What America owns,
and what it costs to keep it.

The argument in one paragraph

America is not deciding whether to build a hundred-trillion-dollar portfolio. It already owns one. The only live question is the renewal rate on assets already in hand — and for sixty years that rate has been falling. Federal investment ran at 6.2% of GDP in 1968 and 2.4% in 2018. Federal R&D was 1.86% of GDP in 1964 and 0.70% by 2020. Over the same window, debt is on course from 101% of GDP this year to 175% by 2056. Borrowing is not the scandal. Borrowing without buying anything is.

Debt raised to buy an asset is a balance-sheet transaction: the liability grows, and so does the asset side. Debt raised to fund consumption is a liquidation — the liability grows alone, and the asset side shrinks by depreciation. The federal government reports the first number every quarter and the second one never.

Part I

The opening balance

Five asset classes carry American capability. Two of them can be priced with published national accounts. Three cannot be priced at all with current statistics — and those three are where the compounding actually happens.

Rank the five asset classes by measured return per dollar, and the ordering runs almost exactly opposite to measurement quality. Roads and bridges — the best-measured asset — return roughly two dollars per dollar. Research returns more than ten. Early childhood returns seven to thirteen percent annually, compounded over a working life. Institutions and alliances return an amount nobody has ever attempted to compute, which is why they are the easiest to spend down and the hardest to defend in a budget hearing. A country that manages only what it measures will, over sixty years, systematically starve its highest-return assets. That is a description of what happened, not a hypothesis.

Part II

What taxpayers actually put in

The renewal rate is the whole story, and it is documented. Two series tell it.

Part III

Comparisons at scale

Abstract percentages of GDP persuade nobody. Four programs, each with a documented price tag, show what this kind of money has actually bought.

The rail comparison, told honestly

The standard version of this argument holds up a photograph of a Chinese bullet train and asks why America has nothing like it. The facts behind the photograph are real: China’s high-speed network passed 50,000 km at the end of 2025, carried 4.28 billion passenger trips that year, and reaches 97% of cities above half a million people. The United States operates no line at 250 km/h or above.

But the photograph is only half the ledger. The United States did not fail at rail. It made a choice — private freight over public passengers — and on the freight side that choice has been comprehensively vindicated. American railroads move about 40% of long-distance ton-miles at rates 44% lower in real terms than in 1981, funded entirely out of private capital. Europe moves passengers on rail and freight on roads; America does the reverse, and the energy arithmetic favors America.

What the comparison does establish is narrower and more damning than the usual version. It is not that America chose roads. It is that having chosen roads, America stopped paying for them. A country that builds one network brilliantly and then declines to maintain it has not made a strategic choice. It has made an accounting error.

Part IV

The fifty-year end state

Three renewal rates, run from 2026 to 2076. This is scenario arithmetic with every parameter stated, not a forecast — and the honest result is more modest than advocacy usually claims.

The central benefit-cost ratio is about two to one. That is a good public investment. It is not a miracle, and any page that tells you infrastructure returns ten dollars on the dollar is selling something. Roads and pipes are worth building because two-to-one beats the alternative uses of the money, not because they are magic.

The tenfold returns are real, but they are somewhere else — in research, and in children. Jones and Summers find that a dollar of R&D generates more than ten dollars of economy-wide benefit. Heckman’s follow-up of the Perry Preschool cohort reports two distinct quantities that are easy to confuse: an annual social rate of return of 7–10% — an internal rate of return, comparable to an equity yield — and a benefit-cost ratio of $7–12 per dollar in present value at a 3% discount rate. The 13% figure comes from the later birth-to-five analysis. These are not two ways of saying the same thing; the second is what the first produces once benefits are integrated over a lifetime and discounted back.

Perry was a small, intensively delivered program for a highly disadvantaged 1960s Michigan cohort. Its returns are an average on the population with the most headroom. A universal national program spends its last dollar on a child who would have done well anyway, and that dollar returns far less.

The Accord therefore books child investment at a marginal return of roughly 1.1–1.2× on the incremental dollar, not at Perry’s average — stated as a discounted benefit-cost ratio, the same convention as every other return on this page. (The Accord’s investment gate is a discounted benefit-cost ratio above 1.0 at the stated real rate; an undiscounted multiple would smuggle in a far lower implied return.) The distance between 1.2× and $7–12 is not a rounding difference. It is the entire reason scaling trial results to national programs fails, and holding that line is what lets the Accord cite Heckman at all. Where that figure comes from — and why it turns out to be a floor rather than a ceiling — is Part V.

