The Greenium: What a Green Bond Saves a Government in Yield
Germany issues green and conventional federal bonds that are identical except for the green label. The yield gap between the twins is the best measure of the greenium there is.
A green bond promises that the money raised goes to something climate-friendly, a railway, a grid upgrade, a flood defence. Investors say they will accept a slightly lower yield to hold one. That yield discount has a name, the greenium, and measuring it is harder than it sounds. Compare a green bond from one issuer against a conventional bond from another and you are not measuring a green premium, you are measuring everything that differs between the two: credit risk, maturity, coupon, liquidity, tax treatment. Germany solved this problem on purpose. When it issues a green federal bond, it issues a conventional twin alongside it with the same coupon and the same maturity date, differing only in the green label and the issue size. The yield gap between the two twins is about as close to a controlled experiment as sovereign debt markets allow.
The twin trick
The reason the German design matters is identification. Most greenium studies fit a model to a whole curve of bonds and read off the gap a green label seems to leave behind, after controlling for everything else with regressors. That works only as well as the controls do, and the greenium is a couple of basis points, small enough that a sloppy maturity adjustment can manufacture or erase it. A basis point is one hundredth of a percentage point, so we are arguing about fractions of a tenth of a percent on the yield.
The twin design removes the modelling. The green 10-year Bund maturing 15 August 2030 (DE0001030708) and its conventional twin (DE0001102507) share an issuer, a coupon of zero, the same maturity date, a tax status, and a quote convention. The German Finanzagentur set them up that way deliberately, so that the only economically meaningful difference left is the green label and the fact that the green leg is smaller. Subtract one yield from the other and most of what usually confounds a greenium estimate cancels out by construction. No curve to fit, no maturity term to get wrong.
The estimand is simple:
greenium (bp) = (conventional yield − green yield) × 100
A positive number means investors accept a lower yield to hold the green twin. The headline of this project is that difference, averaged across pairs and time, with honest uncertainty around it. Everything else, the panel regression and the dynamics, is corroboration stacked behind the twins, not the lead.
What the data is
The numbers here are live. Every trading day the Bundesbank publishes a workbook of reference prices and yields for every listed Federal security, and it labels the green bonds explicitly. The pipeline downloads those monthly workbooks back to September 2020, reads each daily sheet, picks out the green bonds, and matches each one to its conventional twin by identical coupon and maturity. That match is unique by design, so the twin set falls straight out of the data rather than a hand-typed list of identifiers. The current panel is ten green-conventional pairs, from a five-year Green Bobl out to the 30-year Green Bund 2053, roughly 8,600 twin-days of observations between September 2020 and June 2026.
A useful by-product: the same workbook carries each bond’s issue volume, so the liquidity control later is real outstanding-amount data, not a proxy. That turns out to matter a lot.
What the spread says
Pooled across all ten pairs and the whole sample, the twin greenium is 2.1 basis points, with a block-bootstrap 95% interval of 1.8 to 2.4 and a Newey-West t-statistic north of 25. The greenium is positive on 99.6% of pair-days, and all ten pairs have a positive average. A sign test across pairs rejects the idea that this is noise (p ≈ 0.002). Cross-checking with a panel regression that clusters its standard errors by pair, the conservative way to count ten pairs as ten pieces of information rather than thousands of correlated daily observations, gives 2.1bp with a wider interval of 1.0 to 3.2, still clear of zero. Excluding the COVID and 2022 rate-shock windows barely moves it, from 2.1 to 1.95bp.
The shape across maturities is the cleaner result. The greenium is largest at the short end, about 3.2bp for the zero-to-five-year bucket, and settles around 1.5bp for everything longer. The very shortest Green Bobl carries the widest gap, north of 5bp. That term structure lines up with what the Finanzagentur and outside analysts report on these bonds, and it makes economic sense: a given yield give-up is worth more, in price terms, on a short bond than a long one, so equal investor willingness-to-pay shows up as a wider yield gap at the short end.
Is it green, or is it just small?
Here is where the twin design earns its keep, and where the answer gets uncomfortable for the simple story. The green leg of every pair is much smaller than its conventional twin. The Green Bund 2030 has about €11bn outstanding against more than €30bn for its conventional sibling, and a bond that trades less yields a little more to compensate. That liquidity penalty pushes the green yield up, which works against the greenium. So the raw 2.1bp is, if anything, a floor on the pure green preference once you strip liquidity out. Or so the usual argument goes.
The data says the opposite. Because the twins have identical maturities and issue dates, they are the same age and equally on or off the run, so the only liquidity dimension that varies within a pair on a given day is issue size. Put the log of amount outstanding into the panel and the greenium does not widen. It collapses. The liquidity-adjusted estimate is −1.0bp with an interval of −5.4 to +3.4, straddling zero, and the implied liquidity gap is about 3.1bp. Read plainly: the raw greenium is real and robust, but essentially all of it is explained by the green bonds being smaller and less liquid. Once you compare like-for-like on size, the pure preference for the green label is statistically indistinguishable from zero, and the point estimate even tips slightly negative.
