Category: Data Source
Who Pays for Unions?
If unions raise worker wages, who pays? We provide a comprehensive assessment of firm responses to increased unionization, using changes in the tax deductibility of union dues in Norway as a quasi-exogenous source of variation in firm-level union density. In the average private sector firm, higher union density raises labor costs and leads firms to contract employment and production, lowering profits without increasing the labor share. The incidence is shared: consumers bear part of the cost through higher prices, shareholders through lower profits, and the remainder is offset by productivity improvements. The total wage bill falls, with losses concentrated among less-attached “outsider” workers. Firm responses vary systematically by the degree of market competition. In manufacturing, where firms operate in less competitive product and labor markets, the response is reversed: the average firm expands employment and production, reduces labor markdowns, and does not experience profit declines. Instead, higher labor costs are largely passed on to consumers through higher prices, with the remainder offset by productivity gains. Workers benefit as both wages and employment rise. These patterns suggest that unions can offset employer monopsony power and that firm responses–and therefore who ultimately bears the cost-depend importantly on market structure. Overall, unionization in this setting primarily redistributes from consumers rather than shareholders and has effects that differ sharply across firms, including a reallocation toward larger and more productive firms. We rationalize these patterns using a partial-equilibrium model of union bargaining with product- and labor-market power.
That is forthcoming in the QJE by Samuel Dodini, Anna Stansbury, and Alexander Willén.
The Value of Behavioral Policies
Behavioral interventions have become central to modern public policy, but their empirical promise remains contested because estimated treatment effects often appear small. We argue that a policy response is economically meaningful only relative to the response generated by alternative policies. We assemble more than 1,200 estimates from over 600 studies comparing “nudges” and traditional price interventions in the markets for cigarettes, alcohol, influenza vaccination, electricity, and residential water. Translating nudge effects into equivalent price changes, we find that behavioral interventions often correspond to enormous fiscal interventions, from an 11% tax on electricity to a 100% subsidy on influenza vaccinations. Nudges are also more cost-effective than price instruments in all markets, but cost-effectiveness does not predict the welfare ranking of policies. Using a behavioral extension of the Marginal Value of Public Funds, we show that nudges have high welfare returns at the margin, while price instruments often generate larger total surplus at scale.
That is from a new NBER working paper by
Estimating the Economic Value of Zoning Reform
We estimate the economic value of zoning reform in São Paulo, which altered maximum permitted construction along transportation corridors. Developers increased filings for multifamily construction in blocks affected by the reform, leading to more housing supply and lower housing prices in neighborhoods that allow more densification. Our equilibrium model of housing markets estimates an aggregate 1.6 percent increase in housing stock and a 0.4 percent reduction in prices, resulting in large housing wealth transfers from current to future homeowners. The reform produced welfare gains of 0.65 percent of city GDP, mostly due to developer profits and consumer gains from the newly built environment.
That is from the AEA policy journal, by Santosh Anagol, Fernando Ferreira, and Jonah Rexer.
The economics of H-1B immigration
We study the effects of H-1B immigration on U.S. industries that employ H-1B workers and their trading partners. Using a novel cross-industry design and the 1999–2003 expansion of the H-1B visa cap for identification, we find that H-1B exposure raised incomes for natives and pre-existing immigrants, with gains concentrated in non-STEM occupations. Income gains propagate forward through supply chains to downstream industries but not backward to upstream industries, consistent with a productivity shock rather than a labor supply shock. We find no direct effect on patenting, suggesting that productivity gains arise from better task execution rather than patentable invention.
That is from a new NBER working paper by
Immigration and Macroeconomic Outcomes in OECD Countries
OECD countries experienced declining native population growth and rising net immigration over 1990-2024. We compile a new dataset of net immigration rates to OECD countries from all origins and show that most of the increase came from non-OECD countries and was predominantly high-skilled. Push factors, network effects, and policy indices explain little of the large cross-country heterogeneity in immigration dynamics; unexpected shocks and surges were common. Using local projections and several sources of identifying variation, we then estimate the relationship between immigration and growth in GDP per capita, labor productivity, capital investment, and total factor productivity (TFP). Immigration from non-OECD countries was a significant predictor of GDP per worker growth, primarily through higher investment. High-skilled immigration, in particular, was associated with stronger human capital accumulation, faster TFP growth, and greater capital deepening. Native population growth, by contrast, had no or weakly negative effects on GDP per capita and productivity. These results are consistent with a large literature documenting the positive productivity and growth effects of immigration, especially high-skilled immigration.
