Rebuilt from primary sources · 30 July 2026

German pensions, and what actually changed

Deutsche Bank Research published “Private pension reform in Germany — a bold move towards capital markets” on 19 February 2026. This page rebuilds it: every chart re-fetched from the original provider, both of its models re-implemented and checked against its published outputs, and its description of the draft law checked against the law that Parliament actually passed six weeks later.

model checks: —
3
pillars, one of which does almost all the work
Public transfers are 66% of the income of Germans aged 66+
€127bn
federal top-up to the state pension in 2026
24.3% of the core federal budget
14.7m
Riester contracts left, from a 16.6m peak
Falling every year since 2017
1.0%
cost cap on the new standard product
The paper says 1.5% — Parliament cut it
Start here

The paper describes a draft. The draft is now law — and it changed.

DB Research were writing about the cabinet bill of 17 December 2025. The Bundestag passed an amended version on 27 March 2026, the Bundesrat consented on 8 May, and it was promulgated as the Altersvorsorgereformgesetz in BGBl. 2026 I Nr. 156 on 29 May 2026. Products go on sale from 1 January 2027. Four of the paper's headline parameters moved:

Cost cap: 1.5% → 1.0%

And it is not a fee on assets. § 2a AltZertG defines Effektivkosten as a reduction-in-yield computed like the total-cost indicator in Annex VI of Delegated Regulation (EU) 2017/653 — all-in, look-through, including the fund's own TER. That is a materially tighter constraint than the paper's reading.

Grants: max €480 → €540, child bonus rate quadrupled

50 cents per euro on the first €360 (was 30 cents to €1,200), then 25 cents to €1,800. The child bonus rate went from 25 cents per euro to €1 per €1 — though the €300-per-child cap is unchanged, so the maximum did not move. A parent of one child saving €25 a month now collects €450 of state money on €300 saved.

The self-employed are in

The paper says they are “mostly ineligible”. New § 10a Abs. 1 Satz 5 EStG brings in anyone under 67 with trade or professional income who files a tax return, plus employees in professional pension schemes. Cost to the exchequer: €350m a year.

Still opt-in — but that may not be the end of it

The paper's central criticism survives intact: there is no auto-enrolment, despite the European Commission recommending it in November 2025. On the third pillar as legislated, take-up is the binding constraint.

Since the paper was written, though, the debate has moved somewhere the paper does not go. The Alterssicherungskommission reported on 23 June 2026 recommending a compulsory funded component inside the statutory pension itself — an extra 2 percentage points of contributions, split employer/employee, phased from 2028, on the Swedish premium-pension model. The coalition committed on 2 July to implement the report "fully and promptly", with legislation targeted for the end of 2026. If that happens it would dwarf the Altersvorsorgedepot, and the opt-in problem would be solved by going round it rather than through it.

Part 1

How the German system works today

Germany's retirement system is conventionally described as three pillars. In practice it is one large pillar and two small ones. The statutory pension is pay-as-you-go: today's contributions pay today's pensions, and nothing is saved. The other two are funded — money goes into assets and grows.

Pillar 1 · pay-as-you-go

Gesetzliche Rentenversicherung

  • Compulsory for employees. 18.6% of gross pay up to €101,400 (2026), split evenly employer/employee.
  • 40.1m active insured; 26.1m pensions in payment.
  • Federal budget tops it up by €127.4bn in 2026 — a quarter of the core budget.
  • Average pension actually paid: €1,289 a month net of health and care premiums, across all pension types — roughly €1,470 on a gross basis.
Pillar 2 · occupational

Betriebliche Altersversorgung

  • Employer-sponsored, voluntary for the employer in most cases.
  • 51.9% of employees covered (2023) — 18.1m people, whole economy incl. public-sector schemes. The rate has been drifting down since 2020.
  • €728bn of assets at end-2023, but that is the private sector only (aba); a companion series prints €704.6bn. The two figures are on different scopes — worth knowing before dividing one by the other.
  • Politically hard to expand: more of it means higher non-wage labour costs.
Pillar 3 · private

Riester → Altersvorsorgedepot

  • Voluntary, subsidised, individual. This is the pillar being reformed.
  • 14.66m Riester contracts at end-2025; a fifth to a quarter of them dormant.
  • Median private pension claim per household that has one: €20,300.
  • From 2027: no guarantee required, up to 100% equity, and a 1.0% cost cap — on the standard product only; everything else is uncapped.

