Conversable Economist

Thomas Chow poses the questions in “Kristin Forbes on ‘wargaming’ for the
next crisis,” subtitled “The MIT professor and former BoE MPC member speaks about QE, scenario analysis and the ar of conducting monetary policy in the ‘fog of war'” (Central Banking, May 14, 2026, free registration required). Forbes has a new book out, The Art of Monetary Policy: Lessons from Sun Tzu for Central Banks. Here are a few points that caught my eye:

Economic shocks are happening faster, so advance scenario planning has become more important

Even during the global financial crisis (GFC) in 2008, the major bank runs evolved slowly over days, weeks and months. But what we saw with SVB [Silicon Valley Bank] in 2023 was that bank runs can happen in a matter of hours. Policy-makers don’t have the luxury of going back to the office, calling people together, pondering options and then putting together a plan. Even in the GFC when things were happening quickly, there was this saying: “let’s just get to the weekend”. Because if you could get to the weekend, you had until Asia opened to come up with a plan. Now you may not have the luxury of getting to a weekend, so you really need to do scenario planning, ahead of time, and be ready to respond much more quickly than needed in the past. …

I also think it’s important for central banks to start treating scenario analysis as ‘business as usual’ so that a hypothetical scenario does not becomes political or create a headline such as “the Bank of England is predicting a war in Iran”. We don’t want central banks to look like they’re predicting major geopolitical events, or showing a preference for a specific outcome related to a political decision. If central banks don’t do scenario analysis regularly, there can be a tendency to interpret the selection of scenarios as political. And that’s not the point. Central banks should be thinking about and planning for all sorts of outcomes. Planning for a scenario does not mean they have a desire for it to happen.

The Bank of England has separate committees to decide on monetary and financial issues, and those committees have some external members outside the BoE.

I think the restructuring the BoE did a number of years ago – to have one committee for monetary policy and one for financial policy – makes a lot of sense. One reason builds on the last point – that this can make clear what the goals are from asset purchases based on which committee makes the decision. Another reason is that monetary policy and financial stability require different expertise. By having two committees, you can have people on each committee who specialise in that area. Then you have the right people around the table when you’re making decisions, without creating such a large committee that it becomes less effective.

Another reason why this structure works so well is because you have a portion of each committee that is internal, who see the big picture of everything that is going on in the central bank. But then you also have a portion of each committee that is external, who just focus on that specific function of the committee, but who also bring in very different views. The externals often have other jobs, and you get more diversity of opinion than if you just have people who work full-time at the central bank and are spread very thin across a lot of other responsibilities. This works particularly well at the BoE as they encourage the externals to have independent views. You can vote against the governor in the MPC, you need to testify regularly to explain what you’ve added to the discussion, and you’re supported in expressing your different views publicly.

The temptation that central bank’s face to take big steps, and worry about consequences later

The one principle highlighted in the book that we haven’t talked about is evaluating the short- and long-run trade-offs. There is a tendency in the ‘fog of war’ to take very large policy actions because you want to avoid the worst outcome and send a strong signal of your commitment – for example, Hank Paulson’s quote about using the ‘bazooka’ to stop fire sales during the GFC. But one lesson we’ve learned since then is that the very large policy actions may not yield the optimal outcome. The ‘big bazooka’ may have big costs afterwards. These newer tools for monetary policy have costs that didn’t receive sufficient attention during recent crises. For example, large asset purchases leave central banks with large balance sheets, which have had substantial fiscal costs in some countries. If you don’t need such a large volume of asset purchases, you should avoid them. You should better calibrate the response and stop the programme as soon as you can. You should also think about the exit strategy. For example, when you purchase assets, consider how you will unwind them when the battle is over – which can affect exactly what you buy. Even in the ‘fog of war’, in the moment of panic, think carefully about the costs and benefits of your actions – including the costs when the war ends.

For a majority of Medicare and Medicaid recipients, the government program is paying a private health insurance company to provide the service from a private firm. In the Spring 2026 issue of the Journal of Economic Perspectives, Tim Layton, Luca Maini, and J. Michael McWilliams tell the story of Medicare in “Substitutes for Success? Public versus Private Competition in Medicare Advantage.”  Mark Shepard and Jacob Wallace tell the story of Medicaid in  “Understanding Medicaid Managed Care: The Procured Competition Model.”  (I work as Managing Editor of the JEP.)

In general, the hope of both programs is that having private health insurance companies competing with each other will provide higher quality and/or lower costs for recipients of Medicare and Medicaid. But of course, the two programs have quite different focuses.

Medicare is primarily focused on over-65 Americans. Layton, Maini, and McWilliams describe traditional Medicare program like this:

Medicare, the program that provides publicly- financed universal health insurance coverage for nearly all Americans over the age of 65, is typically characterized as a single-payer health insurance system, with government officials administratively setting fee-for-service prices and cost-sharing levels for every medical procedure and service that physicians and hospitals can provide, actuaries setting premium rates, and essentially all private healthcare providers accepting Medicare patients and payment from the program. As such, it is quite similar to single-payer health insurance systems in many European countries. It is also a simple, relatively easy-to-understand program and one of the most popular government programs in history, with over 80 percent of Americans having a favorable opinion of the program.

However, 57% of Medicare recipents are not in this traditional program. Instead, they are enrolled in what is technically Medicare Part C, commonly known as Medicare Advantage. For them, Medicare buys a health insurance policy from a private company. In theory, the idea was that Medicare would spend about the same amount on a person, no matter which option they chose. In this case, the benefit from getting private insurance through Medicare Advantage would arise if the private insurance companies could run more efficiently, use some of those efficiencies to offer additional benefits and also to earn a profit.

