PD, LGD and EAD: How Banks Calculate ECL Under IFRS 9

A step-by-step guide for finance teams, credit risk modellers, and IFRS 9 practitioners who need to go beyond the formula and understand where the inputs actually come from — and where auditors push back.

✓ Written by Prima Consulting’s advisory team  ·  ✓ Serving GCC, Europe & APAC  ·  ✓ Actuaries + CPAs + CFAs

TL;DR

The pd lgd ead ifrs 9 framework is how banks calculate expected credit loss under IFRS 9 — but knowing the formula (ECL = PD × LGD × EAD) is not the same as knowing how each input is built. This article covers how probability of default is estimated using point-in-time vs through-the-cycle methods, how loss given default accounts for collateral recovery and haircuts, and how exposure at default handles drawn and undrawn balances. You’ll also get a worked retail loan example with a sensitivity table. If your ECL model is under audit scrutiny, start here.

The ECL Formula Finance Teams Get Wrong Every Year

Most finance teams can write ECL = PD × LGD × EAD on a whiteboard. Fewer can explain why the PD they’re using is point-in-time rather than through-the-cycle, or why their auditors keep asking about the LGD haircut applied to residential collateral. That gap — between knowing the formula and understanding the inputs — is exactly where IFRS 9 impairment calculations break down in practice.

The EBA’s 2023 IFRS 9 monitoring report found that variability in PD modelling practices is one of the primary drivers of divergence in ECL estimates across EU institutions. Not model philosophy. Not economic forecasts. PD inputs.

This article walks through each component in plain terms, shows a complete worked example, and covers the audit challenges that come up most often.

What this article covers:

• How PD, LGD, and EAD are derived — not just defined
• A full retail loan ECL calculation with a sensitivity table
• The most common audit findings on these three inputs

If you want the broader IFRS 9 framework before getting into the components, that’s a good place to start. But if you’re here for the inputs, let’s get into them.

What Is Probability of Default (PD) in IFRS 9?

PD is an estimate of the likelihood a borrower will default over a given period. For Stage 1 assets, that period is 12 months. For Stage 2 and Stage 3 assets, you use lifetime PD — which means you need a full term structure of default probabilities, not just a single annual rate.

Here’s the part that trips up most teams: under IFRS 9, PD must be a best estimate, not a conservative one. That’s the opposite of Basel, where PD estimates are deliberately cautious. If your team is feeding Basel regulatory PD directly into your IFRS 9 model without adjustment, you’re almost certainly over-provisioning — and your auditors will ask why.

Point-in-Time vs Through-the-Cycle PD: Why the Difference Matters

Through-the-cycle (TTC) PD reflects average default rates across a full economic cycle. Banks use this for regulatory capital because it smooths out volatility. But IFRS 9 requires point-in-time (PIT) PD — an estimate conditioned on current economic conditions and the forward-looking outlook.

The practical difference is significant. During a recession, your PIT PD will spike well above your TTC PD. During a recovery, it’ll fall below it. The BIS Basel discussion paper makes this explicit: because IFRS 9 uses PIT estimates, accounting ECL tends to exceed regulatory expected loss in stress periods — sometimes by a wide margin.

Most banks convert their TTC PDs to PIT using macroeconomic overlays. The common approach links historical PD observations to macro variables like GDP growth and unemployment, then applies forward-looking forecasts to shift the curve. That forward-looking adjustment is what auditors scrutinise hardest.

How Banks Estimate PD for a Retail Loan Portfolio

For a standard retail mortgage or personal loan portfolio, PD estimation typically starts with internal default history — usually at least five years of loan performance data, segmented by risk band, product type, and origination vintage.

From that history, banks derive an average annual default rate per segment. That becomes the baseline TTC PD. They then apply a scalar — derived from regression analysis linking macro variables to historical default experience — to shift it to PIT.

One thing to watch: if your historical data is short or comes mainly from a benign credit period, your baseline PD will be understated. That’s not a modelling choice — it’s a data problem, and it’s one the World Bank flagged as a consistent challenge for banks in developing markets.

Loss Given Default in IFRS 9: Where Collateral Assumptions Go Wrong

LGD is the percentage of the exposure you expect to lose after recoveries — expressed as: LGD = 1 − recovery rate. A loan with a 40% recovery rate has an LGD of 60%.

This sounds simple. It’s not.

Recovery rates depend on the type and quality of collateral, the legal environment for enforcement, expected time to recovery, and the costs of the recovery process itself. Each of these involves judgment, and that judgment has to be documented, consistently applied, and defensible under audit.

How LGD Accounts for Collateral Recovery

For secured retail loans — say, a mortgage on a residential property — LGD is driven mainly by the collateral value at the point of default, minus enforcement costs and time discounting.

The standard approach uses a haircut on the current collateral value to arrive at the expected recovery amount. A property worth $500,000 today might be assigned a 70% haircut recovery factor, giving you an expected recovery of $350,000. If the outstanding loan balance at default (EAD) is $400,000, your LGD = ($400,000 − $350,000) / $400,000 = 12.5%.

