Why debt matters

Abstract

Your retirement portfolio likely holds some bonds. In the hypothetical absence of federal government debt, those bonds would be corporate debt invested in capital ventures that yield future returns. You receive a return because an actual capital investment is made into the economy. That investment increases capital per worker and wages while lowering borrowing costs for a home mortgage.

In reality, of course, 40% of your bond allocation likely holds various denominations of U.S. government debt. Naturally, those assets represent actual savings from your perspective. However, most government debt is used for spending programs or tax cuts that benefit current and past generations, rather than investing in long-term public assets like early childhood education or infrastructure. You earn a return simply because the U.S. Treasury must offer it to compete with alternative investments. So, most government debt redirects savings from actual investments for the future to support past and current consumption. Government debt “crowds out” private investment.

But crowding out is just one problem. As debt grows, as it is projected to do for the United States, bond markets eventually lose faith in the ability of the government to make full payments in real terms. Such “bank runs” are infrequent but violent. They are fully rational and so they cannot be educated away or mitigated with financial regulation. Debt crises have brought down superpowers.

It is tempting to immediately inquire about the exact empirical evidence—How much does debt crowd out private investment? What is the nation's fiscal capacity before a bank run emerges?—but it would make little sense without having a solid model for interpretation.

Unfortunately, serious modeling of government debt on the economy is largely absent inside DC policy circles.1 But the tools have existed for a long time and are widely used in academics. Paul Samuelson (1958) and Peter Diamond (1965) introduced the lifecycle overlapping-generations (OLG) model that accommodates a wide range of government policies. The OLG model requires being explicit instead of using heavy discretion to map from policy to non-OLG model input. Welfare analysis is also rigorous.

Both “crowding out” and “bank runs” (fiscal limits) naturally emerge. OLG is the workhorse model of modern public finance and macroeconomics, so much so, that the top five general interest journals in economics expect papers to use this framework when analyzing the impact of fiscal policy on macroeconomic variables. OLG is as basic as the standard laws of thermodynamics in physics; debates happen only around the edges. But OLG has an “entry fee” due to the cost of learning public finance and computational methods. The richer the OLG model, the higher these costs. So, instead, DC policy circles largely use the simpler “neoclassical growth” model developed by Ramsey (1928), even for longer-term projections. It is a mistake with potentially dire circumstances.

The “neoclassical growth” model, developed by Ramsey (1928), assumes that the entire economy can be represented by a single infinitely-lived household. This Übermensch decides how to allocate consumption between her current and future selves. So, any shift in resources, including debt, conducted by an invasive external government between those selves is simply undone by the Übermensch, a property known as Ricardian equivalence. Debt, therefore, has no impact on anything: not on consumption, capital, wages, interest rates, not on anything. Besides lacking a debt channel, the Ramsey model cannot be used to model most spending programs or progressive taxes. But the neoclassical growth model has one advantage: It is easy to solve.

The Congressional Budget Office's (CBO) CBOLT model (CBO, 2025a) enhances the Ramsey model by incorporating an additional reduced-form rule (i.e., ad-hoc rule) between capital and federal debt, wherein a $1 increase in deficits reduces capital by $0.33. Established in 1974, the CBO provides impartial information on budgetary and economic issues rather than performing cost-benefit analyses or serving as the nation's indicator for future long-term financial crises. The CBO, along with the Joint Committee on Taxation (JCT), houses some of the nation's leading experts in budget analysis. However, neither the CBO nor the JCT is mandated to thoroughly model the long-term financial implications of growing federal debt. If they were, such analysis would likely dominate almost all other aspects of their legislative functions.

CBO is institutionally risk averse and adaptive in their long-term projections. As such, they will unlikely ever see a crisis in front of them.

Indeed, the CBO does not necessarily view CBOLT or its related internal models as the primary means of capturing debt dynamics. For short-term estimation, the CBO employs a forecasting model. For long-term forecasting, despite receiving less attention, the CBO has access to an OLG lifecycle model of their own. But CBOLT gets all the attention, and it is leading to a misunderstanding of the impact of debt on the economy. Its $0.33 reduced form offset rule is not shorthand for reasonable economics modeling. Its main benefit is simply being easy to solve.

Understanding why reduced-form models like CBOLT are problematic points us to the need to interpret the empirical evidence through the OLG model, where policy is more precisely modeled. It is only then that we see the true damaging long-run effects of federal debt.

CBO estimates that a 1% increase in debt-to-GDP ratio increases the government's borrowing cost by 2 basis points. To be sure, the central estimate in the literature appears to be more than twice that value, including calculations by economists generally associated with the “left” and “right” (Salmon, 2025). But there is a bigger problem: proper estimation controls and interpretation.

