The Work of Chad Jones

One of the great economists of our times

Chad Jones is one of the few people who I feel in awe of. Writing this review of his work felt, at the outset, like climbing up a cliff of glass. It is so clearly written, and on such important topics, that to compress any piece of it felt like a slight; even blasphemy. My overview should not be taken as comprehensive — I encourage referring to the original papers, and even reading everything which he has ever written. He is worth it.

Work:

Jones’s work seeks to explain why economic growth occurs, why some countries are richer than others, the role of research and development in economic growth, what we should do about artificial intelligence, and incorporating life, death, and birth into economic models.

Jones first worked on endogenous growth models. Growth theory began in earnest with Solow (1957), who broke down production into two factors, labor and capital, with labor multiplied by A, representing technology. Because capital has declining marginal returns and we must pay for depreciation, a steady state exists in which there is no growth. In order to accurately describe economic growth, we must have exogenous technological change. Exogenous technological change is obviously unrealistic, and more importantly it did not well explain the world. The decreasing returns to capital implied that poor countries should be rapidly converging to the level of rich countries. As Bob Lucas notes, since the return to capital should be much higher in poor countries, we should expect capital to flow to poor countries.

To explain this, economists sought to endogenize idea creation. The rate of technological growth is determined by our investment into finding new ideas. Takeoff into infinite technological growth in finite time is prevented by companies either not receiving the full social value of their inventions (Romer) or by businesses being able to take the business of firms which invested earlier (Aghion-Howitt). Depending on the rate at which ideas spread, and how they spread across borders, income in different countries can permanently diverge.

Chad Jones’s first papers challenged this new paradigm. In “Time Series Tests of Endogenous Growth Models”, he argues that the new models give strong predictions about government policy changes. If government policy permanently changes the return to investment, or openness to trade, then we should expect it to show in the growth rate in GDP per capita. Since it hasn’t, “either by some astonishing coincidence all of the movements in variables … have been offsetting, or the hallmark of the endogenous growth models … is misleading.” (p. 496)

In “R&D-Based Models of Economic Growth”, he shows that if there are decreasing returns in finding new ideas, then in the long-run we revert back to the simple story of the Solow model. Economic growth is once more determined by variables commonly taken to be exogenous, like population growth. Years later, he would return to the topic with Nicholas Bloom, John Van Reenen, and Michael Webb, and test “Are Ideas Getting Harder to Find?”. We have continued to find new ideas, but only at the cost of continuously raising research inputs. Between the 1930s and today, total factor productivity has risen by about 2% every year, while the effective number of researchers has increased nearly 25 times over. In computer chips, Moore’s Law (that the density of computer chips doubles every year) has held, but a doubling now requires 18 times as many researchers as it did in the 1970s. We face the spectre of eternal stagnation, without a substantial change in how we find ideas.

Ideas are of profound importance to economic growth in Jones’s telling. Ideas are the quintessential example of a public good – once invented, everyone is able to copy the ideas, without necessarily paying the inventor. As Thomas Jefferson wrote, “He who receives an idea from me, receives instruction himself without lessening mine”. Since finding the idea took considerable investment, we underproduce ideas relative to the social optimum. Jones and John C. Williams (1998) estimate that, at a minimum, the optimal rate of investment into research and development is two to four times higher than the actual level of investment. This implies that the socially optimal tax rate may be much lower than commonly estimated. Typical optimal tax papers assume that there are no spillovers from labor, and that taxation should simply maximize government revenue. Allowing for ideas to be disproportionately generated by high-earners, as Jones turns this on its head, and suggests that, if anything, marginal tax rates on the very highest bracket should be smaller than those with less income!

