RESEARCH DOSSIER
Marginal Capital Consumption in Underwriting Acceptance
Abstract. Marginal capital consumption is the change in an insurer's required risk capital attributable to acceptance of a specific contract into a specified portfolio. This monograph investigates the decision-relevant difference between prospective capital requirements and the attribution of existing capital, explains why dependency and the shape of the aggregate loss tail can reverse underwriting preferences, and analyzes how contractual reinsurance, scarce risk-bearing resources and model uncertainty alter an acceptance decision. The research combines mathematical derivation, reproducible synthetic scenarios, critical comparison of capital-allocation literature, and comparative analysis of insurance supervisory frameworks. A discrete loss experiment demonstrates that identical expected policy losses can produce materially different marginal unexpected-loss capital and that a quantile-based criterion can rank contracts differently from expected shortfall. A second experiment isolates the nonlinear effect of exhausted reinsurance layers. The analysis establishes the conditional nature of capital pricing and shows why an underwriting decision requires both an economic contribution measure and a check against binding prudential, liquidity and concentration constraints. The conclusions apply to a documented model and decision horizon; no confidential insurer data or empirical industry effect size is asserted.
Keywords: Underwriting; marginal capital consumption; economic capital; solvency capital; expected shortfall; value at risk; Euler allocation; portfolio dependence; reinsurance; risk appetite; capital pricing.
Introduction and conceptual framework
Insurance underwriting converts a proposed exposure into a contingent contractual liability. Acceptance simultaneously changes the insurer's expected cash flows, distribution of adverse outcomes, use of reinsurance and remaining risk-bearing capacity. The economic meaning of a policy therefore resides partly in its relationship to the existing portfolio. A contract can be attractive in isolation and undesirable within a concentration of correlated contracts; conversely, accepting a moderately volatile exposure may improve diversification.
The principal scientific object is the incremental capital consequence of a single acceptance decision. The decision unit can be one policy, a negotiated line on a syndicated placement, an endorsement or a renewal relative to its expiration baseline. The reference point must be explicit. At renewal, continuation of the old contract is not always the relevant counterfactual: a future book without renewal and a book with a new term can differ in exposures, premium and treaty conditions. The unit of measurement in this study is a monetary amount, measured at a common valuation date and over an explicitly specified risk horizon.
Three questions organize the analysis. How can capital consumption be measured when the candidate and incumbents share loss drivers? Under what conditions can a contract with a positive expected margin impose excessive demand on tail capacity? How does the effectiveness of reinsurance and uncertainty about the modeled distribution change the attribution and the decision?
The foundational distinction concerns the objects being compared. Expected claims and related liabilities are funded through the valuation and funding of obligations. Economic risk capital represents a selected measure of adverse deviations or adverse balance-sheet change. Regulatory capital follows the applicable jurisdiction's definitions, aggregation rules and eligibility requirements. Available capital is the stock of resources recognized under the specified basis. A cost of capital is a rate or required compensation for using capital over time. These quantities have different dimensions and institutional meanings. Neither premium nor a reserve is automatically an increment of capital.
The risk measurement literature provides a disciplined starting point. Artzner et al. (1999) organize risk measurement around coherence and discuss failure of diversification under quantile-based rules. This research uses that published abstract for its conceptual scope, then derives the relevant distinctions explicitly below. A coherent loss-based risk measure is monotone, translation equivariant, positively homogeneous and subadditive. For losses, translation equivariance adds a deterministic loss to the measured risk. These are mathematical properties of a chosen functional; they do not guarantee that the estimated distribution, risk horizon or model inputs represent an insurer accurately.
A scientific assessment of acceptance requires an incremental experiment rather than a comparison of unrelated standalone metrics. Both portfolios are valued on the same basis. The candidate is added to exactly the same initial portfolio, using identical exogenous states wherever possible. Any changes in premiums, recoverables, expenses, assets or reinsurance terms attributable to acceptance belong in the with-contract case. Assumptions that change independently of the proposed policy belong in a new baseline or an explicitly identified stress case.
