Same Automation, 11.7% Growth or 2.2%: The Saving Rate Decides

The same path of automation produces 11.7% annual growth when households choose how much to save and 2.2% when the saving rate stays at the 2025 U.S. level of 23%. [Fact] A slowdown that looks like a limit of the technology can be a shortage of capital. In the same model, the wage falls to between a quarter and a half of its no-automation trend for two decades.

The paper is "Saving for the Machines," NBER Working Paper 35964 by Jesús Fernández-Villaverde of the University of Pennsylvania, dated October 2026. It is a calibrated theory paper, built on the task framework of Acemoglu and Restrepo, in which every task becomes automatable between 2025 and 2045.

Capital supply sets the size of the boom

Automation moves tasks from labor to machines and so raises the capital share, from one-third today toward one under full automation. [Fact] A household that saves optimally keeps reinvesting roughly the same fraction of what machines earn: the ratio of investment to capital income moves from 0.69 to about 0.55 while the return on capital climbs from 11% to 52%. [Estimate] In U.S. data the ratio has stayed between 0.50 and 0.78 since 1950. [Fact] Because machines earn a larger share of output, the saving rate rises with the capital share, from 15% of output on impact to 54% in the limit under the benchmark calibration. [Estimate]

With the saving rate frozen at 23%, the arithmetic reverses. When tasks substitute one for one, growth in 2045 reaches 6.5% instead of 22%; when tasks are strong complements, no boom appears at all. [Estimate]

Task structure sets the duration and the wage

Substitutability across tasks decides how long the boom lasts and what happens to pay. When 85% of today's labor tasks become automatable, the economy looks fully automated until about 2040. After that, under strong complementarity, growth returns to near 2% by 2065 and the tasks left with humans earn scarcity rents: the wage, after falling to little more than a quarter of trend in 2040, rises to nine times trend. [Estimate] Under full automation that recovery never comes, because labor cannot earn more than the cost of the machine that replaces it.

Long bonds price the boom on impact. The ten-year real yield jumps from 5.1% to 13.6% the year the automation path becomes known, while the short rate does not move. [Estimate]

Who gains and what compensation costs

Split households into owners of capital and hand-to-mouth households that consume their labor income, at the 30% share Kaplan, Violante and Weidner measured for the United States. Owners gain the equivalent of 244% of their consumption under unit substitutability; hand-to-mouth households lose 40% to 41%, and 65% to 66% under strong complementarity. [Estimate] A fixed saving rate does not spare them: the loss is 42% to 68%. [Estimate] A balanced-budget tax on net capital income that keeps labor income on trend peaks at 3.5% of net capital income in the first case and 22% in the second, then fades below 1.5% by 2065, because the transfer grows at 1.6% a year and output at 11% to 22%. [Estimate]

What it means for which occupations

The paper names factors, not job titles. Its dividing line runs between people who own capital and people who live on wages, and the wage-only group loses in every calibration. For economists, the testable claim is that the saving rate should track the capital share; investment stuck near a fifth of GDP while the capital share climbs would mean the fixed-saving world. For financial and investment analysts, the asset-price signature is a jump in long real yields with no move in the short rate or the stock market on impact. For personal financial advisors, the household split is the client question.

What to prepare

Under unit substitutability, automation starts with a consumption boom, higher wages and a dip in saving; under strong complementarity, consumption falls at once, saving jumps and the wage is below trend by 2032. Watch which of the two opens. The compensation tax peaks around 2035 to 2042 and fades after, so transfer design is a decision for the next decade.

Limits

This is a calibrated model, not measured data: the automation path is assumed, the household either saves optimally or at a fixed rate, and capital ownership does not change hands. The author lists political economy as the main omission, since a wage at a quarter of trend for two decades may not be permitted by voters. The paper is a working paper without peer review.

Sources

• Fernández-Villaverde, J., "Saving for the Machines," NBER Working Paper 35964, National Bureau of Economic Research, October 2026: https://www.nber.org/papers/w35964

About this analysis

This article was written with AI assistance from the full text of the NBER working paper (75 pages). Occupation links point to our job pages; the mapping from the paper's factors to those pages is ours.