The sovereign green-bond greenium

Since 2020 the German Finance Agency has issued green Federal securities each paired with a conventional twin of identical coupon and maturity — the cleanest natural experiment in sustainable finance. The only thing that differs within a twin pair is the green label (and the smaller, less-liquid green leg). The yield difference is the greenium. This report estimates it model-free from the twins, adjusts it for liquidity, and stress-tests it with placebo pairs.

↗ open the interactive dashboard — time series with a confidence band, term structure, and raw-vs-adjusted.

1. Why twins identify the greenium

A naive greenium — "green bonds yield less than other bonds" — is confounded by everything that differs across bonds: maturity, coupon, issuer, credit, tax. The German twin bond concept removes all of that: each green Bund is issued alongside a conventional Bund with the identical coupon and maturity date, differing only in the green label and issue size. So for a twin pair on a given day,

greenium (bp) = (conventional yield − green yield) × 100,

is a like-for-like difference with no curve-fitting and no convention mismatch (both legs share the Bundesbank annual / actual-actual quote basis). This is the headline estimator; the liquidity-adjusted panel is corroboration, not the lead.

2. The headline: a small, positive, consistently-signed greenium

The greenium is largest at the short end and compresses along the curve — the term structure on the dashboard runs from the short-tenor pairs down to the 30y green Bunds, consistent with published figures of a few basis points, larger at the short end.

3. Liquidity: how much is a green preference vs a liquidity premium?

The green leg is smaller and less liquid than its conventional twin, so part of the raw spread can be a liquidity premium rather than a green preference. We run a panel of yields with pair×date fixed effects (each green bond compared only to its own twin on the same day) and control for issue size — the only liquidity dimension that varies within a twin-day cell (age and on/off-the-run are identical across twins by construction).

4. Dynamics

5. Robustness & refutation

The honest tests that could have killed the result:

• Placebo pairs. A conventional bond's deviation from the conventional curve should be ≈ 0 — and it is (), with a CI spanning zero. So the greenium is not a mechanical maturity/curve artefact. Green legs, by contrast, sit below that same curve.
• Liquidity-only model. The greenium that survives at equal liquidity is — reported alongside the raw spread so the liquidity share is explicit.
• Sign stability. twin pairs carry a positive mean greenium (sign-test p = ): a real greenium is consistently signed.
• Stress windows. Results are reported with and without the COVID and 2022 rate-shock windows; .
• Uncertainty. Every level carries a stationary block-bootstrap CI and a Newey–West HAC standard error — the greenium is a few bp, so naive iid intervals would badly overstate precision.

6. What this is — and is not

• Sovereign-specific. The clean twin identification exists for German Bunds. It does not transfer to corporate green bonds, where credit risk, covenants and use-of-proceeds verification differ materially.
• Small, noisy magnitudes. A few basis points is near market microstructure noise; conclusions are about averages with bands, not precise daily values.
• Liquidity vs preference. Even with twins, the residual blends a genuine green preference with a liquidity premium; we bound but cannot fully separate them without proprietary order-flow data.
• Not investment advice. The greenium is a research quantity, not a trade.

7. Data & pipeline

Primary source: Deutsche Bundesbank, "prices and yields of listed Federal securities" (daily per-ISIN yields and issue volumes; the green bonds are labelled, and the twin set is rebuilt from those labels). Every headline number traces to a value in site/data/*.json; see SOURCES.md. Seed = .