3 minute read

Interest rate risk in the banking book (IRRBB) is one of the fundamental components of Basel III regulations. Supervisors require banks to evaluate the impact of standardized interest-rate shocks on Net Interest Income (NII) and Economic Value of Equity (EVE). These shocks represent a series of pre-specified movements to the yield curve.

But real-world interest rate changes are not arbitrary shifts.
They occur in a macro-financial system where:

  • GDP and inflation respond to policy,
  • deficits and debt interact with long-term rates,
  • the yield curve embeds expectations about future monetary actions.

So a natural question emerges:

What if we embed the Basel IRRBB scenarios inside a full macroeconomic model?
What if we allow GDP, inflation, debt, and the policy rate to react consistently to these scenarios?

This project builds exactly that end-to-end pipeline using modern Bayesian tools in Python.

👉 Full technical notebook:
https://thomasmartins.github.io/IRRBB_SVAR_forecasting/notebooks/MAIN.html

1. Overview of the pipeline

The project integrates four components:

  1. Bayesian VAR (BVAR) to capture macro-financial dynamics
  2. SVAR identification using sign restrictions to simulate macroeconomic scenarios
  3. Basel III IRRBB yield-curve scenarios mapped into level/slope constraints
  4. NII and EVE calculations under both regulatory and structural shocks

Everything is estimated with PyMC + BlackJAX, producing full posterior uncertainty.

The codebase is modular (in /src), and the notebook walks through the full system from raw data to final NII/EVE estimates.

2. Data: the macro-financial panel

The dataset includes:

  • Real GDP growth (q/q annualised)
  • Inflation (annualised)
  • Government deficit (% GDP, sign-corrected)
  • Debt-to-GDP
  • ECB policy rate
  • Yield curve level, slope, curvature factors

These variables represent the minimal set needed to link:

  • macroeconomic variables
  • monetary and fiscal policy
  • term structure movements

into a multivariate system.

3. Bayesian VAR (BVAR)

We estimate a VAR with:

  • Minnesota-style priors (shrinkage) on coefficients
  • HMC/NUTS sampling via PyMC and BlackJAX
  • Full posterior distributions for all parameters

This gives a flexible, stable, and regularized representation of macro-financial dynamics suitable for conditional forecasts.

4. SVAR identification with sign restrictions

Causality is hard to disentangle in reduced-form VAR shocks.
This is why we impose sign restrictions on impulse responses to analyze macroeconomic scenarios.

Monetary tightening shock

  • Policy rate ↑ on impact
  • Yield curve flattens
  • Inflation ↓ (lagged)
  • GDP ↓ (lagged)

Fiscal expansion shock

  • Deficit increases
  • Level of the curve ↑ (moderately)
  • GDP ↑ (lagged)
  • Inflation ↑ (lagged)
  • Debt ↑ over the medium run

Posterior-accepted structural shocks generate clean IRFs: they show the expected disinflation and flattening after a monetary hike, and the expected demand-side effects after a fiscal expansion (increased GDP and inflation).

5. Basel III IRRBB scenarios

The ECB prescribes six regulatory yield-curve shocks:

  • Parallel up / parallel down
  • Steepener / flattener
  • Short-end up / short-end down

These are mapped into factor constraints on the:

  • level
  • slope
  • curvature

of the yield curve.

For example, “Parallel up”: the level increases by 200 bps

This allows the Basel scenarios to be imposed as future path constraints inside the BVAR.

6. Conditional forecasts

To simulate scenarios, I use Waggoner–Zha conditioning:

  • Start with the unconditional BVAR forecast distribution of future macro variables.
  • Impose linear constraints on future level/slope values.
  • Compute the conditional mean + covariance for the entire system.

This gives model-consistent macro reactions to Basel-style yield shifts.

A key insight from the results:

Basel scenarios barely move GDP or inflation
— exactly as expected, because these are not macroeconomic scenarios, but regulatory curve shocks.

Structural SVAR shocks, however, show clear macro dynamics.

7. NII and EVE: linking macro-finance to bank risk

Finally, I translate the conditional yield-curve paths into IRRBB metrics:

NII (Net Interest Income)

A flow measure: how earnings change over 1 year due to repricing.

EVE (Economic Value of Equity)

A stock measure: present value of the banking book, approximated with durations.

The implementation is stylised but realistic:

  • Repricing buckets for NII
  • Pass-through from market rates to customer rates
  • Duration-based EVE impacts
  • CET1 normalization

Some of the insights we get:

  • Parallel-up increases NII, decreases EVE
  • Steepener/Flattener produce similar EVE due to the simplified duration structure of our model balance sheet
  • Monetary tightening slightly reduces NII and increases EVE

8. Full notebook and repository

  • đź“’ Jupyter notebook:
    https://thomasmartins.github.io/IRRBB_SVAR_forecasting/notebooks/MAIN.html

  • 📦 GitHub repo:
    https://github.com/thomasmartins/IRRBB_SVAR_forecasting

9. Final thoughts

This approach allows us to analyze IRRBB in a macroeconomic context, connecting curve movements to policy scenarios and macroeconomic variables.

This project demonstrates how Bayesian econometrics, SVAR identification, and IRRBB regulatory analytics can be combined into a coherent workflow.

You May Also Enjoy