Risk Aggregation, Stress Testing and Capital Management

Introduction and Context

Enterprise risk management (ERM) is concerned not only with identifying individual risks but also with understanding how risks interact across products, legal entities, geographical markets and business lines. A bank may face credit risk in its lending portfolio, market risk in its trading book, operational risk in its payment systems, liquidity risk in funding markets and strategic risk from competition or regulation. These risks cannot be managed adequately by simply adding separate risk reports. Their aggregation requires a coherent measurement framework, dependable data, explicit assumptions about dependence and strong governance.

Risk aggregation, stress testing and capital management are therefore interconnected. Aggregation estimates the distribution of enterprise-wide losses. Stress testing examines how that distribution and the institution’s capital position behave under severe but plausible conditions. Capital management determines how much loss-absorbing capacity is required, how it should be allocated and whether the institution is being compensated for the risks it accepts.

The central managerial challenge is that diversification is not constant. Correlations can rise sharply during crises, liquidity can disappear, operational failures can occur simultaneously with market disruption, and model relationships estimated in normal times may fail in the tail. Graduate-level ERM therefore requires both quantitative modelling and informed judgement.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain the purpose and architecture of enterprise-wide risk aggregation.
2. Distinguish marginal, stand-alone, incremental and diversified risk measures.
3. Analyze diversification, dependence and tail dependence.
4. Design scenario, sensitivity, stress and reverse-stress tests.
5. Relate economic capital to regulatory capital and available financial resources.
6. Apply risk-adjusted performance measures to business and capital-allocation decisions.
7. Evaluate model, governance, ethical and implementation risks in capital management.

1. The Architecture of Enterprise Risk Aggregation

1.1 From risk silos to an enterprise view

A risk silo measures one exposure or one business unit independently. An enterprise view asks four additional questions:

• How large is the exposure relative to the whole institution?
• What common drivers affect several portfolios?
• Which risks are naturally offsetting and which are reinforcing?
• How much capital is required for the combined loss distribution?

The aggregation process normally contains five layers:

A sound framework defines a common risk taxonomy, common data standards, consistent time horizons and a clear treatment of diversification benefits. It also prevents double counting. For example, a credit spread movement may appear in both market-risk and credit-risk calculations; the framework must specify whether the effect is assigned to one category, jointly modelled or reconciled through an adjustment.

1.2 Loss distributions and risk measures

Let L denote a non-negative loss over a specified horizon. A risk measure converts the loss distribution into a decision quantity. Value at Risk at confidence level α is the α-quantile of loss:

Here, P is the probability operator and α is commonly 0.95 or 0.99. VaR answers: “What loss threshold will not be exceeded with probability approximately α over the stated horizon?” It does not describe the average severity beyond that threshold.

Expected Shortfall (ES), also called conditional VaR in some contexts, measures the average loss in the tail:

For continuous distributions, this is equivalent to the average of tail quantiles:

The first expression is intuitive but requires care for discrete distributions, where several definitions exist. The second expression is a general quantile-based representation under appropriate integrability conditions. ES is often preferred for capital analysis because it reflects tail severity and, under standard conditions, is subadditive.

A risk measure should be interpreted together with its horizon, confidence level, currency, loss definition, valuation basis and treatment of recoveries. A one-day trading VaR cannot be compared directly with one-year insurance capital without adjustment.

2. Aggregation, Diversification and Dependence

2.1 The variance–covariance framework

For two loss variables L₁ and L₂, the variance of total loss is:

where σ₁ and σ₂ are loss standard deviations and ρ₁2 is their correlation. The formula assumes finite variances and a linear additive loss representation. It is exact for variance regardless of distribution, but it does not by itself determine VaR or ES unless the distributional structure is also specified.

For n risks with exposure or allocation weights collected in vector w, and covariance matrix Σ, portfolio variance is:

The covariance matrix must be symmetric and positive semidefinite for a valid variance model. Negative correlations may create diversification, while positive correlations increase aggregate volatility. However, volatility diversification is not identical to tail diversification.

2.2 Stand-alone, marginal and incremental risk

Stand-alone risk is the risk of a business unit considered independently. Aggregate risk is the risk of the combined institution. Diversification benefit may be represented as:

where R_i is the chosen stand-alone risk measure and R_total is the aggregate measure. This expression is meaningful only when all measures use the same horizon, confidence level and loss convention.

