Every quantitative statement about credit risk rests on three parameters. Probability of default (PD) is the likelihood that a borrower fails to meet its obligations within a period, usually one year. Loss given default (LGD) is the share of the exposure the lender loses if default occurs, after collateral, guarantees, recoveries and the costs and delay of collecting them. Exposure at default (EAD) is the amount the lender expects to be owed at the moment of default, including the part of any undrawn commitment the borrower draws before failing. Their product, PD times LGD times EAD, is expected loss, the average loss the lender should plan for and price in; the variability around it is unexpected loss, which capital exists to absorb. The same three numbers, estimated under different conventions, drive regulatory capital under the Basel internal ratings-based approach, the loan loss allowance under IFRS 9 and CECL, stress test projections, risk-based pricing and portfolio limits, which is why an auditor who understands them can read almost any credit risk report.
This guide defines each parameter and the default definition they depend on, distinguishes the estimation conventions (through-the-cycle versus point-in-time, twelve-month versus lifetime, downturn versus expected), works the expected loss arithmetic on a portfolio, explains how Basel, IFRS 9 and CECL use and constrain the parameters, describes how they are estimated and validated, gives internal audit’s program, and closes with a worked example from Lakeshore Bancorp, the $9 billion regional bank used in examples across this site. The credit risk guide covers the wider subject, the model risk management guide the framework the models live in, and the credit risk for internal auditors guide the practitioner’s introduction.
In this guide
- The three parameters and the default definition
- Estimation conventions: TTC and PIT, twelve-month and lifetime, downturn
- The expected loss arithmetic on a portfolio
- How Basel uses the parameters: the IRB approach and its floors
- How IFRS 9 and CECL use them: the allowance
- Stress testing, pricing and limits
- How the parameters are estimated
- How they are validated
- Internal audit’s program
- A worked example: Lakeshore Bancorp’s CECL allowance audit
- Questions about PD, LGD and EAD
- Related guides
The three parameters and the default definition
All three parameters depend on what counts as a default, and the definition is a choice with consequences. The Basel framework defines default as either of two events: the bank considers the obligor unlikely to pay its credit obligations in full without recourse to actions such as realizing security, or the obligor is more than 90 days past due on any material credit obligation (national supervisors may use 180 days for some retail and public-sector exposures). Indicators of unlikeliness to pay include non-accrual status, a specific provision or charge-off, a distressed restructuring, bankruptcy or similar protection, and the sale of the obligation at a material credit-related loss. IFRS 9 presumes default no later than 90 days past due unless rebutted and otherwise lets the entity use its internal credit risk management definition; CECL does not define default at all, since it measures lifetime loss rather than default events, but PD/LGD methods under CECL must define it to work. A looser definition (180 days) lowers PD and raises LGD (because borrowers who cure between 90 and 180 days are excluded from defaults, and the ones left are worse); a tighter one does the reverse. The product is less sensitive than the parts, which is why validation tests expected loss as well as PD and LGD separately.
