Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Solvency Capital Add-ons and Supervisory Model Challenges

Abstract. Supervisory capital add-ons address cases in which a risk-sensitive solvency framework, an approved internal model, a system of governance, or specified long-term valuation adjustments fail to reflect an insurer's actual risk profile adequately. The scientific problem is simultaneously legal, actuarial, institutional, and informational: the supervisor must identify a material deviation, establish the authority to intervene, quantify a defensible correction, ensure that the requirement preserves policyholder interests, and determine when remediation warrants withdrawal. This monograph examines those decisions through a close reading of European Solvency II and its Delegated Regulation, model validation standards, prudential policies in the United Kingdom, insurance supervisory principles issued by the International Association of Insurance Supervisors, and selected Canadian and United States comparators. It derives the economic consequences of a supplementary SCR through balance-sheet identities and illustrative calculations, distinguishes deficiencies in risk measurement from deficiencies in risk governance, and analyzes uncertainty in loss distributions, dependence, reinsurance, management actions, long-term guarantees, and group capital. Original synthesis identifies the critical relationship between evidence, corrective measurement, ongoing verification, and legal accountability. The analysis is documentary and comparative; the examples are constructed rather than estimates of any insurer's capital position. EU amendments scheduled to apply January 30, 2027 are analyzed prospectively. The work establishes no new named theory or regulatory doctrine.

Keywords: Solvency capital add-on; supervisory model challenge; Solvency II; solvency capital requirement; internal model; capital calibration; model risk; risk profile deviation; supervisory review; actuarial validation; standard formula; governance risk; insurance group; capital adequacy; policyholder protection; regulatory discretion; prudential remediation; risk margin; Solvency UK; own risk and solvency assessment; tail dependence.

Introduction

The use of a regulatory capital formula or an approved internal model presupposes a continuing correspondence between the distribution of an insurer's economic risks and the assumptions used to convert those risks into a capital requirement. That correspondence may weaken as underwriting portfolios change, risk dependencies intensify, assets and liabilities become more sensitive to market movements, or model governance deteriorates. A solvency number generated by an established procedure remains a calculation; its supervisory authority derives from whether the procedure remains adequate for the actual insurer.

In the European Union, Article 36 supervisory review supplies the institutional setting and Article 37 capital add-ons provides an exceptional legal correction to an inadequate Solvency Capital Requirement (SCR) (European Parliament and Council, 2009, as amended). The legal instrument distinguishes discrepancies in standard-formula assumptions, undermeasurement by approved models, governance failures and specified adjustments or transitional measures. Its operation is a supervisory decision requiring reasons and recurrent review. An add-on changes the capital demand faced by the insurer; it does not itself repair data, change the insurer's contractual promises, create liquid assets, or assure that supplementary own funds will be raised.

The scientific questions are threefold. Under what evidence and legal conditions is an additional requirement justified? How can a correction be calculated when the model used to identify the deviation may itself be incomplete? What makes the correction accountable, temporary where appropriate, and comparable across different insurance and legal systems? Answering requires more than reproducing a legal list. The investigation follows the causal chain from risk exposures and modeling assumptions to supervisory diagnosis, numerical intervention, remediation, and policyholder consequences.

Scope and method

The primary jurisdictional analysis addresses EU Solvency II as applicable on November 14, 2026. Subsequent reform under Directive (EU) 2025/2 is discussed as a prospective change scheduled to apply from January 30, 2027 (European Parliament and Council, 2025). The UK is an independently governed comparator under the Prudential Regulation Authority's 2024 capital add-on policy. IAIS standards supply international principles; Canadian and US instruments illustrate related but legally distinct capital supervision. No equivalence of enforcement powers or regulatory ratios is assumed among these regimes.

The research method integrates doctrinal reading of original instruments, comparative institutional analysis, a risk-distribution interpretation of regulatory capital, constructed arithmetic examples, and critical evaluation of published statistical and actuarial research. Each factual regulatory proposition is traced to a directly linked primary source; scholarly claims are tied to their authors and publication years. Original reasoning is identified as analytical synthesis rather than represented as an empirically established general law. The document uses no supervisory confidential data, nonpublic insurer model files, or fitted loss samples.

An important distinction in the literature concerns model outputs and their verification. Gaigall's (2021) model-change test shows that statistical comparisons of outputs can form part of validating changes to an internal risk model; Dacorogna's (2023) validation analysis emphasizes calibration, dependence assumptions, stress tests, and probability integral transforms. Both are useful for understanding how supervisors can challenge a model, while neither converts a test result automatically into a lawful capital add-on. The legal threshold and the scientific diagnostic question remain separately identifiable.

Interpretive boundaries

A regulatory SCR is calculated under a specified regime and reference date. The one-year 99.5% calibration of Solvency II is a formal target for quantifiable risks borne by basic own funds; it is not an observed annual insolvency probability or a guarantee that every extreme scenario has been modeled (European Parliament and Council, 2009, Article 101). In the following chapters, “corrected SCR” means an illustrative estimate of what the SCR would be if specified assumptions adequately represented the actual risk profile. “Add-on” means the supplementary capital requirement imposed by a competent authority under the regime's legal conditions. The legal categories, model weaknesses and managerial remedies remain distinguishable throughout.

Supervisory Model Assessment

Supervisory review as a continuing assessment

Supervisory review begins with the insurer's strategies, processes, reporting and risk governance and extends to technical provisions, investment rules, own-fund quantity and quality, capital requirements and ongoing compliance with internal-model standards. These elements appear together in the Solvency II supervisory review process (European Parliament and Council, 2009, Article 36). The supervisor therefore assesses a measurement system situated within the insurer's management and financial condition. An internal-model capital number cannot be judged independently of the data-generating business and the controls that produce it.

