Reducing the Federal Debt

The four numbers that carry the argument, before any of the detail. Debt held by the public was $30,172 billion at the FY2025 close, the last audited fiscal-year close. Under CBO’s February 11, 2026 baseline and no policy change it reaches 120.2 percent of GDP in FY2036 with the unified deficit still widening at 6.7 percent of GDP. Merely holding the ratio flat costs 2.39 percent of GDP, permanently — $1,117 billion a year at full strength. And net interest, the line nobody votes on, grows from $1,039 billion in FY2026 to $2,144 billion in FY2036 on its own.

The number in the headline

1.1 Two debts, not one

The federal government reports two debt figures, and they are not interchangeable. Total public debt outstanding, often called gross debt, is every Treasury security in existence. It includes the securities held by the government’s own trust funds — Social Security, Medicare Part A, the military and civil service retirement funds — which are, in economic substance, one part of the government owing another part. Debt held by the public strips those out. It is what the Treasury owes to everyone who is not the Treasury: households, pension funds, insurers, mutual funds, banks, foreign central banks, and the Federal Reserve.

The gap between the two is intragovernmental holdings. In CBO’s own February 2026 projections it is $7,335 billion at the end of FY2026 — gross federal debt of $39,430 billion against debt held by the public of $32,095 billion, both measured on the February 11, 2026 baseline. That is not a rounding difference. A reader who takes the gross number from a headline and divides it by GDP will produce a debt-to-GDP ratio twenty-three points higher than the published one, and will then compare it against ratios computed on the other basis.

1.2 What crossed forty trillion, and when

The figure that crossed forty trillion was gross debt. The Treasury publishes it daily in the Debt to the Penny series, and the first print above the line came at the record date of August 18, 2026, posted the following morning. The House Budget Committee marked the milestone the same day. Five months earlier, in March 2026, the same series had crossed thirty-nine trillion.

Debt held by the public, the series this document projects, stood at $30,172 billion at the end of fiscal year 2025, which is the last audited fiscal-year close and the launching point for every projection that follows.

1.3 Why this document measures debt held by the public

Three reasons, and they are practical rather than ideological. First, debt held by the public is the number that has to be sold into a market and refinanced at a market price, so it is the number that generates the interest expense on the government’s income statement. Second, it is the series the Congressional Budget Office projects and the series every published debt-to-GDP comparison uses, domestically and internationally, so it is the only basis on which this work can be checked against anyone else’s. Third, intragovernmental holdings move for reasons that have nothing to do with fiscal policy — a trust fund drawing down its balance reduces gross debt while the government’s actual position is unchanged — which makes gross debt a noisy measure of the thing we are trying to measure.

1.4 The first two-clock problem

CBO’s February 2026 baseline projects gross federal debt of $39,430 billion at the end of fiscal year 2026. Gross debt printed at $40,047 billion on August 18, 2026 — six weeks before that fiscal year closes, and $617 billion above a level the projection did not expect to be reached until the year was over. The projection was six months old at the time.

This is not a criticism of CBO. The agency publishes its assumptions and its cut-off dates in the front of every volume, which is precisely what allows the comparison to be made at all. It is an illustration of the rule this Institute applies to every figure it prints: a number is a measurement taken at a moment, by someone, under stated conditions. When any of those three change, the number changes, and a document that carries only its own publication date gives the reader no way to see it.

The reporting layer and the underwriting layer

Since 2017, USAFacts — the nonpartisan civic data initiative founded by Steve Ballmer after he left Microsoft — has published an annual Government 10-K. The idea is exactly what the name says. The Securities and Exchange Commission requires every public company to file a Form 10-K describing its financial position, its results, and its risk factors. Government has no such obligation, so USAFacts writes one on the government’s behalf, consolidating federal, state and local finances into a single statement in the form an investor would recognize.

It is the right instinct, and it is the missing half of the work. A 10-K is a reporting document. It is backward-looking by construction, it presents results as they occurred, and it does not adjust. That discipline is the source of its authority: a 10-K is credible precisely because it does not tell you what management wishes had happened.