Marginal risk contribution measures the sensitivity of total risk to a small change in an exposure. For volatility-based portfolio risk:

The component contribution is:

Under differentiability and positive portfolio risk, the component contributions sum to total volatility by Euler’s theorem. For VaR or ES, Euler allocation can be used when the risk measure is sufficiently differentiable, often through conditional expectation or simulation-based derivatives. In practice, allocation results can be unstable when positions are illiquid, distributions are discontinuous or model inputs change abruptly.

Incremental risk is the change in total risk caused by adding or removing a position:

It is especially useful for limit decisions and acquisitions. A position can have substantial stand-alone risk but low incremental risk if it offsets existing exposures. Conversely, a small position can have high incremental risk if it loads on an already concentrated factor.

2.3 Dependence beyond correlation

Correlation measures average linear co-movement. It does not fully describe dependence in skewed or heavy-tailed distributions. Two risks may have low ordinary correlation but become highly connected during a crisis. Sources of hidden dependence include:

• common macroeconomic factors;
• shared borrowers, suppliers, custodians or technology providers;
• collateral and funding feedback loops;
• common model assumptions;
• procyclical behaviour, such as margin calls and forced sales;
• legal or regulatory actions affecting several entities simultaneously.

Copulas separate marginal distributions from dependence structures. If F_i is the marginal distribution of loss L_i, a joint distribution may be written conceptually as:

where C is a copula and θ represents dependence parameters. This separation is useful but not automatically reliable. A copula is only as credible as its calibration, economic interpretation and validation.

Tail dependence describes the probability that one variable is extreme when another is extreme. Lower-tail dependence for variables X and Y can be defined as:

Upper-tail dependence is defined similarly as u→1−. A Gaussian dependence model with correlation less than one has zero asymptotic tail dependence, even though it may show ordinary correlation. Student-t dependence can produce positive tail dependence. This distinction matters when estimating joint defaults, market losses or liquidity outflows.

2.4 Worked example: diversification and concentration

Assume two business units have annual loss standard deviations of 40 and 30 monetary units, with correlation 0.25. Their aggregate standard deviation is:

The sum of stand-alone standard deviations is 70, so the volatility-based diversification benefit is approximately 14.32 monetary units, or 20.46% of the undiversified sum. This result is not a guarantee of tail diversification. If correlation rises to 0.80 during a stress period, aggregate standard deviation becomes:

The benefit falls to approximately 3.52 units. Management should therefore avoid embedding a stable normal-period correlation in capital planning without testing correlation deterioration.

3. Stress Testing and Reverse Stress Testing

3.1 Purpose and design principles

Stress testing evaluates the effects of adverse conditions on risk, earnings, liquidity and capital. It is not a forecast of the most likely future. Its purpose is to expose vulnerabilities, challenge assumptions and support decisions before losses occur.

A credible programme combines:

1. Sensitivity tests, which change one factor, such as interest rates or default rates.
2. Scenario tests, which change several linked variables according to a coherent narrative.
3. Historical stresses, which replay or adapt past episodes.
4. Hypothetical stresses, which explore plausible but unprecedented combinations.
5. Reverse stress tests, which start with institutional failure and identify conditions capable of causing it.

Scenarios should be severe but plausible, internally consistent and decision-relevant. A scenario that increases unemployment, property prices and borrower credit quality simultaneously may be mathematically convenient but economically incoherent. Scenario design should specify the initial condition, shock path, management response, time horizon and feedback effects.

3.2 Translating scenarios into losses

A scenario engine maps risk drivers to financial outcomes. For a simple credit portfolio, expected loss can be represented as:

where PD is probability of default, LGD is loss given default and EAD is exposure at default. Units must be consistent: if EAD is in millions of currency, expected loss is also in millions. The formula is a simplified expectation and does not capture unexpected loss, dependence, timing, discounting or recovery uncertainty.

A stress scenario may change each parameter. For example, a recession can increase PD, reduce collateral values and therefore increase LGD, and raise EAD through drawn credit lines. The combined effect is often greater than the effect of changing only one parameter.

For market positions, scenario valuation should revalue instruments under shocked curves, spreads, prices, volatilities and basis relationships. For operational risk, scenarios may estimate losses from system outage, cyberattack, fraud, litigation and remediation costs. For liquidity risk, the analysis should include cash-flow timing, collateral calls, deposit outflows, market access and asset-sale haircuts.

