Loss given default in plain language

How to study this topic

Loss given default, or LGD, is the bank's estimate of the portion of exposure that may be lost if default happens, after considering recoveries, collateral, guarantees, costs, timing and workout outcomes. The phrase may sound technical, but the banking meaning is practical. A bank uses this concept to decide how much risk it is taking, how much loss it may face, how to price or limit exposure, when to refer a case, how to monitor portfolios and how to explain credit decisions.

Keep the scope banking-first. This chapter is about lending, borrowers, counterparties, credit obligations, portfolios, limits, collateral, repayment, arrears, provisions, capital, governance and customer treatment. It is not a payments chapter. Transaction behaviour may appear only as account conduct evidence where it genuinely supports credit risk.

A learner should finish this topic able to explain it to a credit analyst, business analyst, data engineer, model validator, underwriter, collections manager and auditor. Plain language matters because credit-risk models are not useful if only modellers understand the answer.

The plain language meaning

LGD answers the question: if default happens, what percentage of the exposure might the bank lose after recoveries? The bank may recover some money through collateral, guarantees, restructuring, collections, legal action or sale of the exposure. LGD focuses on the part that may not come back.

The concept must always be tied to purpose. A number without purpose is dangerous. A PD used for application decisioning may differ from a PD used for IFRS 9, capital or portfolio monitoring. An LGD used for pricing may differ from a downturn LGD view. An EAD for stress testing may differ from day-to-day exposure monitoring. Scorecards also differ by product, portfolio and decision point.

Plain language does not mean weak explanation. It means the explanation should remain accurate without hiding behind formulas.

Source systems and evidence

Useful source areas include collateral value, loan-to-value ratio, seniority, guarantees, recovery history, collection costs, workout period, foreclosure process, write-off data, settlement outcomes, legal costs, and economic conditions. A credit-risk chapter depends on controlled data because every input can change the risk view. Income evidence, arrears data, utilisation, collateral value, limit data, customer identity, bureau information and case outcomes all need clear source ownership and timing.

The bank must know which source is authoritative. A front-end application may show what the customer declared. A document-verification platform may show what was evidenced. A core loan system may show what was booked. A collections system may show what happened after stress started. A risk engine may show the rating and score. Finance may show impairment treatment. These views must be mapped, not casually mixed.

The source question becomes sharper when AI and ML are used. If a model learns from a field, the bank must know whether that field is complete, stable, corrected, overridden, refreshed and legally usable for the purpose.

Key inputs and calculations

Key inputs include collateral value, loan-to-value ratio, seniority, guarantees, recovery history, collection costs, workout period, and foreclosure process. The calculation must define the time horizon, observation period, product scope, customer population, exclusion rules and treatment of missing values. Without these, the same concept can be calculated in several different ways and produce conflicting results.

For LGD, the bank needs defaulted exposure history, recovery cash flows, collateral data, costs, time to recovery, cure rules and discounting logic. LGD is often harder than it looks because recoveries may arrive slowly and workout data may be incomplete.

The formula is only one part. Banking interpretation, data quality and governance decide whether the output is safe to use.

Controls before use

Controls should include defaulted exposure sample, recovery cash-flow validation, collateral valuation policy, discounting approach, cost treatment, and cure treatment. These controls must run before the concept feeds a model, decision engine, risk report, provisioning process or strategy review. Technical checks are not enough. The bank must check whether the value makes sense in credit terms.

Important checks include population coverage, source reconciliation, date alignment, treatment of overrides, treatment of outliers, missingness analysis, validation against realised outcomes and monitoring of drift. Credit-risk models can fail slowly. They may not throw a system error. They may simply become less aligned with borrower behaviour or economic conditions.

The bank should also define who can override the model, how overrides are recorded, and how override performance is reviewed. Human judgement can improve credit decisions, but undocumented override culture can damage validation.

How it supports bank decisions

Common uses include expected loss calculation, IFRS 9 ECL, CECL loss estimation, capital models, pricing, collateral policy, collections strategy, and portfolio stress testing. Each use has a different control requirement. A model used for internal portfolio insight may need one level of governance. A model used for credit approval, regulatory capital, provisioning or customer treatment needs much stronger evidence, validation and monitoring.