| Parameter | Definition | Units | Main drivers | Typical range |
|---|---|---|---|---|
| PD | Probability that the obligor defaults within the horizon (usually twelve months) | Percent per period | Financial strength, leverage, cash flow coverage, industry, management, behavior (payment history, utilization), macroeconomic conditions | 0.03 percent for the strongest corporates to 20 percent or more for distressed borrowers; retail unsecured 1 to 8 percent; mortgages 0.2 to 2 percent in normal conditions |
| LGD | Share of EAD lost if default occurs, after recoveries net of costs, discounted to the default date | Percent of EAD | Collateral type, value and coverage; seniority; guarantees; jurisdiction and legal process; cure rates; time to recovery; economic conditions at default | 10 to 25 percent for well-secured real estate lending; 35 to 45 percent for senior unsecured corporate debt; 60 to 90 percent for unsecured retail and subordinated debt |
| EAD | Amount owed at default: drawn balance plus the expected draw on undrawn commitments (the credit conversion factor), plus accrued interest and fees | Currency | Current balance, facility limit, utilization behavior before default, amortization, product type | CCFs from 10 percent for uncommitted facilities to 100 percent for irrevocable commitments; borrowers typically draw more as they approach default |
| Expected loss (EL) | PD times LGD times EAD | Currency (or percent of exposure as the EL rate) | All of the above | A few basis points to several percent of exposure |
| Unexpected loss | Variability of loss around EL at a confidence level; what capital covers | Currency | PD and LGD volatility, correlation between obligors, concentration | A multiple of EL that rises with correlation |
Estimation conventions: TTC and PIT, twelve-month and lifetime, downturn
The same borrower can carry several PDs at once, each correct for its purpose. A through-the-cycle (TTC) PD estimates the long-run average default rate of borrowers with the obligor’s characteristics across an economic cycle; it is stable, moves little with conditions and is what Basel requires for regulatory capital (the long-run average of one-year default rates for each grade), so that capital does not swing with the economy. A point-in-time (PIT) PD estimates the probability of default over the coming period given current and forecast conditions; it rises in a downturn and falls in a boom, and it is what IFRS 9 and CECL require because an allowance should reflect expected losses now, not on average. A twelve-month PD covers the next year; a lifetime PD covers the remaining life of the instrument, built from a term structure of marginal PDs by year (conditional on survival to that year) and used for IFRS 9 stage 2 and 3 assets and for all CECL assets. Downturn LGD is the LGD expected in adverse conditions, which Basel requires for capital because collateral values and recovery rates fall exactly when defaults rise (the dependence of LGD on the state of the economy is why expected loss in a downturn exceeds average PD times average LGD); accounting frameworks use an LGD consistent with the forecast scenario instead. The model risk management guide’s test of assumption governance applies to every one of these choices, because a model switched from TTC to PIT, or a lifetime PD built from an optimistic macroeconomic forecast, moves the allowance without any change in the loans.
| Use | PD | LGD | EAD | Horizon |
|---|---|---|---|---|
| Basel IRB regulatory capital | Through-the-cycle; long-run average by grade; floors apply | Downturn; floors apply under A-IRB; supervisory values under F-IRB | Drawn plus CCF on undrawn; supervisory CCFs under F-IRB; floors under A-IRB | One year (maturity adjustment in the formula) |
| IFRS 9 allowance | Point-in-time; twelve-month for stage 1, lifetime for stages 2 and 3; forward-looking, probability-weighted scenarios | Scenario-consistent expected LGD | Expected exposure over the horizon, including expected draws and prepayments | Twelve months or remaining life |
| CECL allowance | Point-in-time lifetime (where a PD/LGD method is used; other methods estimate loss rates directly), with a reasonable and supportable forecast then reversion to history | Lifetime expected | Amortized cost over the contractual life, adjusted for prepayments; unconditionally cancellable commitments excluded | Remaining contractual life |
| Stress testing (CCAR, DFAST, ICAAP) | Point-in-time conditional on the scenario path, by quarter | Scenario-conditional | Projected balances under the scenario | Nine quarters (US) or the scenario horizon |
| Pricing and RAROC | Point-in-time or TTC by policy | Expected | Expected | Life of the facility |
| Limits and portfolio management | Current grade PD | Expected | Current and committed | Ongoing |
The expected loss arithmetic on a portfolio
The arithmetic is simple and the sensitivities are not. Take a $100 million commercial portfolio in five segments. Expected loss per segment is PD times LGD times EAD; the portfolio EL is the sum; the EL rate is EL over EAD. The table works it, then shows what happens when conditions worsen: PDs rise by half and LGDs by ten points, which is a mild downturn by historical standards, and EL more than doubles because the two parameters move together.