3.3 Capital impact and management actions

A stress test should not stop at a loss estimate. It should show the effect on profit, retained earnings, regulatory capital, liquidity buffers, risk limits and business strategy. A simplified capital ratio is:

Both numerator and denominator may change under stress. Losses reduce capital; risk-weighted assets may increase because exposures migrate to higher-risk grades; market movements may affect valuation reserves; and management actions may reduce assets or raise capital. A static balance-sheet stress and an assumed-management-action stress should be reported separately.

Management actions must be realistic. A plan that assumes all assets can be sold at book value during a market-wide crisis is not credible. Actions should specify timing, legal authority, execution capacity, market impact and possible reputational consequences.

3.4 Reverse stress testing

Reverse stress testing begins with a defined failure condition, such as breaching a minimum capital ratio, exhausting the liquidity buffer or becoming unable to meet obligations. It then identifies combinations of events that could produce that outcome.

The method is valuable because conventional scenarios may be anchored to familiar history. Reverse testing asks: “What would have to happen for our strategy, controls or capital plan to fail?” It can reveal concentration in a particular counterparty, dependence on short-term funding, excessive model reliance or an unrealistic recovery plan.

Reverse tests should not be interpreted as precise probabilities. Their value lies in identifying vulnerabilities, thresholds and preventive indicators. Results can support risk appetite, early-warning indicators, contingency funding plans and recovery options.

4. Economic Capital, Regulatory Capital and Financial Resources

4.1 Economic capital

Economic capital is the amount of capital management believes is required to absorb unexpected losses over a stated horizon and confidence level while maintaining a target solvency standard. A simplified formulation is:

If the risk measure is ES or VaR, expected loss is subtracted to distinguish ordinary anticipated losses from unexpected losses. The treatment of expected loss must be consistent with pricing, provisions and earnings. Economic capital may also include capital for model uncertainty, stress losses, reputational effects or risks not captured by the primary model.

Economic capital is an internal management construct. It supports pricing, limits, portfolio selection, strategic planning and capital allocation. It is not necessarily equal to the capital required by supervisors or accounting standards.

4.2 Regulatory capital

Regulatory capital is determined by applicable laws, supervisory rules and reporting requirements. It is designed to protect depositors, policyholders, creditors and financial-system stability. Regulatory frameworks typically define eligible capital, risk-weighted assets, deductions, buffers, leverage constraints and liquidity requirements.

Regulatory capital can differ from economic capital for legitimate reasons. Regulation may use standardized risk weights, prescribed stress assumptions, floors, add-ons and conservative definitions. Internal models may recognize diversification that regulation limits or excludes. Conversely, economic capital may include strategic, reputational or concentration risks not fully represented in minimum regulatory calculations.

4.3 Capital adequacy and capital planning

Capital management must distinguish:

• Available capital: resources legally and operationally available to absorb losses.
• Required capital: capital implied by economic, regulatory and internal risk standards.
• Target capital: desired capital after considering uncertainty, growth and market confidence.
• Management buffer: amount held above minimum requirements.

A capital plan projects these quantities under base and stressed conditions. It should integrate earnings, dividends, issuance, maturities, asset growth, risk migration, acquisitions, model changes and management actions. Capital is scarce, but holding too little can create insolvency and forced deleveraging; holding too much can reduce returns and strategic flexibility.

5. Risk-Adjusted Performance and Capital Allocation

5.1 Why accounting profit is insufficient

A business unit can generate high accounting profit while consuming disproportionate capital or exposing the institution to severe tail loss. Capital allocation therefore links return to risk. Common measures include:

A practical numerator may be net income after expected loss, operating costs, taxes and allocated funding costs. The denominator should reflect the risk measure and horizon used in the decision. A unit should be compared with a hurdle rate that compensates for the institution’s cost of capital and strategic requirements.

Other measures include return on risk-adjusted capital (often used similarly to RAROC), economic profit and risk-adjusted return on capital. Economic profit can be expressed as:

If risk-adjusted profit is 18 million, economic capital is 120 million and the hurdle rate is 12%, economic profit is 3.6 million. The business creates value only if it earns more than the capital charge.

