AI and the Natural Rate: Puzzle or Policy Challenge?

What Is Driving the Rise in r*?

Jens H. E. Christensen and Glenn D. Rudebusch analyze whether the roughly one-percentage-point rise in the U.S. natural real interest rate, r*, since 2020 can be explained by news concerning government debt, artificial intelligence, or monetary policy. The authors conduct a high-frequency event study covering January 2020 through September 2025. They calculate cumulative movements in r* during three-day windows i.e. the day before, day of, and day after, surrounding:

1. 52 major U.S. fiscal-policy events.
2. 25 releases of important AI models from OpenAI, Google, Anthropic, and xAI.
3. Scheduled FOMC meetings.

The dependent variables are four daily, market-based estimates of r*, derived from alternative dynamic term-structure models using Treasury and TIPS data. Each measure represents the expected average real short-term interest rate five to ten years ahead, after removing estimated term premiums.

Rather than estimating the average response to a quantified surprise, the authors ask how much of the total post-2020 rise in r* accumulated within the selected event windows. Using four different term-structure models provides a robustness check against dependence on any single estimate of the unobservable natural rate.

Findings

• AI news does not explain the rise. Across the four models, r* declined cumulatively by approximately 23–35 basis points around major AI releases. Thus, the market response runs opposite to the hypothesis that AI news increased expected productivity and capital demand sufficiently to raise r*.
• Fiscal news has a limited effect. Fiscal-event windows produce small positive cumulative movements, accounting for approximately 20% of the observed increase in r*. The authors caution that this may be an upper-bound estimate because they deliberately use a broad set of fiscal events.
• FOMC meetings do not explain the increase. Although movements around FOMC meetings capture much of the pre-2020 secular decline in long-term rates, post-2020 FOMC windows capture none of the subsequent rise in r*, they instead show a modest downward tendency.

Conclusion
The results do not demonstrate that fiscal policy or AI has no long-run effect on r*. Rather, they show that market-based estimates of r* did not rise systematically when the selected fiscal and AI information became public. The authors acknowledge that news may diffuse gradually rather than within three-day windows, AI releases may differ in their economic implications, and the estimates of r* are inherently model-dependent. Nevertheless, taken at face value, the results leave the post-2020 increase in the natural rate as an important empirical puzzle.

Can Fiscal, AI, or Monetary News Explain the Rise in r*
Authors: Jens H. E. Christensen, Glenn D. Rudebusch
From: Federal Reserve Bank of San Francisco, Brookings Institution

AI, aggregate demand, and the natural interest rate

Luca Fornaro and Martin Wolf examine how AI-enabled automation affects aggregate demand, inflation, employment, inequality, and the natural interest rate. The authors develop a theoretical macroeconomic model with three central features:

1. Firms choose whether to perform production tasks with labor or automate them with capital.
2. Workers consume a larger share of their income than capital owners, making income distribution important for aggregate demand.
3. Nominal wage rigidities allow output and employment to fall below potential and give monetary policy real effects.

AI progress is modeled as an increase in the share of tasks that can be automated. The authors compare alternative monetary-policy responses, examine the zero lower bound, and use calibrated numerical simulations to illustrate the transition toward greater automation. They then introduce employment subsidies such as  reductions in labor taxes to study how fiscal policy can complement monetary policy.

Findings

AI raises productivity and potential output, but it also transfers income from relatively high-spending workers to higher-saving capital owners. Without sufficient policy support, the resulting decline in consumption demand can reduce output, employment, and investment producing an “AI slump” despite higher productivity.

Expansionary monetary policy can instead generate an AI boom, but it faces two constraints:

• Short run: Investment in automation raises r*, while labor displacement can create inflationary pressure and worsen the trade-off between inflation and employment.
• Medium run: Once the investment boom subsides, weaker worker consumption lowers r*, potentially pushing it below zero and creating an automation-driven liquidity trap.

Employment subsidies or lower labor taxes can support workers’ incomes and aggregate demand while reducing firms’ labor costs and inflation. The broader conclusion is that AI’s macroeconomic outcome is not predetermined: whether it produces a boom or a slump depends substantially on the monetary–fiscal policy mix.

Macroeconomic Policies for AI
Authors: Luca Fornaro, Martin Wolf
From: CREI, Universitat Pompeu Fabra, University of St. Gallen