Individual policy incremental risk
Counterfactual capital measurement
Let X be the aggregate discounted adverse loss for the existing portfolio, and Y the candidate policy's additional loss, initially under an additive gross-loss representation. Positive numbers denote losses. For a given capital functional K, the finite change is
A useful economic convention for this study is K(Z) = rho(Z) - E[Z], where rho is a monetary risk measure and the mean is finite. This measures unexpected loss relative to the modeled mean. Other capital definitions, particularly regulatory balance-sheet measures, must retain their own valuation rules. Expected profit is considered separately in the economic decision. Deducting expected losses twice would understate capital or overstate profit.
Stand-alone capital K(Y), a finite increment, and a contribution to an existing portfolio answer different questions. Stand-alone capital quantifies a hypothetical isolated exposure. The finite increment measures the effect of committing to a particular contract. An allocation assigns a share of total risk capital to assets, policies or business units. The incremental comparison does not require every policy's assigned contribution to sum to portfolio capital. Full allocation, in contrast, is desirable for reconciled performance reporting.
For fractional line size t, define f(t) = K(X+tY) with t from zero to one. A local derivative f'(0) estimates the initial change per proposed unit of exposure; f'(1) describes the local change at the fully accepted position. When f is continuously differentiable,
This identity clarifies the approximation error when a derivative substitutes for a finite acceptance. When K is convex and differentiable along the line, the initial slope is no greater than the average slope, and the final slope is no smaller. This conclusion follows directly from convexity. Discontinuous contractual thresholds and nonconvex capital rules require direct recalculation rather than this inequality.
Euler allocation and attribution
For a differentiable degree-one homogeneous functional G(w)=K(sum of w_i X_i), Euler's theorem gives G(w)=sum of w_i times its partial derivative with respect to w_i. The latter product is the Euler contribution. The theoretical treatment by Tasche (2007) connects this decomposition to local improvement in return on risk-adjusted capital, as well as practical estimation for different risk measures. Denault (2001) approaches the division of diversification benefits through axiomatic allocation, including gradient-based attribution. The publisher abstract establishes that general connection; its full proofs are outside the evidence examined here.
The underwriter's transaction is a discrete commitment, often with minimum premium, nonseparable terms and a fixed expense. Euler attribution is useful for a small scalable line and for subsequent management reporting; it cannot automatically price a large irreversible commitment at the starting derivative. The full change is particularly important when an acceptance moves exposure into a different underwriting authority tier or exhausts available treaty recovery.
For a smooth continuous portfolio distribution, Tasche (2001) relates a quantile derivative under suitable regularity conditions to conditional expectations at the aggregate quantile. In loss convention, the local VaR derivative in direction Y takes the form E[Y given X=q], when the conditions hold. For centered capital subtract E[Y]. This is a conditional quantity on a boundary, usually estimated by smoothing or a suitable analytical density rather than an ordinary count of exact quantile matches. It is poorly determined in sparse tails and need not exist at a distributional atom.
An ES derivative instead uses aggregate tail membership in the continuous differentiable case, E[Y given X exceeds q], with the mean subtracted for centered capital. This provides a different economic signal because exposures that matter beyond the quantile receive weight. At atoms or ties, a properly weighted tail and possible directional derivatives must replace a naive strict-exceedance average. Recompute the complete finite difference where local attribution is ambiguous.
Balance-sheet representation
An underwriting decision moves assets and liabilities as well as loss exposure. For a terminal balance-sheet loss B measured in a consistent accounting or economic valuation system, the general comparison is K(B_with)-K(B_without). This remains valid even if B_with is not B_without plus an independent policy claim variable. New acquisition expenses may immediately reduce available capital. Premium receivables may carry credit risk. Currency of expected recoveries may change hedging requirements. Long-duration insurance products may interact with market risk, lapse behavior and options embedded in liabilities.
A useful decision report includes both required capital change and available capital change. Write A0 and A1 for eligible resources before and after the transaction, and C0 and C1 for the corresponding requirement. Absolute headroom moves by (A1-A0)-(C1-C0). If a management target requires A at least m times C, the buffer moves by (A1-A0)-m(C1-C0). The ratio A/C is another view, but its numerator and denominator cannot safely be interpreted in isolation. This accounting of headroom turns capital consumption into a test of the actual constraint rather than a comparison of incompatible measures.