LGD is central because it influences the severity side of credit loss. Two borrowers may have similar PD but very different LGD because collateral, seniority, guarantees or recoveries differ. Banks cannot understand credit loss with PD alone.

The output should be explainable to the decision owner. If a credit manager cannot understand why the model changed a view, the bank has a governance problem even if the statistical result is strong.

Provisioning, capital and portfolio view

Credit-risk concepts matter beyond individual decisions. They feed portfolio monitoring, risk appetite, concentration review, stress testing, capital planning and expected credit loss. The same borrower-level evidence can roll up into management and regulatory views.

IFRS 9 expected credit loss uses credit-risk movement to decide whether a financial asset remains in a 12-month expected loss view or moves into lifetime expected loss because credit risk has increased significantly. That makes data timing, default definition and risk movement important. A poor risk signal can affect provisions and reported performance.

Capital frameworks also care about credit-risk parameters. PD, LGD and EAD are not classroom abbreviations. They are part of how banks understand expected and unexpected loss, risk-weighted assets, portfolio quality and risk-adjusted return.

Customer and conduct impact

Credit-risk models affect customers. They can approve credit, restrict credit, increase price, reduce limit, refer for manual review, trigger collections activity or identify customers needing support. A wrong model can harm a customer by granting unaffordable credit or by denying reasonable credit without a fair basis.

Fairness is not solved by removing obvious sensitive fields. Proxies can appear through income, geography, employment type, channel usage, account history, product access or missing data. The bank should review whether the model treats customer groups fairly and whether limitations are visible to decision-makers.

Human review is important where impact is high. A score can support judgement, but the bank remains accountable for the decision.

Validation and monitoring

Validation should review concept, data, methodology, calibration, discrimination, stability, limitations and use. It should not only ask whether the model predicts. It should ask whether the model is suitable for the bank's portfolio, product, policy, customer population and decision purpose.

Monitoring should compare predicted outcomes with actual outcomes. For PD, defaults should be compared with predicted default rates. For LGD, realised loss and recovery experience should be reviewed. For EAD, drawdown and exposure at default should be tested. For scorecards, approval quality, referral rates, override outcomes and portfolio performance should be monitored.

If performance deteriorates, the bank may need recalibration, redevelopment, policy adjustment, overlays, tighter controls or temporary restrictions. Model monitoring must have a response path.

Common mistakes

The first mistake is confusing the parameter with the whole credit decision. PD, LGD, EAD and scorecards are decision inputs, not the entire decision. The second mistake is using weak labels. The third mistake is ignoring data timing. The fourth mistake is assuming a score remains valid after products, policy or economic conditions change.

Another mistake is treating a model as neutral because it is quantitative. Credit models reflect historical data, policy decisions, source availability, customer access and economic cycles. The bank must challenge whether the model learned credit risk or learned old operating patterns.

The final mistake is failing to explain limitations to users. A model can be useful with limitations if users understand them. A model becomes dangerous when limitations are hidden.

Practical banking example

A bank has two borrowers with similar default likelihood. One loan is secured by strong collateral with clear legal enforceability. Another is unsecured with weak recovery history. Their PD may be similar, but LGD can be very different. That difference changes expected loss and pricing.

The practical point is that credit-risk models support judgement. They should make banking decisions more consistent and evidence-based, but they should not remove responsibility from the bank.

Bank-ready checklist

Before using this concept in production, ask: is the definition clear, is the population right, are sources authoritative, are time windows correct, are labels validated, are overrides governed, are missing values explained, are customer impacts reviewed, and can the result be reproduced later?

Then ask: is the model calibrated, independently reviewed, monitored, documented and approved for this exact purpose? Is it being used for approval, pricing, provisioning, capital, collections or internal insight? Has the bank matched controls to impact?

If the answers are strong, the concept is bank-ready. If not, the model may still produce numbers, but the bank should not treat those numbers as trusted credit-risk evidence.

