Ida May Fuller paid into Social Security for three years. She collected benefits for thirty-five. That gap between what she put in and what she took out was not a mistake — it was the design. The 1935 Act paid the first generation before there was a first generation's worth of contributions to pay them with. Someone had to cover the difference, and by construction it wasn't Fuller's cohort. It was every cohort that came after her, folded quietly into a payroll tax rate that never once appeared on the federal balance sheet as debt.
That's the first unbooked liability. It has a name in the actuarial literature — the legacy debt, or the closed-group transition cost of the founding generation. Estimates place its present value at $29.7 trillion. That is not the famous Social Security "unfunded obligation" number you see in headlines about future insolvency — this is the backward-looking bill for a decision made once, in 1935, to launch the system on a pay-as-you-go basis instead of pre-funding it. Every worker since has served a share of that debt through payroll withholding, without a bond ever being issued, without a debt-service line ever appearing in a budget document.
The second liability is structurally identical and gets almost no attention at all: the tax expenditure budget. Every preferential rate, deduction, or exclusion the tax code grants is revenue the Treasury chose not to collect — which is fiscally indistinguishable from the Treasury collecting it and mailing a check back. The Joint Committee on Taxation puts this year's tax expenditures at $2.3 trillion, projects $11.7 trillion over the next five years, and the concept itself is old enough that Stanley Surrey coined the term in 1968, when it already ran to 4.5 percent of GDP. It has been as high as 9 percent. It sits at 6.4 percent today.
Chart 13A puts the two side by side. They are the same species of liability — a government choosing not to book a transfer as debt — measured on different clocks. One is a single historical accident, now fully accrued. The other resets every year, indefinitely, by design.
To be precise about what's being compared: this is not an argument that Social Security is unfunded. It isn't. Payroll contributions cover 91.2 percent of current program income — the system is, overwhelmingly, workers paying for their own eventual benefit, exactly as the contributory design intends. The legacy debt is the exception carved out of that design, not evidence against it: a real, one-time gap between what the founding cohort paid in and what it drew out, a gap that should have been issued as explicit federal debt in the years it was actually incurred and retired on a schedule, rather than folded permanently into everyone's payroll tax rate. Tax expenditures carry no equivalent contributory relationship at all — no worker, of any income level, pays a premium to earn a capital-gains preference. One liability sits on top of a real contract, partially violated once. The other has no contract underneath it anywhere.
What should have happened
The counterfactual for Social Security is straightforward, and Treasury has done exactly this kind of thing before. The 1935 transition cost should have been issued as an explicit, amortized federal obligation — a Treasury bond series dedicated to funding the founding cohort's benefits, serviced out of general revenue like any other piece of the public debt, and retired on a fixed schedule. That would have done two things a hidden payroll surcharge never can: put the true cost of the political choice on the books where voters and bond markets could see it, and separated it cleanly from the ongoing, actuarially-linked contributions of everyone who came after. Instead, the transition cost was hidden inside a tax that looks, to this day, like a personal pension premium. It isn't one. Roughly a third of what today's payroll tax collects is still debt service on an obligation nobody ever called debt.
The counterfactual for tax expenditures is equally straightforward and considerably more radical in what it would require politically: score them exactly as Congress scores an appropriation. Put each one through the same PAYGO test, the same sunset review, the same annual up-or-down vote that a housing subsidy or a SNAP allotment gets. Chart 13B shows why that would matter. The preferential rate on capital gains and dividends alone cost $225 billion in a single year — more than the entire earned income credit and the Affordable Care Act exchange subsidies combined. Step-up basis at death and the pass-through deduction add another $119 billion between them. None of these three provisions faces the reauthorization scrutiny that a housing voucher program faces every single appropriations cycle.
The asymmetry, in one picture
Chart 13C shows the tax expenditure budget's share of GDP across nine decades. It is not a flat line and it is not a rounding error. It moved from under one percent in the New Deal years to roughly nine percent before the 1986 reform clawed some of it back, and it sits at 6.4 percent now — larger, on its own, than most G7 countries' entire defense budgets as a share of output.
