Conversable Economist

It is of course impossible to predict the effects of new AI technologies on total economic growth in a precise way, but there are strong reasons to believe that the effect will be modest for a decade or two, anyway. Charles Jones Jones explores some of the reasons in “AI and Our Economic Future,” in the most recent Journal of Economic Perspectives (Summer 2026). His arguments readily accept that AI capabilities both have grown very rapidly and will continue to grow very rapidly: indeed, Jones begins by stating that “Artificial intelligence (AI) will likely be the most transformative technology of the modern era.” But he also suggests that growth of this particular technological capability will add to total economic growth in only a modest way for the medium term.

(In the shade of this parenthesis, I will note that I work as Managing Editor of the Journal of Economic Perspectives. Because both the JEP and this post are freely available, I have no financial conflict of interest here. But I confess an intellectual conflict of interest: I do think that the stuff published by the journal I help to run is interesting.)

As a starting point, consider this graph of US economic growth since 1870. The red dashed line shows a growth rate of 2% per year. The blue solid line shows actual growth of average US income per person. You can see economic events like the fall of the Great Depression, the bounceback during World War II and after, but amidst all the events of the last 150 years, 2% growth on average is a pretty good description of what has happened.

Jones writes:

Consider the astounding innovations that underlie the graph. In the 1870s, Thomas Edison’s experiments with electric lighting were just getting underway. Fifty years later, electrification had transformed the economy, both in factories and in city life. Throughout the 150 years, innovations such as the internal combustion engine, airplanes, vacuum tubes, antibiotics, transistors, semiconductors, personal computers, and the internet profoundly changed living standards. Many of these innovations are what economic historians call “general purpose technologies,” whose transformative effects extend throughout the economy. Many also automated some of the tasks involved in creating new ideas—say, through improvements in scientific tools and equipment—raising the productivity of research and idea generation. Yet apparently none of these innovations changed the long- run growth rate of the US economy.

To put it another way, it may be that the extraordinary run of innovations of the last 150 years, a disproportionate number of them with US roots, were needed just to keep the economy growing at that 2% annual pace. But the idea that a fast-growing new technology with broad implications for work and products across the economy will juice economic growth beyond the historical average just hasn’t happened.

Indeed, the US economy has a vivid example in the past few decades of a general purpose technology with sweeping implications for work and the economy that became amazingly and dramatically cheaper. As Jones points out: “You have on your desk roughly 100 million times the computing power that was available on the best computers from the early 1970s. But you and I are not 100 million times more productive …” If you had known back in 1970 that computing power would become 100 million times cheaper in the next half-century, would you have predicted that the US economy would just keep tootling along with 2% growth? In hindsight, of course, it’s easy to say “yes, it’s all obvious to me.” Or would you have been more likely to believe that most workers would become obsolete and the economy would be unrecognizably changed?

The “O-ring theory of development” also known as a theory of weak links is one way to organize thinking on this subject. Jones explains:

An iconic illustration of weak links arose with the accident that destroyed the space shuttle Challenger in 1986. The post-crash investigation found that the key design flaw was in the O-rings: doughnut-shaped pieces of rubber that are designed to create a seal between two interlocking parts. Challenger launched during record-low temperatures causing the O-rings to become brittle and leak, which caused the space shuttle to explode. Partly inspired by this episode, Kremer (1993) developed his O-ring model of economic development.

Basically, the O-ring theory points out that the speed of any complicated process will be limited by its most fragile or difficult-to-replace part. There are probably certain parts in your car that could handle a speed of 300 miles-per-hour; but there are other parts that cannot handle that speed, and they limit how fast you can go. From this perspective, the question for total economic growth is not how fast AI will progress, but what other non-AI factors will limit the speed of growth.

Say that AI tools make all computer software infinitely cheap–essentially a price of zero. Well, software spending is now about 2% of the US economy. If firms and consumers could have and use all the software they want at a price of zero, how much would the total economy expand? Sure, it would be a boost for output. But it wouldn’t be infinite growth. Of course, if 100% of all production could be automated at near-zero cost, then the gains for consumption (everything for nothing!) could be very large. But that outcome is hard to imagine. For example, AI is not going to create more land or natural resources, so even if AI can give us all infinite numbers of on-screen entertainment channels, some imperatives of scarcity and geography will continue to apply for producing energy and physical goods. Organizational and political change in response to AI capabilities can be a limiting O-ring factor, too.

So what about scenarios between 2% of everything being automated and (essentially) free, and 100% of everything being automated? Jones offers some suggestive but pursely illustrative calculations based on assumptions about AI substituting for large and smaller parts of output; for example, one calculation suggests that even if AI could substitute for 50% of economic inputs by replacing all cognitive labor, total GDP would rise by 19%. And of course, in a world of O-ring change, the process of AI making these changes of this size is not going to happen over a year or two, but instead over a few decades.

Again, none of this is to say that AI will not advance rapidly and cause major changes, only that from the perspective of total economic growth, these changes are likely to appear relatively modest for years, even for a couple of decades.

The Jones paper has a number of other insights, but I’ll just add one of them here. In discussions of AI, there is enormous focus on jobs and economic output, but considerably less focus on possible gains to health. One of the great strengths of AI is to focus on a well-defined problem (think chess, or a math proof), and then to draw on a vast body of knowlege to suggest possible answers. Working with AI tools, it seems at least possible that researchers will be able to speed up their progress on addressing health risks of cancer, heart disease, high blood pressure, diabetes, Alzheimer’s, and a wide range of other health risks. Maybe you aren’t at all sure that AI is worth the social disruption just for the economic growth. But what if AI-based research offers a chance for the average person now alive to enjoy several decades more of healthy life? It could be that the non-economic gains from AI-based research (better health? clean energy?) turn out to be more important for human well-being than its effects on total economic growth.

 

The US unemployment rate was 4.2% in June, which is low by historical levels. Nonetheless, lot of workers, and perhaps expecially young workers right out of college, are experiencing the US economy as a situation where it’s hard to get a job. For example, in the “The Job Market for Recent College Graduates,” I reviewed evidence that relative to other education groups, the unemployment rate for recent graduates is higher than usual, while in certain jobs, the job growth for those age 22-27 seems lower than usual. In “Are the Entry-Level Jobs Drying Up for Young Adults?” I discussed flows between unemployment, being out of the labor force, and taking a job. There’s are some bits of new evidence on each question.