Part V

The marginal dollar

Every dollar the public raises is a dollar somebody else does not keep. There is a level of taxation past which the last dollar collected does less good in public hands than in private ones. That level exists. The question is where it sits, and whether America is anywhere near it.

The tool for this is the Marginal Value of Public Funds — for each policy, what its beneficiaries would pay for it, divided by its net cost to government after long-run budget effects. Hendren and Sprung-Keyser computed it for 133 US policy changes over half a century, with a comparison rule that is exactly what is needed here: moving a dollar from policy B to policy A raises welfare if and only if the first number beats the second.

The insight that follows reframes the whole question. A tax cut is just another policy with an MVPF. So the value of leaving the marginal dollar with high earners is not a vague counterfactual. It is a computable number, and it is the hurdle rate that public spending has to clear.

What the dynastic dollar is worth

For a top-rate tax cut, willingness to pay is the transfer itself and the net cost is the revenue lost, less whatever comes back through behavioral response. That response is governed by the elasticity of taxable income, which the standard survey places at 0.12–0.40 in the best long-run estimates, with 0.25 as the mid-range. At a combined top marginal rate of 42.5%:

The hurdle comes out at 1.10 to 1.73, central 1.23 — the same 1.1–1.2× that intuition suggests as the point where public collection stops paying. That convergence is worth noticing, because it means the intuition is sound and the framing is inverted.

1.1–1.2× is not a ceiling on what public investment should return. It is the floor set by the alternative. Public spending should continue expanding until the return on its last dollar falls to roughly 1.2. Everything above that line is money the national fiduciary is right to collect.

One refinement makes the hurdle lower still. The MVPF weights the recipient’s willingness to pay at 1.0, treating a dollar of consumption by a billionaire as socially identical to a dollar anywhere else. Any social welfare function that declines to do that pushes the effective hurdle below 1.23.

How far above the hurdle is the Accord?

Every category except adult transfers clears the hurdle at its low end. Even the worst-performing use the Accord funds heavily — ordinary roads and pipes — returns 1.7 times what the same dollar is worth left in place. Adult transfers are the one place where the inflection genuinely bites, and that is a design signal rather than an embarrassment: it is precisely why the Accord’s architecture routes through capability rather than cash.

But does the private dollar build anything?

The MVPF asks what a dollar is worth to whoever holds it. It does not ask what the dollar builds, and for national capability that second question matters more.

Start from the equilibrium fact: a dollar of genuinely new capital formation is worth roughly a dollar, because the pre-tax return equals the required return plus depreciation. The social surplus above that is only the wedge — taxes collected on the return, plus spillovers. So the social product of the dynastic dollar depends almost entirely on what share of it becomes new capital at all.

Bidding for assets that already exist is not capital formation. It is a transfer to the seller. The $200 million painting is the clean case: the artist is dead, the canvas already exists, and national productive capacity is precisely unchanged by the transaction. The same logic applies, less vividly, to buying already-issued equity and to bidding up land.

This is a weaker piece of evidence than the MVPF calculation and should be labeled as such. Mian, Straub and Sufi establish an aggregate correlation across four decades, not a clean causal estimate of what the next marginal dollar does. Some incremental dynastic wealth genuinely does fund venture capital, new issues and productive expansion. The range above is wide for that reason. But even the assumption most generous to dynastic capital puts the social product of the marginal private dollar below one. The forty-five-year historical record of where released capital actually went — the other half of this question — is NBS-2, the private side of the ledger.

Does the capital leave?

This is the objection that does the most work in public argument and the least work in the data. Young, Varner, Lurie and Prisinzano examined 45 million tax records covering 3.7 million filers earning over $1 million between 1999 and 2011. Their finding: tax flight is occurring, but only at the margins of statistical and socioeconomic significance. In an average state with 9,000 millionaires, a one-point tax increase is expected to cost 23 of them — attrition of about a quarter of one percent. Millionaires migrate at 2.4% a year. People earning around $10,000 migrate at 4.5%. The elite are less mobile than the poor.

And that is the state-level number, which is the upper bound on the federal question. The United States taxes on citizenship rather than residence. Leaving the country does not leave the tax base; renunciation triggers an exit tax. Moving from New Jersey to Florida is cheap. Expatriating is not.

The Laffer constraint is real and it is simply not binding here. At every elasticity in the credible range, the revenue-maximizing combined top rate sits between 62% and 85% against a current 42.5%. Only at an elasticity around 1.0 — far outside the 0.12–0.40 literature — does the peak fall below where America already stands.

The MVPF literature reports average returns on policies already enacted. It does not give the schedule along which marginal returns decline as a program scales. This analysis can show that the current position sits well above the hurdle. It cannot locate the inflection point.