But no practitioner stops at the 10-K. In diligence, the filed statements are the left-hand column. The work is in the columns to the right — the ones labeled with an adjustment and a reason. Normalize for the items that will not recur. Rebase for a different measurement window. Strip out the double counts. Carry the adjusted figure forward and see what it does over the projection period. That is underwriting, and it is a different job from reporting.

2.1 A two-clock note on the 10-K itself

The most recent edition is the 2026 Government 10-K, and it covers fiscal years 2014 through 2023. The publication clock reads 2026; the measurement clock on the newest fiscal year in the book reads 2023. USAFacts is explicit about why: delayed Internal Revenue Service data meant the report could not be published by Tax Day for the first time, and the organization filed its own version of an SEC Form 12b-25 — the notification a public company files when it will miss a deadline — before releasing the report two weeks later.

That is exactly correct practice, and it is worth saying plainly, because the point of the two-clock rule is not that stale numbers are a sin. Stale numbers are unavoidable when someone else controls the measurement. The sin is carrying them silently. USAFacts carried them out loud, and filed the notice. The distance between the two clocks is also the reason this document builds its projections on CBO rather than on the 10-K: the debt at the end of fiscal 2023, as reported there, was $33.1 trillion gross. Thirty-four months later the same series printed above forty.

The baseline, as filed

Everything that follows is built on one document: the Congressional Budget Office’s The Budget and Economic Outlook: 2026 to 2036, published February 11, 2026 and reposted with corrections on March 24, 2026. The tables below reproduce its Tables 1-1, 1-2 and 1-3 for the years this document projects. No figure in this section is the Institute’s. They are reproduced so that every adjusted number later in the document can be traced back to a published line.

3.1 Four measurement clocks inside one publication date

A single date on the cover of a budget projection conceals the fact that its components were frozen at four different moments. CBO states all four in its own front matter. A reader who takes February 11, 2026 as the as-of date for the whole volume will be wrong about every one of them.

Table 3.1. Source: Congressional Budget Office, The Budget and Economic Outlook: 2026 to 2036, February 11, 2026, front matter. The widest gap inside this single volume is eighty-three days.

3.2 Revenues, outlays, and the total deficit

Table 3.2. CBO baseline: revenues, outlays and the total deficit, FY2026–FY2036 ($ billions). Source: CBO, February 11, 2026, Table 1-1. GDP is the fiscal-year nominal series CBO uses for its own percentage-of-GDP lines; using a calendar-year series here will not reproduce CBO’s published percentages.

3.3 The primary deficit, net interest, and the debt

The total deficit splits into two pieces that behave completely differently. The primary deficit is what the government spends beyond what it collects, before any interest. It is the piece policy controls. Net interest is the price of the borrowing already done, and it compounds whether or not anyone acts — across this window it grows from $1,039 billion to $2,144 billion, more than doubling, on a debt stock that rises by eighty-six percent. Other means of financing is the reconciling item between the deficit and the change in debt: student loan revaluations, coin seigniorage, and changes in the Treasury’s cash balance.

Table 3.3. CBO baseline: the deficit decomposed and the debt path, FY2026–FY2036 ($ billions). Source: CBO, February 11, 2026, Tables 1-1 and 1-3. Debt held by the public at the end of FY2025 was $30,172 billion, which is where the projection in Section 4 begins. Positive primary deficit means a deficit; negative other means of financing reduces borrowing.

3.4 Two published deficit series, and which one belongs in a debt roll-forward

CBO publishes the deficit twice. Table 1-1 is unadjusted. Table 1-2 removes the distortion created when October 1 falls on a weekend and certain payments shift into the prior fiscal year. The two series differ in five of the eleven years in this window, by as much as $178 billion in a single year.

Table 3.4. Where the two published primary-deficit series diverge ($ billions). Source: CBO, February 11, 2026, Tables 1-1 and 1-2.

The model

4.1 The roll-forward identity

Debt at the end of a year is debt at the start, plus the primary deficit, less whatever savings a scenario assumes, plus the interest accrued during the year, plus other means of financing. The only complication is that interest is charged on the debt as it stands through the year, and the debt as it stands through the year depends on the interest. The identity is therefore implicit and has to be solved rather than simply accumulated:

4.2 The interest rate is backed out, not assumed

The effective interest rate on debt held by the public is not an input to this model. It is solved for, year by year, as the rate that reproduces CBO’s own published net interest line on CBO’s own published debt path. That choice removes an entire category of argument: no reader has to accept the Institute’s view of where rates are going, because the Institute does not have one here. The solved rates run from 3.34 percent in FY2026 to 3.92 percent in FY2036, against CBO’s stated 3.4 percent rising to 3.9 percent — both measured on the February 11, 2026 baseline.

