How PD, LGD, and EAD feed capital calculations

How to study this topic

PD, LGD and EAD turn borrower risk, loss severity and exposure behaviour into controlled inputs for expected loss, unexpected loss and capital sensitivity. Read this as a banking control chapter, not as a technology marketing chapter. The useful question is whether the bank can trust, explain, control and monitor the result in credit-risk capital measurement.

The topic should stay close to banking evidence, policy ownership, customer outcome, regulatory expectation, data quality, operational process and audit trail. If the explanation drifts into generic AI language, it loses the reason this card exists.

A strong learner should be able to explain the model role, the evidence, the control owner, the human review point and the wrong-outcome risk to a business analyst, compliance analyst, credit-risk manager, developer, tester, auditor and senior risk owner.

Plain language meaning

In plain language, how pd, lgd, and ead feed capital calculations is about turning messy banking evidence into a controlled banking answer. The bank must separate observed fact, interpretation, model output and business action.

The model may classify, retrieve, summarise, estimate or rank, but the bank decides whether to approve, refer, investigate, escalate, communicate, provision, report or remediate.

Confidence in wording is not confidence in source quality, model design, policy authority or control effectiveness. Banking AI needs proof, not just fluent output.

Where it sits in the bank

This topic normally touches risk, compliance, product, operations, legal, data, model governance, technology and internal audit. Ownership must be explicit for source, policy, model, output and action.

The relevant population includes the banking cases described by the topic, including clean cases and edge cases: missing evidence, vulnerable customers, unusual products, disputed outcomes, local regulatory differences and manual overrides.

The same AI output can be low risk in internal learning and high risk when it affects a customer, control conclusion, finance number, regulatory response or audit file. Purpose matters.

Evidence and source material

Relevant evidence includes probability of default, loss given default, exposure at default, rating grade, collateral value, facility limit, drawn balance, credit conversion factor, maturity, default definition, recovery history, and risk-weighted asset output. These sources are not equal; official records, user-entered values, derived values, draft documents and approved policy need different trust treatment.

Timing matters because credit outcomes, policy versions, model versions, approval thresholds and customer status change. A correct answer for one date can be wrong for another date.

Evidence must be traceable to source, owner, version, date, access permission, transformation, retrieval path and limitation.

Data quality and control checks

Controls should include parameter governance, rating system approval, calibration evidence, portfolio mapping, collateral validation, EAD treatment, capital reconciliation, and model validation. These controls stop weak evidence from being treated as strong evidence and stop model output moving faster than governance.

Quality means authority, completeness, business meaning, lineage, label quality, fairness, privacy, security, citation quality, output review and customer impact.

When evidence fails, the response should be known: block, limit, refer, escalate, fallback, sample, remediate or retire the use.

How AI and ML can be adopted

Useful adoption includes supporting risk segmentation, detecting rating migration, challenging parameter stability, finding collateral exceptions, and monitoring EAD behaviour. These are support uses first; they improve search, classification, prioritisation, explanation, drafting and monitoring.

A bank should move from internal research assistance to controlled decision support and then to restricted automation only where validation, monitoring, accountability and fallback are mature.

The model role should be named with a verb: search, summarise, classify, estimate, recommend, refer, block, approve, escalate, report or communicate. Each verb has a different risk level.

Decision boundary and human judgement

The model may support judgement, but it should not erase judgement. The user must know whether the output is guidance, evidence, a draft, a score, a ranking, a referral trigger, a monitoring signal or a proposed communication.

Human review is useful only when the reviewer sees source evidence, reason codes, citations, limitations, model version, prompt context, retrieved documents and the policy rule being applied.

Overrides and corrections should be captured because they may reveal source gaps, policy ambiguity, retrieval weakness, model limitation or training needs.

Customer, compliance and conduct impact

The direct wrong outcome is understating capital, overstating returns, misreading portfolio quality or failing regulatory challenge. That is why the bank should not judge AI only by speed, test-set accuracy or user satisfaction.

A wrong model or unsupported GenAI answer can affect approval, decline, referral, complaint handling, compliance review, audit response, customer communication, collections, provisioning, capital, controls testing or regulatory reporting.

Conduct control asks what happens to the person, obligation, report or control affected by the answer. If the output creates pressure, exclusion, delay, weak disclosure or unfair treatment, it is a real banking risk.

Validation and monitoring

Validation should review concept, data, methodology, source quality, prompt design, retrieval quality, limitations, output behaviour and approved use.