Bruce Fallick of the Cleveland Fed discusses “The `Low-hire, Low-fire’ Labor Market” (Economic Commentary 26-17, August 4, 2026). The green line on the graph shows the percent of new hires each month as a share of total employment; the maroon line shows the quit rate each month as a share of total employment; the blue line shows the layoff rate. (The pandemic years are left out, because the labor market goes whacky during that time.) Notice that layoffs have trended down a bit over time, but not moved much. However, in the last few years the hire and quit rates have both been dropping.

A high quit rate (that is, leaving a job voluntarily) is sometimes used as a measure of whether the job market is strong, because a strong job market means that people have more options for voluntarily leaving one job and moving to another. Conversely, you can see the quit rate and the hire rate both falling during the Great Recession from 2008-2010; the labor market was weak, so voluntary departures were less common. But it’s unusual to have the hire rate and quit rate falling at a time of relatively low unemployment.

Fallick looks back at other data sources and finds that this pattern of decline in the “fluidity” of the US labor market goes back at least to the 1980s. But this pattern seems to be evolving over decades, and its underlying causes are not clear. One possible reason is that the general aging of America means that the share of workers in mid-career and older is high. If such workers are less likely to leave voluntarily, then quit rates would be lower and the need to hire would decline as well.

Another piece of labor market evidence comes from Ingrid Chen, Marianna Kudlyak, and Riva Mikhlin at the San Franciso Fed, who discuss “Job-Finding Anomalies of the Current Expansion” (FRBSF Economic Letter 26-20, August 5, 2026). Their starting point is that working-age adults can be divided into three groups: the employed who have jobs, the unemployed who don’t have a job but are actively looking, and the out-of-the-labor for who don’t have a job but are not actively looking. One can look at movements between these categories. The authors explain this way:

The job-finding rate for unemployed people is typically 25–30%, much higher than the typical job-finding rate for the nonparticipation group of 4–5%, showing that the unemployed are far more likely to find a job (Figure 1). This is not surprising because a large share of nonparticipants are retirees or students who are not looking or not interested in a job. However, even though their job-finding rate is relatively low, nonparticipants make up two thirds of all people who transition from nonemployment to employment each month on average. This is mainly because the pool of nonparticipants is much larger than the pool of unemployed individuals.

The graph is a little tricky to read, because the green line is being measured on the left-hand axis and the blue line on the right-hand axis. But looking at the green line and the right-hand axis, you can seem that the movement from unemployment to a job has dropped from about 30% to in 2023 to about 25% at present. So yes, it has become harder for someone who is actively looking for a job to find one.

Now break down these patterns by age group, and look at the time period from January 2023 to January 2026. As the green bars show, movement from unemployment to a job drops off for all groups. However, movement from out-of-the-labor-force into a job drops off especially strongly for younger adults. To put it differently, for young adults entering the labor market, it has indeed become more difficult to find a job than it is for those in other age groups.

None of these patterns suggest that it’s a new Great Depression out there. But the patterns do suggest that the labor market is less fluid, which makes it hard for those trying to gain a foothold.

I have been the Managing Editor of the Journal of Economic Perspectives since the first issue in Summer 1987. The JEP is published by the American Economic Association, which decided back in 2011–to my delight–that the journal would be freely available online, from the current issue all the way back to the first issue. You can download individual articles or entire issues, and it is available in various e-reader formats, too. Here, I’ll start with the Table of Contents for the just-released Summer 2026 issue, which in the Taylor household is known as issue #157. Below that are abstracts and direct links for each of the papers. I plan to blog more specifically about some of the papers in the few weeks, as well.

_____________________

Symposium on Artificial Intelligence

“AI and Our Economic Future,” by Charles I. Jones

Artificial intelligence (AI) will likely be the most transformative technology of the modern era. What if machines—AI for cognitive tasks and AI plus advanced robots for physical tasks—can perform every task a human can? This essay makes three main points. First, even though US growth rates have been stable at roughly 2 percent per year for 150 years, it is distinctly possible that automating intelligence leads economic growth rates to accelerate. Second, this acceleration is likely to be slowed by the presence of “weak links.” While we each have access to 100 million times more transistors on our desktop computer than people in the 1970s, we are not 100 million times more productive. Computers can invert matrices at lightning speed, but we humans must still decide what matrix to invert, what hypothesis to test, and so on. Accelerating economic growth requires the vast majority of the weak links to be automated away, which delays the large gains. Finally, even though weak links slow the benefits, they may actually speed up the risks. When a chain is only as strong as its weakest link, damaging one link in the chain can be very costly. A powerful AI that is superhuman at software engineering could be misused by a bad actor to do substantial harm by hacking the financial system or a virology lab.

“The Emerging Market for Intelligence: How Firms Buy and Sell AI,” by Mert Demirer, Andrey Fradkin, and Nadav Tadelis

    We describe the emerging business-to-business market for large language model (LLM) inference and document key empirical patterns in its supply, pricing, and dynamics, using data from OpenRouter. First, supply has expanded rapidly: the number of commercially available models, model creators, and inference providers has grown sharply, driven heavily by open-source entrants. Second, the price of intelligence has fallen roughly a thousandfold, and open-source models now cost about 90 percent less than comparable closed-source ones. Third, the market is highly dynamic, with frequent turnover among leading models and creators. Fourth, we document substantial horizontal and vertical differentiation: no single model dominates across use cases, and demand for intelligence varies widely across applications. We place these patterns in historical perspective alongside earlier general-purpose technologies.

Symposium on Tariffs

“US Tariff Policy since 1789,” by Miguel Acosta, Lydia Cox, Andrew Greenland, John Lopresti, Christopher M. Meissner, Martin Rotemberg, and Sharon Traiberman

We use a newly assembled, tariff-line-level dataset spanning the full history of US trade policy to revisit the evolution of tariffs since 1789. We document the institutional shift from Congressional setting to multilateral negotiation, the steady growth in granularity of the tariff code alongside expanding administrative capacity, and the under-appreciated role of specific tariffs, which feature prominently throughout US history and whose ad valorem equivalent moves mechanically with prices. We discuss the implications of this relationship for how we interpret past liberalization episodes and for empirical identification of tariff effects. Finally, we outline how the new data can advance research on the political economy of tariff-setting and on the macroeconomic and distributional consequences of trade policy.