5.2 Allocation principles

Capital allocation should be coherent with risk measurement. Desirable properties include:

• Full allocation: allocated capital sums to enterprise capital.
• Diversification recognition: offsetting exposures receive appropriate benefit.
• Risk sensitivity: capital changes when material risk drivers change.
• Decision usefulness: managers can understand and influence the allocation.
• Stability: allocations do not change excessively because of estimation noise.
• Avoidance of double counting: related risks are not charged twice.

Allocating all diversification benefits mechanically can create incentives for business units to appear correlated with others, even when that relationship is unstable. Senior management may therefore use both marginal allocations and conservative floors for concentrated or poorly modelled risks.

6. Integrated Illustrative Case: Meridian Financial Group

The following is a fictional illustrative case.

Background

Meridian Financial Group is a diversified financial institution with commercial lending, residential mortgages, securities trading and payments operations. Its internal capital model reports economic capital of 8.4 billion currency units, compared with regulatory capital requirements of 7.5 billion. The apparent surplus is 0.9 billion.

Problem

The board observes that the surplus depends heavily on diversification assumptions. The model uses normal-period correlations and gives a substantial benefit to the combination of lending and trading risks. Meanwhile, 28% of wholesale funding matures within three months, and the payments division relies on a single cloud-service provider.

Analysis

Risk management develops three scenarios:

Under the combined crisis, credit losses are 2.1 billion, market losses are 1.3 billion, operational losses are 0.4 billion and liquidity costs are 0.6 billion. These figures are not simply added to economic capital because they represent stressed losses over a specified scenario rather than a calibrated unexpected-loss quantile. However, the stress reduces projected regulatory capital by 3.2 billion after tax and other adjustments. Risk-weighted assets increase by 4.0 billion because credit quality deteriorates.

Suppose pre-stress eligible capital is 12.0 billion and risk-weighted assets are 80.0 billion. The initial capital ratio is:

After the stress, eligible capital is 8.8 billion and risk-weighted assets are 84.0 billion. The stressed ratio is:

The decline is caused by both loss absorption and risk-weight inflation. The result remains above a hypothetical 9% internal trigger but leaves little room for model error, further deterioration or management constraints.

Approach and solution

Meridian takes five actions. First, it replaces a single correlation matrix with a factor-based model and overlays a stressed-dependence scenario. Second, it introduces a reverse stress test in which the capital ratio falls below 9% or the liquidity buffer is exhausted. Third, it limits reliance on diversification benefits by applying a governance-adjusted capital floor. Fourth, it diversifies cloud providers and increases recovery testing for payment operations. Fifth, it changes performance reporting so business units receive capital charges based on marginal stressed risk rather than only normal-period stand-alone risk.

Outcome

The revised capital plan retains a larger management buffer, reduces short-term funding dependence and identifies an early-warning threshold based on credit migration, collateral calls and funding spreads. The board accepts lower near-term reported RAROC in exchange for greater resilience and more credible recovery capacity.

Learning points

The case demonstrates that aggregation is not merely an arithmetic exercise. Dependence can change under stress, capital ratios can deteriorate through numerator and denominator effects, and operational and liquidity risks can amplify credit and market losses. It also shows why stress testing must lead to specific decisions rather than remain a compliance report.

7. Governance, Model Risk and Ethical Considerations

Risk aggregation is vulnerable to model risk: incorrect specifications, biased data, unstable parameters, coding errors and inappropriate use. Validation should test conceptual soundness, data quality, implementation accuracy, sensitivity, stability, back-testing where applicable and performance under stressed conditions. Independent review is particularly important when a model produces large diversification benefits.

Governance should define ownership of risk taxonomies, assumptions, scenario approval, model changes, overrides and escalation. Boards and senior executives must understand the economic meaning of model outputs rather than treating them as objective facts. Reports should disclose confidence levels, uncertainty ranges, exclusions and material judgement.

Ethical issues arise when capital models encourage excessive risk-taking, hide losses through optimistic assumptions or allocate diversification benefits to reward politically powerful units. A model that understates risk can produce apparently superior performance and inappropriate bonuses. Compensation should therefore consider stressed outcomes, deferred losses, conduct events and the sustainability of returns.

Data privacy, cybersecurity and model confidentiality also matter. Aggregating enterprise data can create surveillance and access risks. Controls should enforce purpose limitation, segregation of duties, retention standards and auditable access.