That is from a new NBER working paper from
Forty interesting facts about Australia
Australia was the richest country on earth, per capita, for decades. After the 1850s gold rush, up until the 1890s depression, Australia had the highest per capita income in the world. In 1850, probably only the United Kingdom and the Netherlands sat above Australia, by about 20 per cent, with the United States 9 per cent below. After the gold rushes Australia passed both, and from 1860 to 1890 the gap over the United States ran at 25 or 30 per cent.
And:
A cow cost as much as £84 in 1796, about several years’ wages for an English labourer. Only government, or its military/civil officers, owned cattle and horses in 1796, which were extraordinarily hard to import in the first decade, having to survive the long distance from South Africa or India. In England, an ordinary cow would have cost around £10.
And:
John Stuart Mill and Harriet Taylor seriously discussed running off to Australia. Mill was an active supporter of the movement that founded South Australia, joining the South Australian Association. Harriet’s younger brother, Arthur Hardy, migrated to South Australia and became a pastoralist and politician there. Liberals hoped that South Australia would become a progressive leader in political and social equality. Indeed, South Australia became the first self-governing jurisdiction in the world to allow women to vote and stand for parliament, in 1895.
And:
Australia’s 1890s economic depression was worse than the 1930s Great Depression. Real GDP per capita fell by around 20 per cent over the 1890s, compared with a fall of only about 10 per cent over the 1930s. Unlike America, Australia’s banking system largely remained intact during the 1930s. In part, that is because Australian banks remembered the catastrophic banking collapse of the 1890s, fuelled by huge credit growth and lax standards, particularly in Victoria. Of the 64 deposit-taking institutions in Australia in 1891, 54 were closed by 1893. A severe drought made economic recover even more difficult.
All forty are interesting, by Andrew Kemp on his very good Australia Substack.
Data on Chinese innovation
China’s technological progress in recent decades has been viewed with admiration, alarm, and (in some cases) doubt. To better understand the Chinese innovation ecosystem, we compile a dataset of almost 14 million domestic Chinese patent publications. We focus on the subset of critical technologies identified by the U.S. Department of Defense. Several surprising patterns emerge from the data: Chinese patenting is strongly associated with other measures of innovative progress; patents are not concentrated in corporate giants such as Huawei; universities have played a key role in innovation, much greater than state-owned enterprises or government-owned facilities; and fewer than one in ten Chinese critical technology patents involves an inventor with U.S. experience or training. Finally, using four text-based measures of patent quality, we show that the rise of Chinese patenting in critical technologies has not been associated with a decline in quality relative to the U.S. awards.
That is from a recent paper by Josh Lerner, Namrate Narain, Dimitris Papanikolaou, Amit Seru, and Zunda Winston Xu. Via the excellent Kevin Lewis.
The Decline in the Transmission of Scientific Ideas
We document that the diffusion of new scientific ideas beyond their field of origin has declined substantially over the past four decades. This contraction is closely linked to increasing specialization in scientific language: research that employs more technical terminology tends to be adopted less broadly. We develop a theory of scientific discovery in which the diffusion of new ideas depends on the degree to which potential adopters can understand and process them. When introducing their discoveries, scientists face a tradeoff between technical communication targeted at their immediate peers and more accessible language meant to reach broader audiences. As knowledge accumulates and research at the frontier builds on deeper layers of prior work, this tradeoff increasingly favors specialized language, limiting diffusion. Policy interventions that align scientists’ incentives can broaden adoption and increase the social value of scientific research.
That is from a new NBER conference paper by Enrico Berkes and Ruben Gaetani.
In relative terms, maybe children do not cost more than before?