What your state pension is, arithmetically

The German formula is unusually legible. Every year you earn Entgeltpunkte — pay points. Earn the national average wage and you get exactly one point. Earn half, you get 0.5. At retirement, your points are multiplied by a single national euro amount, the aktueller Rentenwert, which is set by regulation each July.

monthly pension = Entgeltpunkte × Zugangsfaktor × Rentenartfaktor × aktueller Rentenwert
Zugangsfaktor = 1.0 at the standard retirement age (−0.3%/month early, +0.5%/month late) · Rentenartfaktor = 1.0 for an old-age pension · aktueller Rentenwert = €42.52 from 1 July 2026 · average wage 2026 = €51,944 · contribution ceiling = €101,400

The benchmark politicians argue about — the Standardrente — is 45 years at exactly average pay: 45 points, or €1,913 a month gross from July 2026. Very few people have that record. The average payment per pensioner is €1,289 a month, but that is net of health and care premiums and averages across old-age, disability and survivors' pensions — about €1,470 gross. Compare like with like before drawing the gap.

Part 2

Why the pay-as-you-go pillar is under pressure

A pay-as-you-go system is a claim by one generation on the next. Its arithmetic depends on the ratio of people paying in to people drawing out — and on Germany's demography, that ratio is deteriorating for structural reasons that no plausible immigration path reverses.

Germany's fertility rate fell to 1.32 children per woman in 2025 — Destatis published that on 1 July 2026, and it is the lowest since 2006. The baby-boom cohorts are retiring now. On Destatis's latest projection the number of people aged 20–64 per person aged 65+ falls from 2.44 in 2026 (2.58 in 2024) to 1.93 by 2040. The chart runs on the Eurostat vintage, which has not yet picked up the 2025 German print.

A correction worth making

The DB note says there will be “only two contributors to the statutory pension for each retiree” by 2040, down from 2½. That is the demographic support ratio (20–64 year-olds per person 65+), not the pension system's own ratio. On the Rentenversicherungsbericht's own equivalence basis the statutory scheme is already at 1.89 contributors per pensioner in 2025, falling to 1.60 by 2039. (On a simple headcount of active insured to pensioners it is 1.87.) The system is further along than the sentence implies.

Two levers hold the system together, and the law currently pins both. The Sicherungsniveau vor Steuern is legally floored at 48%. Read the definition carefully, because the headline number is not what most people assume: § 154a SGB VI compares the standard pension to the average wage with both sides net of social-insurance contributions, and only then before income tax. Take the gross figures — €1,913 × 12 against a €51,944 average wage — and you get 44%, not 48%. The netting on both sides is what makes the ratio 48. The contribution rate has sat at 18.6% since 2018. Something has to give: the government's own projection has contributions rising to 20% from 2029 and the replacement rate slipping below the floor once it expires, while federal transfers keep climbing.

Part 3

Where retirement income actually comes from

This is the single chart that makes the case for reform. Germany is not unusually poor in old age — it is unusually concentrated. Two-thirds of the income of Germans aged 66 and over is a public transfer. Capital income contributes 10%. In Denmark it is 22%, in the United States 23%.

Note what the Netherlands and the United Kingdom do instead: they lean on occupational pensions, at 39% and 30% of older people's income. Sweden splits the difference. Germany has essentially neither a large occupational pillar nor a large private one, so when the public pillar is squeezed there is nothing structural to take up the slack.