Dependence and tail contribution
The mechanics of dependence
Let X and Y be integrable losses. Even when their marginal distributions are known, the joint distribution determines aggregate risk. In the special case of finite second moments, variance of X+Y equals Var(X)+Var(Y)+2Cov(X,Y). If centered losses are jointly normal with standard deviations s_X and s_Y and correlation r, any fixed-confidence normal VaR or ES of their centered sum is proportional to sqrt(s_X squared+s_Y squared+2r s_X s_Y). For a positive proportionality constant, the marginal change is increasing in r. This is a property of the stated joint normal family, not a universal ordering under arbitrary dependence.
Insurance loss distributions can have common catastrophe, legal, inflationary and economic drivers. The conditional damage distribution during a severe event may bear little relation to ordinary-period claims correlation. Tail dependence is therefore a substantive model assumption. A spatially remote exposure can still share business interruption suppliers, cloud service providers or a common reinsurer. A nominally different product can concentrate legal exposure to the same defendant or policy wording. Dependency mapping must distinguish peril, insured group, geographic footprint, litigation trigger and recovery counterparty.
Value at risk at confidence alpha is the lower quantile q_alpha(Z)=inf{z:F_Z(z) >= alpha}. Expected shortfall is the integrated upper quantile:
For integrable losses it admits the representation q + E[(Z-q)_+]/(1-alpha) with q equal to a quantile at alpha. A direct verification splits the integral into the constant q over the relevant probability interval and the excess above q. This formulation allocates the correct fractional probability at a tied quantile. Acerbi and Tasche (2002) emphasize why handling discontinuities correctly matters for coherence. The journal publisher record was inaccessible in this session; the open manuscript and independent journal metadata were used.
Subadditivity of ES implies K(X+Y)-K(X) <= K(Y) for centered ES, since expectations add. It does not imply that the increment must be positive. Negative marginal centered capital can arise from diversification or hedging of the existing stochastic exposure. Its interpretation requires a consistent balance-sheet loss definition and a test of robustness. A small or negative increment inferred from one model is not a permanent supply of financing.
Explicit scenario experiment
Consider four mutually exclusive states with probabilities 0.94, 0.04, 0.015 and 0.005. The units are synthetic monetary units; the candidates A and B are alternative decisions, not simultaneous policies. Each has expected claims of 0.95. The first is exposed principally in the severe baseline states, while the second pays in the moderate state. Cash flows other than claims are held fixed. These numbers are constructed for analytical discrimination and represent no actual insurer.
| State | Probability | Existing X | Candidate A | Candidate B |
|---|---|---|---|---|
| Ordinary | 0.940 | 0 | 0 | 0 |
| Moderate | 0.040 | 20 | 0 | 23.75 |
| Severe | 0.015 | 80 | 30 | 0 |
| Extreme | 0.005 | 200 | 100 | 0 |
The baseline mean is 3.00. Each with-policy mean is 3.95. At alpha=0.975, the upper 2.5% contains the extreme state, the severe state and 0.5 percentage points of the moderate state. The lower quantile is 20 for X and X+A, but 43.75 for X+B. The weighted baseline tail mean is (0.005 times 200 + 0.015 times 80 + 0.005 times 20)/0.025 = 92.00. With A it is 130.00; with B it is 96.75.
| Measure | Existing | With A | With B |
|---|---|---|---|
| Expected loss | 3.00 | 3.95 | 3.95 |
| VaR 97.5% | 20.00 | 20.00 | 43.75 |
| ES 97.5% | 92.00 | 130.00 | 96.75 |
| Centered ES capital | 89.00 | 126.05 | 92.80 |
| Incremental centered ES capital | 0.00 | 37.05 | 3.80 |
This experiment answers the second research question directly. Identical expected claims coexist with an almost tenfold difference in incremental centered ES capital. The centered VaR changes are -0.95 and 22.80 respectively, so the quantile-based ranking favors A even though A has a much greater adverse-tail contribution at the same confidence level. The discrepancy is a feature of the chosen loss distribution and measurement rule. It is neither a regulatory calculation nor a general ranking of contracts. At a different confidence level, the ranking could change.