So the defensible claim is bounded: at present allocations, the marginal public dollar in every major Accord category except adult cash transfers returns between 1.7 and 8 times what the same dollar is worth left with top wealth holders. That America is “far” from the inflection is an inference from the size of that gap, not a measured distance to it — and it is the honest form of the claim.

There are two ways a country gets this wrong. Tax heavily and invest past the threshold where returns accrue, and the last dollar is waste — the overreach every critic of big government warns about. Tax too little and invest too little, and crumbling infrastructure and an unhealthy populace first limit growth, then demand remedial spending that returns nothing.

A government that taxes the highest holders and spends on viable investments grows the country. The Accord defines that balance mathematically — the hurdle in this section is the definition — and strives to come close to it, bounded by tax recovery on one side and delivery capacity on the other.

Part VI

What can be priced below the line

The environment as a whole has no price. Two pieces of it do, and the larger one is not an asset at all.

Carbon is a liability, and it is the largest single number on this page

Climate damage has a federal price. EPA’s 2023 methodology puts the social cost of a ton of CO₂ at $190 under a 2.0% near-term Ramsey discount rate, with $120 and $340 at 2.5% and 1.5%. The underlying model was published in Nature and gave $185. American energy-related CO₂ emissions were 4.904 Gt in 2025, up 2% on the year. Multiply and discount over the same fifty years as everything else on this page:

At the central estimate, carbon damages accruing over the module’s horizon come to $29.3 trillion. That is 1.4 times the entire public capital stock of the United States. It is also larger than the full fifty-year cost of the Accord’s investment program, which has a present value of $22.9 trillion.

The country is running an off-book liability bigger than the on-book asset, and bigger than the bill for fixing the on-book asset. No infrastructure argument on this page is as large as this one.

One honest caveat, prominently. The social cost of carbon is the most politically contested number in federal economics. The interagency working group that produced it was disbanded in January 2025 and EPA was directed to reconsider the estimate. The Accord publishes EPA’s 2023 central value because it is the most recent figure produced through a formal, peer-reviewed process, and publishes the full range beside it. A reader who prefers $120 gets $18.5 trillion, which does not change any conclusion on this page.

The National Park System

The parks have a measured value, and it is not visitor spending. Haefele, Loomis and Bilmes surveyed American households on willingness to pay and found a total economic value of $92 billion a year — $62B for lands, waters and historic sites, $30B for National Park Service programs. Most of that is non-use value: 95% of respondents said preserving the parks mattered to them whether or not they ever visit.

That is a flow, so it has to be capitalized before it can sit on a balance sheet. At a 2% real discount rate the perpetuity value is $4.6 trillion; at 3%, $3.1 trillion. Over the module’s fifty-year horizon rather than in perpetuity, $2.9 trillion.

Two things bound the number. It is a stated-preference estimate, and stated preference carries a known hypothetical bias — people say they would pay more than they do pay. And the National Park System is roughly 85 million of some 640 million federal acres. This line is a floor on a floor.

Health capital

Yes, the health system has a value, and booking it as hospital buildings would miss it entirely. The asset is health itself. Murphy and Topel valued American longevity gains directly: cumulative gains after 1970 were worth $61 trillion to men and $34 trillion to women, and over 1970–2000 rising life expectancy added about $3.2 trillion a year to national wealth — roughly half of GDP, none of it counted anywhere.

Their return calculation is the striking one. Additional medical spending after 1970 offset only 36% of the value of the longevity it bought — an implied return on the marginal health dollar of about 2.8 to 1, in the same range as public infrastructure and well short of research. Hospital plant is already inside the BEA produced-capital figure above; booking the buildings again would double-count.

The caveat is serious enough to state twice. That work runs to 2000 and is in the paper’s own dollars. Since roughly 2010, American life expectancy has stalled and then fallen. The stock stands; the flow almost certainly does not, and may now be negative. Health capital is the one line on this balance sheet where the United States may currently be posting a loss.

Overseas basing

Plant replacement value for overseas installations is knowable — the Department of Defense publishes it in its Base Structure Report — and it is the wrong number. In fiscal 2005, when the split was reported cleanly, total plant replacement value was about $658 billion with over 81% of it inside the United States, implying roughly $125 billion overseas. The current figure should be pulled from the FY2025 report before publication and is logged as an open item.

But whatever that number turns out to be, it prices concrete, and the concrete is the cheap part. The asset is access: the standing legal right to operate from someone else’s territory. Replicating the reach of the overseas posture without the agreements that permit it would cost a multiple of the physical plant, and in several theatres could not be bought at any price. That is why overseas basing belongs with alliances rather than with buildings, and why the balance sheet treats a lapsed agreement as a write-down even though no asset was sold.