Compare the political treatment. Social Security's legacy debt was baked into a payroll tax that funds an "entitlement" — a word that, in current usage, functions as an accusation. Every reform conversation starts from the premise that the benefit is the thing under negotiation: raise the retirement age, adjust the COLA formula, means-test the payout. Nobody proposes retroactively un-paying Ida May Fuller, but the framing places the burden of adjustment entirely on the recipient side of the ledger.
The tax expenditure side runs the opposite framing. A capital gains preference or a step-up-basis exclusion is called "tax policy" — treated as the neutral baseline, not a transfer, so that removing it isn't "cutting a benefit," it's "raising taxes." The 2017 and 2025 rounds of tax legislation expanded this side of the ledger substantially — new exemptions for tip and overtime income, an added senior deduction, factory expensing — with none of the "entitlement crisis" language that accompanies even routine Social Security actuarial updates.
Both are the state declining to collect. Both were created by legislative choice, not natural law. Only one gets called a crisis.
Projecting the difference
If the 1935 transition cost had been financed as explicit Treasury debt from the outset, amortized over — for illustration — a 60-year horizon at historical average Treasury borrowing costs, today's OASDI payroll tax rate could very plausibly be a meaningfully smaller share of covered wages than the 12.4 percent combined rate workers and employers pay now, with the difference visible every year as a debt-service line item rather than buried inside a "contribution." Whether that number is one point or three points of payroll depends on assumptions about the interest rate path and amortization schedule used, and that modeling — a proper actuarial simulation, not a napkin estimate — is the next piece in this series.
The tax expenditure side of the counterfactual is more direct: if the $2.3 trillion currently exempted annually were instead collected and a portion returned as explicit, means-tested, sunset-reviewed transfers — the way the earned income credit already is — the federal government would not be poorer. It would simply know, in a way it currently doesn't force itself to know, who receives $29.7 trillion worth of transfer over time and why. That is the entire point of Haig-Simons comprehensive income accounting: not to raise or lower the tax burden by definition, but to put every transfer, however delivered, on the same ledger, subject to the same test of public justification.
Four architectures, one test
The coherence test this piece is applying to Social Security and the tax expenditure budget — no claim without contribution, no confiscation of contribution without a corresponding claim — also applies cleanly at the level of individual estate and investment planning, where it's easier to see because the mechanisms are more visible and the actors fewer. Four structures, all lawful, all in active use, show how differently that test resolves depending on design rather than intent.
Architecture one: step-up basis at death, held directly. An individual holds appreciated property — stock, real estate, a business — until death. Unrealized gains accumulated over a lifetime are never taxed to the decedent, and the heirs' basis resets to fair market value at the moment of death, erasing the built-in gain entirely rather than deferring it. This is the mechanism behind the $59.7 billion annual figure already shown in Chart 13B. It fails the coherence test outright: a real economic gain existed, was never realized as income to anyone, and is permanently exempted from tax rather than merely deferred.
Architecture two: pass-through entity with a basis election. The same appreciated assets are instead held inside a family limited partnership or LLC taxed as a partnership. Ordinary income the entity earns — rent, interest, realized gains from asset sales — is taxed annually to the partners with no exception; there is no shelter on current yield. But unrealized appreciation on assets the entity still holds gets the same basis reset at the death of a partner as architecture one, now applied at the entity level through a formal election, and frequently compounded by valuation discounts of 20–40 percent on the decedent's partnership interest, since a minority stake in an illiquid, closely-held entity is worth less on paper than its pro-rata share of the underlying assets. This structure doesn't just preserve the step-up benefit — it can enlarge it. It fails the coherence test more thoroughly than architecture one, because the discount subtracts value from the taxable estate that was never actually lost.