The economic interpretation is that acceptance changes the portfolio's exposure to expensive outcomes, not merely the mean or the frequency of ordinary claims. Underwriting that uses only expected loss or a quantile approximation can ignore the magnitude of already rare events. The selected risk measure should be documented together with supplementary event limits. A mathematical example is sufficient to refute a universal claim that equal expected losses imply equal capital consumption, but it is insufficient to estimate how frequently such reversals occur in practice.
Reinsurance and effective marginal retention
A ceded contract changes the mapping from gross loss to retained loss. Let R(s) be the contractual recovery in state s, adjusted as appropriate for covered losses, limit, attachment, exclusions, aggregate exhaustion and recoverability. With a portfolio-wide treaty, the incremental recovery is R(X+Y)-R(X); it is generally not R(Y). Allocating a fixed ceded percentage to each contract can therefore misstate the capacity actually consumed.
Consider a synthetic single-period aggregate excess layer attaching at 100 with a limit of 100. For gross loss z, recovery equals min(max(z-100,0),100). Assume no premium for the purpose of the loss-only example, full and immediate recoverability, and a treaty that covers all modeled claims. X produces retained states 0, 20, 80 and 100. X+A produces retained states 0, 20, 100 and 200. The extreme gross state rises from 200 to 300, but recovery remains 100 because the limit is exhausted.
At alpha=0.975 the baseline retained ES is 72.00, mean 2.50 and centered capital 69.50. After acceptance the corresponding figures are 104.00, 3.30 and 100.70. Incremental net centered ES capital is 31.20 rather than 37.05 gross. This comparison depends on recomputing the retained baseline and candidate distribution under the same treaty. The difference reflects modeled recovery in the severe state; the most extreme additional loss is fully retained. One cannot infer the price of protection from the reduction in capital because ceded premiums, friction, default and recovery timing were deliberately omitted.
Retentions and limits introduce kinks in scenario cash flows; terms involving reinstatement, aggregate annual deductibles and occurrence aggregation can introduce additional discontinuities or state dependence. The actuarial model must represent wording faithfully. In particular, two contracts can both be technically covered but compete for the same aggregate protection. A contract consuming a treaty's last available capacity can raise the retained risk of other business. A renewal that changes attachment or counterparty security requires a fresh portfolio comparison.
The institutional relevance is clear in ICP 16.7 (IAIS, 2024), which links underwriting policy with capital, risk appetite, pricing and reinsurance, including transfer effectiveness in adverse circumstances. For the EU calculation, Solvency II Article 101 recognizes risk mitigation subject to proper inclusion of accompanying risks. Economic modeling of a counterparty failure under a catastrophe may therefore matter even where the treaty nominally transfers exposure. The statutory eligibility of a specific arrangement requires jurisdiction-specific investigation, not an inference from this general principle.
Capital pricing
From expected margin to economic contribution
Let m(Y) be the incremental expected profit after claims, acquisition, administration, relevant reinsurance costs and other modeled cash flows on a consistent horizon, before the designated capital charge. In a one-period simplified setting with a positive incremental required capital and charge rate h, the capital-adjusted contribution is m(Y)-h times Delta K. A binding legal eligibility rule remains separately operative. For a positive Delta K, dividing expected profit by incremental capital gives a marginal return ratio; the numerator and denominator must use compatible time, tax and valuation bases.
Suppose candidates A and B each generate one monetary unit of positive expected profit after identical expense and premium assumptions adjusted for their equal expected claims. At an illustrative one-period capital charge of 0.10 per unit, A's economic contribution is 1-3.705=-2.705, whereas B's is 1-0.380=0.620. The apparent profitability of A does not survive the cost of consuming adverse-tail capacity. If management also imposes a maximum per-acceptance incremental capital of 5, B passes that threshold and A fails, subject to all other constraints. Neither the hurdle nor the threshold is an estimate of industry practice.
The economic charge depends on what capital can be released, its funding cost and the binding constraint. If both candidates are already included in planned growth, the appropriate counterfactual may be the substitution of one for another. Where available capital is temporarily fixed, the relevant opportunity cost can be the foregone value of another feasible contract. If capital can be raised, the financing cost and adjustment delay matter. A scarce catastrophe limit creates opportunity cost even when regulatory capital has ample headroom.