Loss after a defined default

Loss given default, or LGD, estimates the share of exposure not recovered after a borrower defaults under a specified methodology. If an illustrative facility has exposure at default of 100 and the bank eventually recovers 70 after relevant costs and discounting, a simple illustration has 30 of loss, or 30% LGD. Production calculations may use different conventions, time horizons and regulatory or accounting requirements. The definition must say what counts as default, what exposure forms the denominator, which cash flows count as recovery, how costs and time are treated, and whether a case is resolved. A percentage without this contract can be misleading.

LGD is conditional on a default event. It is not the probability that a borrower defaults, the expected loss on every performing loan, or the amount a customer owes. A bank might use LGD in credit portfolio risk, pricing, allowance or capital calculations under distinct approved methods. The same observed recovery file may support several estimates, but a model validated for one use cannot automatically serve another. An analyst should identify the purpose and jurisdiction before discussing floors, downturn adjustments or accounting measurement.

Start with a recovery timeline

At default, freeze the exposure under the approved rule and record the event date. The bank may receive borrower payments, enforce a guarantee, sell collateral or write off an amount over months or years. A write-off is an accounting event, not necessarily the end of collections; later recoveries can occur. A collateral sale can incur legal and maintenance costs. A simple sum of cash received divided by the default balance ignores when those flows occur and whether they are attributable to that facility. The methodology identifies relevant direct and indirect costs, discounting and allocation to facilities.

Consider a fictional secured loan with exposure of 100 at default. The bank receives 20 from the borrower after six months and 55 from collateral after two years, while incurring approved recovery costs of 8. The illustrative undiscounted net recovery is 67 and loss is 33, but an approved economic LGD may discount delayed cash flows and apply other conventions. This is not a recommended formula. The important lesson is that recovery amount, timing, cost and denominator must be explicit. A model trained on final write-off flags would miss these distinctions.

Collateral is an uncertain source of recovery

A collateral valuation before default is not a guaranteed recovery amount. Market price may move, enforceability can be limited, senior liens may have priority, and sale costs can be material. Record valuation type, date, currency, legal rights and haircuts under the relevant method. A residential property, vehicle, inventory pledge and unsecured guarantee have different recovery processes. An ML model can use collateral attributes and market conditions to estimate loss severity, but it needs a dataset that links defaults to actual recovery experience and appropriate validation across conditions.

A loan-to-value ratio is a useful input only if the loan balance and valuation are consistent in time and scope. A stale property value can make a facility appear safer during a downturn. A guarantee from a related party may be less valuable when both borrower and guarantor face the same shock. Review concentration and correlation. A model with good average fit can understate tail losses if it has little stressed recovery data. Scenario sensitivity and cautious limitations should be documented rather than hidden behind a precise-looking decimal.

Denominator and facility mapping

The exposure at default can differ from the balance at origination or at a later reporting date. A revolving borrower may draw more before default. Accrued interest and fees may be treated under specific rules. If one customer has several facilities and a single collateral pool, allocate recoveries according to contracts and the bank's approved methodology. A customer-level total divided by one facility's exposure can produce nonsensical LGD. Use stable facility IDs, obligor links and versioned collateral relationships.

Reorganizations complicate the mapping. A facility can be refinanced, sold, transferred to a recovery system or assigned a new identifier. Keep lineage from original default through all successor accounts and recovery events. If a cash receipt is applied to the wrong facility, the model target changes even though the cash is real. Reconcile recovery totals to finance and servicing records, and sample individual cases. A high-level aggregate that balances can still conceal facility-level misallocation.

Incomplete and selected outcomes

Many recent defaults have not completed recovery. Training only on closed workouts selects cases that resolve quickly and can understate loss for difficult cases. Simply assigning unresolved cases a full loss can be equally misleading. The model owner documents a method for censoring or estimating incomplete recovery and tests sensitivity. Compare mature cohorts, recovery age and unresolved share by product and collateral. A model validated on loans defaulted a decade ago may not describe today's portfolio or legal process.