The scope of review includes the ability to withstand possible future events and unfavorable economic developments. The distinction is important for newly concentrated portfolios. A year's current claims experience can remain benign while a business strategy increases correlated catastrophe exposure and narrows reinsurance coverage. A model that faithfully reproduces last year's portfolio may have deteriorated in relevance without any mechanical computational error. It is the evolving fit between model specification and actual financial exposure that demands examination.

The IAIS Insurance Core Principles make enterprise risk management and own risk and solvency assessment central elements of solvency supervision (IAIS, 2024–2026 online text). An insurer's own assessment has a different object from the supervisor's prescription: it evaluates the insurer's forward risk and resources across its strategic horizon, including risks that may be inadequately expressed in a standard regulatory formula. Supervisory challenge gains credibility when internal plans, experience studies, capital management documents and independent model validation can be reconciled into a consistent account of the insurer.

Standard formula and approved model

A standard formula supplies comparability and administrability through prescribed modules, factors, stress scenarios and aggregation assumptions. An approved full or partial internal model promises a better fit for a firm's risk profile, subject to legal standards and ongoing oversight. Under the Solvency II internal-model provisions, the approval and continued use of a model are separate regulatory questions, and Article 119 allows a reasoned supervisory requirement to adopt an internal model when a significant divergence makes the standard formula inappropriate (European Parliament and Council, 2009, consolidated 2025 text).

A standard formula can be wrong for a particular firm even if the formula is well calibrated for the population used to design it. Consider a property insurer whose reinsurance structure contains concentrated credit exposures or an unusual catastrophe aggregate limit. Generic correlations or risk factors may fail to capture the firm's contract-specific losses. Conversely, an internal model may incorporate detailed exposures yet understate dependence or assume future management actions that would be unavailable during simultaneous market and claims shocks. Detail and customization provide potential relevance, while also expanding the number of assumptions requiring verification.

An approved internal model requires a regular cycle of validation, including performance monitoring, specification review, experience comparison, sensitivity, stability and data-quality assessment under Article 124 validation standards (European Parliament and Council, 2009). The Article 241 validation process calls for independence from model development and operation as well as documented validation methods and remediation procedures (European Commission, 2015, as amended). Supervisory challenge should test the model's reliability through the same evidence the undertaking is expected to retain.

Diagnostic objects: data, model, decisions

The object of a model challenge must be named precisely. Data error concerns the accuracy, completeness, timeliness or representativeness of observations. Parameter error concerns fitted values given a specification. Specification error concerns omitted drivers, distributions, nonlinearities, management actions or dependency structures. Implementation error concerns code, interfaces, versioning or reproducibility. Governance error concerns control failures and ineffective model use. These categories are analytical diagnostics in this monograph, and may coexist in a concrete case.

Suppose a claims severity model uses a stable historical inflation parameter while newly written contracts contain significantly higher repair-cost exposure. The observation may indicate input staleness, parameter drift, an omitted cost driver or a portfolio definition problem. The add-on cannot be calculated responsibly from the label “inflation risk” alone. The supervisor must establish the insured population, contract terms, projected ultimate losses, indexation mechanisms, reinsurance recovery behavior and materiality of the deviation.

Research on internal models supports a plurality of checks. Borgonovo, Clemente, and Rabitti (2024) propose complementary sensitivity settings to explain what assumptions drive an insurance model's output. Androschuck and colleagues (2017) show how simulation, dependence structures and proxy models introduce testing and communication challenges. A model can pass a local parameter sensitivity check while still misrepresenting the joint tail. The diagnostic process therefore needs both granular attribution and integrated stress testing.

Governance and the use test

The governance of capital models has an evidentiary function. If senior management decisions, risk appetite statements, reinsurance purchases, underwriting limits and capital planning repeatedly disregard the model's results, the model's stated role may be doubtful. If validation reports identify weaknesses but actions are deferred without accountability, an apparently sophisticated framework may contain material institutional risk. Supervisory evidence should therefore trace whether the model informs decisions and whether identified weaknesses are escalated, budgeted and resolved.

Under Article 37, governance deviations capable of justifying an add-on must meet the particular legal conditions; not every process defect is capitalized. The governance significance assessment considers prudent management, the likely effect on policyholders, proportionality, potential loss and anticipated duration (European Commission, 2015, Article 277). The supervisory explanation should show why other measures are unlikely to cure deficiencies sufficiently in time. An add-on designed to address a failure of risk control consequently rests on a different evidentiary chain from an add-on representing a numerical model underestimation.

Evaluation record and reproducibility

For an auditable supervisory decision, a scientific assessment records the valuation date, data version, portfolio perimeter, model version, validation findings, affected risk drivers, materiality estimates, alternatives examined, supervisory responses and the reasoning behind the final decision. Reproducibility requires that an independent reviewer can trace a numerical change to the assumptions that produced it. There is no requirement here that all confidential information be publicly disclosed; legal confidentiality duties restrict publication of firm-level proprietary inputs. Reproducibility within appropriate supervisory and undertaking controls remains essential.

This leads to the first central finding: the quality of a capital add-on depends on the quality of the diagnosis. A capital amount may look precise while its basis is speculative. A well-documented challenge makes uncertainty explicit, identifies the adverse consequences of the discrepancy and selects a method proportional to available information. It also establishes a measurable path for reversal when the underlying weakness is corrected.