The smooth Euler-based performance relationship in Tasche (2007) is local to a specified homogeneous risk measurement problem. The present comparison uses a full finite change to assess a transaction. Myers and Read (2001) provide a distinct insurance-finance perspective by analyzing surplus allocation through option valuation. The accessible author-affiliated summary supports this methodological contrast, not a reconstruction of their option-pricing formulas. Different allocation criteria can reflect different economic purposes: performance measurement, policyholder security, financing or pricing. A common reporting label does not make them interchangeable.
Multiple horizons and negative consumption
Long-duration contracts may require capital for many future periods. In a consistent projection, the present value of a stream of incremental capital charges is the sum of each period's projected incremental capital times the relevant charge rate and discount factor. A terminal provision or run-off assumption must address capital continuing after the explicit horizon. The calculation should define when capital is held, released and remunerated, and avoid counting an identical capital charge both in product profit and in the decision hurdle.
A negative Delta K poses a distinct problem. Dividing profit by a negative amount reverses the intuitive interpretation of a return ratio; division by zero is undefined. Direct incremental contribution and post-acceptance constraints are preferable in these cases. A documented diversification credit may legitimately influence a quote, but the credit should survive adverse dependency assumptions and remain within the insurer's authority and liquidity constraints. Where a hedge or offset requires another party's continuing performance, the underwriter should test failure of that offset.
Capital charges are decision inputs, not legal or accounting declarations. Internal capital pricing must be reconciled with the selected valuation of liabilities and with actual cost measures. A regulatory risk margin or contractual service measure cannot be substituted mechanically for the internal marginal charge. The objective is a coherent comparative decision under documented assumptions rather than a universal premium formula.
Risk appetite thresholds
A policy-level evaluation must specify the constraint set before ranking risks. At minimum the institution may need to consider eligible capital headroom, economic capital budget, liquidity under stress, gross and net catastrophe accumulations, counterparty concentrations, limits on specific underwriting classes and delegated authority. Each limit has its own valuation and horizon. A contract with positive capital-adjusted contribution remains infeasible if a binding constraint would be violated.
Formally, let a denote a feasible decision such as declining, accepting a smaller line, changing lawful terms, obtaining facultative reinsurance, or accepting the full proposal. Let g_j(a) be a measured post-decision quantity and b_j its corresponding limit. Choose among options satisfying g_j(a)<=b_j and other legal and contractual prerequisites; compare expected economic contribution on the same assumptions. A scalar charge summarizes some opportunity costs but does not replace this feasibility check. When constraints multiply, an optimization of the entire portfolio can allocate capacity more efficiently than sequentially ranking uncoordinated quotes.
The issue is particularly pronounced in decentralized underwriting. Several proposals can pass independently against the same snapshot yet exceed aggregate capacity when accepted together. A defensible process timestamps the portfolio baseline, reserves capacity during quote validity where appropriate, recalculates upon binding, and releases reservations on expiry. The allocation identifier must refer to a unique quote version so that endorsements and duplicate submissions do not double-consume the same capacity. Since risk evolves, these controls also apply between approval and inception.
ICP 16.3 and 16.7 (IAIS, 2024) support the institutional relationship among risk appetite, risk limits, capital and underwriting policy. The specific workflow here is an analytical application to the decision mechanism rather than a claimed supervisory mandate. A high materiality referral, for example, can require independent validation of assumptions before the actual binding decision. Recording an exception should identify the applicable limit, quantitative impact, legitimate authority, expiry and compensating controls.
An institutional capital budget must also respect applicable insurance eligibility, rating, nondiscrimination and market-conduct rules. Portfolio scarcity does not create permission to use legally prohibited classifications or to alter binding coverage unilaterally. Where an exposure is constrained, the appropriate action may be an authorized referral, lawful product alternative or a documented decline according to applicable rules. This study does not establish product-specific legal permission in any jurisdiction.
Model risk and regulatory interpretation
Model validation and uncertainty
Marginal capital is a difference between two risk functionals estimated from related distributions. Simulation error can be substantial relative to the difference, especially at high confidence levels. A paired simulation design uses common random numbers for both portfolios: the same catastrophe event, claim inflation factor, mortality shock or economic state enters both evaluations. Both scenarios are valued separately, preserving changes due to the candidate. Common random numbers often reduce variance of a difference when the outputs are positively related; the reduction is not automatic. For tail estimators an uncertainty assessment should re-estimate both portfolio risk measures on each resample.