Collections strategy shapes outcomes. The bank may offer hardship support, sell nonperforming loans or change enforcement practice. Historic LGD reflects those interventions, not only borrower characteristics. If a new strategy is proposed, an ML model trained under the old one may need recalibration or restricted use. Document policy changes in the development sample and compare outcomes by workout path without claiming causal effects from observational data alone. A borrower who receives an arrangement may have been selected because their case looked recoverable.

Economic conditions and downturns

Defaults and recoveries can worsen together in a recession. Property values fall, sale times lengthen and guarantors can weaken. A model that uses only average benign-period recoveries can understate losses in stressed conditions. The bank should test available downturn cohorts, scenario sensitivity and concentrations, and follow the applicable prudential or accounting methodology for its intended use. Do not mechanically add an arbitrary stress factor to every LGD; document which assumptions already reflect stress and how an additional adjustment was approved.

Market data used in a historical back-test needs its publication vintage. A valuation revised after default cannot be used as if known before a decision. A recovery model predicting LGD at default may legitimately use collateral information known then, while a quarterly allowance model can use newer information available at its reporting date. These are different prediction moments. Store the source and decision cutoffs so the bank can explain why two estimates for the same facility differ.

An LGD model's features and target

Features might include product type, seniority, collateral coverage, borrower segment, jurisdiction, guarantee structure and macro context available at the prediction time. Avoid using future workout outcomes, later sale price or final write-off as inputs. The target should follow the approved realized-LGD calculation, with event and cash-flow lineage. Missing collateral documentation is distinct from an unsecured loan. A model service needs explicit missingness states and a fallback, not a zero value interpreted as no loss.

Evaluation should cover both ranking and calibration. A model may distinguish high- and low-loss cases but systematically underestimate the amount. Compare predictions with mature realized outcomes by product, collateral, vintage and relevant stress segment. Check stability around a new recovery policy or data migration. Outliers deserve investigation: an apparent LGD above 100% can result from costs, denominator or data errors, depending on the methodology. Do not automatically clip unusual values before understanding them.

A worked case file

For one fictional facility, the case file includes contract, balance and accrued items at the defined default date; collateral and guarantee records; recovery receipts with dates; allocated costs; amendments; and the final status. The analyst calculates the observed LGD under the approved convention and compares it with the model estimate made at the earlier prediction moment. If a recovery arrives years later, keep both the prior observed-target version and the updated one. A reviewer should be able to reproduce the arithmetic and explain whether a difference reflects timing, new evidence or a changed method.

Now test a second facility for the same borrower that shares collateral. The bank needs an allocation rule and a record of priority. A single property sale cannot be counted in full against both loans. Test a reversed receipt, a cost correction, a currency conversion and a late guarantee payment. If a recovery system migration creates duplicate cash-flow rows, a simple sum can overstate recovery. Source reconciliation and stable IDs are as important as the statistical model.

Relationship to PD and EAD

Expected loss intuition combines likelihood of default, exposure at the event and severity conditional on default. LGD does not replace the other components. An applicant with low PD can still generate a severe loss if an unusual default occurs; an unsecured small exposure can have high LGD but low monetary loss. A model owner should examine whether PD, EAD and LGD share economic drivers and whether a joint calculation double-counts or overlooks dependencies. The actual use in capital or impairment follows the applicable framework and bank approval.

For a loan approval, a predicted LGD may inform portfolio economics but should not be presented as a complete reason for declining an individual borrower. Credit policy and affordability remain separate. For a finance allowance, recovery timing and expected cash shortfalls matter under the relevant accounting method. For regulatory capital, constraints and supervisory approval may apply. The feature platform can serve common evidence while publishing separately governed parameter versions for each use.

Monitoring and governance

Near-term monitors include missing collateral values, new default volumes, recovery-event delays, model coverage and overrides. Realized recovery performance matures slowly; a quarter of new defaults cannot prove the final LGD calibration. Track vintages and the share unresolved at each age. Compare expected and realized cash flows on sufficiently mature cohorts, and investigate whether a change in economy, legal process, servicing or source coding explains movement. A higher recovered amount can reflect a change in case mix rather than a better model.