Risk Profile Deviations

The significance threshold and legal classification

Solvency II provides distinct grounds under Article 37(1)(a)–(d). Standard-formula deviation is associated with circumstances where requiring an internal model is inappropriate, ineffective, or the model is being developed. Internal-model deviation concerns quantifiable risks captured insufficiently when the model has not been adapted effectively within an appropriate time. Governance deviation concerns failures sufficiently serious to compromise risk identification, measurement, monitoring, management and reporting, with other remedies unlikely to improve matters in time. The fourth category concerns material deviation from assumptions underlying specified matching, volatility or transitional adjustments.

A deviation becomes supervisory-relevant through evidence of its size, nature, durability, sensitivity and potential adverse consequences. Article 276 significance factors directs attention to the nature, type and size of deviations, likely policyholder impact, sensitivity and the duration and volatility of the problem (European Commission, 2015, as amended). The relevant finding is an inadequacy of the actual model or formula in context, not merely the discovery of a difference between two methods.

No model exactly represents every possible future event. The diagnostic question is therefore the practical importance of the difference under the regulatory risk measure and the insurer's exposures. An estimate that changes a minor submodule but leaves total capital and the confidence assessment substantially unaffected can have a different supervisory significance from an omitted catastrophe aggregation that shifts the enterprise tail. Materiality also depends on the availability and liquidity of own funds, claims-paying horizon and time needed for a durable repair.

Difference between risk and solvency resources

Let eligible own funds be OFOF, the unadjusted SCR be KK, and a supervisory add-on be AA. The illustrative coverage ratio after imposition is

Rafter=OFK+A.R_{after}=\frac{OF}{K+A}.

For a synthetic insurer with OF=180OF=180, K=120K=120 and A=30A=30 monetary units, the initial ratio is 180/120=150%180/120=150\%, while the adjusted ratio is 180/150=120%180/150=120\%. No actual 30 units are withdrawn from the asset side merely by making this calculation; the denominator changes because the supervisory capital requirement changes. Compliance may nevertheless require acquiring additional eligible capital, reducing exposures or another legally effective improvement.

The numerator and denominator respond differently to events. Suppose an unrelated adverse event reduces eligible own funds to 165 while the add-on remains 30 and underlying modeled SCR rises to 135. Then R=165/(135+30)=100%R=165/(135+30)=100\%. A supervisor must distinguish the new financial loss from the original risk-model deficiency and check whether the deficiency itself intensified. Automatically attributing every deterioration to the add-on would double count economic effects at the level of explanation, even if regulatory calculations are technically additive under the governing rules.

Dependence and tail underestimation

The risks of an insurer often become more dependent in stress periods. Catastrophe losses can coincide with inflation, asset-price reductions, reinsurance counterparty strain, liquidity demands and operational overload. Models that combine individually plausible risk marginal distributions using weak dependence assumptions can materially understate joint tail losses. Simulation-based aggregation research discusses copulas, tail dependence and proxy-model testing (Androschuck et al., 2017); McNeil, Frey, and Embrechts (2015) provide the statistical foundations of dependence and extreme-risk measurement.

For two illustrative normally modeled independent risk components with standard deviations σ1=60\sigma_1=60 and σ2=80\sigma_2=80, the standard deviation of their sum is 602+802=100\sqrt{60^2+80^2}=100. If the actual correlation is 0.5, the corresponding variance calculation is 602+802+2(0.5)(60)(80)=14,80060^2+80^2+2(0.5)(60)(80)=14{,}800, yielding a standard deviation of approximately 121.66. This is a 21.66% increase in the standard deviation of the sum. The example is explicitly a covariance calculation, not a Solvency II SCR calculation or a general estimate of a 99.5% capital add-on: nonnormal tails, valuation and regulatory aggregation can change the result.

A model challenge in this case would ask whether the dependence assumption is supported by experience, economic mechanisms, exposure geography, common suppliers, reinsurance contracts, catastrophe clustering and reverse stress tests. The same model can appear accurate at an ordinary percentile yet fail at the regulatory tail. Dependence validation must assess plausible joint scenarios as well as pairwise historical correlations.

Structural change and nonstationarity

Historical insurance data may fail to represent current exposures when building codes, cyber infrastructure, climate-related hazards, monetary conditions, claims litigation or distribution channels change. Extrapolation from a past regime assumes sufficient stability in the data-generating mechanism. When that assumption becomes untenable, the question is whether the institution has identified the new process, quantified a credible range of outcomes and reflected the risk in its capital and risk appetite.

A serious capital challenge should separate estimation uncertainty from changes in the insured population. The former may narrow with additional relevant data; the latter may require revised model structure and new external scientific information. Delegated Regulation Article 242 requires at-least-annual comparisons of modeled results against experience and includes output stability analysis and reverse stress testing (European Commission, 2015). Such tests provide avenues for discovering nonstationarity without assuming that any one short observed sample can reveal a 200-year event probability.

Contractual risk mitigation and basis risk

Reinsurance and hedging may reduce the retained loss distribution but can introduce counterparty risk, coverage mismatch, basis risk, reinstatement costs and settlement timing problems. If a model uses gross historical loss data but assumes favorable net retention after treaties, the supervisory review must inspect the legal contract language, limits, attachment points, exclusions, collateral, cross-border enforceability and recoverability. A model that correctly computes the ceded amount under one scenario may still overestimate collection under widespread stress.

A capital add-on can be appropriate where the relevant legal conditions are met and the discrepancy cannot be remedied in time, while improved contract data, exposure limits, collateral, treaty redesign or internal-model changes may be more enduring responses. The effect of a remedy on own funds, recoverables, liquidity and future risk must be assessed separately. The insurer's ability to buy a new reinsurance program should never be treated as equivalent to immediate removal of already incurred residual exposure.

