Estimating the portion of exposure that will not be recovered after default, and translating post-default uncertainty into a disciplined Expected Credit Loss framework.
If Probability of Default tells us how likely a borrower is to fail, Loss Given Default tells us how much of the exposure is actually expected to be lost if that failure occurs. In many ways, LGD is where the abstract idea of credit deterioration becomes financially concrete. A borrower may default, but the economic consequence of that default depends on what happens afterward. Will the institution recover most of the balance through collateral, restructuring, guarantors or cash collections? Will recovery be delayed, costly and partial? Will the asset cure, settle or be written off? Will security that appears strong on paper prove weaker in enforcement? These are not secondary details. They are central to the measurement of loss.
Loss Given Default (LGD) Modelling is the discipline of estimating how much of an exposure will remain unrecovered if default occurs. In ECL, LGD must reflect collateral quality, enforceability, cure behaviour, restructuring outcomes, recovery timing, enforcement costs and forward-looking conditions. A strong LGD framework measures real recovery economics, not just headline security values.
If Probability of Default tells us how likely a borrower is to fail, Loss Given Default tells us how much of the exposure is actually expected to be lost if that failure occurs. In many ways, LGD is where the abstract idea of credit deterioration becomes financially concrete. A borrower may default, but the economic consequence of that default depends on what happens afterward. Will the institution recover most of the balance through collateral, restructuring, guarantors or cash collections? Will recovery be delayed, costly and partial? Will the asset cure, settle or be written off? Will security that appears strong on paper prove weaker in enforcement? These are not secondary details. They are central to the measurement of loss.
LGD modelling therefore occupies a critical place in the ECL framework. It is the component that forces the institution to confront the real economics of post-default outcomes. It asks not merely whether a default event occurs, but what that event ultimately means in cash terms. This is why institutions with superficially similar PD profiles can have dramatically different ECL outcomes. The probability of failure may be comparable, yet the severity of loss after failure may vary widely depending on collateral coverage, legal enforceability, recovery discipline, borrower type, product structure, cure behaviour and time to resolution.
A weak LGD framework can materially distort the allowance. If recoveries are assumed too optimistically, the ECL number becomes understated and management may gain false comfort from security values that are not realistically realisable. If recoveries are assumed too harshly, the framework may overstate loss severity and fail to distinguish between economically protected and unprotected exposures. If timing is ignored, present value loss can be misstated even when headline recoveries are eventually accurate. If collateral is treated as though it were cash, rather than as an uncertain path to potential cash, the model becomes structurally misleading.
This is why LGD modelling must be approached with both analytical rigor and practical realism. It is not simply an exercise in assigning a haircut. It is a structured assessment of recovery economics after default.
This article explores LGD modelling in depth: what it means, how it is constructed, how collateral and recoveries should be treated, why timing matters, how cure and restructuring affect loss severity, and what failures institutions most commonly make when translating post-default behaviour into an ECL estimate.
Loss Given Default is the proportion of exposure that is expected to be lost if default occurs, after taking account of recoveries, collateral realisation, guarantees, settlements, cures and any other cash flows that reduce the ultimate loss.
That definition is important because LGD is often misunderstood as a static unsecured percentage. In reality, LGD is a recovery-based concept. It is not primarily about how bad a default looks at the moment it occurs. It is about what remains unrecovered after the post-default process unfolds.
This means LGD is fundamentally connected to the institution's recovery environment. A default in one portfolio may result in high eventual recovery because security is strong, enforcement is efficient and workout discipline is mature. A similar default in another portfolio may result in severe loss because recovery is slow, collateral is weak, legal action is costly or borrower assets have little realisable value.
LGD therefore is not merely a property of the borrower. It is a property of the interaction between exposure structure, protection features, recovery process and institutional capability.
A common misconception is that LGD can be derived by looking at collateral coverage and subtracting it from exposure. This is too simplistic and often dangerous.
Collateral is not the same as recovery. A property valuation is not the same as cash proceeds. A guarantee document is not the same as enforceable support. An inventory charge is not the same as timely liquidation. Security can reduce loss, but only if it is legally enforceable, operationally accessible, economically realisable and appropriately timed.
A mature LGD framework therefore asks a deeper question: after default occurs, what cash flows are actually expected to be collected, when are they expected to arrive, what costs will be incurred in obtaining them, and how certain are those expectations?