“Labor Market Responses to Tariffs: Frictions, Dynamics, and Policy Responses,” by Rafael Dix-Carneiro and Brian K. Kovak

This article introduces the evidence and associated modeling frameworks contemporary economists use to understand the effects of trade and trade policy on labor markets, with a particular emphasis on labor-market frictions and adjustment dynamics. The effects of trade shocks differ across industries, regions, and occupations, implying the presence of important adjustment frictions in labor markets, and these effects evolve slowly over time, implying the need for dynamic frameworks rationalizing slow transitions. After reviewing the key insights from this literature, we discuss policies aimed at mitigating costs to workers and ensuring that the gains from trade are shared more equitably.

“Should We Tax Trade? A Pigouvian Perspective,” by Arnaud Costinot and Iván Werning

 We develop a simple and intuitive Pigouvian perspective on optimal trade policy. Our approach unifies a wide range of rationales for taxing trade, from the classical optimal tariff argument to contemporary debates about global carbon emissions and geopolitics. We also clarify when trade policy intervention is warranted and when alternative domestic instruments should be used instead.

“The Incidence of Tariffs: Rates and Reality,” by Gita Gopinath and Brent Neiman

In 2025, statutory tariff rates on US imports rose to levels not seen in over one hundred years. What were the implications for prices? On the one hand, shipping lags, exemptions, and enforcement gaps kept the actual implemented rates at only half of the statutory rates, moderating the tariffs’ impact. On the other hand, tariff pass-through to US import prices reached 92 percent, so the United States bore a large share of the costs. We study the incidence of the 2018–2019 and 2025 US tariffs and discuss implications for US sourcing, domestic manufacturing costs, and the dollar.

“Global Imbalances, Tariffs, and Industrial Policy,” by Pierre-Olivier Gourinchas, Gene Kindberg-Hanlon, Manasa Patnam, Lorenzo Rotunno, and Michele Ruta

Global imbalances denote the distribution of countries’ current account balances, identically equal to the difference between two forward-looking aggregate variables: national saving and domestic investment. Industrial and trade policies have traditionally not been considered important drivers of aggregate saving or investment, and therefore of current account balances. The former because most industrial policies are small in scope; the latter because permanent tariffs have no intertemporal effect in the textbook model, with an offsetting appreciation of the real exchange rate. The rapidly growing use of both industrial and trade policies in recent years calls for a reassessment. This paper presents a framework to think about the role of both policies. For industrial policy, we make the important distinction between the traditional sector-specific policies via subsidies or other targeted instruments (“micro industrial policy”) and broader policies (“macro industrial policy”) that aim to promote industrial developments and competitiveness through the deployment of more aggregate instruments such as financial repression, foreign reserve accumulation, or capital controls. A key finding is that micro industrial policy tends to increase external balances if it fails to raise aggregate productivity. By contrast, macro industrial policy can, under some conditions, boost the current account, forcing other countries to adjust. Yet, these policies often come at the cost of suppressed domestic consumption and possibly domestic welfare. Our analysis confirms that tariffs are a weak tool to improve current account balances. Finally, traditional macroeconomic drivers—such as fiscal policy, demographics or credit cycles—remain critical drivers of global imbalances, especially for the United States and China.

“Evaluating the Fiscal and Distributional Implications of Tariffs,” by Kyle Pomerleau and Erica York

The Trump administration has enacted a series of significant taxes on imports, including an across-the-board tariff on nearly all imports. Tariffs are now expected to raise nearly 1 percent of GDP in revenue. Given tariffs’ growing importance as a source of revenue, it is important to understand their implications for federal finances and the wellbeing of households. This paper describes and reviews current revenue and distributional analyses of tariffs.

Symposium on Instrumental Variables

“Correct (and Incorrect) Inference with a Single Instrumental Variable: Practical Takeaways from the Weak Instruments Literature,” by David S. Lee and Jack Porter

 Most empirical economists have encountered the warning that instrumental variables can be “weak,” but the underlying issues—what makes an instrument weak, why weakness distorts inference, and what to do about it—are less widely understood. This article offers an accessible introduction to the weak instruments problem for the common just-identified case of a single endogenous regressor and a single instrument. We explain why the usual two-stage least squares t-ratio and its “±1.96 times the standard error” confidence interval can yield incorrect inferences, much as homoskedasticity-only standard errors do when errors are not homoskedastic. We then describe practical, robust-to-weak-instrument solutions—including the Anderson-Rubin and tF methods—that deliver valid confidence intervals whatever the instrument’s true strength, and we offer some do’s and don’ts, notably why the popular “F greater than 10″ rule has no theoretical justification in this setting.

“Leniency Designs: An Operator’s Manual,” by Paul Goldsmith-Pinkham, Peter Hull, and Michal Kolesár

We develop a step-by-step guide to leniency (a.k.a. judge or examiner instrument) designs, drawing on recent econometric literatures. The unbiased jackknife instrumental variables estimator (UJIVE) is purpose-built for leveraging exogenous leniency variation, avoiding subtle biases even in the presence of many decision-makers or controls. We show how UJIVE can also be used to assess key assumptions underlying leniency designs, including quasi-random assignment and average first-stage monotonicity, and to probe the external validity of treatment effect estimates. We further discuss statistical inference, arguing that non-clustered standard errors are often appropriate. A reanalysis of Farre-Mensa et al. (2020), using quasi-random examiner assignment to estimate the value of patents to startups, illustrates our checklist.