8. Technology, AI and Contemporary Developments

Cloud computing, distributed data platforms and machine learning can improve aggregation speed and scenario analysis. Machine-learning methods may detect nonlinear relationships, identify anomalous exposures and generate alternative stress scenarios. However, predictive accuracy in ordinary periods does not establish reliability in crisis tails. Regime changes, limited crisis data, explainability problems and feedback effects remain serious concerns.

AI outputs should therefore be governed through documented data lineage, challenger models, human review, reproducibility, monitoring for drift and explicit override procedures. Generative AI may assist with scenario narratives or control documentation, but it should not independently determine capital adequacy or obscure the assumptions behind a stress result.

Contemporary practice increasingly emphasizes climate-related financial risk, cyber resilience, intraday liquidity, interconnectedness and non-bank financial intermediation. These risks challenge traditional annual or static capital models because their transmission channels are nonlinear, long-dated and difficult to calibrate. A useful response is not to claim false precision but to combine quantitative ranges, narratives, sensitivity analysis and robust decision rules.

9. Best-Practice Framework

An effective enterprise programme should:

1. Define risk consistently across entities and products.
2. Reconcile exposure data before modelling aggregation.
3. Use risk measures suited to the decision and tail objective.
4. Model common factors, concentrations and changing dependence.
5. Compare normal, stressed and reverse-stressed outcomes.
6. Separate expected loss, unexpected loss and scenario loss.
7. Reconcile economic capital with regulatory capital without conflating them.
8. Allocate capital using transparent and decision-relevant methods.
9. Challenge diversification benefits and apply prudent floors where necessary.
10. Link results to limits, pricing, funding, compensation and strategic choices.
11. Validate models independently and document uncertainty.
12. Escalate breaches promptly and test management actions operationally.

Challenges and Limitations

No aggregation system captures the full risk universe. Historical data may contain few crises; correlations are unstable; tail dependence is difficult to estimate; liquidity and operational losses are not always smoothly distributed; and scenario narratives may omit second-round effects. VaR can ignore tail severity, while ES depends on sparse and model-sensitive observations. Economic capital can create an illusion of precision, and regulatory capital may be optimized without improving actual resilience.

The appropriate response is triangulation: use several risk measures, models and scenario types; compare model output with expert judgement; maintain conservative governance where uncertainty is material; and treat capital as one element of resilience alongside liquidity, controls, governance and business continuity.

Chapter Summary

Risk aggregation converts diverse exposures into an enterprise view of loss, capital and vulnerability. Variance–covariance methods provide useful structure but do not fully capture nonlinear losses or crisis dependence. Diversification benefits must therefore be tested against common factors, concentration and tail dependence. VaR identifies a quantile, whereas ES captures average severity beyond the quantile and is generally more informative for tail capital decisions.

Stress testing examines the consequences of adverse but plausible conditions, while reverse stress testing begins with failure and searches for its causes. Capital management distinguishes economic capital, regulatory capital, available resources and management buffers. Risk-adjusted measures such as RAROC and economic profit connect risk consumption to performance, but they are meaningful only when capital allocations are consistent, transparent and appropriately conservative. Effective ERM integrates models, scenarios, governance, technology and managerial judgement.

Key Terms

Risk aggregation: Combining exposures and loss distributions across risks, entities or business units.

Diversification: Reduction in aggregate risk arising from imperfect dependence among exposures.

Tail dependence: The tendency of variables to experience extreme outcomes together.

Value at Risk: A specified quantile of a loss distribution over a defined horizon.

Expected Shortfall: The average loss in the tail beyond a chosen quantile.

Stand-alone risk: Risk measured for an individual unit without portfolio offsets.

Marginal risk: Sensitivity of aggregate risk to a small change in an exposure.

Incremental risk: Change in total risk caused by adding or removing an exposure.

Covariance matrix: Matrix describing pairwise covariance among risk variables.

Copula: A function that represents dependence separately from marginal distributions.

Stress testing: Analysis of financial effects under adverse specified conditions.

Reverse stress testing: Starting with a failure condition and identifying scenarios that could cause it.

Economic capital: Internally assessed capital required for unexpected loss at a chosen solvency standard.

Regulatory capital: Capital defined under applicable supervisory rules.

Risk-weighted assets: Regulatory exposure measure adjusted by prescribed risk weights.