Part 4

Riester: what the third pillar has been since 2001

Riester was introduced in 2001 alongside a cut to the statutory replacement rate — the explicit bargain being that the state pension would shrink and subsidised private saving would fill the gap. It half-worked. Uptake peaked at 16.6m contracts in 2017 and has fallen every year since.

The design flaws compounded. A statutory guarantee that payouts must at least equal contributions forced providers into low-yielding safe assets. Through the 2012–22 low-rate period that guarantee became nearly impossible to fund once costs were deducted, and new sales collapsed. Costs were high and opaque; grants were conditional on a 4%-of-income minimum contribution that people fell foul of; payouts are fully taxed. Estimated returns range from negative to about 3% nominal on net contributions — often negative in real terms.

The dormancy problem — and how to count it honestly

BMAS estimates that between a fifth and a quarter of Riester contracts no longer receive contributions; the Sachverständigenrat puts it at about a quarter. The DB note narrows that to “20–25%”; this page would not narrow it further.

The gap between contracts and people is the more useful measure of reach — but it needs three caveats the DB note does not give. There were 14.66m contracts at end-2025 and 9.18m people receiving a grant for contribution year 2023: a 5.5m gap, not the 6m DB cite. Those are different units — one person may hold two subsidised contracts — and different years. And the 2023 grant figure is provisional: applications for that year ran to the end of 2025, so it will be revised up. Against roughly 37m eligible people, peak Riester take-up was about 45% of contracts to eligible persons, which is not the same as 45% of people.

Part 5

Germans save a lot. They just save it badly.

German households put away 20.0% of disposable income in 2024 — among the highest rates in Europe and well above the EU-27 average of 14.5%. The problem is not the quantity of saving. It is where it goes.

Over a third of German household financial wealth sits in bank deposits, and 28% more in insurance and pension entitlements — much of it traditional life policies. Together that is 66% of retail financial wealth in instruments that have paid almost nothing for two decades, on the Eurostat 2024 numbers plotted above. (The ECB's quarterly sector accounts, which run to a later quarter and use a slightly wider household sector, put the same pair at 62.9% for 2025-Q4; DB print 65%. All three are saying the same thing.) The chart below rebuilds what that has cost, using ECB interest-rate statistics rather than a vendor index.

This one reproduces DB exactly

Compounding the ECB's monthly interest rates on German household deposits, weighted month by month by the actual outstanding balances in each deposit type, gives +26.0% cumulative from January 2003 to December 2025 — an annualised 1.01%. The DB note prints +26%. Over the same window an equal-weighted mix of German and global equities returned about +770%; DB print +767%.

Part 5b

Who actually holds the money — and is it DB or DC?

Everything above is about the third pillar, because that is what the paper is about. But the money Germany has already saved for retirement does not sit there. It sits in occupational schemes and in the professional funds, and the institutions that hold it are the ones a reform has to move. This part maps them.

The headline answer to the second question is counterintuitive, so it is worth stating plainly before the detail: German occupational pensions are defined benefit, almost universally, even where the product looks like DC.

€728bn
occupational assets, private sector
Year-end 2023, all five vehicles
€295bn
in 91 professional schemes
Germany's second-largest funded pool
€41bn
held by the statutory pension
A liquidity buffer, not a portfolio
0.15%
of occupational assets are true DC
Eight years after the law created it

The money is not where the pillar labels suggest

Pillar one is the system — 40.1m contributors, 21.5m pensioners, a €127bn federal transfer — and it holds essentially no assets. Its entire reserve is a liquidity buffer that §§ 158 and 216 SGB VI require to stay short and liquid, and it is being drawn down: €41.3bn at end-2025, €38.7bn two months later, heading for roughly one monthly outlay by end-2026. That depletion is what forces the contribution rate to 19.8% in 2028.

Meanwhile the 91 berufsständische Versorgungswerke — compulsory schemes for doctors, lawyers, architects and the other regulated professions, whose members are exempt from the statutory pension entirely — hold €295bn and are filed under pillar one in most descriptions. They are the second-largest pot of investable retirement money in the country.