But here’s where it gets harder. That 70% recovery factor needs to reflect forward-looking conditions — not just what the property sold for last year. In a falling housing market, a 70% factor may be optimistic. Auditors will want to see how you’ve stress-tested that assumption and whether your macroeconomic scenario weighting changes the collateral haircut.

LGD Haircuts and Why Auditors Flag Them

The Bank of England’s 2025 IFRS 9 thematic feedback letter specifically called out LGD model challenge as an area needing more investment. Recovery strategies at many banks are treated as stable and reviewed only when individual assessments flag changes. That’s not good enough. LGD models should be backtested against actual recovery outcomes, recalibrated when macro conditions shift, and — critically — documented in enough detail that someone outside the modelling team can reproduce the output.

If your LGD haircuts haven’t been updated since original model build, that’s a red flag. Especially for portfolios with commercial real estate or unsecured consumer credit exposure, where recoveries are highly sensitive to economic conditions.

If you’re working through an ECL model build or model validation, Prima Consulting’s ECL modelling team works across retail, corporate, and sovereign portfolios in the GCC and beyond.

Exposure at Default: Drawn Balances Are Only Half the Story

EAD is the amount you expect to be exposed to at the moment a borrower defaults. For a term loan with a fixed repayment schedule, EAD is relatively straightforward — it’s the outstanding principal balance, adjusted for expected repayments before the default date.

But for revolving facilities, credit cards, and overdrafts, it gets more complicated. Borrowers often draw down more heavily on their facilities before defaulting. That behavioural pattern has to be built into the EAD estimate.

EAD for Drawn Exposures

For drawn balances on a term loan, EAD is typically calculated as:

EAD = Outstanding principal × (1 − expected repayment rate before default)

That repayment rate comes from historical analysis of how much principal borrowers typically repay between the observation date and their eventual default. For a portfolio with a 6% average repayment rate in the period preceding default, a $100,000 balance gives an EAD of $94,000.

EAD for Undrawn Commitments and the Credit Conversion Factor

This is where most modelling teams underestimate exposure. Undrawn credit facilities — the portion of a credit line that hasn’t been used — still represent potential exposure at default.

To capture this, IFRS 9 practitioners use a Credit Conversion Factor (CCF):

EAD = Drawn balance + (CCF × Undrawn balance)

If a borrower has drawn $60,000 of a $100,000 revolving facility, with a CCF of 0.75 applied to the $40,000 undrawn portion: EAD = $60,000 + (0.75 × $40,000) = $90,000.

The CCF is estimated from historical data on how much borrowers draw down before defaulting. For credit cards and revolving facilities, CCFs are often high — sometimes above 0.80 — because financially distressed borrowers tend to max out available credit in the period leading up to default. That’s a behavioural pattern, not just a modelling assumption, and it needs to be supported by observed data.

ECL Calculation Example: A Retail Loan Portfolio Step by Step

Let’s make this concrete. Suppose you’re calculating the Stage 1 ECL for a retail mortgage portfolio with the following characteristics:

• Outstanding balance: $10,000,000
• Weighted average remaining maturity: 4 years
• Stage: Stage 1 (no significant increase in credit risk since origination)
• Effective interest rate (EIR): 5.5%

The ECL Formula Table

Scenario Weighting and Forward-Looking Adjustments

IFRS 9 requires a probability-weighted ECL — which means you don’t just use one set of macro assumptions. You run at least two or three scenarios, assign probabilities to each, and take the weighted average.

Using the same portfolio:

• Base case (60% weight): ECL = $57,495 — stable GDP growth, unemployment at 4.2%
• Downside (30% weight): ECL = $98,000 — GDP contraction of 1.5%, unemployment rising to 6.8%
• Upside (10% weight): ECL = $38,000 — strong growth, unemployment falling to 3.5%

Probability-weighted ECL = ($57,495 × 0.60) + ($98,000 × 0.30) + ($38,000 × 0.10) = $66,297

That’s 15% higher than the base case alone. And that gap — the difference between single-scenario ECL and probability-weighted ECL — is exactly what auditors look for when they ask whether your forward-looking adjustments are material and defensible.

Sensitivity Analysis: How Each Variable Shifts the ECL Output

That last row is the one worth sitting with. A combined 50bps PD increase, 10 percentage point LGD increase, and 5% EAD increase together moves ECL by over 60%. This isn’t a stress test — it’s the kind of shift that can happen in a moderate downturn. Your ECL model needs to handle that without requiring a full rebuild.

IFRS 9 Staging and How It Changes Your ECL Horizon

The three inputs don’t work in isolation — they’re applied differently depending on which IFRS 9 stage a loan sits in. Understanding the IFRS 9 Financial Instruments staging framework is essential before you can finalise your ECL.

Stage 1 uses 12-month PD. Stages 2 and 3 use lifetime PD. That sounds straightforward, but it means your PD model needs to produce not just a single one-year estimate, but a full term structure — marginal default probabilities for each year of the loan’s remaining life.