In work with Jin (2021), we generated a large synthetic data set from an overlapping-generations lifecycle model with aggregate uncertainty. We ran regressions that Gamber and Seliski (2019) used with historical data for the CBO and found strikingly similar results: A positive but small relationship between debt and borrowing rates. However, that result was driven by cyclical shocks: After a negative shock to total factor productivity, the supply of government debt (by fiscal rule) and the demand for government debt (computed in general equilibrium) both increase.2 As a result, a small increase in the interest rate appears to be associated with a large increase in debt.

But the relationship between debt and economic variables along the nation's secular path of increasing debt is what matters. In fact, the CBO does not project any possibility for additional debt runup due to a future recession, although capital markets clearly price it.3 Even with this omission, the problem with contemporaneous fiscal policy is not cyclical; it is the rising secular trend, as argued by Auerbach and Yagan (2024): “Year-to-year feedback has disappeared” (as cited in Brookings, 2024). Commenting on their paper at Brookings, Auerbach continued: “We once had government that was responsive to this problem in both Republican and Democratic administrations and we don't now” (as cited in Brookings, 2024).

Jin (2021) showed that the relationship between the secular rise in debt and capital (and, hence, wages and interest rates) is very strong and not linear. That is the evidence that matters, not reduced-form evidence. Using the same model, Penn Wharton Budget Model (PWBM, 2023) has calculated that U.S. debt cannot exceed 200% under the most optimistic scenario in which capital markets keep believing that the U.S. Congress will eventually create fiscal balance, with levels between 175% and 190% being dangerous. The U.S. is not Japan, which has a household saving rate almost 10 times larger (and, so, a higher net national savings rate) than the U.S. The U.S. is not Europe, which benefits from U.S. innovation.

The U.S. economy faces other challenges at the same time. U.S. fertility rates are falling with more of the tax base shifting to lower-productivity households over time. The worker-retiree ratio is falling. Skill-adjusted labor supply growth is falling as more of the population enters retirement.4 Productivity gains from AI do not reverse this point.5 Even a broad-based income tax, set to maximize (peak Laffer-curve) revenue, would not be sufficient at 200% to cover the nation's bills unless spending is cut, either by defaulting on interest payments or by defaulting on implicit debt.

The U.S. government's explicit debt capacity is also severely limited by pay-as-go entitlement programs. These “implicit debt” programs borrow twice as much from future workers as explicit Treasury debt (PWBM, 2025) and have identical economic and inter-generational impacts per dollar borrowed (Auerbach et al., 1991).6 But reduced-form models like CBOLT do not accommodate implicit debt and are easily gamed. Legislation, like the Social Security 2100 Act (2023), appears to eliminate Social Security's actuarial 75-year shortfall—that is, it appears to eliminate the Social Security trust fund's reliance on general revenue transfers to meet “scheduled benefits.” But it increases implicit debt by more than $1 for every $1 of explicit debt reduced (see PWBM, 2019), thereby harming future generations rather than improving their finances. The CBOLT model fails to capture how growing implicit debt crowds out capital over time.

Without implicit debt, the government's explicit debt fiscal capacity would almost triple.

Reduced-form models like CBOLT also fail to capture budget closure, which requires that debt interest must be paid.7 This failure is a major problem.

To be sure, higher interest rates are modeled to cause faster debt buildup. And more debt is modeled to slow private capital formation. But debt is ultimately just rolled over. Future interest payments are allowed to exceed thresholds, above which bondholders should not hold any reasonable expectation of full repayment in real (inflation-adjusted) terms. In words, there is no closure.

Inserting realistic closure into reduced-form models is challenging due to the absence of household, investor, and firm optimization and price interactions. Using the results from CBO's CBOLT model, Edelberg and colleagues (2025) estimated that taxes would have to be increased or spending cut (or some combination of both) by 3% of GDP in 2054 to stabilize the debt thereafter at 166% of GDP.8 Of course, the tax increase or spending cut required for closure would increase if done inside an optimization model: When closure is implemented in the future, forward-looking markets shift consumption earlier to reduce taxes after closure; that shift reduces capital accumulation even more.

