P7. IFRS 9 Expected Credit Loss for Vietnamese Banks

Risk: From Market Risk to Expected Credit Loss, Part 7. Previously: Low Default Portfolios.

IFRS 9 expected credit loss replaced a rule that only recognised losses after something had gone wrong. The old incurred-loss model waited for a trigger event; IFRS 9 requires provisions from the day a loan is written. For Vietnamese banks the complication is that a second framework — the State Bank’s own loan classification rules — continues to operate alongside it, and the two do not agree.

The three stages

IFRS 9 sorts every exposure into one of three stages, and the stage determines how much loss you provide for.

The jump from Stage 1 to Stage 2 is the expensive one. Moving from a 12-month to a lifetime horizon can multiply the provision several times over for a long-dated loan, and that migration happens before any payment has been missed. Consequently the criteria governing it carry more financial weight than almost anything else in the standard.

Significant increase in credit risk

IFRS 9 deliberately does not define SICR quantitatively. Each bank sets its own criteria, which creates both flexibility and audit exposure.

Common criteria in practice:

• Relative PD change. Lifetime PD has risen by more than a set multiple or absolute threshold since origination. This is the primary criterion at most banks.
• Rating migration. A downgrade of a specified number of notches.
• Thirty days past due. A rebuttable presumption written into the standard — rebuttable, but rebutting it requires evidence.
• Watchlist status. Inclusion on an internal credit watchlist.
• Forbearance. Restructuring granted due to financial difficulty.

Two design points matter. First, SICR is assessed relative to origination, not against an absolute threshold — a loan written as high-risk and still high-risk has not deteriorated. Second, the test must be symmetric: exposures must be able to move back to Stage 1 when credit quality recovers.

Computing the number

The ECL calculation reuses the three components from Part 4, now extended over time and across scenarios.

1. Build the lifetime PD profile — marginal default probability in each future period.
2. Estimate LGD, adjusted for forward-looking conditions.
3. Project EAD period by period, accounting for amortisation and expected drawdown.
4. Multiply the three per period to get expected loss per period.
5. Discount each period’s figure at the effective interest rate and sum.

For Stage 1 the sum covers twelve months. For Stages 2 and 3 it runs to maturity.

Then the whole calculation repeats under multiple macroeconomic scenarios — typically base, upside and downside — with the results probability-weighted. This is what makes IFRS 9 ECL a point-in-time measure requiring the forward-looking PD discussed in Part 5, and why a Basel through-the-cycle model cannot be dropped in unmodified.

The non-linearity is the part people miss. Because the PD-to-macro relationship is convex, the probability-weighted ECL exceeds the ECL computed from the base scenario alone. Running only a central case understates the provision, and auditors test for exactly this.

The Vietnamese position

Two frameworks operate in parallel, and understanding the gap is the practical core of the job.

The regulatory framework. Circular 31/2024/TT-NHNN governs classification of assets for banks, non-bank credit institutions and foreign bank branches, replacing Circular 11/2021/TT-NHNN; it has since been amended by Circular 37/2025/TT-NHNN. Loans are classified into five groups by deterioration, with groups 3 to 5 constituting non-performing debt, and provisioning rates attach to those groups. The method is substantially rule-based, driven by days past due and defined qualitative conditions.

The accounting framework. Vietnam’s IFRS roadmap was approved under Decision 345/QD-BTC in 2020, structured in phases around voluntary adoption from 2022 to 2025 and compulsory application thereafter for defined entity groups. Importantly, that roadmap explicitly excluded banks and other credit institutions, leaving them to a separate roadmap to be issued by the State Bank. Commitment to IFRS was reinforced by the amended Law on Accounting, and a draft circular detailing scope and procedures has been in preparation with an expected effective date of 1 January 2026.

Where this leaves a Vietnamese bank in practice:

Larger banks therefore run both: statutory provisions under the circular, and ECL under IFRS 9 for IFRS-basis reporting where applicable. Reconciling the two is an ongoing operational burden, not a one-time exercise.

Regulatory detail changes. Verify the current status of these instruments and the SBV’s banking-sector roadmap against primary sources before relying on any of it for a decision.

Four problems specific to this market

Short macro series. The PD-to-macro relationship needs a cycle’s worth of data. Most banks have less, as covered in Part 6.

Forbearance distortion. Restructuring programmes that permitted retaining existing loan groups mean historical classification understates true migration. ECL models calibrated across those periods inherit the bias.

Collateral valuation. LGD depends heavily on real estate, where valuation practice is uneven and enforcement timelines are long and variable.

Dual reporting. Running two frameworks requires data in two shapes, and legacy core systems rarely supply both cleanly.

Frequently asked questions

What is IFRS 9 expected credit loss?

It is a provisioning model requiring loss allowances from initial recognition, based on probability-weighted expected losses over 12 months or the asset’s lifetime depending on stage.

What triggers a move to Stage 2?

A significant increase in credit risk since origination. Each bank defines its own criteria, commonly a relative PD change, rating downgrade, or the standard’s rebuttable 30-days-past-due presumption.

Does IFRS 9 apply to Vietnamese banks?

Decision 345/QD-BTC excluded credit institutions from the general roadmap, leaving them to a separate State Bank roadmap. Banks reporting on an IFRS basis apply it; statutory provisioning continues under Circular 31/2024 as amended.

Why use multiple macroeconomic scenarios?

Because the relationship between macro conditions and default is convex. A probability-weighted result exceeds the base-case figure, so a single central scenario understates the provision.

Try it yourself

Build a single-loan ECL calculation, in this order:

1. Take a five-year amortising loan and build its EAD schedule period by period.
2. Assign a marginal PD to each period and an LGD assumption.
3. Compute 12-month ECL for Stage 1 and lifetime ECL for Stage 2.
4. Note the ratio between them — this is the cost of a single stage migration.
5. Repeat under three macro scenarios and probability-weight the results.

Compare that weighted figure to the base case alone. The gap is the convexity effect, and it is the number an auditor will ask about first.

Next in this series: model validation — what regulators actually ask.

Sources

• Circular 31/2024/TT-NHNN on classification of assets of credit institutions
• Circular 37/2025/TT-NHNN amending Circular 31/2024
• PwC Vietnam on Decision 345/QD-BTC and the IFRS roadmap
• KPMG Vietnam on the draft circular and IFRS roadmap status