A 5-year retail mortgage in Stage 2 needs five annual marginal PDs, each reflecting the deteriorating credit condition that triggered the Stage 2 classification in the first place. Most teams can produce an annual PD. Fewer can produce a well-calibrated lifetime term structure that actually reflects the credit trajectory of a deteriorating borrower. That gap shows up in audits as “insufficient granularity in lifetime PD estimation.”

For a fuller look at how IAS 39’s incurred loss approach compares to IFRS 9’s expected loss model, the IFRS 9 vs IAS 39 comparison is worth reviewing. The contrast explains why the shift to point-in-time inputs was so disruptive for Basel-based modelling teams.

Prima Consulting’s advisory team has supported ECL model builds and validations across commercial banks, Islamic finance institutions, and insurance companies in Saudi Arabia, the UAE, and across the GCC. See how we approach IFRS 9 advisory.

The Audit Challenges Banks Keep Running Into

Let me be direct about this: most IFRS 9 audit findings on PD, LGD, and EAD inputs are not about edge cases. They’re about the same issues, repeated, at bank after bank.

On PD:

Auditors consistently find that forward-looking adjustments are poorly documented. The macro-to-PD linkage exists in a spreadsheet somewhere, built by one person three years ago, with no documented methodology. When the macro environment shifts and PD barely moves, the auditors want to know why. “The model doesn’t pick it up well” is not an answer.

On LGD:

The Bank of England’s 2025 ECL thematic review flagged that LGD models at many banks operate on stable assumptions without sufficient periodic challenge. Recovery rates from 2019 get applied in 2025 without review. That’s a governance problem, not a modelling problem — but it fails audits just the same.

On EAD:

CCF estimates for undrawn commitments are frequently drawn from limited historical default data. If your bank has had low default rates historically, you may have very few observations of actual drawdown behaviour before default. The EBA’s monitoring work found this consistently across high-default portfolio reviews.

And then there’s the overlay problem. The EBA’s second IFRS 9 monitoring report found that overlays — management adjustments applied on top of modelled ECL outputs — are becoming an integral part of the ECL framework but are often applied without robust governance or calibration documentation. An overlay that increases ECL by $2M with a one-line justification is not going to survive audit.

If your team is heading into an audit cycle, the place to start is not the ECL number — it’s the documentation trail behind each of the three inputs. For banks looking at technology solutions to manage this, IFRS 9 software for banks has become a practical way to automate the model-to-ledger reconciliation and provide the audit trail auditors need.

The IFRS 9 expected credit loss guide covers the broader governance and disclosure requirements that sit around these inputs — worth reading alongside the technical modelling work.

What You Now Know

• PD must be point-in-time, not through-the-cycle. Basel PDs cannot be used without adjustment. Your forward-looking macro linkage needs to be documented, calibrated, and reviewed at each reporting date.
• LGD is more than a collateral haircut. It requires forward-looking assumptions, recovery timing, enforcement costs, and periodic recalibration against actual recovery outcomes.
• EAD for undrawn facilities depends on the Credit Conversion Factor. That CCF needs to reflect observed borrower drawdown behaviour before default — not a regulatory floor or a management estimate unsupported by data.

The ECL formula is short. The work behind it isn’t. And the gap between a model that produces a number and a model that survives audit — that gap is almost always in the inputs, not the formula.

For a worked example using trade receivables and a provisioning matrix approach rather than the full PD-LGD-EAD model, the ECL model IFRS 9 examples page covers both methods with practical illustrations.

FAQ: PD, LGD and EAD Under IFRS 9

What is the difference between point-in-time and through-the-cycle PD in IFRS 9?

Point-in-time PD reflects current economic conditions and forward-looking forecasts. Through-the-cycle PD averages default rates across a full cycle and is used for Basel regulatory capital. IFRS 9 requires point-in-time PD because it needs a best estimate of loss, not a conservative regulatory buffer.

How is loss given default calculated for a secured retail loan?

LGD for a secured loan is calculated as 1 minus the net recovery rate. Net recovery accounts for the collateral value at default, minus a haircut for market conditions, minus enforcement and legal costs, discounted for the time it takes to realise the recovery. Banks must apply forward-looking adjustments to reflect expected future collateral values.

What is the Credit Conversion Factor and when does it apply to EAD?

The CCF is a multiplier applied to the undrawn portion of a credit facility to estimate how much of that undrawn balance will have been drawn at the point of default. It applies to revolving credits, overdrafts, and credit cards. EAD = drawn balance + (CCF × undrawn balance). CCFs are estimated from historical default data.

How do IFRS 9 ECL calculations differ between Stage 1 and Stage 2?

Stage 1 uses 12-month PD — the probability of default within the next year. Stage 2 uses lifetime PD — the probability of default over the remaining life of the instrument. This requires a full term structure of marginal annual PDs, not just a single annual rate extrapolated over the remaining term.

What are the most common audit findings on IFRS 9 ECL model inputs?

The EBA and Bank of England have both flagged inadequate documentation of forward-looking PD adjustments, LGD models that haven’t been recalibrated against recent recovery experience, CCF estimates based on limited default observations, and management overlays applied without robust governance or methodology documentation.