As noted above, CBO's assumptions seem generally optimistic. Gokhale and Smetters (2024) estimated a fiscal imbalance today equal to 6.6% of the present value of all future GDP, which is the minimum resources needed for the present value of future revenue to equal the current level of debt held by the public plus the present value of future spending. The main difference relative to CBO mostly falls outside the 10-year budget window, with PWBM's bottom-up (microsimulation) approach—used for estimating marriage/divorce, education attainment, fertility by education, labor force participation by skill group, and over 60 other factors—generally producing a lower skill-adjusted labor-supply growth rate in future years relative to CBO. The Government Accountability Office (2024) is also less optimistic than CBO. GAO estimates a debt-to-GDP ratio of 229% by 2054 under their standard assumptions, with that value increasing to 275% with a 1% increase in the weighted-asset borrowing rate, and to 192% with a 1% decrease in the borrowing rate below their standard assumptions (GAO, 2024).

Notice that the loss is compared to the extended baseline in 2054, which CBO projects to have a debt-to-GDP ratio of 155%. Much more importantly, CBO's assessment does not include the tax increase or spending decrease required for closure; if they did, these losses would be much larger.11 Most importantly, it is highly unlikely that closure would even be possible with realistic tax instruments in a comprehensive model with actual optimization.

At a minimum, are we willing to bet the country's future prosperity on the hope that a fixed 33% capital offset rate—despite its associated estimation issues and missing theoretical underpinnings—is correct? If so, are climate change skeptics not then right to assume that uncertainty about the CO2 harm function is a valid reason for downplaying CO2?

Virtually all debt crises in the past century were not fully characterized by a smooth run-up in bond yields. Bond markets typically broke down fast. Even the United States (too big to fail?) got a small taste after Liberation Day, April 2, 2025. The dollar fell, capital fled, and the 30-year bond climbed to almost 5%, despite markets pricing in some probability of a bluff. Without the tariff pause, markets would likely have fallen even more, as expectations firmed up that tariffs were persistent.

Indeed, economic theory predicts a lack of smoothness in bond markets. The Cole and Kehoe (2000) OLG model has become fundamental for understanding the macro dynamic evidence of debt. Two (or more12) equilibriums emerge: one where the government repays (closure) and another equilibrium where self-fulfilling investor beliefs produce a price spiral where the government does not repay (no closure). The key point: The smooth (continuous) nonlinear dynamics between debt and interest rates only exist in the equilibrium with closure. A sudden sharp fall in bond prices and a concomitant jump in interest rates occur when markets move from a common equilibrium belief in closure to a common belief in non-closure. This result is driven by rational, forward-looking actors.

PWBM (2023) calculated “XX date”—the outer bound in which financial markets could rationally believe that they will be repaid—to be around 20 years. But once financial markets lose faith, they can unravel at smaller debt-GDP ratios. It could happen tomorrow.

The role of debt on macroeconomic variables cannot be understood without a model. All the bond dynamics discussed above naturally emerge within the OLG framework. The OLG model is the modern workhorse model of public macroeconomics due to its ability to coherently tie microeconomic decisions with macroeconomic variables and prices. Without the OLG model, mapping between new policies and model inputs is vague and subjective. Moreover, one cannot easily explain the “Liz Truss moment” and “bond vigilantes” without OLG.

The OLG model has been criticized as being too good, too rational. In fact, frictions—adjustment costs, heterogeneous discounting, beta-delta preferences,13 and search costs—are easy to include. The main resistance by some economists seems much more related to entry costs of the model itself.

Contrary to its reputation as being a strict fiscal disciplinarian, the OLG model shows that debt can sometimes “pay for itself” if used to reduce idiosyncratic risks, including mean-test pre-K education (see PWBM, 2021). Our related analysis fully incorporated the new marginal tax rates from income phaseouts and the negative impact of new debt on capital. It also included the fact that many households would have placed their kids in pre-K without the policy, and so the new public transfers are inframarginal. Still, the gains in future productivity to young children of low-income families eventually outweighed program costs. One could not reasonably model pre-K without OLG-like explicitness.

Some other public investment, for example, in infrastructure (mainly repairs) can also produce a high return.14 Overall, though, these high-return projects are limited in size. While the literature considers the marginal value of public funds, the marginal returns quickly diminish in investment size even for large public infrastructure projects. This literature also ignores debt costs.

Policy incrementalism during the past 40 years has simply produced higher levels of debt with spending increases and tax cuts. But options do exist. PWBM has recently shown how a large-scale set of reforms can fully stabilize debt while substantially growing the economy. Old-age poverty is almost fully eliminated. Workers gain from higher wages, lower borrowing costs, and near-universal health insurance (PWBM, 2024). These calculations are done within a consistent framework that fully accounts for all interactions, incentives, prices, and transfers. At a minimum, informed debates about debt require comprehensive and explicit modeling. That modeling is missing today in DC policy circles.