| Segment | EAD | PD (normal) | LGD (normal) | EL (normal) | PD (downturn) | LGD (downturn) | EL (downturn) |
|---|---|---|---|---|---|---|---|
| Owner-occupied CRE, well secured | $30.0m | 1.0 percent | 20 percent | $60,000 | 1.5 percent | 30 percent | $135,000 |
| Investor CRE | $25.0m | 1.6 percent | 30 percent | $120,000 | 2.4 percent | 40 percent | $240,000 |
| Commercial and industrial, secured by receivables and inventory | $25.0m | 2.0 percent | 40 percent | $200,000 | 3.0 percent | 50 percent | $375,000 |
| Commercial and industrial, unsecured | $12.0m | 2.5 percent | 55 percent | $165,000 | 3.75 percent | 65 percent | $292,500 |
| Equipment finance | $8.0m | 3.0 percent | 35 percent | $84,000 | 4.5 percent | 45 percent | $162,000 |
| Portfolio | $100.0m | 1.74 percent (EAD-weighted) | 36.1 percent (default-weighted) | $629,000 (0.63 percent) | 2.61 percent | 46.1 percent | $1,204,500 (1.20 percent) |
Three observations an auditor should carry from the table. The portfolio EL rate (0.63 percent) is higher than the EAD-weighted PD times the EAD-weighted LGD (1.74 percent times 32.9 percent is 0.57 percent), because the segments with the higher PDs also carry the higher LGDs; averages taken separately understate the loss, so a report that quotes portfolio-level parameters should say how they were weighted (the LGD shown in the table is weighted by expected defaults, which is why it reproduces the EL). A 50 percent rise in PD and a ten-point rise in LGD nearly doubles EL (0.63 to 1.20 percent), which is the correlation between the two parameters that downturn LGD exists to capture. And the EAD column is the one most often taken for granted: if the C&I borrowers draw their undrawn lines before defaulting, as they do, the EAD and the EL rise again without any change in PD or LGD.
How Basel uses the parameters: the IRB approach and its floors
The Basel framework offers banks two routes to credit risk capital. Under the standardised approach, risk weights are prescribed by exposure class and, where available, external ratings or loan-to-value ratios; the bank estimates nothing. Under the internal ratings-based (IRB) approach, introduced in Basel II (2004), the bank estimates the parameters and feeds them into a supervisory formula that converts them to a capital requirement for unexpected loss at a 99.9 percent confidence level over one year, with an asset correlation that depends on PD and exposure class and a maturity adjustment. The foundation IRB (F-IRB) approach lets the bank estimate PD only, with supervisory LGDs (45 percent senior unsecured, 75 percent subordinated, lower for recognized collateral) and supervisory CCFs; the advanced IRB (A-IRB) approach lets it estimate LGD, EAD and effective maturity as well. The 2008 crisis showed that internal estimates varied widely across banks for the same risk, and the finalised Basel III package (December 2017, implementation from 2023 with the output floor phasing in to 2028) constrained the approach: A-IRB was removed for exposures to banks and other financial institutions and to large corporates (consolidated revenues above €500 million), which may use F-IRB or the standardised approach; all equity exposures moved to the standardised approach; input floors were set on PD, LGD and EAD; and an output floor requires total risk-weighted assets to be at least 72.5 percent of what the standardised approaches would produce, phasing in from 50 percent. The table gives the input floors.
| Parameter | Exposure | Floor under finalised Basel III |
|---|---|---|
| PD | Corporate; retail mortgages; other retail; qualifying revolving retail (transactors) | 0.05 percent (5 basis points) |
| PD | Qualifying revolving retail (revolvers) | 0.10 percent |
| LGD (A-IRB) | Corporate, unsecured | 25 percent |
| LGD (A-IRB) | Corporate, secured: financial collateral 0 percent; receivables 10 percent; commercial or residential real estate 10 percent; other physical collateral 15 percent (blended for partially secured exposures) | As listed |
| LGD (A-IRB) | Retail mortgages | 5 percent |
| LGD (A-IRB) | Qualifying revolving retail | 50 percent |
| LGD (A-IRB) | Other retail, unsecured 30 percent; secured, by collateral type, 0 to 15 percent | As listed |
| EAD (A-IRB) | All | At least the on-balance-sheet exposure plus 50 percent of the off-balance-sheet exposure using the standardised credit conversion factors |
| Output floor | Total RWA | 72.5 percent of standardised RWA when fully phased in |
In the United States, the IRB approach applies only to the largest banks (the advanced approaches firms), and the agencies’ 2023 proposal to implement the Basel III endgame would have replaced internal credit risk models with an expanded standardised approach; its fate has been contested, and a bank’s current capital rules should be checked rather than assumed. Regional banks such as Lakeshore compute regulatory capital under the standardised approach and use PD, LGD and EAD only for the allowance, stress testing and management, which is the more common situation and the one the rest of this guide assumes. The internal audit in banking guide covers the supervisory frame.