That is a more honest and more interesting finding than a clean positive number would have been. It says the price investors visibly pay for the green twin is mostly a liquidity artefact, not a willingness to forgo return for environmental impact. The adjusted estimate is imprecise, the interval is wide, so I would not bang the table for the negative point. But the direction is clear and it survives the conservative clustering. Anyone quoting the 2.1bp as a pure green premium is quoting a liquidity premium with a green label on it.
Did the greenium shrink?
Yes, and visibly. Fitting a trend to the weekly greenium over the sample gives a compression of about 0.6bp per year, with a bootstrap interval of −0.79 to −0.43 that sits well clear of zero. The annual averages tell the same story without any model: the pooled greenium ran around 4.4bp in 2021, eased through 3bp in 2022 and 2023, and has spent 2025 and 2026 near 1bp. As green issuance grew and the market matured, the gap narrowed. This is the compression that the literature kept asserting, and on the German twins it is real rather than assumed.
Worth separating two things that are easy to conflate. The greenium declines across maturities, the short-end premium described above, and it has declined over calendar time. Both are present here, and they are different facts. A bond bought today still shows a wider gap at short tenors than long ones, and the whole structure sits lower than it did three years ago.
Trying to break it
A result you cannot attack is not a result, so the pipeline runs a refutation suite designed to fail loudly if the greenium is an artefact. The central one is a placebo. If the green yield gap were really just a maturity or curve quirk dressed up as a green effect, then a conventional bond measured against the conventional curve should show the same kind of gap. It does not. The conventional leg’s deviation from the conventional curve sits at essentially zero, with an interval of roughly ±0.006bp straddling zero, exactly what a placebo should do, while the green leg sits about 2bp below the same curve. The greenium is not a mechanical consequence of how the curve is shaped.
The sign-stability check is the cleanest of the lot: all ten pairs are positive on average, and the sign test rejects chance (p ≈ 0.002). The liquidity-only check tells the same story as the panel adjustment, that the gap shrinks toward zero once size is equalised. Taken together the refutations say the raw greenium is real and not a curve artefact, while pointing at liquidity rather than green preference as what is actually being priced.
Why this is Germany only
The original plan carried France as a second issuer, for cross-issuer corroboration. I cut it, for two reasons that point the same way. France has no twin, so any French greenium has to be read off a fitted curve, which is exactly the model-dependent approach the German design exists to avoid. A curve-fitted number on a worse footing does not corroborate a clean one so much as muddy it. And the data is not there anyway: there is no free, machine-readable source for the daily yields of individual French green OATs. The conventional curve is available, but the green bonds’ own yields sit behind commercial terminals, and the Agence France Trésor site blocks automated access. Rather than prop up a weak leg with a stale or synthetic series, the project stands on the German twins, which are the whole point.
Data, methods, and reproducibility
The German yields come from the Bundesbank’s daily “prices and yields of listed Federal securities” workbooks, which carry the ISIN, coupon, maturity, issue volume and reference yield for every listed security and flag the green bonds by name. Both legs of a twin share the Bundesbank annual, actual-actual yield basis, which is what makes differencing the two legs valid without any convention correction, and that assumption is asserted in the test suite rather than taken on faith. The conventional euro-area curve used for the curve diagnostics comes from the ECB Statistical Data Warehouse.
One seed drives the whole pipeline, the bootstraps included, so a clean run reproduces the same JSON to the digit. The fixed-effects panel is a hand-rolled within-transform with cluster-robust errors rather than an off-the-shelf package, and the stationary block bootstrap is written out in full, both so the few-pairs corrections can be read line by line. Inference is deliberately heavy because the signal is small: a block bootstrap and a Newey-West HAC standard error on every level, cluster-robust errors by pair on the panel, headline computed on weekly means with daily and monthly as robustness. When the live Bundesbank workbooks are unreachable, the pipeline falls back to a clearly labelled synthetic panel and the dashboard says so; the figures above are live data, with no banner. Everything refreshes weekly through GitHub Actions, which re-pulls the data, reruns the tests, and redeploys this page.
Limitations
The liquidity result is the honest headline and also the main caveat. The raw greenium is solid, but the split between green preference and liquidity rests on within-pair variation in issue size, and the adjusted estimate is imprecise enough that I will say liquidity explains most of the gap without insisting the residual is exactly zero or negative. A richer liquidity model, with turnover and bid-ask data the public feeds do not carry, could refine that split.
The evidence is single-country by design. The German twins are the cleanest greenium identification available, but they are German federal bonds, and there is no second issuer here to show the result generalises. That is a deliberate trade of breadth for cleanliness, not an oversight.
The maturity labels are derived from each bond’s residual life when it first appears in the data, so a bond first seen well after its issue, like the 30-year Green Bund whose history here starts in 2021, can read a year or two short of its nominal tenor. It is cosmetic and does not touch the spread. And the twin design, clean as it is, measures a German federal greenium specifically. Whether the same near-zero liquidity-adjusted result holds for corporates, other sovereigns, or lower-rated issuers is a separate question this data cannot answer.