The five routes, and what each one is obliged to do

German occupational pensions run through five Durchführungswege. The choice determines the regulator, the investment rulebook, the insolvency backstop, and which promise types are legally available. Two of them are supervised by nobody.

The regulatory arbitrage worth knowing

A Pensionskasse is bound by the Anlageverordnung's quantitative quotas. A Pensionsfonds — doing substantially the same job — is not: it runs on the prudent-person principle under §§ 236–240 VAG. That is why the Pensionsfonds is the highest-equity vehicle, why corporate de-risking transfers head there, and why it is one of only three routes permitted for the Sozialpartnermodell. The Unterstützungskasse escapes supervision altogether, for the technical reason that it grants no enforceable claim of its own — the claim runs against the employer.

Defined benefit or defined contribution?

BetrAVG recognises four promise types, and only the last is DC in the Anglo-Saxon sense. The reason is § 1 Abs. 1 Satz 3 BetrAVG: the employer steht für die Erfüllung ein — stands behind the promise — regardless of which vehicle holds the money. If the Pensionskasse cuts benefits, the employer tops up. That is exactly what happened when the Pensionskasse der Deutschen Wirtschaft cut benefits and the shortfall landed back on sponsoring employers.

Three things a non-German reader gets wrong

Germany is DB by assets, structurally. Direktzusage (46.2% of assets) and Unterstützungskasse (5.6%) can legally host only DB promises — 51.8% of the market before you look at anything else. Most of the rest is beitragsorientierte Leistungszusage, which converts a contribution into a guaranteed benefit and is a defined benefit obligation under IAS 19.

True DC exists and is rounding error. The reine Beitragszusage, created in 2018 to import a Dutch-style collective-DC design, holds about €1.1bn — roughly 0.15% of occupational assets. It requires a collective bargaining agreement, and IG Metall, the largest union, has refused. Only the chemical industry built a real industry-wide scheme.

The public and professional schemes are “adjustable DB”. Versorgungswerke and the VBL/ZVK points models look and are managed like DB, but the benefit can be cut by amending the scheme statute — and Versorgungswerke have done so. Economically that is closer to Dutch collective DC than to a guaranteed corporate plan. Unlike corporate DB, there is no PSVaG and no state guarantee behind them.

The named asset owners

Ranked by assets where disclosed. Confidence flags are on every figure: the professional and corporate aggregates are well documented, the municipal ZVK aggregate is not.

Assets are not additive down the table — the aggregates overlap with the named members. BVK is inside the €295bn Versorgungswerke figure; BVV is inside the €210bn Pensionskassen figure.

Part 6

What the reform actually does

The centrepiece is the Altersvorsorgedepot — an old-age provision account holding funds rather than a guarantee product. Below, each element of the reform as Riester has it today, as the cabinet draft proposed it (the version the DB note assesses), and as the enacted law states it.

Where the design is still weak

Opt-in. Everyone must actively choose. Germany's financial literacy is above the OECD average on the headline score (76 vs 63), but only 35% of adults compare offers across providers, 37% seek independent advice, and 26% cannot correctly answer the OECD's compound-interest question.

Guarantee products survive. 80% and 100% capital-protected products remain certifiable. Given the history, a large share of savers will pick them and earn very little.

“Standard product” is a misnomer. The Bundesrat objected to the draft on exactly this point — it permitted “millions of fund combinations”, so the products would not be comparable to each other. The enacted text did not change that, but it did add a power for the government to create a genuinely single public standard product by decree. No decree exists yet.

Risk class 5 has a fire-sale problem. The SRI is calibrated on past volatility. A fund that breaches the class-5 ceiling in a crisis must be replaced — potentially forcing sales at the bottom and missing the recovery.