Valuation adjustments as a separate causal route

Solvency II's fourth category addresses deviations related to the matching adjustment, volatility adjustment and specified transitional measures. These measures can affect the valuation of long-duration obligations, the measured spread sensitivity of liabilities and the reported level of eligible own funds. Their risk-profile assumptions require separate examination from the ordinary distribution of underwriting losses. The appropriate analysis considers asset and liability cash flows together and tests whether the measure produces a valuation effect that corresponds to the undertaking's actual exposures.

This separation has direct quantitative consequences: Article 37(1)(d) uses a proportionality requirement, and Article 284's adjustment calculation incorporates both own-fund and SCR effects (European Commission, 2015). Treating every capital add-on as simply the difference between two SCR numbers would misstate this special category. The fourth category becomes particularly important during regulatory transitions, when a permitted valuation adjustment may interact with stress behavior, eligibility conditions and changing liability assumptions.

Capital Add-on Decision Conditions

Sequential legal and evidentiary questions

A defensible decision begins with jurisdiction and applicability: which undertaking, legal entity, group, supervisory authority and reporting reference date are involved? It then asks which specific statutory ground is met, what facts establish the deviation and why the discrepancy is significant. The next questions concern available corrective measures, the timeliness and effectiveness of remediation, the method for estimating capital consequences, the statement of reasons and the rules for reconsideration. Under the EU statutory decision rule, the measure is exceptional and must be reasoned; this feature constrains both arbitrary intervention and under-response to documented material risk.

For an inadequate standard formula, Article 280's internal-model test provides examples of when an internal-model requirement may be inappropriate because development resources would be disproportionate to the deviation, or ineffective because an adequate model has not been developed (European Commission, 2015). This matters for insurers whose specific exposures materially escape a standard module but who cannot reasonably build a qualifying model immediately. The add-on can bridge an observed inadequacy without pretending that every firm can or should undertake the same model development program.

For an approved internal model, the sequence differs. The authority must identify insufficiently captured quantifiable risk and the failure to adapt the model within an appropriate timeframe. Model-change governance, remediation milestones and validation results therefore form part of the statutory basis. A supervisor should examine whether an existing approval still accurately describes the model's scope and whether a proposed correction is a major model change requiring separate permission. The PRA internal-model monitoring policy illustrates how a current UK regime connects permissions, major changes, residual limitations and continuing model review (PRA, 2024).

Proportionality in governance cases

Governance deficiencies can create financial risk even when a specific statistical loss distribution is difficult to fit. Examples include ineffective reserving controls, poor independent challenge of external catastrophe models, lack of reliable exposure inventory, unresolved model implementation defects, or weak oversight of delegated underwriting. Still, the legal threshold for a governance capital add-on is stronger than a general conclusion that management could be improved.

The governance-deviation criteria ask the supervisor to examine probable loss, duration and the expected effects on policyholders and beneficiaries; Article 286's calculation rule also permits comparison with sufficiently similar situations subject to documented reasons and confidentiality (European Commission, 2015). The rationale should explain why the specified failure generates material risks, why timely noncapital measures are unlikely to be sufficient, and why the amount is proportionate. Disciplinary sanction, board replacement, underwriting restrictions or improved audit processes may address related problems, but the legitimacy of each power depends on its own law.

The economics of governance-based add-ons differ from the economics of measured omitted risks. A weak model-control system can be a latent source of several kinds of loss, so a supervisor should guard against placing a precise-looking charge on overlapping risks without explaining the overlap. Qualitative evidence can establish an urgent need for action; subsequent data collection and modeling should progressively clarify the capital consequences.

Supervisory discretion and reason-giving

An add-on decision must articulate the factual finding, relevant legal provision, modeling assumptions, selected method, key sensitivities, duration or review expectations, and conditions for potential removal. Decision quality improves when the undertaking can test the numerical case and explain alternative evidence. Written reasons enable internal challenge and any remedies provided by applicable administrative law. The supervisor's legal authority does not remove the need for mathematical traceability.

A useful conceptual distinction is between a measure's immediate effect and its counterfactual justification. The immediate effect is the increase in required capital. Its counterfactual justification asks what the requirement would have been if the actual risk profile had been reflected adequately. For numerical risk-profile deviations, EU Article 282 expresses this logic formally; for governance and valuation-adjustment cases, the legal rules demand additional tests of proportionality or a different calculation. This distinction disciplines comparisons among remedies.

Exceptional circumstances and supervisory practice

The word “exceptional” must be interpreted within the legal text and the facts of a particular undertaking. A low recorded frequency of add-ons does not by itself prove that regulation is too permissive, and a high frequency would not automatically prove more effective policyholder protection. Market composition, supervisory culture, internal-model prevalence, use of noncapital corrective tools and reporting practices affect observed frequency. The EIOPA capital add-on statistics publish EEA aggregate information based on reported Solvency II quantitative templates; the currently described series has reference years 2017–2024 (EIOPA, 2026 access).

Reported capital add-ons are an outcome of both risk profiles and supervisor decisions, making a cross-country causal comparison challenging. A jurisdiction with extensive prior remedial activity may show fewer final add-ons than one imposing a charge sooner. Data on imposed charges must therefore be interpreted alongside supervisory review intensity, reasons, changes in risk profile, time to remediation, insurance market mix and disclosure practices. Without those controls, the percentage of insurers with add-ons is a descriptive statistic rather than an estimate of supervisory effectiveness.

