PD, LGD, and EAD are the three core credit risk parameters that determine how much a bank expects to lose on a loan or credit exposure. Probability of Default (PD) measures the likelihood a borrower will fail to repay, Loss Given Default (LGD) measures how much of the exposure the bank will lose if that happens, and Exposure at Default (EAD) measures how large the exposure is at the moment of default. Together, these three inputs feed directly into Expected Credit Loss calculations under IFRS 9 and Risk Weighted Asset calculations under Basel. The sections below break down each parameter individually and explain how they interact across both regulatory frameworks.
How do PD, LGD, and EAD work together in a credit risk model?
PD, LGD, and EAD work together by multiplying to produce Expected Credit Loss (ECL): ECL = PD x LGD x EAD. This formula is the foundation of every credit risk model. PD tells you the probability a borrower defaults, EAD tells you how much is at risk when they do, and LGD tells you what fraction of that exposure you will not recover. The product gives you the expected monetary loss on a credit position.
What makes this formula powerful is that each input is not a fixed number. All three parameters change over time, vary by borrower type, and respond to economic conditions. A mortgage borrower in a rising interest rate environment carries a different PD than the same borrower in a stable rate environment. The collateral backing that mortgage affects LGD. Whether the borrower has drawn down the full credit limit affects EAD. The model has to capture all three dimensions simultaneously.
The three parameters also feed into different regulatory outputs depending on the framework. Under IFRS 9, point-in-time estimates of PD, LGD, and EAD drive provisioning and Expected Credit Loss. Under Basel, through-the-cycle versions of the same parameters drive Risk Weighted Assets and regulatory capital. This means that the same underlying credit risk produces two different sets of calculations, both of which trace back to PD, LGD, and EAD as their starting point.
Because the outputs of one calculation become inputs for another, changes anywhere in the chain ripple through the entire model. A shift in macroeconomic forecasts affects PD estimates, which affects ECL, which affects provisions, which in turn influences the bank’s capital position. Understanding PD, LGD, and EAD individually is therefore only half the picture. Understanding how they connect is what makes a credit risk model genuinely useful.
What is probability of default and how is it calculated?
Probability of Default (PD) is the likelihood that a borrower will fail to meet their contractual debt obligations within a defined time horizon, typically one year. PD is expressed as a percentage and ranges from near zero for highly creditworthy borrowers to close to 100% for borrowers already showing severe signs of financial distress.
PD is calculated using statistical models that analyze historical default data alongside borrower characteristics and economic conditions. The most common approaches include logistic regression models, which estimate the probability of default based on financial ratios and borrower attributes, and survival analysis models, which account for the timing of default over longer horizons. More sophisticated approaches use machine learning techniques to identify non-linear relationships in large datasets.
Point-in-time versus through-the-cycle PD
One of the most important distinctions in PD modelling is between point-in-time (PIT) and through-the-cycle (TTC) estimates. A point-in-time PD reflects current economic conditions and borrower circumstances. It moves up during economic downturns and down during recoveries. IFRS 9 requires point-in-time PD estimates because provisioning should reflect the credit risk that exists right now, not an average across the cycle.
A through-the-cycle PD smooths out economic fluctuations to produce a longer-term average estimate. Basel’s Internal Ratings-Based (IRB) framework uses through-the-cycle PD for regulatory capital calculations because capital requirements are designed to be more stable and not swing dramatically with every economic cycle. The same borrower can therefore have a different PD depending on which framework is asking the question.
Forward-looking information in PD
IFRS 9 adds another layer of complexity by requiring that PD estimates incorporate forward-looking macroeconomic information. This means your model cannot simply look at historical default rates. It must also factor in forecasts for variables like GDP growth, unemployment, and interest rates, and weight multiple economic scenarios when estimating expected credit losses. This forward-looking requirement is one of the features that makes IFRS 9 PD modelling significantly more demanding than traditional approaches.
What is loss given default and what factors drive it?
Loss Given Default (LGD) is the proportion of an exposure that a bank expects to lose permanently if a borrower defaults. It is expressed as a percentage of the Exposure at Default. An LGD of 40% means that if a borrower defaults on a loan of 1,000, the bank expects to recover 600 and lose 400 after accounting for all recovery costs and the time value of money.
LGD is not simply the inverse of the recovery rate. It accounts for the full cost of the recovery process, including legal costs, time delays in recovering collateral, and the discounting effect of receiving recoveries months or years after the default event. A bank may technically recover 70% of a defaulted loan, but if that recovery takes three years and involves significant legal costs, the effective LGD is considerably higher than 30%.
The main factors that drive LGD include:
- Collateral type and quality: Secured loans backed by liquid, high-value assets like residential property typically have lower LGD than unsecured consumer loans. The enforceability of the collateral claim matters as much as the asset value.
- Seniority of the debt: Senior secured creditors recover more than subordinated or unsecured creditors in a default scenario. The position in the capital structure directly affects expected recovery.
