Stress Testing the Yield Curve: From Regulatory Scenarios to Custom Shocks

IRRBB measurement is only as good as the scenarios you run it against. The BCBS prescribes six standardized shocks, but these are a minimum. Banks that rely solely on parallel ±200bp miss the curve movements that often produce the largest real-world losses. Designing the right set of scenarios — regulatory, historical, statistical, and expert-driven — is where measurement meets judgment.

Why Scenarios Matter

The purpose of stress scenarios is to answer a simple question: "What happens to our balance sheet under plausible but adverse rate movements?"

The answer depends critically on the shape of the rate shock — not just its magnitude. A +200bp parallel shift, a +300bp short-rate shock, and a steepener that raises long rates by 100bp while cutting short rates by 100bp all have the same "total movement" but radically different impacts on a bank's ΔNII and ΔEVE.

Consider a bank funded primarily by short-term deposits and invested in long-term fixed-rate mortgages:

• Under parallel +200bp: ΔNII is negative (deposit costs rise faster than asset income reprices), ΔEVE is sharply negative (long assets lose more value than short liabilities).
• Under steepener (short −100bp, long +100bp): ΔNII may improve (deposit costs fall), but ΔEVE worsens (long-end rates rise, crushing mortgage values).
• Under short rate +300bp: ΔNII is severely negative (massive deposit repricing), but ΔEVE may be moderate (long-end barely moves, so mortgage values hold).

Each scenario reveals a different vulnerability. No single scenario captures the full risk profile.

The BCBS Six Scenarios

The Basel Committee prescribes six interest rate shock scenarios, each designed to capture a distinct mode of yield curve movement:

The shock magnitudes are currency-specific, calibrated from historical rate movements and updated periodically. For USD, the parallel shock is approximately ±200bp, with larger magnitudes at the short end (up to ±300bp) and smaller at the long end.

Beyond Regulation

The six standardized scenarios provide a common language and comparability across banks. But they have significant limitations:

• Fixed shapes: Real curve movements don't follow the prescribed functional forms. The 2022 tightening cycle involved a curve inversion that doesn't map cleanly to any of the six scenarios.
• No tail events: The calibrated magnitudes represent roughly a 99th percentile of historical changes, but the functional form constrains which 99th percentiles are tested.
• No cross-currency basis: For banks with multi-currency exposures, cross-currency basis movements can be as impactful as rate-level changes.

This is why most banks supplement the six scenarios with internal stress tests drawn from three sources: historical episodes, statistical methods, and expert judgment.

Historical Simulation

The most intuitive approach: replay actual historical rate movements and measure the IRRBB impact. Key episodes include:

Historical simulation has the advantage of being credible — "it actually happened" is a powerful argument in ALCO discussions. The limitation: it only tests scenarios we've already seen. The next crisis may involve a curve shape that has no historical precedent.

PCA-Based Scenarios

Principal Component Analysis of historical daily rate changes reveals that yield curve movements are dominated by three factors — shift, twist, and butterfly — which together explain 95%+ of all daily curve dynamics. (See our dedicated PCA article for the full treatment.)

PCA-based scenarios use these eigenvectors to construct statistically grounded shocks:

The advantage of PCA scenarios: they respect the historical correlation structure of rate movements. A parallel shift that moves all tenors by exactly the same amount has never occurred — the short end always moves more than the long end. PCA scenarios automatically incorporate this realistic non-parallelism.

Partial-Information Method

Sometimes a risk manager has a view on specific tenors but not the entire curve. For example: "I want to test a scenario where the 2Y rate rises 150bp and the 10Y rate rises 50bp — but what should the 5Y, 20Y, and 30Y do?"

The partial-information method solves this problem using the PCA decomposition and linear algebra:

The result is a complete curve scenario that is statistically consistent with historical curve dynamics while honoring the risk manager's specific views. The unspecified tenors are filled in using the same correlation structure that PCA extracted from the data — not guessed or linearly interpolated.

This method is particularly valuable when designing scenarios based on macroeconomic narratives: "The Fed hikes the overnight rate by 200bp, the 2Y moves 150bp, but the long end is anchored by quantitative tightening" — specify the 3 known tenors, and let PCA fill in the rest.

Scenario Selection Matters More Than Model Precision

In practice, the most dangerous IRRBB exposures often lie outside the BCBS six scenarios:

• A bear flattener (short rates up sharply, long rates up moderately) — common during aggressive tightening cycles — combines NII pressure with ΔEVE losses.
• A curve inversion followed by normalization — the path dependence matters because NMD behavior and prepayment speeds evolve along the path.
• A basis risk scenario — SOFR and bank funding costs diverge, which the OIS-only IRRBB framework doesn't capture.

The bank's scenario set should be designed to stress the specific vulnerabilities of its balance sheet — not just to satisfy the regulatory minimum.

The Bigger Picture

Scenario design sits at the intersection of quantitative analysis and risk judgment. The BCBS six scenarios provide the regulatory floor. Historical simulation adds credibility. PCA-based scenarios add statistical rigor. Partial-information methods bridge the gap between expert views and statistical consistency.

The ultimate test of a scenario set is not whether it satisfies auditors, but whether it would have warned you in advance of the loss that actually materialized. The banks that navigated 2022 best were those that had already stress-tested aggressive tightening paths — not because they predicted the exact outcome, but because they had built the muscle to think about non-parallel, large-magnitude, multi-quarter rate paths.