Evidence sufficient for decision, and evidence beyond decision

Supervisors may need to act before a complete long-run data set can be obtained. The uncertainty of future tail outcomes is structural, particularly for rare catastrophes and correlated financial shocks. Dacorogna's (2023) discussion explains why reliable direct backtesting of a one-year extreme quantile is limited by sparse observations. The practical burden of proof is accordingly a reasoned triangulation of legal facts, available internal experience, cross-sectional evidence, stress tests, expert judgment and sensitivity analysis. Each component has a specific evidential limit.

The authorial conclusion is that sufficient evidence for action and sufficient evidence for a permanent actuarial model are distinct thresholds. An authority may justifiably require additional capital based on a significant and plausible deficiency while mandating refinement of the underlying analysis. Its decision must preserve the visibility of uncertainty and the terms under which better evidence would change the amount. Such an approach makes supervisory judgment reviewable instead of concealing it inside a single output.

Model Calibration Adjustments

The SCR-difference calculation

For risk-profile deviations falling under Solvency II Article 37(1)(a) or (b), the central calculation in Delegated Regulation Article 282 is the difference between an SCR modified to reflect the actual risk profile and the SCR calculated without any existing or simultaneous capital add-on (European Commission, 2015). Define K0K_0 as the pre-add-on SCR and K∗K^* as the counterfactual adequately specified SCR. The illustrative relationship is

Arisk=K∗−K0.A_{\mathrm{risk}}=K^*-K_0.

The authority's actual lawful decision involves the regulatory conditions, relevant calculation date and validated inputs. The identity shows the direction and economic object of the correction: it is a targeted supplement to an inadequate modeled capital measure, not a general penalty indexed to firm size.

Consider a synthetic company with K0=240K_0=240 million currency units. A corrected distribution and aggregation calculation, based on documented exposure changes and dependent tail losses, yields K∗=292K^*=292 million. The arithmetic difference is Arisk=52A_{\mathrm{risk}}=52 million, increasing the displayed SCR to 292 million. With eligible own funds of 350 million, the pre-add-on coverage ratio is 350/240=145.83%350/240=145.83\%, while the corrected ratio is 350/292=119.86%350/292=119.86\%. These figures describe one hypothetical time point, assume the entire amount is legally eligible and deliberately exclude transitional effects or preexisting add-ons.

A properly executed calculation would separately establish each adjusted quantity, reconcile capital components and make the model reproducible. If the corrected result were only 255 million, the same formula would produce a 15-million supplement. The amount must follow the documented change in measured risk, not a predetermined supervisory percentage. If available data support a credible interval of possible corrected requirements, sensitivity tables should show that uncertainty even when the law ultimately requires a point decision.

Hierarchy of actuarial methods

Article 283's methodological hierarchy first contemplates adjustments to assumptions or parameters, then alternative methods when these are insufficient or inappropriate, and finally comparison with undertakings with similar risk profiles when other approaches fail (European Commission, 2015). The text explicitly demands suitable actuarial and statistical techniques, accurate and appropriate data, and protection of professional secrecy when peers' information is used.

That hierarchy has methodological substance. A parameter adjustment preserves much of the established framework and can isolate the marginal effect of a known deficiency. A structural modification may be needed if the formula omits an important risk driver or uses a poorly suited aggregation mechanism. A peer benchmark can be informative when an undertaking has sparse data, but it carries selection bias and heterogeneity: the chosen peers may differ in asset quality, reinsurance conditions, management options, contract mix or risk concentrations.

A supervisor should document why a simpler correction fails before relying on a more indirect one. In a catastrophe portfolio, a geographic concentration adjustment may be transparent if exposure data are strong and the modeled peril remains stable. If the portfolio's loss dependence with other coverages is also wrong, modifying one concentration parameter alone may not restore aggregate tail calibration. The further the calculation moves from the firm's own observed exposures, the more important uncertainty ranges and sensitivity comparisons become.

Valuation-adjustment arithmetic differs

For the specified adjustments or transitional measures under Article 37(1)(d), Article 284's four-term formula combines changes in eligible own funds and SCR, while Article 285's method examines actual assets, liabilities and risk profiles (European Commission, 2015). Using OF0,K0OF_0,K_0 for the reported position and OF∗,K∗OF^*,K^* for a properly modified position, the algebraic structure of Article 284 can be expressed as

Aadjustment=(OF0−OF∗)+(K∗−K0).A_{\mathrm{adjustment}}=(OF_0-OF^*)+(K^*-K_0).

This expression should be used only for the legal category and inputs to which Article 284 applies. It reveals why a valuation adjustment may have an effect through capital resources as well as required capital.

Illustratively, suppose OF0=300OF_0=300 and K0=200K_0=200, whereas correcting a valuation adjustment to fit actual assets and liabilities gives OF∗=284OF^*=284 and K∗=211K^*=211. The formula yields Aadjustment=(300−284)+(211−200)=16+11=27A_{\mathrm{adjustment}}=(300-284)+(211-200)=16+11=27. Treating the discrepancy solely as an 11-unit change in SCR would miss the 16-unit own-funds effect. This is a constructed numerical identity and does not assert that any specific insurer meets the statutory materiality test.

The example also demonstrates the need to examine apparently improved coverage carefully. A favorable valuation adjustment may increase reported own funds, lower modeled sensitivity or both. If the adjustment is inconsistent with the undertaking's actual assets and liabilities, an add-on may compensate the solvency assessment while the firm corrects its valuation and risk management. This is a more subtle intervention than simply raising a catastrophe risk charge.