Articles

“Stefanie Stantcheva, 2025 Clark Medalist,” by James Poterba and Iván Werning

The American Economic Association awarded the 2025 John Bates Clark Medal to Stefanie Stantcheva of Harvard University. Her research ranges widely in the field of public economics. It includes theoretical contributions to optimal income taxation, empirical contributions to the study of how income tax rates across countries and states affect the mobility of inventors and the level of innovation, and contributions in the development and execution of on-line surveys to measure respondents’ beliefs about economic primitives, such as the distribution of income or the rate of inflation, and the way these beliefs influence policy preferences. She has demonstrated that randomized controlled trials can be embedded within on-line surveys and used to the way informational interventions and other treatments affect economic beliefs and policy preferences.

“Recommendations for Further Reading,” by Timothy Taylor

Back in the pre-pandemic year of 2019, the European Union passed the European Green Deal, setting a goal of having the EU be climate-neutral in its emissions by 2050. Here in 2026, 20% of that 30-year time window has now passed. The government of the EU have a reputation for strong action toward these types of goals. How is it going? Magnus Henrekson,  Christian Sandström  and Mikael Stenkula have edited a collection of 16 essays that provide a cautionary progress report in A Green Entrepreneurial State? Exploring the Pitfalls of Green Deals (Springer International Studies in Entrepreneurship
Volume 49, 2026, open access).

In their introductory overview essay for the volume, Henrekson, Sandström, and Stenkula point out that the EU is embracing a wide array of policies: not just solar and (on-shore and offshore) wind, but also hydrogen, zero-emission cars, improved batteries, and much else. For example, “all new passenger vehicles and light commercial vehicles registered from 2035 must be zero-emission.”

The Hydrogen Strategy envisions the installation of “at least 40 GW of renewable hydrogen electrolysers in the EU by 2030, producing up to 10 million tonnes of renewable hydrogen. REPowerEU further strengthens this by calling for both 10 million tonnes of domestic production and an additional 10 million tonnes of imports by 2030 . Hydrogen is viewed as a key energy carrier for sectors where direct electrification is challenging, such as steel, chemicals, and heavy transport.” For perspective, 40 gigawatts is roughly equal to 25 new nuclear reactors.

The EU Offshore Renewable Energy Strategy includes “60 GW of offshore wind by 2030 and 300 GW by 2050. … To put this in perspective, the 2050 target would be nearly five times the total installed capacity of France’s 57 nuclear reactors, which stood at around 63 GW in 2025.”

In short, these and other proposals are enormous in scale. Several of the papers focus on specific programs. From the descriptions of the editors:

In their essay “Italy’s Superbonus and the Capture of Climate Policy by Modern Monetary Theory,” Luciano Capone and Carlo Stagnaro (2026) analyze how Italy implemented an environmental policy that produced the largest budget deficit in Europe since the Second World War. The program allowed households to claim a 110% tax credit for expenses incurred in improving the energy efficiency of buildings and enhancing their seismic resilience. These tax credits were fully transferable to third parties, such as construction companies or financial institutions. In total, the policy generated costs amounting to approximately EUR 220 billion, equivalent to about 10% of a single year’s GDP, while delivering limited environmental benefits and fostering widespread tax fraud. …

Launched in the early 2000s, Germany’s Energiewende has long been seen as a model for large-scale renewable energy transitions. By rapidly expanding solar and wind power, Germany aimed to replace both nuclear and fossil fuels, inspiring the European Green Deal and its goal of EU-wide carbon neutrality by 2050. Germany itself pledged to reach neutrality by 2045. In “The German Energiewende: A Green Deal Template or Planned Failure?”, Michel Deshaies (2026) highlights several weaknesses. To replace stable nuclear and fossil-based generation with variable wind and solar, three key conditions are required: large overcapacity, extensive and costly grid expansion, and massive storage capacity. Yet Germany focused mainly on deploying renewables, giving far less attention to grids and storage. Even optimistic projections for hydrogen or other energy carriers cannot eliminate the need for huge renewable capacity. Electricity makes up only about 20% of Germany’s total energy use; the remaining 80%—mainly transport, heating, and industry—still depends on oil and gas. This means deep decarbonization must go far beyond the power sector.
The Energiewende has also driven sharp increases in household electricity prices … leaving Germany with the highest household electricity costs in Europe, around 40% above the EU average. …

In “Explaining Northvolt’s Bankruptcy and the Dilemma of Green Deals,” Christian Sandström (2026) examines the unintended consequences of green industrial policies by analyzing the bankruptcy of Northvolt, Europe’s largest initiative to establish an independent battery manufacturing facility. Founded in 2017 and growing rapidly to nearly 6,000 employees by 2023, Northvolt struggled to scale production and remained dependent on Chinese suppliers—undermining EU’s ambition of strategic autonomy. In March 2025, the company filed for bankruptcy.

For me, these kinds of grand plans offer a fascinating blend of technological optimism that these highly aggressive goals are in fact achievable (maybe some are, but not all), along with occasional outbreaks of technological pessimism that households just need to become accustomed to using less energy. Several political economy papers seek to diagnose the political process that is leading to these outcomes.

In his contribution entitled “Behavioral Political Economy and Environmental Policy: Explaining Persistent Deviations from Efficient Policies,” Jan Schnellenbach (2026) highlights that the design and implementation of Green Deals are almost never guided by neutral efficiency considerations alone. Instead, they are deeply shaped by behav-ioral dynamics biases, heuristics, and expressive political behavior—which distort both citizen preferences and policy choices. … As a result, citi-zens often embrace “bliss beliefs” about the environment, views that signal virtue or identity rather than reflect careful cost–benefit assessments. By contrast, dissenting opinions are stigmatized as unvirtuous and carry high social costs that few are willing to bear. Politicians, in turn, respond to these expressive preferences … According to Schnellenbach, this dynamic renders Green Deals vulnerable to emotionally salient but economically inefficient outcomes. …

In their contribution entitled “Raiders of the Entrepreneurial State: A Baptist and Bootlegger Analysis,” Jeffrey Muldoon and Derek Yonai (2026) critically examine the notion of an “entrepreneurial state” (Mazzucato 2013), and reframe its implica-tions for innovation, regulation, and rent-seeking through the lens of the “Bootleggers and Baptists” theory developed by public choice scholars. Muldoon and Yonai argue that instead of stimulating productive entrepreneurship, state-led industrial policy opens abundant opportunities for “plunder” by well-connected insiders (“Bootleg-gers”) who exploit the state’s moral legitimacy, often provided by well-meaning “Baptists” advocating the common good.