4.3 Controls

A model that cannot reproduce the published baseline when told to change nothing is not a model. Run with zero savings, the roll-forward is required to land on CBO’s own debt and interest figures for all eleven years. It does.

Table 4.1. Control checks: the model against the published baseline. The maximum error on an eleven-year, thirty-trillion-dollar debt roll-forward is under three billion dollars, or one hundredth of one percent. It is rounding in CBO’s own published tables, not model error: CBO reports whole billions, and in FY2031 the published revenue, outlay, primary deficit and interest lines differ by one billion dollars among themselves for the same reason.

4.4 What the model does not do

It does not forecast the economy. GDP, inflation and the rate structure are CBO’s, unchanged in every scenario. It does not model macroeconomic feedback from fiscal consolidation — the effect of a five-percent-of-GDP tightening on growth is real, contested, and outside what can be defended from published scores. It does not model a debt crisis, a rate shock, or a change in the willingness of anyone to hold Treasuries. Each of those would move the answer, and each would require an assumption the Institute cannot source. The model answers a narrower and more answerable question: holding CBO’s economy fixed, what size of primary-deficit reduction lands the debt on a stated path?

The adjustments column

This is the section that makes the document a proforma rather than a summary. Published budget scores cannot simply be added up and dropped into a different projection window. Five things have to be done to them first, and each one is stated here with its size, its direction, and its reason. A reader who disagrees with any single adjustment can back it out and reprice the whole document.

ADJ-1 — Window rebasing

Every option score in the inventory was estimated over fiscal years 2025 through 2034. This document projects fiscal years 2027 through 2036. A policy held constant as a share of the economy raises more over the later window simply because the economy is larger. Using CBO’s own published fiscal-year GDP memorandum totals — $367,905 billion for FY2025–2034 and $397,364 billion for FY2027–2036, both from the February 11, 2026 baseline — the rebasing factor is 1.0801. Every score is multiplied by it. The adjustment is mechanical, it is disclosed, and it adds 8.01 percent to every score it touches.

ADJ-2 — The baseline break

Public Law 119-21, signed July 4, 2025, permanently extended the 2017 individual rate structure and rewrote Medicaid and Marketplace subsidy law. Every option score in the inventory predates it — the scores were measured on December 12, 2024 against a June 2024 baseline. Some options are unaffected; others were scored against a baseline in which the 2017 rates were assumed to expire, and would score differently today. Rather than guess at revised scores CBO has not published, this document flags every exposed option and reports the exposed share of each package as a disclosed measurement, so the reader can see exactly how much of a scenario rests on pre-July-2025 law. That share runs from three percent of the Low package to twenty percent of the Medium package.

ADJ-3 — Exclusivity, to prevent double counting

Several options in the published inventory overlap. Limiting itemized deductions and capping the charitable deduction touch the same tax base; capping federal Medicaid spending and cutting the federal matching rate are two ways of doing one thing; a uniform Social Security benefit and a means-tested benefit reduction for high earners are alternatives, not additions. Each such family is assigned to an exclusivity group, and no package may contain more than one member of a group. The build fails loudly if it does. Six groups are enforced: the value-added tax, itemized deductions, the individual rate schedule, a new payroll tax, the Social Security benefit formula, and structural Medicaid.

ADJ-4 — Interest-rate feedback, deliberately held at zero

There is a good argument that a government which credibly reduces its borrowing pays a lower rate on what remains, and that the effect compounds. The model implements it and it is switched off, at zero basis points, in every published table. The reason is simple: turning it on makes every scenario look better than it is, and the size of the effect cannot be sourced to a published score. Leaving it off means the savings reported here are the floor. Anyone who believes in the feedback can only conclude the scenarios are easier than stated, never harder.