How IFRS 9 and CECL use them: the allowance
The 2008 crisis also discredited the incurred loss model of provisioning, under which a loss was recognized only when a triggering event had occurred, so that allowances were smallest just before losses were largest. Both successors are expected loss models that use the parameters, differently. IFRS 9 (effective 1 January 2018) uses three stages: stage 1 assets, whose credit risk has not increased significantly since origination, carry a twelve-month expected credit loss (twelve-month PD times LGD times EAD); stage 2 assets, whose credit risk has increased significantly (the SICR assessment, with a rebuttable presumption at 30 days past due), carry a lifetime expected credit loss; stage 3 assets are credit-impaired (defaulted) and carry a lifetime loss with interest recognized on the net amount. Estimates must be probability-weighted across forward-looking macroeconomic scenarios and unbiased. CECL (ASC 326, from ASU 2016-13; effective 2020 for larger SEC filers and 2023 for all others) requires a lifetime expected credit loss on every financial asset at amortized cost from the day it is recognized, with no staging: the allowance reflects losses expected over the contractual life, using historical loss experience adjusted for current conditions and a reasonable and supportable forecast, after which the estimate reverts to historical averages. CECL does not prescribe a method; banks use PD/LGD approaches, discounted cash flow, loss-rate, vintage and weighted average remaining maturity (WARM) methods, and layer qualitative factors on top. The parameters therefore appear in two forms under CECL: explicitly, in PD/LGD and DCF methods, and implicitly, in loss rates that are PD times LGD in disguise. The accounts receivable guide covers CECL’s application to trade receivables at corporates, where the same logic applies to a simpler book.
| Feature | IFRS 9 | CECL |
|---|---|---|
| Effective | 1 January 2018 | 2020 (larger SEC filers); 2023 (others) |
| Horizon | Twelve months (stage 1); lifetime (stages 2 and 3) | Lifetime for all assets from recognition |
| Staging | Three stages driven by significant increase in credit risk and impairment | None |
| Forward-looking information | Required; multiple probability-weighted scenarios | Required for the reasonable and supportable period; reversion to history thereafter; single or multiple scenarios |
| Methods | Usually PD/LGD/EAD term structures | Any reasonable method: PD/LGD, DCF, loss rate, vintage, WARM; qualitative factors |
| Default definition | Entity’s definition; 90-day presumption | Not defined; needed for PD/LGD methods |
| Undrawn commitments | Included in EAD where the commitment is not unconditionally cancellable | Separate liability for off-balance-sheet exposures; unconditionally cancellable commitments excluded |
| Day-one effect | Twelve-month loss on origination | Full lifetime loss on origination (the “day-one loss”) |
| Typical audit issue | SICR thresholds set to keep assets in stage 1; scenario weights | Qualitative factor overlays without documented basis; forecast period and reversion choices |
Stress testing, pricing and limits
In stress testing, the parameters become functions of a scenario: PD models are conditioned on macroeconomic variables (unemployment, GDP growth, property prices, rates) so that a severely adverse path produces a quarter-by-quarter default rate; LGD models are conditioned on collateral price paths; EAD reflects projected balances and draws. The projected losses, net of the allowance already held, reduce projected capital, which is how the US supervisory stress tests (DFAST, and CCAR for the largest firms) and the ICAAP connect the parameters to capital planning. In pricing, expected loss is the credit cost that the spread must cover, and risk-adjusted return on capital (RAROC) divides the spread net of expected loss and costs by the capital the exposure consumes, so that a 1 percent PD loan priced at the same spread as a 0.2 percent PD loan is visibly a bad trade. In limits and portfolio management, grade migration (the movement of exposures across PD grades) is the earliest reliable signal of portfolio deterioration, and concentration limits are set on exposure weighted by PD and LGD rather than on gross balances alone. The risk appetite, tolerance and limits guide gives the structure those limits sit in, and the KRI program guide the indicators (migration, past-due trends, utilization) that lead the parameters.