Distinguish process variability within a selected model, parameter uncertainty, structural uncertainty and numerical error. Parameter changes can alter claim frequency, severity or dependence; structural alternatives may include different event catalogs, aggregate policy wording interpretations, or common-shock specifications. Numerical error arises from scenario count, quantile discretization and valuation approximation. A favorable estimate near a constraint is weak evidence when plausible model alternatives produce infeasible results.
For an independent equal-weight scenario sample of size N, an upper tail of probability 1-alpha corresponds to N(1-alpha) observations before treatment of ties. For example, at 99.5% confidence a sample of 20,000 scenarios has approximately 100 tail observations. This calculation describes nominal tail information, not a guarantee of statistical precision or independence. Weighted importance samples require attention to effective weight concentration; resampling dependent event years as independent policy records would understate uncertainty.
The professional guidance in ASOP No. 56 (Actuarial Standards Board, 2019) addresses model checking, sensitivity, reconciliation and validation for covered actuarial services. A decision-oriented validation program should check gross-to-net cash-flow reconciliation, exact policy limits, unchanged baseline scenarios, mathematical limiting cases, and stability under reasonable alternatives. Model documentation should identify the intended purpose, data dates, estimation procedure and interpretation of output. Professional requirements attach according to their scope; ASOP 56 is not an international insurance statute.
Comparative prudential settings
The following comparison identifies distinct capital objects and the implications for marginal calculation. It is bounded to the stated institutional texts examined in November 2026 and does not purport to codify all national implementation or insurer-specific permissions.
In the European Union, Solvency II Article 100 requires eligible own funds covering SCR and permits standard-formula or internal-model calculation within the framework. Article 101 sets a one-year 99.5% VaR calibration of basic own funds and includes underwriting, market, credit and operational risks. Regulatory marginal SCR consequently requires recomputing the prescribed or authorized total measure after acceptance. The economic centered-ES experiment cannot be presented as this statutory calculation. EIOPA's regulatory overview identifies January 30, 2027 as the effective date of the revised framework under Directive (EU) 2025/2; the discussion uses the earlier operative framework, not that future revision.
In the United States, NAIC's RBC overview explains that statutory risk-based capital is a minimum and intervention mechanism differentiated by major insurance industry segment. An incremental RBC calculation should be performed using the applicable formula, filing basis and jurisdiction; this study does not assign a uniform confidence level to RBC. The NAIC ORSA framework gives covered insurers and groups a forward-looking assessment of risks and capital adequacy. It permits a meaningful conceptual distinction between statutory minimum and the institution's own assessment without equating the two.
In Canada, the 2026 Minimum Capital Test describes a P&C capital framework calibrated at a one-year CTE 99% target including a terminal provision, with target requirements net of diversification divided by 1.5 for the minimum. OSFI also identifies implicit within-class diversification in insurance risk factors. Therefore a formula-derived change may embody prescribed diversification assumptions even when the actual concentration differs. An economic assessment of concentrations supplements the statutory measurement rather than replacing it.
For Switzerland, FINMA's Swiss Solvency Test describes a market-consistent total balance-sheet assessment modeling one-year changes and comparing risk-bearing with target capital. Its institutional concept emphasizes the joint movement of liabilities and assets. The overview does not establish the complete formula or current detailed calibration; no exact Swiss marginal capital computation is claimed here.
At the international level, IAIS ICP 16 brings underwriting, transfer and capital within a coordinated solvency-management perspective. ICPs guide supervisory systems; they do not directly impose an identical national formula on all insurers. Comparison reveals a shared need to analyze portfolio-level solvency, but different legal objects, valuation bases, calibration and formula implementation.
Comparative synthesis, limitations and reproducibility
Scientific synthesis
The research establishes that the relevant marginal capital quantity is conditional on both the portfolio and the institutional purpose of measurement. An actuarial distribution can isolate the adverse-loss effects of a candidate; a statutory calculation evaluates a separate legally specified requirement. The capital-pricing layer interprets scarcity and financing consequences. The governance layer determines whether the decision is feasible. Each layer needs its own inputs and a reconciliation to the others.