ADJ-5 — Phase-in

No policy of this size takes effect at full force on day one. Each scenario begins in FY2027 and phases in over five years at twenty, forty, sixty, eighty and one hundred percent, then holds at full strength as a constant share of GDP so the package grows with the economy rather than eroding against it. The phase-in is why the ten-year savings total is meaningfully less than ten times the full-strength annual figure.

Table 5.1. The adjustments, in one place. Three of the five push the reported savings down and one pushes them up. That is not a design goal; it is what the adjustments happen to do. The point of the column is that they are visible.

Three scenarios

Each scenario is defined by where debt held by the public ends up in FY2036, as a share of GDP, and by nothing else. The model then solves for the size of the permanent primary-deficit reduction that lands there. Defining the scenarios by outcome rather than by policy is deliberate: it separates the arithmetic question of how large the change must be from the political question of what the change should consist of. Section 7 then shows one way, out of very many, to assemble a package of that size from published scores.

Table 6.1. The three scenarios against the baseline ($ billions unless noted), solved on CBO’s February 11, 2026 baseline. The baseline row is CBO’s own projection carried through the model with no policy change. Interest avoided is the compounding effect described in Section 4.1: it is not a policy choice, it is what stops happening once the debt stops growing as fast.

6.1 Low — halt the rise

Target: debt held by the public at 100.6 percent of GDP in FY2036, which is where it stands at the end of FY2026. The debt keeps growing in dollars throughout; it simply stops growing faster than the economy. This is the least ambitious target that can be described as stabilization, and it is not a soft one.

Table 6.2. Low scenario, year by year (target 100.6 percent of GDP in FY2036; $ billions unless noted). Debt-to-GDP peaks at 102.9 percent in FY2029 before turning. The ratio is above its FY2026 level for nine of the ten action years and only returns to it at the very end of the window, which is what a target defined on the terminal year does.

6.2 Medium — a decisive downward path

Target: 90 percent of GDP in FY2036. This is the scenario in which the ratio is unambiguously falling rather than merely not rising. It requires a permanent reduction of 3.68 percent of GDP, which is roughly fifty-four percent more than the Low package and reaches $1,721 billion a year by the end of the window.

Table 6.3. Medium scenario, year by year (target 90 percent of GDP in FY2036; $ billions unless noted).

6.3 High — back to the FY2019 level

Target: 79 percent of GDP in FY2036. At $16.8 trillion, debt held by the public at the end of fiscal year 2019 equalled 79 percent of GDP — CBO’s own figure, published in Federal Debt: A Primer in March 2020 and measured at the FY2019 close, before the pandemic borrowing began. Returning to that level by FY2036 requires a permanent primary-deficit reduction of 5.03 percent of GDP, $2,348 billion a year at full strength, and drives the unified deficit to 0.2 percent of GDP in the terminal year — essentially a balanced budget.

Table 6.4. High scenario, year by year (target 79 percent of GDP in FY2036; $ billions unless noted). A note on what this target is and is not: seventy-nine percent is the pre-pandemic level, not a long-run average and not a prudential standard. Debt held by the public averaged far less than that over the last half century. The FY2019 reading is used here because it is a real, recent, audited level that the United States actually carried, which makes it a defensible target rather than an aspiration.

The lever inventory

Section 6 establishes how large the change has to be. This section shows that a change of that size can be assembled entirely from options that have already been scored by the Congressional Budget Office, without inventing a single number. The inventory contains 76 levers across revenue, mandatory spending and discretionary spending, with no category excluded — Social Security, Medicare, Medicaid, defense, nondefense discretionary, tax expenditures, tax rates, and the federal balance sheet are all in scope.

7.1 Where the scores come from, and when they were measured

Table 7.1. The measurement clock on the lever inventory. This is the second two-clock finding in the document. The volume’s own file date reads October 2025, eleven months fresher than the scores inside it, which are December 2024 measurements taken against a June 2024 baseline. No newer edition exists: the Institute verified this three ways in August 2026 — against CBO’s Budget Options topic page, against its live options search sorted by date, and against the publication record.

7.2 The intensity dial

CBO scores most options as a range, because most options can be written at more than one intensity. Reducing the Department of Defense budget scores one way at a small cut and another at a large one. Rather than pick a point inside each range by hand — which would make the package a set of seventy-six private judgments — this document sets a single dial, theta, between the low and high end of every range at once, and solves for the one value of theta that makes the package equal the savings the scenario requires. A reader who wants to reprice an entire scenario has exactly one number to move.