This is why LGD modelling is not a simple haircut exercise. It is a recovery economics exercise.
A useful way to understand LGD is to think of it as being shaped by several interrelated elements:
These elements matter because the same nominal recovery rate can imply different LGD outcomes depending on timing and cost. A recovery of 70 percent received quickly is not economically identical to a recovery of 70 percent received after several years of litigation and enforcement expense. Likewise, a portfolio that frequently cures after early default may exhibit different effective loss severity from one where distress becomes terminal.
A professional LGD framework therefore measures not only how much is recovered, but how the recovery path unfolds.
One of the most underestimated aspects of LGD modelling is time. Institutions often focus on eventual recovery percentages and give insufficient attention to the timing of those recoveries. Yet Expected Credit Loss is a present value concept. A delayed recovery is worth less than an immediate one, even if the nominal amount is the same.
Timing matters for several reasons.
First, discounting reduces the present value of later cash flows.
Second, long recovery periods often imply greater uncertainty and greater operational burden.
Third, delay may signal deeper workout difficulty, which itself can correlate with lower ultimate recovery.
Fourth, legal and enforcement costs tend to accumulate with time.
A robust LGD framework therefore includes recovery timing explicitly. It does not stop at asking what proportion will be recovered. It also asks when and through what path the recovery will arrive.
This point is especially important in secured lending, where institutions sometimes become overconfident because nominal collateral values appear high, even though actual liquidation may take years.
The distinction between secured and unsecured exposures is one of the most visible drivers of LGD, but it should be handled carefully.
Secured exposures often have lower LGD than unsecured ones because collateral provides a potential source of recovery. But "secured" is not a uniform condition. The quality of security varies enormously. A first-charge mortgage over a liquid, well-documented asset is very different from a second-charge claim over specialised equipment in a weak market. The same legal term can conceal very different recovery realities.
A mature LGD framework therefore avoids treating secured and unsecured as a simple binary. It examines:
For unsecured exposures, the framework must place greater emphasis on expected borrower cash flow recovery, guarantor support where applicable, settlement patterns and historical workout experience.
In both cases, the true question is not whether the exposure is labelled secured, but how much protection is economically real.
A recurring weakness in credit frameworks is the tendency to treat collateral as a source of reassurance rather than as a recovery mechanism that must itself be analysed critically.
This distinction matters. Collateral often creates psychological comfort in origination discussions, but LGD modelling requires a more exacting perspective. The institution must ask:
Only after these questions are considered can collateral be translated into expected recovery. A valuation report is not an LGD model. It is one input into a recovery estimate.
If default data supports PD modelling, recovery data supports LGD modelling. A strong LGD framework is usually built, where possible, on observed post-default outcomes: collections, settlements, collateral realisations, guarantor proceeds, cure events, write-offs and the timing of those events.
This is important because recovery behaviour is often more idiosyncratic and operationally driven than default behaviour. Two institutions holding similar portfolios may experience different LGD outcomes because their workout effectiveness, legal processes, collateral tracking discipline or settlement practices differ materially.
For this reason, empirical LGD should ideally be informed by the institution's own recovery history where sufficient and reliable data exists. External benchmarks may sometimes supplement, especially in sparse-data portfolios, but internal post-default experience usually carries the strongest relevance.
The challenge, of course, is that recovery data is often fragmented across systems, legal files and manual workout records. This is why LGD modelling so often exposes weaknesses in data architecture. It forces the institution to ask whether it truly knows what happened after default.
A useful conceptual distinction in LGD modelling is between workout-oriented views and market-oriented views.
A workout LGD perspective is based on actual recovery experience through the institution's own post-default processes. It reflects how defaults are resolved in practice: collections, restructurings, collateral sales, settlements and write-offs.
A market-oriented LGD perspective may use externally observed market prices or stressed valuation logic to infer loss severity, more relevant in certain traded or mark-to-market contexts.
In many ECL applications, workout LGD is particularly important because the allowance concerns expected cash shortfalls in the institution's actual holding and recovery context. The institution is usually not measuring what a distressed asset might sell for immediately in a market transaction, but what loss is expected given its own recovery path.
This does not eliminate the relevance of market signals, especially where asset disposals are common or collateral values are market-sensitive. But it does reinforce that LGD should usually reflect the economic reality of the institution's expected workout process.