Consider two broad approaches to reducing carbon taxes. One approach imposes a tax on carbon emissions from all sources, and uses some funding from that tax for public funding of active research and development efforts into clean energy, along with demonstration projects. This approach relies on altering incentives and being at the leading edge of technology, but leaves considerable flexibility in how energy users react. In the other approach, the government hands out large subsidies and/or tax breaks to companies and households who undertake specific actions, along with specifying certain technologies and timetables and trying to set up “national champion” production facilities and firms. This second approach seems to have greater political viability, perhaps because it is also easier for lobbyists to game the system and for politicians to take credit, but it’s far from obvious that it has better outcomes.

The tools that are used for public policy matter more than the announced timelines and goals. It’s legitimate to ask whether the policies that make up the EU green deal (and similar proposals in other countries) are actually working to reduce carbon emissions in a cost-effective manner, or whether they mostly involve handing out subsidies and tax breaks.

Lots of students come to college looking for what I sometimes call a “golden ticket” major–that is, choosing a major that will assure future prosperity and status. Lots of first-year students show up declaring that their career path is to become a doctor, but a certain percentage of them make other plans after getting their butt kicked in their first real college-level intro chemistry or intro biology course. The huge increase in computer science majors in the last decade or so was at least in part driven by a belief that such a major would assure a well-paid career–a belief that has not aged well with the arrival of AI tools.

But more broadly, undergraduate students often have an unrealistic set of beliefs about how certain majors will with high probability lead to certain jobs, which then leads to disappointment. John J. Conlon and Dev Patel explore the issue in their research paper “What Jobs Come to Mind? Stereotypes About Fields of Study” (recently published online in Quarterly Journal of Economics, June 22, 2026). They write:

Almost half of employed U.S. college graduates report that their job is not closely related to their degree, and mismatch between major and occupation strongly predicts job dissatisfaction, lower salaries, and the rate at which graduates say they regret their choice of what to study. What explains these patterns? One possibility is that they primarily reflect uncertainty—for instance, about undergraduates’ future job preferences—resolving into ex post regret. But a second possibility is that they in part reflect systematic ex ante mistakes, for example, due to student misperceptions about the consequences of major choice. Identifying the nature and magnitude of any such mistakes would point toward policy interventions to improve student outcomes. …

We test this prediction first through surveys run in partnership with an academic program at the Ohio State University (OSU). Our respondents are first-year undergraduates who are undecided about their major. The main survey questions simply ask students their beliefs about the share of U.S. graduates with different majors who are working in various careers. We find large and systematic stereotyping: students overestimate the preva-lence of almost every major’s representative career. For example, the average student believes that 53% of art majors work as artists (17% do), that 47% of journalism majors work as journalists (4% do), that 38% of political science majors work as lawyers (16% do), and that 43% of psychology majors work as counselors (21% do). Very similar patterns appear in students’ beliefs about majors they are versus are not likely to pursue and in their beliefs about the careers they themselves are likely to have …

To provide evidence on the breadth and persistence of such stereotyping, we analyze a large-scale, nationally representative sample of U.S. college first-year students including more than 9 million respondents over 40 years. We find that dramatically more first-years expect to attain their major’s representative career than actually end up working in that job: 63% of biology majors expect to be doctors (versus 23% in reality), 62% of psychology majors expect to be counselors (versus 21%), 65% of art majors expect to be artists (versus 17%), 42% of communications/journalism majors expect to be writers or journalists (versus 4%), and so on. Pooling across majors, these facts combine to produce large gaps between the expected and actual careers that college graduates pursue. For professions that are primarily the rare-but-representative outcome of particular majors—doctor, counselor, journalist—two to four times more college first-years expect to work in these jobs than actually do. In contrast, many fewer expect to be teachers, work in business, or be nonemployed than ultimately are, be- cause these are the common alternatives to the representative career of many majors. These overall gaps—amounting to 40,000 to 200,000 students a year—appear largely unchanged since at least the 1970s.

The authors carried out an experiment at Ohio State University, a “light touch intervention,” where some students were given actual information on the share of student in a given major that ended up in certain high-profile jobs. The student who got the information were likely to remain undecided longer about declaring a major, and also to end up in a different major than the one toward which they had originally been leaning.

My own sense, based on not much (if you need a fancy name, call it “casual empiricism”), is that the faculty who are teaching undergraduates are often not well-informed about the actual job market outcomes of their students. But faculty want to be encouraging, so if someone says that they want to be an artist, journalist, lawyer, or doctor, they are counselled that a certain major is a good way to head toward that goal. It’s a rare faculty member who would have data available on what those who majored in their subject actually end up doing, and who would tell students that for almost half of them, their eventual job will not be closely related to their major.

The potential benefit of having the government require that certain jobs require an official license is quality control and protection. Personally, I rather like knowing that my nurse or doctor or dentist has gone through courses and training. The potential cost is that requirement to have a government license can become a way for those who have a job to limit the number of people who can do that job, and thus to push up their own wages, rather than offering benefits to consumers. What economists call a “Baptists-and-bootleggers” coalition can form (for more background, see here and here). The Baptists and the bootleggers both supported legal limitations on alcohol: the Baptists for moral concerns, and the bootleggers because legal limits made their business more profitable. In every job, there is inevitably going to be an example or two where a customer is treated in a way that is personally harmful or financially costly or both. A coalition will form where one set of participants wants occupational regulation to reduce the risk of these harms, while existing workers in the field will favor occupational regulation to limit the competition they face.