Independent validation checks concept, data, target calculation, model performance and use. A model trained on unsecured consumer loans should not silently score a new secured commercial portfolio. If a collateral feed is stale, the bank can use a documented fallback or refer the estimate; it should not supply a reassuring old value without a freshness signal. Model and finance owners approve material changes in their respective uses, and the release record pins source, feature, model and calculation versions.

Incident example

Suppose a recovery feed duplicates receipts after a platform upgrade. The observed LGD targets in a recent dataset fall sharply, making a new model appear stronger. Before deployment, reconciliation against the finance ledger finds duplicate receipt IDs. The owner stops the training release, corrects the source mapping and reruns evaluation on a fixed dated sample. If the same feed affected prior allowance or capital calculations, the bank scopes those uses separately. Preserve the original and corrected results for audit. Retraining on flawed targets would have embedded a data defect in model parameters.

A similar problem can arise when a property valuation is joined using today's customer map rather than the default-time collateral link. A reviewer samples facility-level source evidence and checks priority and effective dates. A reverse lineage query finds other estimates using the bad join. The response includes affected reports and decisions, not just a repaired ETL job. A robust LGD process can explain where the recovery came from and which point-in-time information the model used.

Recovery timing as a model target

Two defaults can produce the same eventual undiscounted recovery but very different economic results. One may pay within months, while the other takes years of legal and servicing work. A model could predict recovery cash flows by period rather than one terminal percentage. That approach needs enough mature observations, a clear treatment of unresolved cases and a consistent discounting convention. A terminal LGD model can be simpler but may hide timing uncertainty. Choose the method for the approved decision and validate it against the cash-flow question the bank needs to answer.

For example, imagine two collateral sales that each yield 60 on an exposure of 100. The first settles after six months with modest costs; the second settles after four years with substantial legal expense. A simple gross-recovery percentage treats them alike. An economic loss method may not. If the bank changes the timing assumption in an allowance calculation, compare effects on a fixed portfolio and document the accounting rationale. A model should not infer timing from later case-status fields unavailable at the prediction date. Preserve cash-flow event timestamps and revisions so a validator can reconstruct outcomes.

Segmentation and sparse samples

LGD varies across collateral and product structures, but making too many small segments can produce unstable estimates. A bank may pool similar exposures or use a hierarchical approach under its approved methodology. Check that pooled loans genuinely share recovery drivers, and inspect important segments individually. A high average fit across unsecured consumer loans can conceal poor estimates for a small secured product. Sparse stressed defaults create additional uncertainty; the report should show counts, unresolved share and sensitivity rather than a spurious precise percentage.

For a new product with little recovery history, an external benchmark or expert adjustment may be considered under bank policy, but its comparability and limitations need scrutiny. Differences in legal jurisdiction, seniority, collection strategy and economic period can make a benchmark misleading. Record the rationale, approval and monitoring trigger. As actual outcomes mature, compare them with the estimate and revise the model under controlled change. Do not present borrowed parameters as if they were calibrated on the bank's own product.

Recoveries, write-offs and customer treatment

Accounting write-off, sale of a distressed loan and release of a borrower are distinct events. A sale price may provide an observed recovery under a defined methodology, while later recoveries belong to a purchaser. A write-off does not necessarily end customer obligations or collections activity. The LGD dataset should mark which legal and economic rights the bank held during the recovery window. The model team must avoid counting proceeds twice across a sale and a subsequent ledger entry. Finance, legal and servicing owners validate the mapping.

Loss modeling should not drive inappropriate pressure on customers. A predicted high LGD is a portfolio estimate, not a finding that a borrower will refuse to pay. Collections actions follow customer, hardship and conduct policies with human accountability. If a model helps prioritize outreach, validate whether the action is supportive and whether it creates unequal or harmful treatment. Record interventions because they affect later recoveries; otherwise a model may mistake the impact of an aggressive strategy for a natural borrower characteristic.