Architecture three: a foundation holding corpus permanently. Assets are irrevocably endowed into a foundation — a legal person that does not die, so there is no death-tax trigger to plan around in the first place. The endowing act is itself a taxable event: the assets were already taxed as ordinary income before being contributed. Thereafter, the corpus grows inside the foundation untaxed on paper gains — the same treatment any unrealized holding gets before sale — while every distribution to a beneficiary is taxed in full as ordinary income at the point of receipt, provided the foundation's accounting distinguishes distributed principal (already taxed once, not taxed again) from distributed growth (taxed on realization). Done honestly, this passes the coherence test: each dollar is taxed exactly once, at whichever point it first becomes realized income to a person. The mechanism that most often breaks this symmetry in practice is cross-border — a beneficiary's country of residence may apply throwback or accumulated-income surcharges precisely because it wants its own claim on growth realized inside a foreign entity it never got to tax as it accrued.
Architecture four: capital deployed into an operating business, no wrapper advantage sought. Capital — including capital held inside a structure like architecture three — is invested directly into a productive operating company: a factory, a manufacturing line, an active business with employees. Every layer is taxed at the point it's realized and nowhere else: corporate profit at the standard corporate rate, no accelerated shelter beyond ordinary depreciation any operator in the sector takes; a manager's or director's wages at ordinary personal rates, with no capital-gains treatment applied to labor income; profit distributed to the capital provider taxed — or exempted — under whatever the standing corporate distribution rule happens to be in that jurisdiction, applied identically to every investor in the country, not engineered for this specific investor. Nothing in the structure is built to survive a future change in that distribution rule; if the exemption is narrowed or repealed, the entity simply pays more, the way it would have from day one under the new rule. This is the cleanest pass of the four: no discount, no reset, no lock-in against future policy — just ordinary tax at each point value is actually created and realized.
The pattern across all four: the legal form doing the sheltering — a death-triggered basis reset, a valuation discount, favorable cross-border treatment, a locked-in preferential rate — is available in rough proportion to the capital and professional access required to acquire it. A wage earner has no equivalent instrument for any of these. That's not an argument that any of the four is illegal or even, in the case of three and four, improperly designed — it's the same observation the tax-expenditure and legacy-debt material already makes about the federal ledger, now visible one balance sheet at a time: coherence requires applying the same test to a $30 million estate plan that this piece applies to a $30 trillion federal liability.
The point
This isn't an argument that Social Security is illegitimate or that every tax preference is indefensible. Some of both are well-justified transfers a democratic society chooses to make. The argument is narrower and, I think, harder to dismiss: the same accounting standard has to apply to both, or the debate isn't about economics — it's about which beneficiaries have the lobbying power to keep their transfer off the books. Convergence, not conspiracy: nobody had to plan this asymmetry. Concentrated, well-organized recipients of tax expenditures defend their transfer year after year in obscure committee markups nobody watches. Diffuse Social Security recipients get a headline "crisis" every time the trustees' report comes out. That outcome doesn't require coordination. It only requires letting one set of numbers stay invisible while the other stays permanently in the news.
Nothing happens by accident.
Legacy debt estimate from Leimer (2016), "The Legacy Debt Associated with Past Social Security Transfers," Social Security Bulletin 76(3), as summarized in CRR Issue Brief 19-9. Tax expenditure figures from the Joint Committee on Taxation's FY2025–2029 tax expenditure report (JCX series) and the Tax Policy Center's ranking of top tax expenditures, FY2024 (JCT/Treasury estimates, updated January 2024). Social Security funding ratio (91.2% payroll-tax financed) from the SSA 2025 OASDI Trustees Report. GDP-share historical benchmarks (1968: 4.5%; 1985: ~9%; 2024: 6.4%) from CEP Web, "Advancing Scrutiny: 55 Years of Tax Expenditures Reporting." Chart 13C interpolates between documented benchmark years and should be read as illustrative of trend and magnitude, not as a year-by-year historical series — a full reconstruction from annual Treasury/JCT reports is a candidate for a follow-up piece. The projected payroll-tax counterfactual is directional; a full actuarial amortization model is needed before publishing a specific point estimate.