RAROC: Risk-adjusted return divided by allocated risk capital.

Economic profit: Risk-adjusted profit less a capital hurdle charge.

Model risk: Risk of loss from incorrect, misused or poorly implemented models.

Management buffer: Capital held above minimum requirements for uncertainty and resilience.

Review Questions

5 Short Answer Questions

1. What is the difference between stand-alone risk and incremental risk?
2. Why can correlation fail to describe crisis dependence adequately?
3. What question does Expected Shortfall answer that VaR does not?
4. Distinguish economic capital from regulatory capital.
5. Why should a stress test include management actions and capital-ratio effects?

5 Long Answer Questions

1. Explain the architecture of an enterprise risk-aggregation framework, including data, dependence, measurement and governance.
2. Compare variance–covariance aggregation with factor and copula approaches. Discuss assumptions and limitations.
3. Explain how stress testing and reverse stress testing complement ordinary risk measurement.
4. Discuss the principles that should govern capital allocation across business units.
5. Evaluate the usefulness and limitations of RAROC as a strategic performance measure.

3 Analytical / Application Questions

1. Two business units have loss standard deviations of 50 and 40 million and correlation 0.30. Calculate aggregate standard deviation using the variance–covariance formula. Explain why the result should not be interpreted as a complete estimate of tail risk.
2. Design a combined stress scenario for a bank exposed to mortgage credit risk, securities trading, wholesale funding and cyber risk. Specify the transmission channels and the capital and liquidity indicators that should be monitored.
3. An institution reports substantial diversification benefits but uses normal-period correlations, has concentrated exposures to one sector and relies on short-term funding. Recommend a validation and capital-governance programme, including at least one reverse-stress test.

Multiple Choice Questions with Answers

Question 1

Which measure focuses on the average loss beyond a specified loss quantile?

A. Expected Shortfall

B. Standard deviation

C. Expected profit

D. Risk-weighted assets

Correct Answer: A. Expected Shortfall

Question 2

If two loss risks have positive correlation, all else equal, their aggregate variance is generally:

A. Lower than both individual variances

B. Increased by the covariance term

C. Equal to the smaller individual variance

D. Unaffected by their dependence

Correct Answer: B. Increased by the covariance term

Question 3

A reverse stress test begins by identifying:

A. The most likely quarterly forecast

B. A failure condition and scenarios that could cause it

C. The average historical return

D. The regulatory reporting date

Correct Answer: B. A failure condition and scenarios that could cause it

Question 4

Economic capital is primarily intended to measure:

A. Only accounting provisions

B. Internally assessed capacity for unexpected losses

C. The market value of all liabilities

D. A fixed statutory tax charge

Correct Answer: B. Internally assessed capacity for unexpected losses

Question 5

A portfolio has loss standard deviations of 30 and 40, with correlation 0. The aggregate standard deviation is approximately: √(30² + 40²)

A. 10

B. 50

C. 70

D. 1,200

Correct Answer: B. 50

Question 6

Which statement about VaR is correct?

A. It directly measures the average severity of all tail losses

B. It is a quantile of loss over a specified horizon and confidence level

C. It is always subadditive for every distribution

D. It does not require a loss definition

Correct Answer: B. It is a quantile of loss over a specified horizon and confidence level

Question 7

The simplified credit expected-loss formula is:

A. PD + LGD + EAD

B. PD × LGD × EAD

C. PD ÷ LGD × EAD

D. EAD − PD − LGD

Correct Answer: B. PD × LGD × EAD

Question 8

A business produces risk-adjusted profit of 18 million, economic capital of 120 million and a hurdle rate of 12%. Its economic profit is:

A. 3.6 million

B. 14.4 million

C. 18.12 million

D. 21.6 million

Correct Answer: A. 3.6 million

Question 9

Which is a principal limitation of normal-period correlation estimates?

A. They can never be calculated from data

B. They may understate dependence during crises

C. They always imply perfect dependence

D. They measure only accounting profit

Correct Answer: B. They may understate dependence during crises

Question 10

A capital ratio is calculated as:

A. Eligible capital divided by risk-weighted assets

B. Risk-weighted assets divided by eligible capital

C. Expected loss divided by revenue

D. Revenue divided by liquidity outflows

Correct Answer: A. Eligible capital divided by risk-weighted assets