Monitoring should look for drift, bad citations, outdated sources, repeated corrections, unfair outcomes, high override rates, weak explanations, user misuse, data leakage and missing audit trail.

When performance deteriorates, the response may be recalibration, retrieval tuning, source cleanup, stricter guardrails, retraining, manual review, restricted use, incident escalation or retirement.

Diagram walkthrough

The diagram follows five control steps: Borrower and facility, PD LGD EAD, Risk-weight logic, Capital view, and Governance evidence. Read it left to right as a controlled banking flow from evidence or question through AI support and into accountable use.

Each box is a control point. A bank should be able to name the owner, source, rule, limitation and retained evidence at every step.

Bank-ready checklist

Before production use, check purpose, source authority, population, date, output role, customer impact, compliance impact and reproducibility.

Then check access control, validation, monitoring, override governance, audit evidence, fallback rules, incident response and business ownership.

If those controls are weak, the model may still produce an answer, but the bank should not treat the answer as trusted banking evidence.

Source anchors for accurate study

Basel credit-risk principles frame credit risk around a suitable credit-risk environment, sound credit granting, administration, measurement, monitoring and adequate controls.

The Basel Framework uses probability of default, loss given default and exposure at default as core credit-risk components for internal ratings based credit-risk measurement.

IFRS 9 is effective for annual periods beginning on or after 1 January 2018 and includes expected credit loss impairment requirements for financial instruments.

CECL under US GAAP estimates expected credit losses over the contractual life using historical experience, current conditions, and reasonable and supportable forecasts.

NIST AI RMF is a voluntary framework for managing risks to individuals, organisations and society from AI systems across design, development, use and evaluation.

US banking model-risk guidance expects model purpose, input quality, assumptions, limitations, validation, monitoring, governance, controls and effective challenge to be proportionate to model materiality.

Federal Reserve SR 26-2, dated 17 April 2026, supersedes SR 11-7 and SR 21-8 and attaches revised interagency guidance on model risk management for banking organisations.

The 2026 revised model-risk guidance states that generative AI and agentic AI are not within that guidance scope, while traditional statistical, quantitative and non-generative/non-agentic AI models are covered.

The EU AI Act treats AI systems used to evaluate the creditworthiness of natural persons or establish a credit score as high-risk; Union-law fraud-detection uses and prudential capital-requirement uses are carved out.

Three estimates with different units

Probability of default, loss given default and exposure at default answer different questions. PD estimates the chance of a defined default over a stated horizon for a defined borrower or facility population. LGD measures loss severity conditional on default under a specified recovery and discounting convention. EAD estimates the exposure present if default occurs, including treatment of undrawn commitments where relevant. Their product is a familiar intuition for expected loss, but actual regulatory capital calculations depend on the applicable jurisdiction, approach, parameters, constraints and supervisory rules. A banking analyst should not present one simple multiplication as the complete capital formula.

Machine learning may help estimate a component, but the approved use must be explicit. A PD model that ranks loan applicants is not automatically calibrated for a prudential capital calculation. A collections LGD model trained on settled recoveries may be biased toward older defaults and may not represent current conditions. An EAD model for revolving facilities needs to account for the drawn balance and future utilization before default. Each component requires its own population, outcome, horizon, data quality and validation. Common source data does not make the three estimates interchangeable.

Define default before estimating it

A PD target needs an event definition, observation date and follow-up period. Delinquency, restructuring and qualitative triggers may be treated under applicable rules; a generic servicing flag is not enough. An obligor with several facilities may have a different unit from a facility-level reporting table. A recently originated loan cannot be counted as a non-default over a full twelve-month horizon before the year has elapsed. A loan sold or prepaid during the window needs a stated treatment. If a rule version changes, compare labels on a dated cohort and explain whether historical training data are harmonized.

The feature vector must precede the default event. A collections note written after default or a later bureau update cannot enter an application PD model. Availability time matters as well as effective time. Retain the source and feature versions used in training and live scoring. A model validation report should test discrimination, calibration and stability by product and relevant borrower segment. A good overall ranking metric can hide poor calibration in a segment that carries material exposure.

Estimate loss conditional on default

LGD depends on recoveries, costs, collateral, guarantees, seniority and time. Define gross exposure at the chosen default event, recovery cash flows, workout costs, discounting and the endpoint for an unresolved case. A recovery received years later cannot be treated as known at default for an earlier prediction. Losses from unresolved recent defaults may be incompletely observed. A model trained only on closed workouts can underrepresent difficult, long-running cases. The team needs a policy for incomplete recovery histories and evidence that the development sample matches the intended portfolio.