- Economic conditions at the time of default: Property values, asset market liquidity, and general economic conditions at the point of default affect how much a bank can recover. LGD tends to increase during economic downturns when asset values are depressed and buyers are scarce.
- Borrower type and industry: Corporate defaults in capital-intensive industries may leave more recoverable assets than service businesses. Retail borrowers present different recovery dynamics than corporate counterparties.
Like PD, LGD estimates differ between IFRS 9 and Basel frameworks. IFRS 9 requires point-in-time, forward-looking LGD estimates that reflect current and expected future conditions. Basel IRB models use downturn LGD, which represents the loss expected during adverse economic conditions. This deliberate conservatism in Basel LGD is designed to ensure capital adequacy even during stress periods.
What is exposure at default and why does it change over time?
Exposure at Default (EAD) is the total amount a bank is exposed to at the moment a borrower defaults. For a fully drawn term loan, EAD is straightforward: it is approximately equal to the outstanding balance. For revolving credit facilities, overdrafts, and credit cards, EAD is more complex because borrowers can draw down additional funds before they formally default, which means the exposure at the point of default may be higher than the current balance.
EAD changes over time for several reasons. Loan balances amortize as borrowers make scheduled repayments, reducing EAD on term products. On revolving facilities, borrowers in financial distress often draw down available credit before defaulting, a behavior known as increased credit limit usage. This means that for products with undrawn commitments, EAD at the moment of default is frequently higher than the current drawn balance.
To account for this, banks model a Credit Conversion Factor (CCF), which estimates what proportion of currently undrawn commitments will be drawn down before default. A revolving credit facility with a current balance of 500 and a limit of 1,000 does not have an EAD of 500. The CCF adjusts upward to reflect the expected additional drawdown, producing a higher and more realistic EAD estimate.
EAD also varies by product type. Derivatives and off-balance-sheet items require more complex EAD modelling because their exposure value fluctuates with market movements. A derivative contract may have minimal current exposure but significant potential future exposure depending on how underlying rates or prices move. Capturing this potential future exposure is an important part of EAD modelling for trading book and treasury products.
How do PD, LGD, and EAD differ between IFRS 9 and Basel frameworks?
PD, LGD, and EAD are used under both IFRS 9 and Basel, but the two frameworks calibrate these parameters differently because they serve different purposes. IFRS 9 uses point-in-time, forward-looking estimates to calculate provisions that reflect current and expected credit conditions. Basel uses more conservative, through-the-cycle or downturn estimates to calculate regulatory capital that remains stable across economic cycles.
The table below summarizes the key differences in how each parameter is treated:
- PD under IFRS 9: Point-in-time, incorporating multiple forward-looking economic scenarios and probability-weighted outcomes across the lifetime of the exposure for Stage 2 and Stage 3 assets.
- PD under Basel IRB: Through-the-cycle, representing a long-run average default rate that does not fluctuate significantly with short-term economic conditions.
- LGD under IFRS 9: Point-in-time and forward-looking, reflecting expected recovery rates under current and forecast economic conditions.
- LGD under Basel IRB: Downturn LGD, calibrated to recovery rates observed during periods of economic stress, deliberately conservative to protect capital adequacy.
- EAD under IFRS 9: Expected exposure at the point of default, incorporating forward-looking assumptions about drawdown behavior and repayment patterns.
- EAD under Basel IRB: Regulatory EAD using standardized or internally modelled Credit Conversion Factors, subject to regulatory floors and output floor constraints under Basel IV.
These differences create a practical challenge for banks running both frameworks simultaneously. The same loan portfolio produces two different sets of PD, LGD, and EAD estimates, which then feed into two different calculation chains. Reconciling IFRS 9 provisions with Basel Risk Weighted Assets requires a clear understanding of where and why the parameters diverge.
Basel IV has added further complexity here. Under the revised Internal Ratings-Based framework introduced through CRR3 in the EU from January 2025, banks previously running IRB models under Basel III now need to recertify or rebuild those models to meet new output floor constraints and enhanced governance standards. This means that EAD, PD, and LGD models that were acceptable under Basel III may require significant rework to remain compliant, adding to the modelling burden that risk teams are managing in 2026.
Understanding the distinction between IFRS 9 and Basel parameters also matters for stress testing. When you run a credit risk stress test, you need to be clear about which version of PD, LGD, and EAD you are stressing and what the downstream effect is on provisions versus regulatory capital. A stress scenario that significantly increases point-in-time PD will have a direct and immediate effect on IFRS 9 ECL. Its effect on Basel RWA depends on whether the stressed PD exceeds the through-the-cycle estimate used in the IRB model. Getting this distinction right is what separates a meaningful stress test from one that produces misleading outputs.
At ElysianNxt, we treat PD, LGD, and EAD as connected parameters within a single credit risk calculation chain rather than separate inputs for separate regulatory engines. Our .NXT platform supports both IFRS 9 and Basel calculations on the same governed data and model environment, so the relationship between provisions and regulatory capital remains traceable and consistent. If you want to see how this works in practice, our team is happy to walk you through it. Basel IV credit risk is a good starting point to understand how we approach the full parameter chain.