The first research question is answered by the paired finite-difference comparison, supported by differentiable attribution where its assumptions hold. The scenario experiment answers the second question by providing a concrete loss distribution in which contracts with equal means show different tail contributions and different rankings under quantile and integrated-tail measures. The treaty experiment answers the third by exposing the dependence of net capital on aggregate protection and its exhaustion. Model uncertainty qualifies each conclusion because a different distribution or wording mapping changes the underlying experiment.
Several competing explanations deserve attention in empirical application. A deterioration in capital-adjusted returns after accepting a risk may result from overpriced reinsurance, a change in fixed expenses, worse claims selection, true dependency concentration, or changed model assumptions. A retrospective association between acceptance and loss outcomes does not isolate capital consumption. The comparison should hold the model version fixed for attribution, distinguish resulting realized experience from an updated forecast, and explicitly record the reason for changing any other input.
Selection effects also complicate retrospective evidence. Declined proposals generally lack observed insured loss experience; accepted business reflects prior underwriting thresholds. An analysis restricted to accepted contracts may underestimate the variability of rejected proposals. The causal interpretation of real-world data therefore requires a stated selection mechanism, independent exposure evidence where lawful and available, and a validation strategy for counterfactual risk. This manuscript does not estimate an acceptance treatment effect.
A portfolio decision may depend on the order of contracts. In general, the sum of sequential finite changes telescopes to the total change, but different ordering reallocates that total among contracts. An insurer reporting historical consumption should preserve its sequence and snapshot dates. Allocations for comparable business-unit performance can use a reconciled allocation convention independently of order-dependent transaction accounting. This resolves an operational tension without elevating either technique into a universal allocation doctrine.
Reproduction and evaluation
The four-state experiment is exactly computable by enumeration; no random sample, proprietary dataset or fitted parameters are used. The following pseudocode describes a weighted empirical calculation. Probabilities are checked to sum to one; states are sorted by loss in descending order and the upper-tail probability is filled fractionally. The same scenario states are reused for every counterfactual.
def capital(losses, probabilities, alpha):
assert len(losses) == len(probabilities)
assert abs(sum(probabilities) - 1.0) < 1e-10
assert 0 < alpha < 1
remaining = 1 - alpha
tail_sum = 0.0
for loss, prob in sorted(zip(losses, probabilities), reverse=True):
take = min(prob, remaining)
tail_sum += loss * take
remaining -= take
if remaining < 1e-12:
break
mean = sum(z * p for z, p in zip(losses, probabilities))
return tail_sum / (1 - alpha) - mean
p = [.94, .04, .015, .005]
x = [0, 20, 80, 200]
a = [0, 0, 30, 100]
b = [0, 23.75, 0, 0]
xa = [u + v for u, v in zip(x, a)]
xb = [u + v for u, v in zip(x, b)]
retained = lambda z: z - min(max(z - 100, 0), 100)
assert abs(capital(x, p, .975) - 89) < 1e-8
assert abs(capital(xa, p, .975) - capital(x, p, .975) - 37.05) < 1e-8
assert abs(capital(xb, p, .975) - capital(x, p, .975) - 3.8) < 1e-8
assert abs(capital(list(map(retained, xa)), p, .975)
- capital(list(map(retained, x)), p, .975) - 31.2) < 1e-8
A realistic replication would require contract effective dates and terms, premium and expense valuation, geocoded or otherwise mapped exposures, paid and incurred loss development, event and systemic factor identifiers, current ceded program wording and recoverable assumptions, eligible capital and required capital definitions, and a reproducible scenario generator. A data dictionary should specify units and temporal relationships, along with data lineage and material exclusions. An external evaluation should define sampling periods and out-of-time tests before reviewing results.
A decision validation should compare model predictions with subsequently observed experience over a suitable horizon, accounting for claim development and exposure changes. Tail events are scarce, so event reconstruction and sensitivity analysis may be necessary in addition to experience tests. One should reproduce the exact prior risk calculation from its historical input snapshot before assessing model drift. This preserves the distinction between an originally reasonable decision and a later adverse outcome.