For an overview of the economics (and thus the tradeoffs) of occupational licensing, a useful starting point is the essay by Janna E. Johnson in the Winter 2026 issue of the Journal of Economic Perspectives, “Occupational Licensing in the United States.” For an essay about competition in the market for physician care–a case where some regulation seems justified to me, but the regulation has also been used to limit supply–a useful starting point is Joshua D. Gottlieb and Sean Nicholson, “Physician Competition: Entry and Substitution,” in the Spring 2026 issue of the Journal of Economic Perspectives. As they point out:

From 1980 to 2025, the US population grew by 50 percent (World Bank 2025). The population above age 64 grew by 140 percent, while that above age 80 nearly tripled; these groups use a disproportionate share of health care and might thus be a better proxy for demand than total population. Over the same 1980–2025 period, the number of first-year positions in traditional US medical schools that award the Doctor of Medicine (MD) degree increased by only 34 percent. This smaller increase is unlikely to reflect a lack of students interested in becoming doctors; in 2025, there were 2.3 times as many applicants to US MD programs as available positions. This ratio has exceeded 2 every year since 2003. The relatively small flow of entering physicians shows up in the stock: the United States has 2.7 practicing physicians per 1,000 population versus an average of 3.8 for OECD countries. The average annual growth rate of physicians per capita in the United States between 2000 and 2022 (0.8 percent) is about one-half of the average growth rate for OECD countries as a whole (1.5 percent).

These patterns are consistent with the belief that regulation of physicians including specialists) isn’t just about assuring quality, but is also about restricting supply to keep the pay of physicians high. (In the shade of these parentheses, I will add that I have worked as Managing Editor of the JEP for 40 years now, and so am perhaps psychologically addicted to the belief that the articles are of interest.)

Jonathan Hartley and Morris Kleiner provide some US and international perspective in “Analyzing Occupational Licensing Across Nations” (Federal Reserve Bank of Minneapolis, Staff Report 685, July 7, 2026). They present a familiar figure of the rise of occupational licensing in the US economy over time.

The rules and limits and tradeoffs of occupational licensing affect a much larger share of the US workforce than labor union membership; after all, only about 10% of total US workers belong to a labor union–and only about 7% of private-sector workers do so.

The new contribution of Hartley and Kleiner is to collect international data on occupational licensing from a wide array of sources. (The colors of the bars in this figure refer to geography: western Europe, Latin America, Africa, and so on.) The level of variation here is interesting: for example, the US has considerably more occupational licensing than Sweden, France, Italy, or the UK, but less than Japan or Germany. Because the sources of this data come from a wide array of surveys in different countries, I’m hesitant to overinterpret some of the comparisons with countries like Nigeria or India, where a large share of the workforce is in the “informal,” untaxed and unregulated economy. But at a minimum, there isn’t a clear pattern where higher-income countries are more likely to have high levels of occupational licensing.

The case for occupational regulation is often a pretty simple one: something bad happened to someone, somewhere, and a regulation might have prevented it from happening. The costs that make it harder for workers to enter certain jobs and preserve higher wages for those already in those jobs, leading to higher costs for consumers, are less clear. Some forms of occupational licensing surely make sense, but the potential tradeoffs are also real.

When I talk with college students, and even high school students, a number of them have a profound sense that it is tremendously hard to find a job coming out of college. The level fear seems to me exaggerated, but some underlying shifts in the young-adult labor market seem quite real.

Here’s a figure from the Federal Reserve Bank of New York showing unemployment rates for recent college graduates and other comparison groups. The light blue line shows the unemployment rate for recent college graduates age 22-27. If you look back through the 1990s and early 2000s, you see that the unemployment rate for this group is typically below the black line, which shows the overall unemployment rate. Also, the unemployment rate for recent college graduates is below the line for all young adult workers age 22-27, but above the ilne for all college graduates (of all ages).

But these patterns are shifting. Even before the pandemic unemployment spike, the unemployment rate for college graduates age 22-27 was about the same as the overall unemployment rate; since the pandemic, it’s higher than the unemployment rate. It’s still true that the age 22-27 college graduates have a lower unemployment rate than for that age group as a whole, but the gap is shrinking. Also, the gap between the unemployment rate for age 22-27 college graduates vs. college graduates of all ages seems to be getting wider.

This figure does not show a job apocalypse for recent college graduates. The unemployment rate for the group is still just a little over 5%. But it’s a change. As another sign of the change, consider some figures on total jobs available in a couple of occupations. This is from “What is really happening to jobs? Separating AI hype from reality,” by Neale Mahoney, Erika McEntarfer, and Karsen Wahal (Stanford Institute for Economic Policy Research, Policy Brief, July 2026).

The top figure shows customer service jobs and the bottom figure shows software developer jobs for different age groups. At the launch of ChatGPT in late 2022, the employment level for all age groups is set to an index number of 100, so that the figure shows the change in jobs before and after that date. Before the launch of ChatGPT, the number of jobs for different age groups climbed pretty much in unison. After ChatGPT, the number of jobs for older workers either continued to rise or levelled out, while the jobs for younger workers under the age of 30 dropped off. Again, it’s a change.

There seem to be three broad hypotheses for the shift in employment prospects for recent college graduates. One is the emergence of new AI tools, which may be able to perform a share of the work that was previously done by recent college graduates. Second, the years after the pandemic also involved a dramatic shift to work-from-home arrangements, and the lack of in-person connectivity may have made some firms less confident that they could provide useful and remote oversight for young and inexperienced workers. Third, recent college graduates are the ones who went through high school and/or college years during the pandemic and its aftermath, which may have led to lower-than-typical acquisition of learning and social skills in a way that hinders their job prospects. Of course, all three of these can be true to some extent.

I have also heard comments that even among the recent college graduates who find jobs, many feel as if the job is not using all their skills, but I don’t know how to evaluate this concern. Yes, there are anecdotes about young college-graduate ride-share drivers. But lots of college graduates for a long time now have ended up taking jobs as junior office workers who take minutes of meetings, pass along messages, setting up Zoom meetings, and doing the first draft of Powerpoint presentations. I don’t know of systematic data on whether the issue of underemployment is better or worse.

I offered some previous thoughts on entry-level jobs for young adults here, focused on data about shifts from unemployment to employment for young adults with different education levels.

For those who can be fuzzy on global geography, like me, it’s perhaps useful to say that the Persian Gulf is a body of water bordered by eight countries: Bahrain, Iran, Iraq, Kuwait, Qatar, Saudi Arabia, the United Arab Emirates (UAE), and Oman. These countries include some of the world’s most prominent oil exporters, who have shipping ports for their oil on the Persian Gulf. However, oil tankers leaving the Persian Gulf (and headed for the Arabian Sea, the Indian Ocean, and destinations around the world) need to travel through the Strait of Hormuz, a channel that is about 100 miles long, but with shipping lanes for big oil tankers that at the narrowest point are only about two miles wide in each direction. The chokepoint is bordered by Iran to north and UAE and Oman to the south.