Not every default ends in terminal loss. Some exposures cure. Some borrowers resume payment after temporary disruption. Some accounts regularise after early distress. Some restructurings produce meaningful recovery of value.
This has an important effect on LGD. Where cure is common and sustainably observed, the effective severity of default loss may be lower than in portfolios where distress almost always ends in write-off or deep impairment.
A mature LGD framework therefore considers whether cure is part of the observed post-default path and how it should be incorporated. This does not mean optimistic assumptions should be made casually. Cure should be based on evidence, not hope. But where historical experience shows real cure behaviour, ignoring it can overstate severity.
At the same time, institutions must distinguish between genuine cure and temporary normalization. If defaults are "cured" only to re-default shortly thereafter, the apparent benefit may be illusory. This is why cure logic and LGD logic need to be aligned with the institution's broader default and recovery definitions.
LGD modelling becomes particularly nuanced where restructurings or concessions occur after distress.
A restructured exposure may recover more value than an immediately written-off one, but that does not automatically mean loss severity is low. The modified cash flows may be slower, more uncertain or reliant on assumptions about borrower recovery that must be tested carefully. Some restructurings merely delay recognition of eventual loss. Others genuinely preserve economic value.
This means LGD frameworks should not treat all restructurings alike. They should examine:
A strong LGD framework understands that post-default recovery can happen through more than one route, but each route must be measured in present value and probability-weighted terms rather than accepted at face value.
LGD is not always stable across economic conditions. In many portfolios, recoveries worsen during downturns. Collateral values fall. Liquidation periods lengthen. Buyers become scarce. Legal processes slow. Guarantors weaken. Distressed asset sales realise lower proceeds. Borrowers that might have been rescued in benign conditions fail under system-wide stress.
This matters enormously for ECL because forward-looking estimation should not assume that recovery experience observed in benign periods will necessarily hold during adverse macroeconomic conditions.
A mature institution therefore asks whether LGD should vary with scenario conditions, either explicitly through scenario-based recovery assumptions or implicitly through calibrated conservatism. This is especially relevant in secured portfolios linked to property, commodities, business assets or cyclical sectors.
An LGD framework that treats collateral recovery as constant across the cycle may understate loss precisely when forward-looking prudence matters most.
Just as with PD, segmentation is critical in LGD modelling. A single severity rate applied across a broad and diverse population can create substantial distortion.
Relevant segmentation dimensions may include:
The purpose of segmentation is not to produce unnecessary complexity. It is to ensure that recoveries are estimated on economically coherent populations. Residential mortgages, unsecured consumer loans, invoice financing, machinery-backed loans and project exposures do not recover in the same way. Treating them as though they do creates an LGD framework that is numerically neat but economically weak.
Although LGD is a distinct component, it should not be modelled in total isolation from exposure behaviour. In some portfolios, severity may depend partly on the size or nature of exposure at default.
For example:
This does not mean LGD and EAD should always be jointly modelled in a highly technical sense. But it does mean the institution should remain alert to the fact that loss severity is sometimes influenced by the state of exposure when default occurs.
One of the most practical aspects of LGD modelling is the application of haircuts and costs to expected recovery sources. This is particularly important for collateral-based recoveries.
A raw asset valuation rarely represents the cash that will actually be realised after default. The institution must consider:
These factors can materially reduce effective recovery even when headline collateral values appear strong. A mature LGD framework therefore incorporates them transparently rather than assuming that recorded collateral value will convert cleanly into recovery proceeds.
This is one of the places where practical credit realism matters most. The model should reflect what recovery teams actually experience, not what origination teams hoped would happen.
Because ECL is a present value measure, recoveries expected in the future should be discounted appropriately. This point is often acknowledged in theory but applied unevenly in practice.
Discounting matters because:
A strong LGD model therefore captures both amount and timing. Institutions that model only ultimate recovery percentages may underestimate loss where recoveries are slow or uncertain. This is especially relevant in portfolios with lengthy legal enforcement, distressed real estate processes, insolvency-driven recovery or complex restructuring outcomes.
Where recovery data is rich, empirical LGD estimation can be powerful. But many institutions face data limitations, low-default portfolios or structural changes that make purely empirical estimation difficult. In such cases, expert-adjusted LGD may be necessary.