The evidence design is analytic and comparative, not a systematic review of every publication. Open mathematical manuscripts and official regulatory texts establish the principal claims; abstract-only and institutional summary sources support only the limited statements identified in the evidence ledger. The manuscript does not report company-level estimates, jurisdiction-by-jurisdiction legal compliance findings, or an empirical optimum. Model examples isolate mechanisms, and the regulatory comparison is deliberately restricted to cited provisions and official overviews. These limits preserve the distinction between demonstrable mathematical results and empirical generalization. The authorial contribution lies in the interdisciplinary interpretation, constructed counterexamples, explicit finite-difference comparison and integration of underwriting controls with capital pricing. No new named theory or priority claim is advanced.
Conclusion
Marginal capital consumption is the transaction-specific change in required capital under a defined portfolio, valuation basis, probability model and decision horizon. Its economically useful expression is the difference between a with-contract and a without-contract calculation. A derivative is a local approximation; Euler attribution is a reconciled decomposition under appropriate mathematical conditions. Each serves a distinct analytical purpose.
The aggregate loss tail determines the significance of dependency. The synthetic evidence shows that equal expected claims and equal expected profit can coexist with substantially different requirements for risk-bearing capacity. Quantile-only and integrated-tail criteria can also rank the same candidates differently. Exposure to already severe states must therefore be assessed with a measure that is informative about the decision's adverse-outcome consequences.
Contractual risk transfer should be evaluated within the entire ceded portfolio. Attachment, exhaustion, recoverability and timing govern whether additional gross exposure creates additional net risk. Economic valuation of the capital consumed must include a cost or opportunity-cost interpretation and a direct test of constraints. Positive expected margin alone is insufficient to conclude that an acceptance creates value.
The comparative legal evidence supports a broader distinction among regulatory capital, economic capital and available resources. Jurisdictions use different statutory definitions and supervisory methods. An insurer's internal assessment can supplement statutory measurement, but acceptance must remain feasible on its actual applicable legal and institutional basis. The scientific result is a portfolio-conditional interpretation of underwriting decisions, with transparent limits on inference, reproducible arithmetic and traceable evidence.
References
Tasche, D. (2007). Capital Allocation to Business Units and Sub-Portfolios: The Euler Principle. arXiv:0708.2542 [q-fin.PM]. Source
Acerbi, C., & Tasche, D. (2002). On the Coherence of Expected Shortfall. Journal of Banking & Finance, 26(7), 1487-1503. Open manuscript arXiv:cond-mat/0104295. Source
Tasche, D. (2001). Conditional Expectation as Quantile Derivative. arXiv:math/0104190. Source
Artzner, P., Delbaen, F., Eber, J.-M., & Heath, D. (1999). Coherent Measures of Risk. Mathematical Finance, 9(3), 203-228. DOI: 10.1111/1467-9965.00068. Source
Denault, M. (2001). Coherent Allocation of Risk Capital. Journal of Risk, 4(1), 1-34. Source
Myers, S. C., & Read, J. A., Jr. (2001). Capital Allocation for Insurance Companies. Journal of Risk and Insurance, 68(4), 545-580. Source
International Association of Insurance Supervisors. (2024). Insurance Core Principles and ComFrame. Online tool, ICP 16, especially 16.3 and 16.7 (December 2024 framework). Source
European Parliament & Council of the European Union. (2009). Directive 2009/138/EC (Solvency II), Article 100, General Provisions. EIOPA Single Rulebook. Source
European Parliament & Council of the European Union. (2009). Directive 2009/138/EC (Solvency II), Article 101, Calculation of the Solvency Capital Requirement. EIOPA Single Rulebook. Source
European Insurance and Occupational Pensions Authority. (n.d.). Solvency II. Regulatory overview. Source
National Association of Insurance Commissioners. (n.d.). Risk-Based Capital. Insurance Topics. Source
National Association of Insurance Commissioners. (2026, July 28 update). Own Risk and Solvency Assessment. Insurance Topics. Source
Office of the Superintendent of Financial Institutions. (2026). Minimum Capital Test (MCT) Guideline (2026). Chapter 1 and section 4.1, Canada. Source
Swiss Financial Market Supervisory Authority. (n.d.). Swiss Solvency Test (SST). Source
Actuarial Standards Board. (2019). Actuarial Standard of Practice No. 56: Modeling. Effective October 1, 2020. Source