What remains genuinely unpriced

Natural capital beyond the parks — the working forests, croplands, watersheds and fisheries — has no US national account. The World Bank prices it on a methodology consistent across 146 countries and states plainly in its own documentation that a country-built account would be more accurate. America has never built one. What exists instead is damage data: extreme weather caused more than $180 billion of it in 2024 alone. The Accord’s externality-pricing architecture is the mechanism for putting the missing prices on the books; until it does, this line is honestly blank.

Institutions have no number at all. Their yield shows up in the rate at which the United States borrows and the willingness of the world to hold dollars, and they are the only asset class that can be destroyed faster than it can be measured.

Alliances have no number either, for the reason given above: they are leased capability — forward basing, coalition burden-sharing, intelligence access, market access — and the lease is worth far more than the fixtures.

Grid hardening and national defense infrastructure are general-fund overdue obligations. They are never funded from externality-pricing trusts, and they require no pay-for. A government that cannot keep the lights on and the borders secure is not economizing; it is failing at the thing it exists to do.

Part VII

The comparison everyone is actually making

China out-invests the United States roughly two to one as a share of output (gross capital formation ~43% vs ~22% of GDP), and its government sector now outspends ours on government-performed R&D. Its demographic position is also materially worse and cannot be fixed with capital: by 2050, fewer than two working-age adults per person over 65, against more than two and a half here. The balance-sheet reading is narrow: the contest is not about who lays more concrete — on concrete, China has already won and the win is worth less than it looks. It is about which country compounds the assets that compound: people, research, institutions, and alliances. On those, the race is undecided and the American position is deteriorating by choice.

The full treatment — the five-track scoreboard and the forfeit dashboard — is at The forfeit conditions →

Part VIII

What this analysis does not establish

The human capital figure is a ratio transfer from cross-country OECD work, not a US-specific lifetime-income calculation. It should be treated as an order of magnitude and nothing finer; rebuilding it on the BEA-affiliated US series is the open item.

The fifty-year scenarios model produced public capital only. They do not model human capital accumulation, and they do not attempt to convert the Heckman or Jones–Summers micro-returns into macro projections. Scaling returns from a small randomized trial to a national program is not warranted, and the Accord does not do it.

The elasticity of output to public capital is the single load-bearing parameter, and the literature it comes from has a publication-bias problem the authors themselves document. The depreciation rate (2.8%) is a judgment parameter pending replacement with BEA’s implied rate — results move materially with it. The model excludes crowding-out of private investment and diminishing returns at high capital ratios; all would reduce the Accord result. The Interstate annual-value figure is model-based, gross of maintenance, and currently sourced through secondary reporting.

The carbon liability is not modeled dynamically. It applies a constant social cost to a constant emissions rate for fifty years. Emissions will not stay constant in either direction, and the social cost of carbon rises over time in EPA’s own tables, so the figure should be read as an order of magnitude on current behavior rather than a projection of damages actually incurred.

The parks figure is stated preference and carries hypothetical bias. Health capital is not inflation-adjusted and stops at 2000. Overseas basing is quoted from a twenty-year-old edition of the Base Structure Report. The marginal-dollar analysis in Part V reports averages on enacted policies, not the declining schedule a scaling program walks down — the inflection point cannot be located from it, only the distance above the hurdle.

Natural capital beyond the parks, institutions and alliances carry no number at all. They are the largest source of error in this analysis, and the error runs in the direction of understatement: the published sheet is a floor. Because park and hospital physical plant already sit inside the BEA produced-capital figure, the priced lines are published separately and never summed into a single net-worth total until that overlap is quantified.

Finally: this is a stock account. It consumes flow outputs from the Accord’s fiscal engine; it does not re-derive them, and the baseline GDP path is CBO’s, not the Accord’s — if the case only worked on our own growth numbers, it would not be a case.