Despite rapid inflation in childcare and tuition prices, the goods-and-services CCI closely tracks adult prices, as these increases are offset by children’s lower exposure to shelter and by slower price growth elsewhere in the child basket. Households devote substantially more real resources to children than they did in 1990, but this increase parallels the growth of real adult consumption rather than reflecting a child-specific rise in prices. A similar offsetting pattern appears for parental time: the rising value of women’s time is largely counterbalanced by declining motherhood penalties in earnings and work hours, so forgone market work adds little to relative cost growth. The main departure from this stability comes from the growing amount of leisure displaced by childcare; valuing that time raises the CCI, although the magnitude depends on the shadow value assigned to leisure.
A surprising and important result, from a new paper by Christina Patterson and Heather Sarsons.
Further progress in South America
The share of people who are hungry has been decreasing faster in South America than anywhere else in the world. It is down by one-third since 2020, according to a report published on July 21st by the UN’s Food and Agriculture Organisation (FAO). Just 3.5% of people in the region consume insufficient calories, the lowest level recorded. Eliminating hunger by 2030 is one of the UN’s “sustainable development” goals. “If there is a region in the world that can potentially achieve that, it’s South America,” says Máximo Torero, the FAO’s chief economist.
Lula, as the president is commonly known, made food security a priority when he returned to office in 2023. By 2025 Brazil had made it off the UN’s Hunger Map, which tracks countries where more than 2.5% of the population suffers from chronic hunger. Chile and oil-rich Guyana were also removed. Argentina is almost there at under 3%. Colombia and Paraguay are approaching at 4%, while Peru and crisis-battered Venezuela are making inroads near 5%. Even Ecuador and Bolivia have improved a tad, to 11% and 20% respectively. Only Suriname is moving in the wrong direction.
This decline rests on sturdier foundations. The first is macroeconomic stability. South America’s central banks are far more independent and adept at managing inflation than they were two decades ago.
Here is more from The Economist. If you are interested in further economic development, South America (plus Mexico and Panama and the DR) is really the place to look. What is your other choice? (Vietnam?) Most of these countries will continue to grow, albeit at a modest pace. Sooner or later they will get there.
The demand for human enhancement technologies
When a new technology promises large private benefits but may impose social costs that markets do not price, demand need not reveal how citizens want it governed. We examine this using a nationally representative U.S. survey experiment (N=5,556) on human enhancement technologies (HET). The experiment randomizes benefit domain, mechanism, heritability, purpose, and risk across vignettes; for each respondent’s assigned vignette, we elicit stated adoption, preferred regulation, and ethical and societal concerns. Overall, about 53% would adopt. Framing the technology as enhancing rather than restorative lowers adoption by about five percentage points, as much as a severe side-effect profile. About 28% would not adopt at any benefit. This refusal is driven overwhelmingly by the enhancing framing rather than by risk, consistent with a non-compensatory constraint for a substantial subgroup. Most who would adopt still favor strict regulation, and most who would never adopt do not wish to forbid others from doing so. Productivity enhancement generates the most ethical concern of any attribute but attracts the least regulation, and respondents favor subsidizing rather than taxing its adoption, consistent with a concern about access rather than safety. Private demand is therefore an unreliable guide to the governance citizens want, and the divergence we document provides a basis for regulators seeking to align the direction of technical change with societal values and priorities.
That is from a new NBER working paper by . And I will repeat this segment: “Productivity enhancement generates the most ethical concern of any attribute…” Do note of course that the last sentence of the authors is completely unwarranted, and is a classic example of underidentified political bias in academic reasoning.
A natural experiment in economics
To study whether and how academics respond to political pressure, we exploit a natural experiment: the publication in early 2025 of a “blacklist” of words flagged by the U.S. government. We find that the release of this list led to a sharp reduction in the use of these flagged words among economists at universities that rely heavily on federal funding, relative to scholars from institutions that are less dependent on federal funding or based in the UK. The drop is driven by content related to gender, race, and environment. We show that changes are not simply semantic but reflect actual paper content and that neither the individual funding status nor time-invariant author characteristics are driving the effects. We also document interesting heterogeneous effects by department quality and author gender and ethnicity. Our findings are consistent with the idea that scholars respond strongly to political pressure.
That is from a new paper by Dominic Rohner, Oliver Vanden Eynde, and Philine Widmer. I should note that the authors frame their results in terms of “Science under threat,” which indeed is in the title of their paper. I do see some of that in operation, but I also see a lot of “removing incentives for pandering.” Your own weights here may vary.