Chokepoints can cause problems, and not just because of armed conflict. Some readers may remember that five years ago in 2021, a container ship called Ever Given ran aground and shut down traffic through the Suez Canal for six days. Now the Strait of Hormuz is giving the world a lesson about how supply shocks affect oil prices.

About 90% of international freight trade moves over the oceans. What are some other main chokepoints for global ocean shipping? Stephan Maurer, Luke Heath Milsom, and Ferdinand Rauch offer insights in “The fragility of the global trading system” (London School of Economics Centre for Economic Performance, Discussion Paper 2186, May 2026). A readable short overview of the paper is available at the blog of the LSE Business Review. In the overview, Maurer, Milsom and Rauch write:

[W]e measured sea routes between 455 ports with a statistical trade model that lets us simulate what would happen if a major route closed, or if a new one opened. We simulate shutting three shipping lanes: the Panama Canal (which connects the Caribbean Sea with the Pacific Ocean), the Suez Canal (between the Red Sea and the Mediterranean Sea) and the Strait of Malacca (between the Indian Ocean and the South China Sea). We also looked at opening two new routes: a year-round Northwest Passage (connecting the Atlantic and Pacific oceans through the Arctic north of Canada) and a proposed canal across Thailand’s Kra Isthmus. (This scheme would lower the distance for many shipment routes between China or Japan and Europe, although the Thai government is reported to be considering a land-bridge project instead of a canal.)

As geography suggests, the Panama Canal is most important for the US economy, because the alternative is that some shipping would instead need to travel around the southern tip of South American. The Suez Canal is important for European trade, because otherwise shipping would need to be re-routed around the southern tip of Africa. (Although Suez is most important of all to the economy of Egypt; the authors estimate that a closure of Suez would reduce the total GDP of Egypt by 8%.) The Strait of Malacca (between Malaysia and Indonesia, with Singapore at its southern end where a key channel is only about two miles wide) is most important for China’s international shipping trade. Otherwise, shipping between China and India, Africa, the Middle East, and Europe would need to be rerouted further to the east through an alternative passage like the Sunda Strait (between the Indonesian islands of Sumatra and Java, but also shallow and narrow for some large modern ships) or the Makassar Straight (between the Indonesian islands of Borneo and Sulawesi, a longer detour but also deeper water).

The authors also look at the possibilities from opening two new trade routes. If ships could pass freely though a Northwest Passage, it would be possible to move trade between the Atlantic and Pacific Oceans without either going through the Panama Canal or around the southern tip of South America. A proposed canal in Thailand across the Kra Isthmus would potentially be quicker than the Strait of Malacca or its alternatives–but would also be a huge infrastructure project.

In a global economy with near-instantaneous communication and rapid air travel, it can sometimes feel as if the details of geography, like chokepoints for ocean shipping, don’t matter all that much. Which is true enough–until they do.

 

Tim Sablik acts as interlocutor in “Interview with Christiane Baumeister,” subtitled “On measuring the effects of oil shocks, changes in global energy markets, and forecasting tail risks” (Econ Focus: Federal Reserve Bank of Richmond, Third Quarter 2026). Here are some of the comments that caught my eye:

On her career choice:

My father was a high school economics teacher, and I loved to spend time in his office browsing through his books. … When I had to pick my field of study in high school, I picked economics, and I never looked back.

On a pass-through of higher oil prices:

When thinking about the passthrough of oil prices to consumer prices, you can divide that into a direct and an indirect mechanism. Given that households do not directly consume crude oil, the first stage of passthrough really happens in oil-related products. The price passthrough is reflected in energy goods and services that form part of the consumer basket, and price developments in these oil-related products then are summarized by the energy component of consumer prices. These effects tend to show up very quickly, typically within the same month of the shock, and they also drive up headline inflation, roughly one-for-one with the share of the energy component.

Indirect inflationary pressures are the result of higher energy prices increasing the cost of inputs for firms. As a result, the oil shock can also have an effect on the production of non-energy goods and services. If firms decide to pass those costs on to consumers, that will show up in core inflation. This component follows a more staggered process, since not all firms raise their prices at the same time. Historically, it can take about three to six months to reach the peak response.

On the current shock to oil prices:

Typically, an oil price shock is the result of different drivers, and what matters in a crisis episode is the relative importance of these various determinants. Sometimes supply dominates, sometimes demand, and every historical event is different. But I think what we’re experiencing right now is probably the cleanest example of an oil supply shock that we’ve had in decades. It really follows the blueprint of a classical supply shock: There’s a war in an oil-producing country or region where production facilities and energy infrastructure get destroyed, and, in this case, a major waterway gets blocked. That leads to a loss of oil output, which then induces a spike in oil prices. What sets the current crisis apart is the sheer size of the supply disruption and the fact that it has affected all the countries in a region. … Looking at the longer run, I expect prices to stay elevated for a prolonged period, even after the reopening of the Strait of Hormuz. Not only will it take considerable time to start production and exports up again — think of all the energy infrastructure that has been destroyed, and all the oil tankers that are somewhere else in the world — but oil inventories will have to be refilled. So, there will be delays in getting more oil online, and at the same time there will be sustained demand from two sources: current oil consumption and stock rebuilding. I think those forces will bolster higher oil prices for a considerable period, at least until the end of 2027.

On the importance of elasticities:

 Throughout my entire journey of researching oil markets, the one thing that I came away with is that elasticities are really key. I think a lot of progress has been made to estimate them better, but there’s room for more. And this is where my thinking has evolved. It is really important to shift our focus from the aggregate to the disaggregate and study heterogeneity across countries — how oil producers and oil consumers adapt to oil shocks. Understanding the differences in country-specific supply and demand elasticity is very important in the beginning of an oil shock, but also for understanding the transmission to the macro economy. How does an oil shock propagate through the economy at the sectoral level? There’s a lot of heterogeneity hidden in the aggregates that we tend to look at, and to really understand the aggregate dynamics, it’s important to dig deeper and look under the hood at what’s happening at the firm and consumer level.

