IFRS 9 PD Calibration: TTC vs PIT

Probability of default (PD) sits at the centre of every credit risk framework. IFRS 9 and Basel both need one — but they need different flavours of the same number, and confusing them is one of the fastest ways to lose an audit.

Two definitions, one word

• Through-the-cycle (TTC) PD is a long-run average default rate for a rating grade, smoothed across economic conditions. It is the regulatory-capital PD under the IRB approach: stable, comparable across time, and slow to move.
• Point-in-time (PIT) PD is the actual expected default probability right now, given today's macro state. It reacts to unemployment, GDP, house-price growth, and the rest of the cycle. It is what IFRS 9 wants for expected credit loss (ECL).

The same borrower with a stable rating can have a wildly different PIT PD in a recession than at the peak of a cycle. Both PDs can be right — they're just answering different questions.

Why IFRS 9 forces you to pick

IFRS 9 asks for an unbiased, forward-looking, probability-weighted estimate of expected credit losses. That is a PIT quantity by construction. A TTC PD plugged directly into the ECL formula produces a number that is stable, defensible in normal times, and systematically wrong at cycle turns — under-reserving into a downturn and over-reserving into a recovery. Regulators noticed. Audit firms noticed. Boards noticed.

Three ways to build a PIT PD

1. Start TTC, apply a macro overlay. Most established IRB banks already have a TTC PD. The pragmatic route is to keep it and layer a macroeconomic overlay — typically a satellite regression of observed default rates on GDP, unemployment, and sector-specific variables — that scales the TTC PD up or down to today's cycle position.

Strengths: leverages existing rating systems, integrates with capital, easy to backtest overlay layer. Weaknesses: the overlay does most of the heavy lifting and its calibration window matters enormously.

2. Build a PIT PD directly. Estimate default probability on borrower + macro features together, using logistic regression or a machine-learning model. The output is PIT by construction.

Strengths: internally consistent, avoids the TTC-to-PIT conversion. Weaknesses: harder to reconcile to Basel PDs, and rating stability suffers if not deliberately smoothed for the rating system.

3. Segment by portfolio. Retail: PIT is straightforward (behavioural scores + macro overlay). Corporate: PIT is harder because default events are rare, so many banks keep TTC ratings and rely on macro overlays and forward-looking scenarios for the PIT view. SME sits in between.

Forward-looking scenarios: where PIT becomes probabilistic

IFRS 9 requires probability-weighted PDs across multiple macro scenarios — typically base, upside, and downside. This is where "point in time" quietly becomes "expected across a distribution of futures".

The common pitfalls:

• Two scenarios and no distribution. A base and a downside with 80/20 weights is not really a probability-weighted view; it is a stress test.
• Scenario weights that never change. If your downside weight is the same in 2019 and 2023, your model is not really forward-looking.
• Overlays on top of overlays. A satellite that already loads on unemployment, plus a management overlay that adjusts for the same unemployment, will double-count. Governance and documentation matter.

Reconciling to the regulatory PD

Boards and auditors will ask, sooner or later, why the IFRS 9 PD is 2% and the Basel PD for the same borrower is 1%. The clean answer is a PIT–TTC reconciliation: a documented decomposition that starts from the TTC PD, adds a cycle adjustment, adds a forward-looking overlay, and lands on the IFRS 9 PIT PD. Reproducing this bridge every reporting cycle is one of the cheapest defences against an audit surprise.

What good looks like

• One canonical TTC rating system, with its own calibration and monitoring.
• A clearly documented PIT overlay or model, with a stable estimation window and a defined re-calibration cadence.
• Multi-scenario forward-looking macros with weights that respond to the cycle.
• A visible PIT–TTC bridge, produced and reviewed each quarter.
• Backtesting of both the TTC ratings and the PIT overlay, with breaks investigated rather than smoothed away.

IFRS 9 is not going to soften. If anything, the direction of travel — climate overlays, geopolitical scenarios, sector-specific downturn assumptions — is towards more forward-looking judgement. A PD framework built to make TTC and PIT explicit, reconcilable, and separately validated ages better than one that quietly blurs the two.