A possible source of increasing research productivity is artificial intelligence. With Philippe Aghion and Ben Jones, Chad Jones wrote the formative work on modelling artificial intelligence. If AI enters into the discovery of new ideas, then it is possible to get a “singularity”, or infinite growth in finite time. If, however, production is constrained by sources which are hard to automate, then we should expect the gains to go to the scarce factor. This follows from Baumol’s theorem (1967), where productivity growth in one sector increases the opportunity cost, and hence the wages, of a stagnant sector. This artificial intelligence could be misaligned, however, and do things which might destroy humans. Jones (2016) and Jones (2025) investigate how much we would be willing to pay to prevent catastrophe. With log utility, we are remarkably unconcerned about existential risk. If our coefficient of relative risk aversion is greater than two, then we would be willing to spend a very large portion of GDP on risk-mitigation. The exception is if technology portends to increase life-span, in which case we should be more willing to take risks of catastrophe.

He has also studied life, death, and well-being. “The End of Economic Growth? Unintended Consequences of a Declining Population” explores the consequences of falling birth rates. Since, in many models, a greater number of people leads to a greater number of ideas, a falling birth rate leads to living standards stagnating as we slide toward extinction. What’s more, if we make a standard assumption that parents derive utility from their children’s well-being, it is possible for the optimal outcome to be human extinction, if we wait too long. There are only two stable outcomes – one where living standards grow without bound, and extinction. In “Population and Welfare” (with Mohamad Adhami, Mark Bils, and Pete Klenow) he proposes extending GDP to take into account differences in the number of people able to experience heightened living standards.

Jones, with Robert Hall, had an important conceptual influence on Daron Acemoglu, Simon Johnson, and Jim Robinson’s theory of institutions as the fundamental cause of differences in long-run growth. In “Why Do Some Countries Produce So Much More Output per Worker than Others”, they attribute growth to particular sources. Greater physical capital intensity or human capital attainment accounts for only a small part of differences in productivity. They attribute the rest to “social infrastructure”, a precursor to the institutions of Acemoglu, Johnson, Robinson. What’s more, since the rate of capital accumulation is to some degree endogenous to social infrastructure, differences in output directly attributable to capital intensity may all ultimately be due to social conditions.

Jones has also had an influence on the study of production networks in the economy, with “Intermediate Goods and Weak Links”. Beginning with Michael Kremer’s “The O-Ring Theory of Economic Development”, economists have recognized that production occurs in many steps, errors in any one of which can wreck the entire product. Improvements can then have a non-linear impact on production. For example, if government mismanagement of power plants raises the cost of electricity, this reduces output in banking and construction. If it is harder to finance and construct projects, then it is harder to improve productivity in electrical generation. In addition, productivity in a firm depends on performance along multiple dimensions. He gives the example of how “textile producers require raw materials, knitting machines, a healthy and trained labor force, knowledge of how to produce, security, business licenses, transportation networks, electricity, etc. These inputs enter in a complementary fashion, in the sense that problems with any input can substantially reduce overall output. Without electricity or production knowledge or raw materials or security or business licenses, production is likely to be severely curtailed.” This can be seen as fundamentally analogous to his work on AI. Growth might be constrained not by the rate of progress in what we improve best, but in what we can improve least. These frictions can lead to profound and permanent misallocation.

Jones, with Hsieh, Hurst, and Klenow, has contributed to our understanding of misallocation in the labor market. Between 1960 and 2010, the fraction of doctors and lawyers who were white men fell from 94% to 62%. If people’s innate abilities were the same over time, then many people were being sorted into the wrong jobs. They estimate that 44% of U.S. economic growth between 1960 and 2010 was due to the better matching of workers and jobs. Jones, Brouillette, and Klenow estimate, however, that the gap in welfare for Black Americans, compared to white Americans, is considerably higher than the gap in consumption.

Lastly, in the mid-2000s Jones wrote several papers to explain rising expenditures on healthcare. The first of the papers emphasizes the marginal utility of a dollar spent rising with technological improvements. Medical technology allows us to do things which were not available at any price in the past. Thus, people will spend more on healthcare as a percentage of their incomes as technology advances. The other, with Robert Hall, explores many of the same themes as his work on AI risk mitigation. As our consumption rises, we value the time in which to consume more, because the first unit of consumption in each period does not have declining marginal returns.