Collateral values are time-sensitive. An appraisal, market estimate and realized sale proceeds are not the same. Record valuation date, source, haircut or recovery assumption and legal enforceability. A secured mortgage and an unsecured personal loan need different treatment. A model might predict realized loss severity from property and loan characteristics, but legal process, economic downturn and servicing strategy affect realized recoveries. Validation should examine stressed periods and concentration, not only average historical loss.

Exposure can grow before default

For a term loan with a fixed schedule, EAD may be relatively close to a future outstanding balance, subject to repayments, interest and other contractual treatment. For a revolving line or card, a customer can draw additional amounts before default. A model estimates this behavior on a defined horizon and facility type. It should use limit, current drawn amount, utilization history and policy changes available at observation time, not the eventual balance as an input. A line reduction or freeze by the bank can affect utilization, so historic outcomes reflect the incumbent intervention policy.

An example makes the distinction visible. At the model date, a facility has a limit of 10,000 and a drawn balance of 4,000. At default months later, the drawn amount is 7,000. The historical 7,000 is a target for EAD validation, not information the model had on the earlier date. If the bank changed line limits during the period, compare cohorts carefully. Treat unused commitments under the applicable capital rules and product contract; a generic assumption that every undrawn unit is drawn is not a model estimate.

From parameter to capital engine

A capital engine receives approved parameters, exposure attributes, risk mitigants and rule versions. It applies the relevant standardized or internal-ratings approach and constraints, rather than accepting any model output with the right field name. An analyst should trace which exposures qualify for each treatment, how missing parameters fall back, and how capital calculations reconcile to the portfolio and finance records. Regulatory definitions and floors can constrain modeled values. Local implementation and approval determine the actual formula and permissions.

Consider a change in the PD model. Compare old and new PDs on a fixed exposure snapshot, then examine risk-weighted assets or other capital outputs under the actual approved engine. A small mean PD change can move a concentrated segment or threshold. Keep the exposure and LGD/EAD assumptions fixed to isolate the PD effect, then analyze combined changes separately. Document product exclusions, overrides and manual adjustments. A model performance improvement does not by itself authorize a capital reporting change.

Reconciliation and validation

For one reporting date, reconcile eligible facility counts, balances, collateral, parameter coverage, calculation failures and final reported totals. Sample an exposure and trace its source attributes, PD, LGD, EAD, capital treatment and aggregation. A source correction after the reporting date should create a marked restatement process under policy; it should not overwrite the original evidence. Compare model estimates with mature observed outcomes at appropriate horizons and note sparse or stressed segments. Independent validation challenges assumptions, data quality, calibration, use and limitations.

Monitoring needs separate clocks. Input distribution and score coverage can be checked quickly; defaults and recoveries mature slowly. A sudden rise in missing collateral values may change capital outputs before any performance metric moves. The owner should investigate source and policy changes before retraining a model. Scenario analysis can show sensitivity to downturn conditions without claiming it predicts a precise future. Capital is an aggregate prudential measure, not a customer-level approval rule.

A worked integration test

Choose a term loan, revolving line and secured facility. For each, freeze an as-of balance, limit, collateral record, applicable model versions and capital rule version. Compute expected parameter inputs and the capital engine output under the bank's approved methodology. Test a missing PD, stale collateral, a changed limit, a default event and a corrected exposure. Confirm which cases are calculated, referred or excluded, and reconcile totals before and after. The specific numeric result depends on jurisdiction and approved approach; the test verifies the chain from source and model to controlled reporting.

This chain should preserve distinct evidence for parameter development, model approval, production use and reported capital. A bank can use ML to improve estimates where permitted and validated, while remaining accountable for rule interpretation, data integrity and reporting. PD describes likelihood, LGD severity and EAD exposure at the event; their usefulness depends on precise definitions and a reproducible path through the actual capital calculation.

Model components under economic stress

Observed defaults, utilization and recoveries can move together during stress. A downturn can raise PD, increase draws on revolving facilities and reduce collateral recovery values. If separate models are calibrated only on benign periods, multiplying their central estimates may understate correlated stress. The bank should document how its applicable capital methodology treats downturn conditions and validate component behavior in available stressed cohorts. Scenario analysis can reveal sensitivity to common drivers, but a scenario is an assumption set rather than an observed forecast. An analyst should not independently layer ad hoc stress factors on top of parameters already calibrated for that purpose without checking for double counting.

