2026). Report on the Cybersecurity Insurance Market. Reporting year 2025. NAIC market report
Lloyd's Market Association. (2020, December 17). Lloyd’s cyber coverage requirements – clarification for excess layer and facultative reinsurance policies. LMA20-057-IR. LMA bulletin
European Insurance and Occupational Pensions Authority. (2022, September 22). EIOPA publishes supervisory statements on exclusions related to systemic events and the management of non-affirmative cyber exposures. EIOPA announcement
OECD. (2026). Examining the need for government-supported financial protection for natural hazard, cyber and infectious disease risks. In Financial Protection Against Catastrophic Risks. OECD Publishing, Paris, March 5. OECD report chapter
Nelson, A., Rekhi, S., Souppaya, M., & Scarfone, K. (2025). Incident Response Recommendations and Considerations for Cybersecurity Risk Management: A CSF 2.0 Community Profile. NIST SP 800-61 Rev. 3. DOI: 10.6028/NIST.SP.800-61r3. NIST final publication
European Parliament and Council of the European Union. (2016). Regulation (EU) 2016/679 of 27 April 2016 (General Data Protection Regulation). Official Journal L 119, 1–88, May 4, Article 33. EUR-Lex legal text
Bogdanova, A. (2026). Cyber Insurance. Insurance by Bogdanova. Aisentica Development. Parent Science repository manuscript consulted; live page retrieval unavailable during this session.