Sources

Supporting literature

1. US Bureau of Economic Analysis, Fixed Asset Accounts, Table 1.1, current-cost net stock, year-end 2024. Series K1WTOTL1ES000, K1GTOTL1ES000, K1GSTLC1ES000, K1GTOTL1EQ000, K1YTOTL1ES000, K1NTOTL1EQ000.
2. Federal Highway Administration, The Dwight D. Eisenhower System of Interstate and Defense Highways, Part III (final Interstate Cost Estimate, 1991). Annual-value estimate as reported in commentary on TRB Special Report 329 (2019) — secondary sourcing, flagged in Part VII.
3. Liu, G., “Measuring the Stock of Human Capital,” in Jorgenson, Landefeld & Schreyer (eds), Measuring Economic Sustainability and Progress, NBER Studies in Income and Wealth 72, 2014. Method: Jorgenson & Fraumeni (1989, 1992). See also Abraham & Mallatt, NBER WP 30136 (2022); Christian, BEA working paper (2016); BEA Working Paper 2026-6.
4. Bom, P. R. D. & Ligthart, J. E., “What Have We Learned from Three Decades of Research on the Productivity of Public Capital?”, Journal of Economic Surveys 28(5), 2014. Cross-check: Calderón, Moral-Benito & Servén (World Bank), elasticities 0.07–0.10 post-correction.
5. American Society of Civil Engineers, 2025 Report Card for America’s Infrastructure, March 2025, including the Energy category report.
6. Congressional Budget Office, The Budget and Economic Outlook: 2026 to 2036 and long-term projections through 2056; CBO, Federal Investment, 1962 to 2018 (2019), as presented in Aspen Economic Strategy Group (2021).
7. National Center for Science and Engineering Statistics, NSF 23-339 (2023) — federal R&D share series.
8. Press, L., “Seeding Networks: the Federal Role,” Communications of the ACM, 1996; National Research Council, Funding a Revolution (1999).
9. University of Texas at Austin Energy Institute, US electric infrastructure replacement value (2016); ASCE 2025 Energy category (transformer age and lead times).
10. Association of American Railroads, Freight Rail Facts & Figures (~40% of long-distance ton-miles); Federal Railroad Administration reports ~28% of total freight movement on a different denominator — the discrepancy is noted rather than silently resolved.
11. Jones, B. F. & Summers, L. H., NBER WP 27863 (2020); Heckman et al., Journal of Public Economics 94 (2010) and the 2016 birth-to-five analysis; Griliches (1958).
12. China series: Congressional Research Service IF11667 (May 2026); RAND RBA3372-1 (May 2026); ITIF (April 2025); OECD MSTI (March 2025); China Railway via UIC (2025).
13. World Bank, The Changing Wealth of Nations 2021 and methodological FAQ.
14. Murphy, K. M. & Topel, R. H., “The Value of Health and Longevity,” Journal of Political Economy 114(5), 2006, 871–904.
15. Haefele, M., Loomis, J. B. & Bilmes, L. J., “Total Economic Valuation of the National Park Service Lands and Programs,” Harvard Kennedy School Working Paper RWP16-024, 2016.
16. US Environmental Protection Agency, Report on the Social Cost of Greenhouse Gases: Estimates Incorporating Recent Scientific Advances, November 2023. Underlying model: Rennert et al., “Comprehensive evidence implies a higher social cost of CO₂,” Nature, 2022. The Interagency Working Group was disbanded in January 2025 and EPA directed to reconsider the estimate.
17. US Energy Information Administration, US Energy-Related Carbon Dioxide Emissions, 2025, released 21 July 2026.
18. US Department of Defense, Base Structure Report, FY2005 edition (total plant replacement value ≈$658B, >81% domestic); the current edition should be substituted — logged as an open item.
19. Hendren, N. & Sprung-Keyser, B., “A Unified Welfare Analysis of Government Policies,” Quarterly Journal of Economics 135(3), 2020, 1209–1318. Interactive data at policyimpacts.org.
20. Saez, E., Slemrod, J. & Giertz, S. H., “The Elasticity of Taxable Income with Respect to Marginal Tax Rates: A Critical Review,” Journal of Economic Literature 50(1), 2012, 3–50.
21. Diamond, P. & Saez, E., “The Case for a Progressive Tax,” Journal of Economic Perspectives 25(4), 2011, 165–190. The 42.5% is the combined marginal rate on income, payroll and sales at federal, state and local level, not the top federal income-tax rate.
22. Mian, A., Straub, L. & Sufi, A., “The Saving Glut of the Rich,” NBER Working Paper 26941, 2020, revised 2025.
23. Young, C., Varner, C., Lurie, I. Z. & Prisinzano, R., “Millionaire Migration and Taxation of the Elite: Evidence from Administrative Data,” American Sociological Review 81(3), 2016, 421–446.

Scenario arithmetic, not a forecast. All parameters stated; every computed figure regenerates from three short transparent models with no dependencies (core scenarios, priced lines, marginal dollar), maintained alongside this page. Changes to any parameter propagate to all tables and both charts — the page is regenerated, never hand-edited. Part 2 of the module — Did anything trickle down? — weighs the private side of the same ledger.