How the parameters are estimated
PD estimation starts with a rating or scoring model that ranks obligors by risk: for corporates, a scorecard or statistical model combining financial ratios (leverage, coverage, liquidity, profitability), qualitative factors (industry, management, market position) and, increasingly, behavioral and external data; for retail, application and behavioral scorecards built by logistic regression or machine learning on the lender’s own default history. The ranking is then calibrated to default probabilities by grade using observed default rates over a long history (Basel requires at least five years of data for corporates and the long-run average for the TTC estimate), with a margin of conservatism for data limitations. External agency default studies, pooled data consortia and vendor models supplement thin internal histories, and the mapping from internal grades to external ratings is itself an assumption to validate. LGD estimation uses the lender’s workout history: for each resolved default, recoveries (collateral proceeds, guarantees, cash) less workout costs, discounted back to the default date, as a share of EAD; the estimate is segmented by collateral type, loan-to-value, seniority and product, cured defaults are handled explicitly (a cure is a zero-loss outcome that lowers average LGD but may be excluded from the downturn estimate), unresolved defaults need an assumption, and the discount rate and the treatment of incomplete workouts materially change the result. Most mid-size banks have too few defaults for a statistical LGD and rely on collateral haircut approaches (expected recovery equals collateral value times a haircut less costs) benchmarked against industry studies. EAD estimation for revolving products measures the share of the undrawn commitment that defaulted borrowers had drawn in the year before default (the credit conversion factor), segmented by product and utilization; for amortizing loans EAD is the scheduled balance adjusted for prepayment; for CECL and IFRS 9 lifetime measures, EAD becomes a path over time. The vendor model validation guide applies to the many banks whose PD and allowance models are bought rather than built.
How they are validated
Validation asks three questions of each parameter model: does it rank risk correctly (discriminatory power), are its numbers right on average (calibration), and does it stay right (stability and ongoing monitoring). For PD, discriminatory power is measured by the Gini coefficient or the area under the ROC curve (an AUC of 0.5 is random; corporate rating models typically achieve 0.7 to 0.85), calibration by comparing predicted and observed default rates by grade with a statistical test of whether the differences could be chance (binomial or Hosmer-Lemeshow tests), and stability by population stability indices and migration analysis. For LGD, back-testing compares predicted and realized loss rates on resolved defaults, by segment, with attention to the downturn adjustment and to whether recent workouts contradict the collateral haircuts. For EAD, realized credit conversion factors on recent defaults are compared with the model’s. Beyond the statistics, validation reviews conceptual soundness (does the model’s logic make sense for the portfolio), data quality and representativeness (does the development sample resemble the current book), the default definition’s application, overrides (how often rating analysts override the model and in which direction, since systematic upgrades at origination are a known pattern), implementation (does the production system compute what the model documentation says) and the macroeconomic linkage used for stress and allowance purposes. The model risk management guide gives the framework-level tests and the model risk audit guide the model-level ones.
Internal audit’s program
Internal audit does not re-validate the models; it tests that the parameters are governed, that validation did its job and that the numbers reach the uses correctly. The program below fits an allowance (CECL or IFRS 9) audit or a credit risk model audit, and it assumes co-sourced quantitative support where the function lacks it, as the model risk management guide recommends.