Stressing the integration

An LGD estimate can feed a capital or impairment calculation with PD and EAD, but the integration should be tested end to end. Select a sample exposure and pin its default definition, outstanding balance, collateral link, LGD model version and downstream method. Introduce a stale collateral value and verify that a missingness or fallback state reaches the calculation rather than being converted silently to zero loss. Introduce a currency mismatch and check conversion timing. Reconcile the aggregate output to the source portfolio. An individual model validation does not prove that the combined reporting engine applies the right parameter to the right facility.

Release acceptance

Prepare examples for an unsecured resolved loan, a secured loan with delayed sale, a revolving facility, an unresolved workout, shared collateral, a guarantee, a reversed receipt and a missing valuation. Write expected target and input state under the approved methodology. Verify that training excludes later recovery outcomes from earlier features, that prediction services reject unsupported products, and that reporting calculations reconcile to dated exposure totals. Test source outage and fallback. An LGD estimate becomes useful when its event, cash flows, costs, timing, population and governed use are all visible to a reviewer.

The reviewer should also inspect one case whose workout spans a model change. The original estimate belongs to its dated model and information set, while updated collateral and recovery evidence may justify a later estimate. Keep both rather than replacing the earlier value. Compare the model's predicted cash-flow path with the actual receipts and costs as they mature, then determine whether a difference reflects a model limitation, a policy intervention or a source error. This case gives validation a concrete way to challenge LGD without assuming the final outcome was known on day one.

A second review should trace a guarantee payment to the correct facility and date. If the guarantee is disputed or payment is pending, the feature must not present the prospective amount as cash already recovered. Document the legal status, expected timing, source evidence and model assumption. This example tests whether the training target and production estimate apply the same recovery definition.

Banking practice note on data definition

For loss given default, data definition matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

The practical discipline is to keep source fact, credit-risk interpretation, model estimate and business action separate. Source facts come from areas such as collateral value, loan-to-value ratio, seniority, guarantees, and recovery history. The credit-risk interpretation explains what those facts mean. The model estimate turns that interpretation into a score, probability, severity, exposure or ranking. The business action decides approval, referral, monitoring, support, pricing, limit or reporting treatment.

This separation helps the bank challenge the result properly. When a score changes, the bank can ask whether borrower behaviour changed, source data changed, policy changed, model calibration changed or economic conditions changed. That is the kind of clarity serious banking AI needs.

Banking practice note on model use

For loss given default, model use matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on customer impact

For loss given default, customer impact matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on portfolio control

For loss given default, portfolio control matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on validation

For loss given default, validation matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on monitoring

For loss given default, monitoring matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on governance

For loss given default, governance matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on audit evidence

For loss given default, audit evidence matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on economic conditions

For loss given default, economic conditions matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Banking practice note on human judgement

For loss given default, human judgement matters because credit risk is not only a model output. It is a banking responsibility that connects borrower behaviour, product policy, capital, provisions, customer treatment, collections, finance reporting, risk appetite and audit evidence. If one part is weak, the model result can look precise while the decision remains fragile.

Loss after a defined default

Loss given default estimates the share of exposure not recovered after a borrower meets the default definition, under a specified horizon and measurement convention. An illustrative exposure of 10,000 with 6,000 net recovered suggests a 40% realized loss share, but timing, collection cost, collateral sale proceeds and discounted cash flows can change the measure. Specify whether the example is a simple teaching ratio or an approved accounting or regulatory calculation.

Take two defaulted facilities with the same balance. One has collateral whose sale may take a year; another has no collateral but early partial repayment. Record default date, exposure, recovery events, costs and observation maturity. A case still in collection cannot be treated as final zero recovery. Assess model calibration on appropriately mature default cohorts, with sensitivity to changing collateral values and collection practices. A recovery process change can move outcomes without a change in borrower creditworthiness.

LGD belongs to a defined exposure and population. A bank should not apply a secured-mortgage recovery assumption to an unsecured card portfolio merely because both have a "loss" field. Validate joins between facility, collateral and legal entity and test cases with shared or released collateral. Preserve source evidence and expert overrides so a reviewer can trace how an estimated recovery became a portfolio measure.

Primary sources for further study

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