Certification is provisional for longer than it looks. The general rule is a two-year revocation reservation on a self-certified product. But a transitional rule (§ 14 Abs. 7 AltZertG) extends that to four years for every application filed up to 31 December 2028 — which is the entire launch window. And certification against the new rules cannot be granted before 1 January 2027, so providers cannot get 2027 products approved during 2026 after all.

Part 6b

The cost of costs

This is the reform's real economics, and DB's Figure 10 in interactive form. A percentage fee sounds small; compounded over a 40-year accumulation phase it is the difference between a pension and a rounding error. Every input below is yours to change.

Method: contributions monthly in advance; the fee is applied multiplicatively, so the net growth factor is (1 + gross) ÷ (1 + cost) — that is how an ad-valorem charge on assets actually works, and it reproduces DB's published outputs to within about 1%. “Cumulative fees” sums the annual charge levied on the running balance, not the compounded opportunity cost. The Rentenfaktor converts capital into a monthly annuity: 25 means €25 a month per €10,000 of capital, a deliberately conservative assumption.

Part 7

What the systems Germany is copying actually look like

The DB note benchmarks against six schemes. Two of them are not third-pillar schemes at all, which matters for what Germany can hope to learn from them.

Two classification corrections

Sweden's premium pension is not a private pension. It is a mandatory, funded, individual-account slice of the public system — 2.5% of pensionable income, alongside 16% going to the pay-as-you-go part. Its 62% coverage and 0.05% fees come from being compulsory and state-run, not from good product design. Sweden's genuine voluntary third pillar lost its tax deduction in 2016 and is residual.

Poland's PPK is occupational. Employer-run, auto-enrolled, with employer and state money in it. Poland's actual third pillar is IKE/IKZE. PPK is a closer analogue to UK auto-enrolment than to the Altersvorsorgedepot.

Two lessons stand up, though both need care with denominators. First, defaults and compulsion drive coverage far more than tax incentives do. Sweden's premium pension reaches 62% of the whole population because it is compulsory for everyone with pensionable income; UK auto-enrolment reaches 89% of eligible employees (about half the 16–64 population, and that figure is Great Britain rather than the UK); purely voluntary personal pensions in France and the UK reach 7–10% of the population. Those are three different bases — but no voluntary scheme anywhere in the comparison gets near the compulsory or auto-enrolled ones.

Second, scale drives costs down — with the caveat that the three numbers usually quoted side by side are not the same measure. AP7's 5 basis points is a fund management fee. Estonia's 61 is an asset-weighted TER. Germany's 100 is an all-in reduction-in-yield including the fund's own TER, administration, distribution and, for insurance wrappers, insurance costs. On a like-for-like basis Germany's cap is tighter than the raw comparison makes it look. The direction of the finding survives; the size of the gap does not.

One nuance the DB note gets wrong: it attributes Poland's participation jump to auto-enrolment being “introduced in 2023”. Auto-enrolment has applied since PPK launched in 2019. 2023 was the first of the statutory re-enrolments that happen every four years — everyone who opted out is put back in and has to opt out again. The next is April 2027. That is the mechanism worth copying, and it is a different one.

Part 8

The macro argument, and how much it is worth

Beyond individual outcomes, DB argue that a funded pillar creates a pool of genuinely long-term capital — money locked up for thirty years or more, which is what finances scale-ups and infrastructure. The cross-country pattern is real: economies with big household claims on insurers and pension funds have deeper equity markets.

A scope note on this chart. DB put the US at 198% of GDP; on the OECD's harmonised national-accounts basis it is 143.5%. The difference is definitional — the Federal Reserve's flow-of-funds measure sweeps in IRAs, annuities and government employee retirement funds that the OECD's sector S.129 for the US does not. Germany's 74.3% reproduces DB's 75% almost exactly. The UK does not report pension funds separately in the OECD dataset, so it is omitted rather than guessed at.

How much new money would actually reach European equities?

DB's back-of-the-envelope gets to “more than €8bn a year”. It is a chain of assumptions, each of which is arguable — so here it is with every link exposed.