| Area | Tests | Evidence | Finding that recurs |
|---|---|---|---|
| Default definition | Policy definition matches the regulatory and accounting requirements; applied consistently in the data (sample defaults and non-defaults near the boundary); cures and returns to performing handled per policy | Policy, default flag logic, sample of loan files | Definition in policy differs from the flag coded in the data warehouse |
| Data | Development and calibration data complete and representative; reconciliation of the modeled portfolio to the general ledger; collateral values current; default and recovery histories complete | Reconciliations, data lineage, collateral valuation dates | Loans missing from the model population; stale collateral values inflating recoveries |
| Segmentation and grades | Segments homogeneous and documented; grade definitions applied; override rates and direction monitored | Segmentation rationale, override log and analysis | Override rate above 20 percent, nearly all upgrades, unanalyzed |
| Parameter governance | PD, LGD and EAD assumptions inventoried with owners, sources, review dates and approval; changes in the period documented, independently reviewed and their effect quantified | Assumption inventory, change log, committee minutes | Material assumption changes approved by the model owner alone |
| Validation | Validation performed on schedule by an independent function; covers discrimination, calibration, stability, conceptual soundness, implementation and overrides; findings tracked and closed | Validation reports, issue log | Validation re-runs the developer’s tests; no back-test against realized losses |
| Forward-looking inputs | Macroeconomic scenarios sourced, approved and consistent across the allowance, stress testing and planning; scenario weights documented; forecast period and reversion method (CECL) justified | Scenario documentation, weights, sensitivity analysis | Allowance scenario more optimistic than the capital plan’s baseline |
| Qualitative factors (CECL) | Each factor tied to a documented risk not captured by the model, with a quantified basis and directional consistency; the overlay’s size relative to the modeled allowance trended | Q-factor documentation, allowance committee minutes | Overlay larger than the modeled result, with narrative support only |
| Use and reporting | Parameters flow correctly into the allowance, capital, stress test, pricing and limit reports; reconciliation of the allowance roll-forward; board reporting shows migration and coverage | Report reconciliations, board packs | Pricing tool uses parameters two versions behind the allowance model |
A worked example: Lakeshore Bancorp’s CECL allowance audit
Lakeshore Bancorp, the $9 billion publicly traded regional bank holding company used in examples across this site, has a commercial portfolio of about $4.1 billion, of which commercial real estate is the largest component, alongside consumer and mortgage lending; it computes regulatory capital under the standardised approach and uses PD, LGD and EAD in its CECL allowance (a top-tier vendor model, rated High in its 126-entity audit universe and audited annually), its capital stress test and its pricing tool. The FY26 allowance audit, 380 hours with 120 hours of co-sourced quantitative review, covered the model’s parameters, the qualitative overlay and the flow of the result into the financial statements, which the external auditor also tests as a critical audit matter.
| Test | Finding | Rating and action |
|---|---|---|
| Default definition applied in the data | Policy defined default at 90 days past due or non-accrual; the data warehouse flag used non-accrual only, so 23 loans past 90 days but still accruing (a practice permitted for well-secured loans in collection) were excluded from the default history; the effect on segment PDs was small (under 3 basis points) but the inconsistency was undocumented | Medium; flag aligned to policy; PD recalibration confirmed immaterial |
| CRE LGD collateral haircuts | The investor CRE LGD used a 25 percent haircut on appraised values, benchmarked to a 2019 industry study; the bank’s own six CRE workouts since 2023 showed realized recoveries 31 to 38 percent below appraisal, with office properties worst; the vendor model allowed segment-level override and none had been applied | High; haircut raised to 35 percent for office and 30 percent for other investor CRE pending more data; allowance effect about $4.2 million; disclosed to the audit committee and the external auditor |
| EAD for revolving C&I lines | Credit conversion factor of 50 percent on undrawn commitments; the bank’s eleven C&I defaults since 2022 showed average utilization at default of 78 percent against 54 percent twelve months before, implying a CCF near 52 percent; assumption supported | No finding |
| Macroeconomic scenario and reversion | The allowance used a single baseline forecast with a four-quarter reasonable and supportable period and straight-line reversion over four quarters; the capital stress test’s baseline differed from it in unemployment by 0.6 points; the difference was documented and explained by vintage, not approved | Low; scenario governance to require a single approved baseline across uses |
| Qualitative factors | Nine Q-factors added $11.3 million (24 percent) to a modeled allowance of $46.8 million; six were quantified against named data; three (economic uncertainty, concentration, model limitations) were set by judgment with narrative support, and the concentration factor duplicated risk already in the CRE segment PDs | Medium; the three factors re-based with quantified anchors; concentration factor reduced to avoid double-counting |
| Override analysis | Rating overrides on 14 percent of commercial grades at the last annual review, 80 percent upgrades by one notch, analyzed by credit review but not reported to the model risk committee | Low; override reporting added to the committee pack |
| Flow to the financial statements and board reporting | Allowance roll-forward reconciled; coverage ratio and grade migration reported quarterly to the board risk committee | No finding |
The report’s message to the audit committee was that the model was sound and the parameters that mattered most were the ones the bank’s own recent experience could inform: the CRE LGD finding came from six workouts anyone could have read, and the Q-factor finding from asking what each overlay measured. The chief audit executive noted that the vendor model’s validation report had passed the collateral haircut as “consistent with industry benchmarks” without comparing it with the bank’s realized recoveries, which became a point for the vendor model validation program, and that the output floor and IRB debates in the capital rules were irrelevant to a standardised-approach bank whose real exposure to the parameters ran through its allowance. The bank audit universe guide shows where the allowance, credit review and model risk entities sit, and the credit risk guide the wider program around them.