Quantile uncertainty and finite simulation

An internal model approximates a probability distribution; the tail is inferred from simulated scenarios or fitted components. If the Monte Carlo sample contains NN independent draws, the nominal expected number above the 99.5th percentile is 0.005N0.005N. At N=10,000N=10{,}000, that is 50 observations. At N=2,000N=2{,}000, it is 10. The actual count is random; dependence, scenario construction, proxy approximation and model error can further affect reliability. A simulation sample size alone does not validate the economic realism of its tail.

The paired-sample bootstrap approach proposed by Gaigall (2021) addresses testing whether modeled capital requirements change significantly as a model evolves. model calibration and stress validation discussed by Dacorogna (2023) provides complementary diagnostics. The tools answer distinct questions: numerical stability, changes in modeled quantiles, distribution fit, tail plausibility and coverage of business-specific hazards. A capital add-on decision should explain which weakness each tool tests and how the results affect the chosen capital correction.

Excessive reliance on a point estimate makes an add-on sensitive to calibration choices. A documented scenario set can expose this dependence without making the supervisor adopt a new legal risk measure. For example, using a range of plausible catastrophe frequency and severity assumptions might yield corrected SCRs of 275, 292 and 318 million against the same 240-million base. The implied arithmetic differences are 35, 52 and 78 million, respectively. The middle value is illustrative; the surrounding values demonstrate sensitivity, not confidence limits derived from a statistical fit.

Partial internal models and aggregation boundaries

Partial models can cover selected risks or parts of a business, leaving other modules under a standard formula. The regulator must examine the integration method, allocation of risks between modeled and nonmodeled perimeters, double counting, omitted interaction and assumptions about diversification. A risk captured in one component may interact with the residual standard-formula capital through the combined loss distribution. Merely attaching an isolated charge to a module can fail to reproduce that interaction.

If two components are strongly positively dependent, an omitted interaction can increase aggregate risk more than the independent component's own stand-alone amount. If a revised model reduces risk in one component and increases it in another, a credible correction should trace both changes. Androschuck et al. (2017) underline the importance of justifying simulation architecture and proxy-model performance. The challenge is to correct the enterprise risk measure while preserving a legible explanation of the assumptions altered.

Capital model, financial condition and liquidity

A capital supplement can lower a ratio without immediately changing reported own funds. Raising cash through new eligible equity may increase available resources, but obtaining it can be difficult in distress. Reinsurance can reduce net retained losses but may create credit and settlement exposures. Asset disposals can improve liquidity while realizing losses or changing market-risk capital. The science of capital adequacy therefore requires the calculation of the add-on to be embedded in the insurer's wider economic balance-sheet forecast.

The undertaking's actual compliance options depend on legal capital eligibility, permissions, group funding constraints, exposure reductions and market conditions. A supervisory add-on is best evaluated by tracking whether the resulting choices make the insurer more resilient under the specified adverse scenarios. This is an analytical assessment of incentives and financial dynamics, not a substitute for the statutory computation.

Remediation and Review

The correction is supervised through time

Under Article 37's review-and-removal rule, supervisory authorities review an imposed add-on at least annually and remove it when the undertaking remedies the deficiencies that caused it (European Parliament and Council, 2009, as amended). This rule links an exceptional capital requirement to an identifiable factual condition. Its presence creates a governance obligation for both sides: the insurer should demonstrate correction; the authority should reconsider whether the original grounds and amount persist.

In internal-model and governance cases, Article 37 also expressly requires the supervisor to ensure that the insurer makes every effort to remedy the deficiencies. A credible remediation plan therefore needs accountable owners, technical deliverables, a complete data trail, an independent verification route, target dates and criteria demonstrating that the updated model is operating effectively. A new number generated by an updated model does not by itself prove that the defect was resolved.

The independent validation requirements and validation tools support an evidence-based closure process (European Commission, 2015). If a model was deficient because inflation assumptions were stale, closure should show that data sourcing, parameter governance, sensitivity review and portfolio monitoring now reflect current exposures. If the defect involved catastrophe tail dependence, closure requires tests of the revised dependence structure and its performance in joint extreme scenarios. If the defect involved board oversight, the insurer should demonstrate operationally effective controls, not merely revised policy wording.

Remediation alternatives and their limits

Capital can serve as a protective measure while deeper weaknesses are being repaired. Other legally available supervisory measures may include improved reporting, tighter risk limits, enhanced independent validation, changes to reinsurance, restrictions on underwriting, corrective governance and requirements for model redevelopment. The specific powers and procedures vary by jurisdiction. Under Article 36's corrective oversight, the EU supervisory review framework requires powers to remedy identified deficiencies; Article 37 sets the additional conditions for a capital add-on (European Parliament and Council, 2009).

A remedy must address the causal mechanism. If an insurer's model fails due to deficient source data, adding capital while leaving the data collection system unchanged can increase its temporary buffer but will not make subsequent capital calculations reliable. If counterparty recoveries are overstated, stronger collateral or more conservative recoverability assumptions may be needed. If risk aggregation understates common shocks, redefining diversification assumptions, stress scenarios and reporting can improve both the model and managerial decisions.

Supervisory decisions can also create incentives. A temporary, measurable add-on may motivate investment in controls that otherwise compete for budgets with revenue projects. Conversely, a charge that is difficult to remove may discourage transparent identification of model weaknesses and encourage superficial modeling adjustments. The design of the decision should therefore reward demonstrable risk reduction and validation rather than numerical engineering to produce a lower SCR.

Continuing quantitative verification

A before-and-after comparison should hold constant the factors necessary to isolate the corrected weakness. Suppose a catastrophe-model change increases estimated tail loss, while a new reinsurance treaty simultaneously reduces the retained exposure and an asset de-risking program increases available capital. A lower post-remediation SCR may be attributable to all three actions; the supervisor should disaggregate the effects rather than infer that model improvement alone caused the change.

