AI and the Treasury Debt Market | American Enterprise Institute
The market for Treasury debt is challenging to interpret even when the environment is calm, much less when multiple geopolitical and technological disruptions are present simultaneously. Two papers released in the past year offer differing perspectives on a consequential unknown in the current market, which is the expected effect artificial intelligence (AI) will have on future productivity growth.
The first of the two relevant papers, co-authored by Isaiah Andrews and Maryam Farboodi, examined market signals in 2023 and 2024 around the releases of updated AI models from five leading developers (OpenAI, Anthropic, Google Deepmind, xAI, and DeepSeek). A main finding was that the nominal interest rates for long-term Treasury and corporate debt fell after these models became public by a statistically significant 12 basis points.
The authors suggest that the most straightforward interpretation of the observed data is market disappointment with early AI results, or possibly relief. Declines in future real interest rates in the bond market are seen as signals that current consumption, financed with borrowing against expected future income, has been dampened. The possible explanations could be that AI might not produce the higher growth some are hoping for or the catastrophe some fear (it is speculated that substantial AI disruption might precipitate a “live-for-today” spending spree).
The second paper, highlighted at Marginal Revolution and amplified in a related Substack post, was written by Howard Kung, Hanno Lustig, and James Paron and offers an alternative take on the findings of Andrews and Farboodi. Instead of market disappointment, they see a favorable “long position” on AI-induced productivity growth. Declines in interest rates for Treasury debt, and the corresponding increase in its value, are said to confirm market optimism for AI-induced growth.
The “long on AI” authors begin by noting that the market for Treasury debt operates with a budget constraint. The stock of accumulated federal borrowing must be repaid at some point with a stream of future primary surpluses, which occur when federal revenue exceeds non-interest spending. Absent a change in policy, CBO projects primary deficits will persist indefinitely. Kung, Lustig, and Paron use the convention of assigning a “terminal value” adjustment after thirty years to offset the level of outstanding debt expected at that time. The present value estimate of this terminal value is $44.6 trillion, which implies investors are expecting either a large fiscal adjustment or else “financial repression” after 2055.
Kung, Lustig, and Paron argue that higher productivity from AI should lead directly to higher valuations of Treasury debt, and thus substantial returns for existing creditors, because higher growth boosts federal revenue more than spending across multiple decades.
To support their analysis, the authors cite the “rules of thumb” the Congressional Budget Office (CBO) posts periodically to inform discussions around alternative growth and budget scenarios. CBO says higher annual real growth of 0.1 percentage point would increase federal revenue by $614 billion over ten years and spending by $51 billion. After factoring in debt service effects, the budget deficit over a decade would fall by $317 billion (the authors draw their assumptions, which are similar to these figures, from an earlier CBO publication).
The revenue growth is tied to “bracket creep,” which occurs because the thresholds used to set progressive tax rates for individuals are indexed to inflation. When wages rise in real terms, more workers get pushed into higher tax brackets.
On the spending side, initial Social Security benefits rise with real growth too, but annual benefit adjustments for current retirees are indexed to consumer inflation. Consequently, higher growth pushes spending up, but with a considerable time lag compared to the revenue adjustment. Kung, Lustig, and Paron estimate the initial elasticity for spending growth relative to economic growth at 0.25, which then rises over time as new retirees account for a larger share of total spending.
Thus, the net fiscal improvement from a boost in productivity can be substantial and, according to the three authors, valuable for Treasury’s existing creditors. With stronger growth, there is less risk of inflation and a lower risk of debt-induced disruptions to the market, which then leads to lower interest rates for Treasury debt. Kung, Lustig, and Paron estimate that a 0.1 percentage point boost in growth would produce a 0.71 percentage point drop in the average nominal interest rate for Treasury debt. The resulting valuation boost for those long on AI would be $1.3 trillion. A 0.5 percentage point jump in annual growth would produce a $6.5 trillion valuation increase. (Andrews and Farboodi see less evidence for the “long on AI” hypothesis, which depends on a risk-related improvement to Treasury pricing that they do not observe in their data).
With so much riding on AI, and no settled consensus on what to expect, the scrutiny on the Treasury market is likely to intensify.