“Progress” is a word that often generates both a lot of positive vibes, and also a lot of pushback? Progress toward what? At what cost? So it’s perhaps useful to specify that when I refer to “progress,” I’m referring to an overall movement toward good things of life: in no particular order, longer and healthier lives; a rising standard of living, especially for the poor; more education for more people; environmental protection; leisure time; personal safety, resources of time and money for people to build community with family, friends, neighbors; and more. Addressing these and other problems of society requires “progress.” Of course, the specifics of how to address social problems are always controversial. But I’d argue strongly that if the solutions are envisioned as a zero-sum game–that is, all about taking from some groups to benefit others–lasting progress is hard to achieve. If the resources of a society are growing, and the question is how to guide and shape the fruit of that growth, progress becomes easier.

Joel Mokyr (Nobel ’25) raises these kinds of questions in his Nobel lecture, “The Past and Future of Innovation: Can Progress Be Sustained?” with the video and slides freely available from at the Nobel website, and a text version now published in the July 2026 issue of the American Economic Review (which requires a library or personal subscription for access). Mokyr’s explanations generally defy attempts at condensation, but here are a few points that caught my eye:

What are the basic underpinnings of economic growth?

[G]rowth can be driven by four different economic phenomena. One is simple capital accumulation, which by definition raises income per capita … . Another is what is known as “Smithian growth”—the gains from trade and specialization. A third is “Northian growth” which occurs when institutional change, more efficient factor markets, and improved property rights make the allocation of resources more efficient. … Finally there is “Schumpeterian growth,” driven by technological progress: making goods cheaper, making them better, or coming up with altogether new goods. In practice, of course, there four processes interact in a myriad of ways and in most cases reinforce one another. But at least logically they are separable.

The problem is that the first three sources of growth all run into diminishing returns–that is, they are not sustainable.

The first three share an important characteristic that I will call the “curse of concavity,” which is a poetic way of thinking of what economists have traditionally called “diminishing returns.” For capital accumulation this is immediate under the standard assumptions of almost any reasonable production function. But the same is true for the gains from trade. When trade is opened up between two nations (or regions) that were not trading before, there are immediate gains. As trading costs fall further, trade widens and deepens and further gains are secured, but as trading costs keep falling the marginal gains keep declining and theoretically further gains go asymptotically to zero as trading costs fall further. The same is true for improved allocations: Once a nation already has “good” institutions, further institutional improvements that make the allocation of resources ever more efficient will yield declining rates of growth. Any economy that depends solely on those three forces eventually will experience a slow-down in its rate of growth. Is the same true for Schumpeterian growth? Is there evidence that economic growth driven by the expansion and diffusion of useful knowledge may peter out?

A standard metaphor for those who worry about progress slowing down is that the “low-hanging fruit” has already been picked. A difficulty with this metaphor is that we aren’t picking fruit here. Instead, it is at least possible that scientific improvements might allow picking large quantities of high-hanging fruit.

To return to the metaphor of low-hanging fruits, in the absence of advances in the underlying understanding of natural phenomena, eventually the picking will become leaner and leaner. But advancing science provides ladders to reach higher and higher into the tree, and the high-hanging fruits may be of higher quality. In various fields, discrete scientific breakthroughs have provided opportunities to altogether new technological breakthroughs.

Once you start thinking along these lines, possibilities open up. What possibilities might become possible with quantum computing? Genomics and biotech? Research into clean energy or battery storage? A new idea is often important because it leads to other ideas. The new AI tools may turn out to be less important in the workplace than they are for generating not-previously considered but fruitful ideas for innovation. For a 20th-century example of an innovation leading to many applications, Mokyr offers the laser:

Perhaps the most versatile tool developed in the twentieth century to help do science (apart from digital computers) was the laser, a true scientific general purpose technology (GPT), which was applied to scores of very different areas of research. One of the most notable scientific breakthroughs driven by this technology was the success of the Laser Interferometer Gravitational-Wave Observatory to confirm the existence of gravitational waves, long predicted by Einstein but not proven till 2015. But the application of lasers to scientific discoveries has been wide ranging, and includes among others MALDI, matrix-assisted laser desorption/ionization (used in mass spectrometry in for instance analytical protein chemistry); LIDAR (light detection and ranging) technology that has application in geology, seismology, remote sensing, atmospheric physics and archaeology. Especially versatile is laser-induced breakdown spectroscopy (LIBS). In addition to multiple industrial uses, it is deployed in analyzing deep-sea sediments or hydrothermal vent fluids, studying elemental composition in unusual environments such as the Mars rover rock analyzer, as well as detect trace elements in nutritional, protein analysis, and material sciences.

As Mokyr points out, it’s perfectly plausible that individual people can find it harder to master the extraordinary realms of knowledge. But this problem arises because the amount of knowledge continues to rise, not because knowledge is levelling off. Ultimately, Mokyr is an optimist about the possibilities for progress, but more pessimistic about human institutions. Here’s his conclusion:

Is economic growth sustainable? My answer is that it is not only sustainable but in fact inevitable if humankind is to cope with the threats that are facing it. The threats of climate change and demographic transition, despite their different features, share three characteristics. One is that they are inexorable, relentless, and in all likelihood irreversible. A second is that they are global in nature: These are not local or regional crises but worldwide ones. Of course their severity is heterogeneous, but practically no part of humanity is exempt of some of its effects. The third is that both are in some way an unintended consequence of economic modernization. Beyond that, these threats are similar in that their respective impacts will be extremely expensive, and thus create heavy pressure on government deficits and the fiscal requirements to deal with them. Yet adapting to these shocks can be and will be possible, if their role as focusing devices will lead to accelerated technological change directed to adapt to the urgent needs of a changing world. There can be little doubt that if the appropriate decision makers will recognize the urgency of the needs, the useful knowledge needed to adapt to a changing world can be generated, given the rapidly expanding frontier of science and the possibilities of integrating robotics, machine learning, and similar general purpose technologies. The real question is whether the institutions—both public and private sectors—will recognize the pending disasters in time and respond appropriately. There is no guarantee that they will.