A useful internal decomposition reports the change due to exposure, assumptions, calibration, aggregation, management actions and capital resources. Interaction terms should be disclosed when decomposition is nonlinear. This is an analytical method of explaining causation, not a new prescribed regulatory capital model. The relevant evidence includes exposure files, scenario definitions, model runs, validation sign-offs, and documented management actions. Independent replication or reconciliation is especially valuable when large reductions are claimed.

Review frequency and event triggers

Annual minimum review under Article 37 is a floor for formal reconsideration, not a reason to ignore material events between review dates. Under the ongoing model-validation rule, data developments and changes in model specification have to be considered as part of a regular process. Significant portfolio transactions, reinsurance exhaustion, new capital instruments, material loss events, revisions of assumptions and governance failures can justify earlier attention under the supervisory framework.

The date of the initial add-on and any later calculation must be retained in the record. Comparing nominal currency values across dates without reconciling exposures, inflation, valuation basis and data versions may create a false narrative about effectiveness. In a long-tail liability insurer, a revised reserve estimate and changed discount curve may explain more of an apparent capital trend than improvements to model architecture. This temporal reasoning protects the scientific validity of the eventual removal decision.

Group concerns and transferability

At group level, an apparently comfortable consolidated solvency ratio may coexist with fragile legal-entity capital positions. Model limitations can arise from intragroup guarantees, reinsurance, liquidity transfers, correlated subsidiaries, or diversification benefits assumed across legally ring-fenced funds. Article 232 group capital add-on provides a group-specific supervisory response where consolidated group SCR inadequately reflects the group risk profile (European Parliament and Council, 2009, as amended).

The supervisor must avoid both overlooking and double counting a deficiency across solo and group calculations. The imposition of an add-on on a subsidiary does not automatically establish that exactly the same amount belongs on top of the group requirement; group aggregation and interaction must be assessed under the applicable rules. Equally, a group model's apparent diversification benefit does not assure ready transfer of eligible resources to the entity where claims become due. The regulatory treatment should be traceable to the relevant group perimeter and applicable solvency methodology.

Disclosure and accountability

Transparency is limited by the legitimate confidentiality of models, contract terms and company-specific supervisory findings. Nevertheless, an intelligible public description can distinguish the legal grounds, scale and broad supervisory purpose of add-ons without publishing proprietary inputs. EIOPA's aggregate reporting on capital add-ons demonstrates one level of transparency (EIOPA, 2026 access). In-depth causal evaluation of the policy requires more detailed evidence than country totals can provide.

A high-quality review record should answer four questions: What deficiency justified the intervention? How was the supplemental amount estimated? Which remedial measures were completed and tested? Why did the authority maintain, change or remove the add-on? These questions preserve continuity across changes in management, supervisory staff, models and market conditions. Requiring those answers is also a practical way to prevent a temporary correction from becoming an unexplained permanent feature of the firm's regulatory capital.

Comparative Capital Governance

European Union: exceptional add-on and shared legal methodology

The EU approach embeds capital add-ons in a detailed sequence of law, delegated calculation rules and ongoing supervisory review. Articles 36–37 define the review and intervention architecture; Articles 276–287 of Delegated Regulation (EU) 2015/35 establish particular assessment and calculation elements. The modification hierarchy and the valuation-adjustment formula demonstrate the regime's insistence that distinct sources of inadequacy require distinct quantitative treatments (European Commission, 2015, as amended).

The regime's strength lies in specifying how an exceptional capital adjustment relates to a target SCR, governance deficiencies and corrective supervision. Its scientific vulnerability lies in the evidential difficulty of establishing a credible counterfactual capital requirement from imperfect models. Supervisory comparison across member states also depends on consistent application, specialist expertise, available data and similar interpretation of significance. EIOPA's annual capital add-on statistics create a public view of use, while remaining insufficient alone to identify causal effects on insurer failures or policyholder protection.

United Kingdom: a distinct post-reform supervisory policy

The PRA's SoP4/24 capital add-on policy sets out the UK approach to circumstances, methods, group usage, decision procedure and continued monitoring and removal, with its November 2024 version effective December 31, 2024 (PRA, 2024). Its companion SoP3/24 internal-model policy addresses permissions, major model changes and ongoing assessment. The UK framework reflects national policy choices made after separation from EU institutional governance and should be interpreted through current UK legislation and the PRA Rulebook rather than treated as directly binding EU law.

The PRA's PS2/24 policy statement discusses an exceptional methodology using the standard formula as a reference for certain internal-model capital add-ons and records consultation concerns about whether the method properly fits annuity and other specialized risk profiles (PRA, 2024). This debate is scientifically instructive: a readily observable benchmark can help supervision when the approved internal model is suspect, but the benchmark itself may have substantial known limitations for the particular undertaking. Benchmarkability and risk sensitivity are related yet distinct virtues.

An appropriate comparison therefore tracks the practical consequences of UK-specific permissions, safeguards and capital add-on methods rather than assuming that similar labels entail identical calculation rules. On either side of the jurisdictional distinction, a model challenge is best justified by explicit evidence that supervisory measures address the insurer's actual risk.

International principles and legal differentiation

The IAIS ICP and ComFrame framework supports risk-sensitive capital assessment, group supervision, ORSA and independent supervisory judgment (IAIS, 2024–2026 online text). International standards establish common expectations and reference points, while the precise legal power to impose an add-on arises from domestic or regional implementation. The difference matters especially in cross-border groups: a consolidated capital metric is not a replacement for local legal authority and coordination.