Part 8b

What comes next, and where the flow actually goes

The paper assesses a third-pillar draft. Six weeks after it was published that draft became law — and three weeks after that, on 23 June 2026, the Alterssicherungskommission reported with 33 recommendations that are substantially larger. The coalition committed on 2 July to implement them in full, with legislation targeted for end-2026 and entry into force in early 2027.

Sizing the mandatory funded pillar

The centrepiece is a gesetzliche Kapitalrente: a parity-financed contribution of two percentage points of gross pay, into individual capital accounts, centrally administered and invested on the Swedish premium-pension model. Work the arithmetic through the contribution base and it is the largest prospective flow anywhere in the German system.

The flow channels, and the one that runs the other way

Not every channel points the same direction. Corporate de-risking is the counterweight, and over the next two to three years it is the larger and more certain of the two.

The structural point

Germany's household and institutional equity exposure is low because the guarantee architecture made it so, not because of preferences. Riester's 100% capital guarantee forced providers into long fixed income. The Anlageverordnung caps risk assets for Pensionskassen. § 1 Abs. 1 S. 3 BetrAVG keeps employers liable and therefore keeps sponsors conservative.

Every significant reform now in flight — the unguaranteed depot, the reine Beitragszusage, the Swedish-model Kapitalrente — attacks the guarantee rather than the quota. That is the correct read of the direction of travel, and it is why the February 2025 increase in the Anlageverordnung risk quota, from 35% to 40%, produced so little: the professional schemes' equity allocations went on falling.

Part 9

Rebuilt vs published: every number checked

Each row is a figure from the DB note set against what the primary source says today. The note's quantitative core holds up unusually well — two of its headline numbers reproduce to the decimal. The soft spots are survey figures, one uncited paper, and a few scope mismatches.

Model validation

Both of the note's quantitative models were re-implemented from scratch and run against the outputs it printed.

Part 10

Sources and method

Nothing on this page is transcribed from the DB note's charts. Every series was re-fetched from the provider's own API or publication, and the two models were re-implemented and validated. Where a figure could not be verified from a primary source it is labelled unverified rather than reproduced.

How it was built

  • Eurostat dissemination API — saving rates, financial balance sheets, GDP, demography, EUROPOP2023 projections.
  • OECD SDMX — Pensions at a Glance income composition; Institutional Investors' Indicators.
  • ECB Data Portal — MIR deposit rates, BSI balance-sheet stocks, Quarterly Sector Accounts.
  • German primaries — BMAS Riester series, BMF subsidy statistics, DRV time series, Bundeshaushalt, Destatis projections, Bundesbank PHF, Deutsches Aktieninstitut, BVI.
  • Legal — BGBl., Bundestag Drucksachen 21/4088 and 21/4996, AltZertG and EStG as amended.
  • International — AP7, Pensionsmyndigheten, PFR Portal PPK, Pensionikeskus, ICI, IRS, HMRC, DWP, IRD/FMA New Zealand.

Caveats

  • DB's Figure 8 uses WFE market capitalisation; this page uses the national-accounts equivalent (ESA F.511 at market value) so both axes come from one source. Levels differ; the pattern does not.
  • Cost figures for the UK, France, Poland and Estonia in DB's Figure 13 could not all be traced to their stated regulators. Only verified values are plotted.
  • The Malmendier et al. (2025) paper underpinning DB's life-insurance return line could not be located. DB themselves label that line “indicative”.
  • OECD pensioner-income data is for 2022 — the latest available vintage.

Take the data with you

Every chart has a data table you can edit in place and a CSV button. You can also take the whole thing at once — the bundle below is exactly what this page is drawing from.

The accompanying archive contains the fetch scripts (fetch_eurostat.py, fetch_oecd.py, fetch_ecb.py), the re-implemented models with their validation suite (models.py — run it and it prints the 16 checks against DB's published numbers), the dataset builders, this page's source, and the raw API payloads under data/raw/ so every number can be traced back to the response it came from.

Source list