This can be appropriate, provided the process is disciplined. The institution should identify:
Expert judgement should not be treated as a weakness in itself. In ECL, some judgement is inevitable. The weakness arises only when judgement is undocumented, inconsistent or used to conceal deficiencies in basic model design.
Some portfolios generate few defaults, and among those defaults, few complete recovery histories. These conditions make LGD modelling especially difficult. Large corporate and project exposures often fall into this category.
Here, institutions may need to combine limited internal history with workout case analysis, collateral characteristics, external evidence and carefully governed judgement. The goal is not to fabricate spurious precision, but to build a severity framework that is as evidence-based as possible given the information available.
Important disciplines in such settings include:
Sparse-data environments demand humility. The institution should know where its LGD certainty ends and judgement begins.
Forward-looking adjustment in LGD is often less discussed than forward-looking adjustment in PD, but it can be equally important.
Macroeconomic conditions may affect:
This means LGD may need scenario sensitivity, especially in portfolios where recoveries are closely tied to asset prices, market liquidity or cyclical sector conditions. For example, a property-backed loan book may experience materially different LGDs under benign and downturn property markets. Similarly, equipment-backed exposures may recover less in a weak industrial environment than in a strong one.
A mature ECL framework therefore asks whether expected severity is stable across scenarios or should vary. If it should vary, that variation must be governed and supported rather than applied informally.
LGD frameworks should be validated as rigorously as PD models, though the nature of validation is different.
Key validation questions may include:
Validation can involve recovery backtesting, workout analysis, cohort review, challenge of collateral assumptions and comparison of model outputs against realised resolution patterns. The objective is not to eliminate uncertainty, because post-default resolution is inherently uncertain. It is to ensure that the model is disciplined, evidence-based and improving over time.
Several implementation failures recur repeatedly.
One is treating collateral value as equivalent to recovery, without reflecting enforceability, timing, costs and forced-sale conditions.
Another is ignoring recovery timing, thereby overstating present value recovery.
A third is using stale or weak collateral data, especially in secured portfolios where valuation freshness matters.
A fourth is underestimating legal and workout costs, which can materially erode net recoveries.
A fifth is failing to distinguish genuine cure from temporary regularisation, leading to over-optimistic severity assumptions.
A sixth is using a single LGD across economically different segments, thereby flattening meaningful differences in security quality, jurisdiction and recovery behaviour.
A seventh is neglecting downturn sensitivity, especially where recoveries depend heavily on asset prices or market liquidity.
These failures are serious because LGD often appears less visible than PD in management discussion, yet it can be equally powerful in driving the allowance. Indeed, in secured portfolios, LGD may be the dominant variable.
Consider two loans, each with an exposure of 100 and recorded collateral value of 120.
At first glance, both appear over-secured and might be assigned low LGD.
But in the first case, the collateral is a residential property with current valuation, clear first-charge status, strong local market liquidity and relatively efficient enforcement. Recovery may indeed be high.
In the second case, the collateral is specialised industrial equipment located in a weak market, subject to uncertain title documentation, lengthy repossession procedure and material disposal costs. Although the nominal collateral value is also 120, expected realised recovery may be far lower and much slower.
A superficial LGD approach may treat the two loans similarly. A mature LGD framework would not. It would recognise that collateral value on paper is only the beginning of the recovery analysis.
A strong institutional LGD framework usually includes the following elements:
The real strength of this structure lies in integration. LGD should not sit as a detached percentage table. It should be part of a wider ECL architecture connecting data, default logic, workout processes, collateral management and forward-looking assessment.
Loss Given Default modelling is one of the most economically revealing parts of the Expected Credit Loss framework. It takes the abstract event of default and asks the practical question that matters most in financial terms: what will actually be lost after the recovery process unfolds? It forces the institution to confront the quality of its protections, the realism of its collateral assumptions, the discipline of its workout process, the time value of delayed recoveries and the uncertainty inherent in post-default resolution.
A strong LGD framework is not satisfied with nominal security comfort or broad historical averages. It looks deeper. It asks how recoveries are achieved, how long they take, what they cost and how they behave under stress. It distinguishes between documentation and realisable protection, between apparent cure and genuine recovery, between headline value and present value.
In that sense, LGD modelling does more than measure severity. It tests whether the institution truly understands what happens after credit failure begins.