The Canadian Guideline A-4 describes supervisory capital targets and forward-looking capital assessment for federally regulated insurers, with differentiated ratios and approaches for life and property-casualty business (OSFI, 2025). The LICAT 2025 guideline states that the Superintendent may direct a life insurer to increase capital notwithstanding compliance with published guideline measures (OSFI, 2025). The Canadian architecture therefore illustrates how a supervisory capital judgment may operate alongside sector-specific target tests, but its legal foundations and numerical requirements are different from EU Article 37.

The US state-based risk-based capital framework uses defined action levels to trigger escalating supervisory responses and expressly cautions that RBC is not a stand-alone measurement of financial solvency (NAIC, 2026). This regime should not be described as an application of the EU SCR add-on formula. The comparison instead shows how a risk-based capital threshold can be combined with other supervisory instruments in a system that places greater emphasis on action levels and state regulatory authorities.

Limits of cross-country ratios

A percentage solvency ratio is a quotient whose numerator and denominator depend on national valuation, asset admissibility, risk factors, capital tiers, calibration horizons, reinsurance recognition and group boundaries. A ratio of 130% under one regime is therefore not automatically stronger or weaker than 125% under another. The comparability problem persists even if both authorities use the phrase “risk-based capital.”

The relevant comparison is institutional and functional. Does the legal framework identify material omissions? Can the supervisor demand better data and models? Is a quantitative supplementary capital measure available when necessary? Are the reasons and remediation terms reviewable? Is there effective intervention before a deterioration becomes irreversible? These questions permit rigorous comparative analysis without inventing a common numeric scale unsupported by law.

The 2027 EU reform boundary

Directive (EU) 2025/2 revises the Solvency II framework, and the relevant national measures generally start applying January 30, 2027 (European Parliament and Council, 2025). The Commission's delegated-acts regulatory register records additional amendments, including Delegated Regulation (EU) 2026/269, published in February 2026 (European Commission, 2026). EIOPA's revised 2026 supervisory review guidelines likewise identify January 30, 2027 as the application date, while the earlier supervisory review guidelines remain the reference before then (EIOPA, 2015; 2026).

An analysis dated November 14, 2026 must therefore keep the baseline and prospective regimes separate. It may examine how reforms could modify proportionality, supervisory practice, reporting or valuation assumptions, but should not describe future provisions as already applicable solely because they have been adopted or published. A capital add-on assessment initiated before transition should also trace which legal version governed the relevant decision and whether any subsequent reassessment is necessary under the new rules.

Comparative synthesis: the conditions of effective intervention

Across the jurisdictions considered, additional capital intervention is useful only insofar as it improves the realistic protection of policyholders and the insurer's capacity to meet obligations. A correct amount with an unverifiable model can become unreliable over time; a well-designed risk model can remain ineffective if organizational control fails; a strong consolidated ratio may conceal a legal-entity funding bottleneck. These are linked mechanisms, not competing definitions of solvency.

The comparative synthesis yields five connected findings. First, the legitimacy of a capital supplement depends on a source-specific legal power and a documented finding of material inadequacy. Second, mathematical corrections require explicit loss distribution, valuation and aggregation assumptions. Third, supervisory model challenge is strongest when it tests model design, data, governance and use together. Fourth, remediation and removal conditions should be established in terms that can actually be tested. Fifth, observed add-on counts and national ratios cannot substitute for evidence of risk reduction and policyholder outcomes. Each is an analytical conclusion supported by the combined legal, statistical and actuarial record; none is presented as a new doctrine.

Conclusion

Solvency capital add-ons are a mechanism for confronting a particular failure of correspondence between a regulated insurance undertaking's actual risk and the risk expressed by its capital calculation or related governance and valuation assumptions. European Solvency II defines exceptional grounds for their use and links decisions to supervisory review, explicit reasons, calculation rules, continuing remediation and at-least-annual reconsideration. Its detailed delegated provisions distinguish numerical SCR underestimation from governance deficiencies and deviations associated with valuation adjustments. Those distinctions are economically consequential because the correction may involve a changed risk distribution, an additional allowance for governance risk, a change in eligible own funds, or several interacting effects.

The principal scientific result of this monograph is that supervisory capital measurement must be evaluated as an evidentiary sequence through time. A credible intervention begins with a material and legally relevant diagnosis; proceeds through an actuarially appropriate, reproducible and sensitivity-tested correction; is accompanied by a remedy matched to the underlying cause; and remains open to adjustment or removal as evidence improves. Differences among legal systems must be respected even where their protective objectives converge. The UK capital add-on policies furnish an important comparative example; Canadian and US practices reveal distinct ways to combine risk-based numbers, internal capital judgment and supervisory action. International principles illuminate common governance objectives without supplying interchangeable national powers.

The illustrative computations clarify the mechanisms. An add-on changes the denominator of an SCR coverage ratio. A corrected risk-profile SCR can be represented through a difference calculation under the relevant EU provisions, whereas a long-term adjustment discrepancy may involve both own-fund and SCR effects. Dependence assumptions, rare-tail simulation and model governance create material uncertainty around any numerical answer. A supervisory decision gains authority when it documents that uncertainty and explains the evidence selected.

The study's conclusions are limited to the legal texts and published research examined as of November 14, 2026. It does not estimate the population-wide causal effectiveness of add-ons, assess any actual insurer's capital requirement or claim independent verification of confidential supervisory practice. Prospective EU changes beginning January 30, 2027 are distinguished from the current baseline. Within those boundaries, the analysis provides a complete interpretive and quantitative account of capital add-ons and supervisory model challenges as instruments of accountable prudential insurance supervision.

References