For model governance, maintain a dependency map of economic inputs used by each component. A macro series revised after a reporting date must not silently alter the archived capital run. A provider outage should invoke a documented contingency for the affected parameter, with counts of exposures and amounts under fallback. Compare capital output before and after correction and assess whether reporting or model use needs formal adjustment. Review overrides as a separate population; a high override rate may signal that one component is not fit for a product or market condition. These checks help explain an aggregate movement instead of attributing it casually to borrower risk.

Banking practice note: definition ownership

For how pd, lgd, and ead feed capital calculations, definition ownership decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from probability of default through parameter governance and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

AI adoption should make weak evidence easier to see, inconsistent treatment easier to challenge, outdated documents easier to detect and operational exceptions easier to route. It must not hide uncertainty behind confident language.

A good implementation records source, owner, version, date, transformation, retrieval path, model version, prompt context, user action, limitation, review decision and monitoring result.

Banking practice note: source authority

For how pd, lgd, and ead feed capital calculations, source authority decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from loss given default through rating system approval and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: effective-date control

For how pd, lgd, and ead feed capital calculations, effective-date control decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from exposure at default through calibration evidence and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: population design

For how pd, lgd, and ead feed capital calculations, population design decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from rating grade through portfolio mapping and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: policy alignment

For how pd, lgd, and ead feed capital calculations, policy alignment decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from collateral value through collateral validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: model purpose

For how pd, lgd, and ead feed capital calculations, model purpose decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from facility limit through EAD treatment and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: approved use

For how pd, lgd, and ead feed capital calculations, approved use decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from drawn balance through capital reconciliation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: human review

For how pd, lgd, and ead feed capital calculations, human review decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from credit conversion factor through model validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: output limitation

For how pd, lgd, and ead feed capital calculations, output limitation decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from maturity through parameter governance and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: fairness and conduct

For how pd, lgd, and ead feed capital calculations, fairness and conduct decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from default definition through rating system approval and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: customer harm

For how pd, lgd, and ead feed capital calculations, customer harm decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from recovery history through calibration evidence and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: regulatory evidence

For how pd, lgd, and ead feed capital calculations, regulatory evidence decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from risk-weighted asset output through portfolio mapping and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: privacy and confidentiality

For how pd, lgd, and ead feed capital calculations, privacy and confidentiality decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from probability of default through collateral validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: access control

For how pd, lgd, and ead feed capital calculations, access control decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from loss given default through EAD treatment and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: audit trail

For how pd, lgd, and ead feed capital calculations, audit trail decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from exposure at default through capital reconciliation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: change control

For how pd, lgd, and ead feed capital calculations, change control decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from rating grade through model validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: monitoring thresholds

For how pd, lgd, and ead feed capital calculations, monitoring thresholds decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from collateral value through parameter governance and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: feedback loops

For how pd, lgd, and ead feed capital calculations, feedback loops decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from facility limit through rating system approval and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: exception routing

For how pd, lgd, and ead feed capital calculations, exception routing decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from drawn balance through calibration evidence and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: incident response

For how pd, lgd, and ead feed capital calculations, incident response decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from credit conversion factor through portfolio mapping and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: committee reporting

For how pd, lgd, and ead feed capital calculations, committee reporting decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from maturity through collateral validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: third-party dependency

For how pd, lgd, and ead feed capital calculations, third-party dependency decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from default definition through EAD treatment and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: training and user behaviour

For how pd, lgd, and ead feed capital calculations, training and user behaviour decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from recovery history through capital reconciliation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: fallback operation

For how pd, lgd, and ead feed capital calculations, fallback operation decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Trace one example from risk-weighted asset output through model validation and into the business use. If the team cannot trace that path without guesswork, the implementation is not mature enough for serious banking use.

Banking practice note: retirement and redevelopment

For how pd, lgd, and ead feed capital calculations, retirement and redevelopment decides whether the bank can trust the evidence, explain the result and defend the action. A model can produce a fluent answer or a precise score quickly, but a bank still has to prove why that output is suitable for credit-risk capital measurement.

Primary sources for further study

• Basel Framework CRE32: IRB risk components
• IFRS Foundation: IFRS 9 Financial Instruments