Table 7.2. The proforma bridge: published scores to the savings each scenario requires ($ billions). This is the table the whole document is built to produce — the numbers as published in the left column, one labeled adjustment in the middle, the adjusted figure on the right, and the variance against the requirement carried to the last column. All three packages close to zero.

7.3 What each package contains

The three packages are nested. Low uses payment rates and base broadening only: it changes what the government pays providers and what the tax code exempts, and it does not create a new tax instrument or change a benefit formula for anyone currently receiving a benefit. Medium adds the structural items — the individual rate schedule, itemized deductions, capped Medicaid, and a uniform Social Security benefit. High adds the two instruments the United States does not currently levy at all: a broad-based value-added tax and a price on greenhouse gas emissions.

Table 7.3. Composition of each package ($ billions, FY2027–2036, adjusted). The last row is ADJ-2 made visible. Twenty percent of the Medium package rests on options scored against a tax baseline that Public Law 119-21 replaced in July 2025. That does not make those scores wrong; it makes them measured on a legal regime that no longer exists, and a reader is entitled to know which ones.

7.4 The largest levers

Sixteen levers do most of the work in the High package. Each row shows CBO’s published range as filed — all measured December 12, 2024 against the June 2024 baseline — the value the intensity dial selects inside that range, and the same figure after window rebasing. Nothing between the second and fourth columns is a judgment about policy; it is arithmetic on a published score.

Table 7.4. The sixteen largest levers in the High package ($ billions). Those sixteen supply $12,130 billion of the $16,489 billion package, or 74 percent of it. Source for every score in the CBO low and CBO high columns: CBO, Options for Reducing the Deficit: 2025 to 2034, December 12, 2024. Where CBO publishes a single figure rather than a range, the low and high columns are equal and the dial has no effect on that line. The remaining fifty-five levers in the package supply the balance.

The sixty percent bookend

Sixty percent of GDP is the number most often cited as a prudential ceiling for a developed economy. It is worth running, and the answer is worth printing, because it is the only way to see the shape of the problem clearly. It is presented here as a bookend rather than as a fourth scenario, because the arithmetic does not support calling it achievable inside this window.

Table 8.1. What sixty percent would require ($ billions unless noted), solved on CBO’s February 11, 2026 baseline.

The last line is the one that settles it. Reaching sixty percent by FY2036 requires the United States to be running a unified budget surplus of 2.9 percent of GDP in the terminal year, and to have sustained a consolidation of 7.34 percent of GDP for most of a decade to get there. The United States has not run a unified surplus of that size since fiscal year 1948. It has run a unified surplus of any size four times since 1970 — fiscal years 1998 through 2001 — and the largest of those, in FY2000, was 2.4 percent of GDP.

What this says, and what it does not

9.1 Six things the document establishes

9.2 Five limitations, stated plainly

The baseline is a projection, not a fact. It is CBO’s, it is published, it is internally consistent, and it is already known to be behind events on at least one line, as Section 1.4 shows. Everything downstream inherits that.

The option scores predate current law. ADJ-2 flags the exposure rather than correcting it, because CBO has not published re-estimates and the Institute will not manufacture them. Three percent of the Low package and twenty percent of the Medium package rest on pre-July-2025 tax law.

There is no macroeconomic feedback. A consolidation of five percent of GDP would affect growth, and growth affects the denominator of every ratio in this document. The direction and size of that effect are genuinely contested, and modelling it would require an assumption that cannot be traced to a published score.

The packages are illustrative, not unique. Seventy-six levers admit an enormous number of combinations that produce the same total. The packages here were assembled to be internally coherent and free of double counting, not to be optimal or recommended. Any reader can build a different package of the same size and the debt path will be identical.

Nothing here accounts for implementation. Scores assume policies work as written. Collection lags, behavioural response beyond what CBO already embeds, administrative capacity, and litigation are all outside the model.

9.3 No position is taken

Sources and measurement dates

Each source below carries the date it was measured, not only the date it was published, wherever the two differ.

Sources and measurement dates. The document carries this as an unnumbered table; it is reproduced here in full.

Related free references and tools