Frequently Asked Questions
What is the difference between expected credit loss (ECL) and unexpected loss, and which one do PD, LGD, and EAD measure?
PD, LGD, and EAD together measure Expected Credit Loss (ECL), which is the statistically anticipated average loss on a credit portfolio over time. Unexpected loss, by contrast, refers to the volatility around that average — the losses that exceed expectations during stress periods — and is what regulatory capital under Basel is primarily designed to absorb. This is why Basel parameters like downturn LGD and through-the-cycle PD are deliberately conservative: they help ensure that capital buffers remain adequate even when actual losses deviate significantly from the expected baseline.
How do I know if my PD, LGD, and EAD models are performing well, and what validation steps should I take?
Model performance should be assessed through a structured validation program covering discrimination, calibration, and stability. For PD models, discrimination is typically measured using the Gini coefficient or Area Under the ROC Curve (AUC), while calibration checks compare predicted default rates against observed default rates over time. LGD and EAD models require backtesting against realized recovery data and actual drawdown behavior at default, and all three models should be subject to periodic stability testing — such as Population Stability Index (PSI) analysis — to detect when the model’s input distribution has shifted enough to warrant recalibration.
What are the most common mistakes banks make when building or calibrating these three parameters?
One of the most frequent mistakes is treating PD, LGD, and EAD as independent model silos rather than interconnected parameters within a single calculation chain, which leads to inconsistencies when the outputs are combined into ECL or RWA. Another common error is failing to account for the correlation between LGD and PD — in economic downturns, defaults tend to rise at the same time as recovery rates fall, meaning LGD and PD are not independent, and ignoring this understates risk. A third pitfall is using insufficient historical data for LGD and EAD calibration, particularly for low-default portfolios where observed defaults are rare and recovery data is sparse.
How does the three-stage IFRS 9 staging model affect which version of PD is used in the ECL calculation?
Under IFRS 9, the staging of an exposure directly determines the time horizon over which PD is applied in the ECL calculation. Stage 1 assets use a 12-month PD, meaning only the probability of default within the next year is considered. Stage 2 and Stage 3 assets require a lifetime PD, which extends the PD estimate across the remaining contractual life of the exposure and must incorporate forward-looking macroeconomic scenarios for the full duration. This means that a borrower migrating from Stage 1 to Stage 2 can trigger a significant jump in ECL even if their underlying credit quality has only deteriorated marginally, because the time horizon for the PD calculation changes dramatically.
What data do I need to start building reliable PD, LGD, and EAD models, and how much historical data is typically required?
At a minimum, you need historical loan-level data covering borrower characteristics at origination, repayment behavior over time, default events with precise dates, and post-default recovery records including amounts, timings, and costs. Regulators and industry best practice generally recommend at least five to seven years of historical data for PD models, and ideally data that spans at least one full economic cycle to capture both benign and stressed conditions. For LGD and EAD, the data requirements are more demanding because recovery processes can take years to complete, meaning you need fully resolved workout cases — not just defaults — to calibrate these parameters accurately, which often requires ten or more years of data for robust estimation.
How does Basel IV's output floor affect banks that use internal models for PD, LGD, and EAD?
Basel IV introduced an aggregate output floor that caps the benefit banks can derive from using their own internal PD, LGD, and EAD models relative to the Standardised Approach. Specifically, RWAs calculated using internal models cannot fall below 72.5% of what the Standardised Approach would produce for the same portfolio. This means that even if a bank’s internally modelled parameters produce lower risk estimates — because their portfolio is genuinely lower risk — the output floor limits how much capital relief they can actually recognize. For banks with well-performing, low-risk portfolios, this can result in materially higher capital requirements than their internal models would otherwise suggest, making the calibration and governance of PD, LGD, and EAD models even more commercially significant.
Can the same PD, LGD, and EAD models be used for both regulatory compliance and internal credit decision-making, or do banks need separate models?
In principle, banks should aim for consistency between regulatory and internal models, but in practice the two sets of requirements often pull in different directions. Regulatory models must meet strict governance, documentation, and conservatism standards set by supervisors, while internal credit decisioning models need to be responsive, granular, and commercially calibrated. Many banks run a dual-use architecture where a single governed model infrastructure produces both regulatory outputs (IFRS 9 ECL and Basel RWA) and internal risk pricing inputs, with clear documentation of any adjustments made for each purpose. The key is maintaining full traceability so that any difference between regulatory and internal parameter estimates can be explained and justified to auditors and supervisors.
Related Articles
- Real Time vs Batch Risk Calculations: Which Architecture Is Better for Basel IV?
- Liquidity Stress Testing Tools: A Buyer's Checklist for 2026
- What is SREP and how does it affect a bank's capital planning?
- 6 questions to ask any Basel IV software vendor before you sign
- 5 things every CFO should know about Basel IV before 2027
This content was generated with the help of AI and it may contain mistakes