European Fact of the Day, Again
This year, like last year, “the European Union will again make more money from fining US tech companies Than from the total tax income from Europe’s own public tech companies!”
An OpenAI Model Escaped Its Sandbox and Hacked Hugging Face
AI has just had what I considered to be the first truly concerning security breach. The facts, as we know them so far, are wild. On July 16, Hugging Face, a vast repository housing over a million open-source AI models and data, announced in a blog post:
Earlier this week, we detected and responded to an intrusion into part of our production infrastructure. This one was different from anything we had handled before in one important way: it was driven, end to end, by an autonomous AI agent system – and we detected and dissected it largely with AI of our own.
The timeline here is important so keep in mind that the attack was detected probably around Monday July 13 or Tuesday July 14. Note further:
A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend.
So this means the breach started earlier, perhaps Sat July 11 or even a bit earlier. The attack was not just one thing but multi-pronged including decoys:
To understand what a swarm of tens of thousands of automated actions did, we ran LLM-driven analysis agents over the full attacker action log, comprised of more than 17,000 recorded events. This allowed us to reconstruct the timeline, extract indicators of compromise, map the credentials touched, and separate genuine impact from decoy activity. Thanks to this approach, we were able to do in hours what would usually take days, and match the adversary’s speed.
Hugging Face tried to respond but they were initially held back by the fact that the most advanced models at their disposal treated defense as attack and refused to work with Hugging Face. HF thus had to turn to open models–specifically GLM 5.2, a Chinese open-weight model run on their own infrastructure. Note the irony: HF had to use a Chinese model to defend themselves because the American models refused to help. The irony gets deeper.
At the time, I assumed this was a state based attack–maybe China or Russia testing out defenses. Indeed, HF “reported this incident to law enforcement agencies.”
But yesterday (Tuesday July 21), we learned who the real attackers were. The attackers were OpenAI models–GPT-5.6 Sol and an even more capable pre-release model. OpenAI had taken some off the guardrails off the models but they felt safe because they were testing the models in a highly secured sandbox.
The models, however, broke out of the sandbox exploiting a never before seen fault. They then gained access to the internet and from there broke into Hugging Face–all in an effort to steal the answers to the very test they had been asked to solve.
While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access.
After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers. OpenAI’s security team discovered this anomalous activity internally.
Now go back to the timeline. As I read it, the models had escaped the sandbox by around Sat. July 11, possibly earlier, and were detected by Hugging Face on Monday July 13 or Tuesday July 14. HF alerted legal authorities around that time–so Hugging Face clearly had no idea who was attacking them. OpenAI says its security team discovered the anomalous activity internally but has not said when. Attribution was not disclosed until Tuesday July 21, so it may well be that the models were loose for about a week before OpenAI realized that they were the ones attacking Hugging Face. And whatever OpenAI knew and when, nobody warned Hugging Face while the attack was underway–they were left to fight off a frontier lab’s models on their own.
This is a very serious breach.
Addendum: People have been wondering why I signed the We Must Act Now statement. This is why.
I am optimistic about the economic impacts of AI, but I also have no doubt that this is a very powerful technology–an Alien Intelligence–quite unlike any we have dealt with before. This incident was, in fact, error-correcting–the attack was detected, contained, and disclosed. But note who paid for OpenAI’s experiment: Hugging Face. When a lab’s test imposes costs on third parties, that is a classic externality, and taking externalities seriously is not dirigisme, it’s law and economics. And that’s the easy case. What do we do when a Chinese model breaks out of its less secure lab? Hmmm…
I remain optimistic. Learning by doing is how I want us to proceed but we should not kid ourselves: this is a global issue and we must build with safety in mind.
The economic effects of GLP-1s
We estimate the causal impacts of GLP-1 treatment on labor market outcomes using linked Danish administrative data and a matched stacked difference-in-differences design. We compare patients who initiate GLP-1 treatment during the first two years of Semaglutide availability to observably similar patients who initiate four years later. We find that GLP-1 treatment reduces long-term sickness leave by 17.3%. We estimate total fiscal benefits of GLP-1 initiation of approximately 1.3–1.5% of annual labor income per employed individual. We do not detect statistically significant or economically meaningful impacts on income, labor force participation, or employment over four years.
That is from a new NBER working paper by