Counterpoint

Abstract

Federal debt can grow over time but not faster than GDP in the long run. Revenue must cover spending plus interest owed on the debt. Calculating the tax increase required to stabilize the debt-GDP ratio after 30 years requires estimating primary deficits (not including interest payments), borrowing rates r and growth rate g for many years after the Congressional Budget Office's (CBO, 2025a) 30-year projection. CBO never reported a stabilizing rate. I am sure that Edelberg et al. (2025) reasonably extrapolated CBO data to estimate their 3% GDP value. But modest changes in assumptions create large differences: A 100% debt-GDP ratio and a 300% debt-GDP ratio could both be reasonably estimated to require the same future 3% of GDP, or even 6% or 9%.1 Using a microsimulation (not OLG) model, where the first decade was purposely aligned to CBO's (including CBO's no growth of discretionary spending2), Gokhale and Smetters (2024) estimated a permanent 6.6% GDP shortfall in 2024 when the debt-GDP ratio was 98%.3 The Government Accountability Office (GAO, 2024) projected a debt-to-GDP ratio of 229% by 2054. CBO (2025b) projected ratios “exceeding” 250% by 2055 under assumptions that are reasonable and even more consistent with the reduced-form literature.

Uncertainty is common in long-term projections, and debates over reduced-form parameter values can be endless. Moreover, a standard insight from the field of finance is that more probability weight should be placed on worse outcomes.4 Otherwise, eliminating the tax system altogether would be optimal.5

But let me pinpoint what I think is the specific disagreement between Louise and me. The risk-free rate itself is under-identified in reduced-form models like CBOLT.6 The modeler picks a time path of interest rates—maybe based on yet another reduced-form estimation7—instead of computing the rate required for bond market demand to equal government supply at each point in time. My sense is that Louise seems fine with that; I am not. We agree with using simple models to study stimulus during short-term disruptions like COVID-19.8 But I disagree that reduced-form models are sensible for longer-term analysis where forward-looking trillion-dollar capital markets require that asset demand and supply add up (equilibrium). If simple statistical relationships, estimated at historical lower levels of debt, applied to an increasing debt path, debt crises would not exist. Even CBO has a limited appetite for this simplicity, capping their reported debt-GDP projections at 250%, which is easily achievable in 30 years with reasonable assumptions.

The OLG model with aggregate uncertainty requires that prices, including interest rates, follow supply and demand. In words, things must add up. The OLG model can also incorporate future demographic changes from a microsimulation model. This responsiveness of the OLG model to demographics is a feature, not a bug. The OLG model can then be tested against historical data, which I described in my essay, before being used to predict future responses. In contrast, reduced-form models can only be weakly “tested” at best; their parameters are derived from historical data without meaningful over-identification restrictions useful for testing.9 Even if estimated well, including handling endogeneity and isolating structural from cyclical effects, reduced-form parameters are not applicable to a very different future that includes an increasing debt path (known as the Lucas critique10).

Unless the U.S. household saving rate increases remarkably, or GDP growth substantially increases, or implicit debt (see my other essay in this series) sharply falls, U.S. explicit debt cannot increase much beyond 200% of GDP. Debt reduces capital and wages. Demand for this level of debt does not exist without higher interest rates. Higher interest payments require more distorting taxes, reducing household savings, capital and wages. Rates climb even more to compete for this limited supply of capital. The cycle repeats. Unraveling is common in insurance markets; it just requires enough initial heterogenous risk-price mismatch. In practice, in past debt crises, forward-looking debt markets unravel before hitting fiscal capacity. Bond markets unravel once they lose hope. PWBM's goal is to get bond markets to respond sooner, to “bend” more before “breaking.”

The 200% value could change. For example, more demand for debt could arise from stablecoins. So, we will model this demand capacity and refresh. Reduced-form models could not provide these insights.

Louise is right that OLG models often fail to capture the top 1% wealth concentration (although the top 10% is doable). An ongoing PWBM project with entrepreneurship and bequests creates this concentration, consistent with data. But the role of debt does not subside.11 More generally, non-retirement savings, including international capital flows, tend to be more sensitive to interest rate movements than retirement savings.

U.S. long-term bond rates should be low due to an aging population.12 Even so, real (inflation-adjusted) 10- and 30-year U.S. bond yields exceed their historical averages despite a worsening growth outlook.13 Nominal yields are close to the UK's, where both inflation expectations and expected real growth rates are only slightly worse than the U.S. Although the U.S. and UK have similar debt-to-GDP ratios, the UK's debt duration is twice as long. If the U.S. duration matched, U.S. long-term rates would be even higher, likely exceeding the UK's, despite similar outlooks. Of course, it is too early to draw conclusions, and, more immediately, markets were pricing a high chance (over 80%) of recent OBBB legislation passing for several months. To be sure, the U.S. has more debt capacity than the UK, largely due to lower U.S. tax rates. Still, the recent UK experience, and the recent U.S. Liberation Day experience, if it were not largely paused, suggests caution. We are far from invincible.

Barring serious policy reform, only time will tell who is right. And this is an argument that I sincerely hope to lose. In two dozen years or so, I hope to get a call from Louise laughing at me, telling me that I was worried about nothing, and that I now owe her lunch at some swanky Dupont Circle location. And, if I am right, I will give her geo-coordinates; just remember to bring a tent and fishing pole when visiting.