Questions about PD, LGD and EAD
What do PD, LGD and EAD stand for?
Probability of default (the likelihood a borrower defaults within a period, usually a year), loss given default (the share of the exposure lost if default occurs, after recoveries) and exposure at default (the amount owed at the moment of default, including expected draws on undrawn commitments). Their product is expected loss.
What is the difference between expected loss and unexpected loss?
Expected loss is the average loss (PD times LGD times EAD), covered by pricing and the allowance. Unexpected loss is the variability around that average at a chosen confidence level, driven by correlation and concentration, and it is what regulatory and economic capital exist to absorb.
What is the difference between through-the-cycle and point-in-time PD?
A through-the-cycle PD is the long-run average default rate for borrowers of a grade across an economic cycle, used for Basel regulatory capital so that capital is stable. A point-in-time PD reflects current and forecast conditions, rising in downturns, and is required for IFRS 9 and CECL allowances.
What are the Basel input floors for PD and LGD?
Under the finalised Basel III framework, PD is floored at 0.05 percent for most exposure classes (0.10 percent for revolving retail borrowers who carry balances), and advanced IRB LGD is floored at 25 percent for unsecured corporate exposures, 5 percent for retail mortgages, 50 percent for qualifying revolving retail and 30 percent for other unsecured retail, with lower floors for secured exposures by collateral type. An output floor of 72.5 percent of standardised risk-weighted assets applies when fully phased in.
How do IFRS 9 and CECL use PD, LGD and EAD?
IFRS 9 computes a twelve-month expected credit loss for stage 1 assets and a lifetime loss for stage 2 and 3 assets using point-in-time, forward-looking parameters. CECL requires a lifetime expected loss on every asset from recognition and permits any reasonable method, in which the parameters appear explicitly (PD/LGD and discounted cash flow methods) or implicitly (loss rates).
What should internal audit test in PD, LGD and EAD models?
The default definition as coded in the data, data completeness and collateral currency, segmentation and overrides, governance of assumption changes, the independence and substance of validation (including back-tests against realized losses), the macroeconomic scenarios and their consistency across uses, qualitative overlays, and the flow of the parameters into the allowance, capital, stress test and pricing reports.
Related guides
- Credit risk: a comprehensive guide — the wider subject.
- Credit risk for internal auditors — the practitioner’s introduction.
- Credit risk fundamentals — the overview.
- How to audit model risk management — the framework the parameter models live in.
- Model risk and algorithm audit — validating individual models.
- Vendor model validation — when the allowance model is bought.
- SR 11-7 and SR 26-2 explained — the model risk guidance.
- 10 things auditors must know about model risk — the vocabulary.
- How to audit accounts receivable — CECL on trade receivables.
- Risk appetite, tolerance, capacity and limits — concentration limits weighted by risk.
- Designing a KRI program — migration and past-due indicators.
- Internal audit in banking — the supervisory frame.
- A bank audit universe worked example — where the allowance and credit review sit.
- Value at risk — the market risk counterpart.
- The Risk Library — every risk type on this site.
- Auditing the Underwriting Process: A Credit Risk Program — policy versus practice, exceptions, authority, appraisals and covenants.
Leave a Reply