One natural concern is that Medicare Advantage would find ways to attract healthier people, and “save” money in that way. This seems to have been a real issue in the past, but not so much in the last decade or so. One piece of evidence compares mortality rates. As the figure shows, Medicare Advantage had lower mortality rates (plausibly becuase they were a healthier group) in the past, but this gap was narrowing from about 2010 to 2019 (perhaps because the expansion of Medicare Advantage was causing less-healthy people to shift over), and has now largely gone away.

However, in fact, the goverment still seems to pay more for a patient in Medicare Advantage than it would for the same patient in traditional Medicare, according to an in-house group called the Medicare Payment Advisory Commission. Thus, Medicare Advantage can avoid charging the co-pays and deductibles of traditional Medicare, and often provide larger benefits as well, because it has more money to do so. A likely underlying reason is that the Medicare Advantage firms have an incentive to find ways to “code” the medical treatments they provide in a way that keeps their costs higher, while benefits for traditional Medicare have not moved much in the last 15 years.

Indeed, Layton, Maini, and McWilliams suggest that given the financial problems of traditional Medicare, Congress has been unwilling to expand benefits for the traditional program. For example, the trust fund for Medicare Part A hospital insurance, funded by payroll taxes, is scheduled to go broke in the next few years. However, Congress has in effect raised Medicare benefits through the back door, by using the formulas that determine what private health insurance companies would receive under Medicare Part C.

The Medicaid program is aimed at low-income families, as well as low-income elderly and disabled. But as Shepard and Wallace write in their JEP article:

Over the past several decades, Medicaid has undergone a major institutional transformation. Historically, states administered the program directly, paying doctors, hospitals, and other providers through a state-run fee- for-service system. Today, about 85 percent of beneficiaries are enrolled in Medicaid managed care, under which states contract with private insurers to provide coverage and manage care. This shift toward outsourcing has been driven by concerns about rising costs, program complexity, and the limits of states’ capacity to administer insurance efficiently. …

Medicaid covers a broad set of benefits, encompassing a wider range of services than most other health insurers. In addition to mandatory coverage of hospital and physician care, states may include “optional” benefits such as dental and vision care for adults (all children receive dental and vision benefits). Medicaid also finances a large share of long-term services and supports for individuals with complex medical or functional needs, including care in nursing homes and home-based care. Coverage is provided with little to no patient cost- sharing, and beneficiaries generally do not pay premiums. …

States specify detailed provisions via managed care contracts that are often hundreds of pages long. These contracts define covered services, patient cost-sharing (essentially zero), rules for provider networks, insurer compensation arrangements, and provisions for quality reporting and oversight. They also specify insurer prices (payments per enrollee) and other compensation provisions, such as risk adjustment. Insurer flexibility is limited to a few features like provider networks and care management rules. Additionally, Medicaid insurers each offer a single plan…

The classic problem here is that states want on one side to assure that private health insurance firms provide access to the full range of Medicaid services, which explains all the rules about what must be provided. On the other hand, state are often expressing hope that that competition between the health insurance companies (and the contracts that the health insurance companies have with health care providers) will lead to cost-cutting pressures.

However, the health insurance companies have a quite limited number of ways that they can compete. In this situation, health insurance firms cannot compete on attracting patients by offering a lower price, or by offering additional health care services. service. The state government controls the amount of entry by other health insuance firms. Shepard and Wallace call this “procured competition.”

The two articles offer much more about the details of how Medicare and Medicaid are intertwined with private health insurance companies, along with discussions of how the contracts and arrangements might be tweaked and renegotiated. Here, I will end by touching on another subject. One sometimes hears advocates of a single-payer national health insurance program in the United States advocate Medicare-for-all or Medicaid-for-all. After all, the argument goes, these programs already provide health insurance for 40% of all Americans. Why not the rest.

However, advocates of this approach often seem to have in mind the old-style traditional versions of Medicare and Medicaid, with government making payments directly to health care providers–and health insurance companies being eliminated. But given the modern form of these programs, supporting Medicare-for-all or Medicaid-for-all means, yes, supporting private health insurance companies under a different set of rules.

Indeed, when when you read about low administrative costs for Medicare and Medicaid, it’s worth remembering that they are outsourcing many of their administrative costs to private health insurance companies. It’s also worth remembering that neither Medicare nor Medicaid is self-supporting. For Medicare, there is a widespread belief that that program is funded by a combination of payroll taxes and the premiums paid by the elderly, but those cover only about 55% of total Medicare costs, and the rest is general federal tax revenues. Medicaid is funded by a mix of federal and state tax money. It would seem peculiar to advocate a sweeping national health insurance reform that sought to re-allocate the insurance premiums now paid by employers on behalf their employees into Medicare, so that Medicare could turn aorund and pay the money to the same private health insurance companies that would have received it anyway.

I admit this story is insider stuff for those with experience of academic seminars for economists. But it made me laugh, so I pass it along. It’s how Jesse Tack, Jisang Yu, and Roderick M. Rejesus introduce their review essay “Recent approaches in agricultural production economics: Where the heck are the prices?” (Food Policy, May 2026).

A scene is unfolding. It’s 2005 and an excited young PhD candidate applied for a job and has been invited to give a job market seminar at a prominent department. They are nervous, but well prepared. Nice new outfit, a couple practice presentations under their belt. Their voice cracks just a smidge as they introduce themselves to the audience and share some personal background on why they are so excited about this job opportunity. By the time they get to the title of the paper they have recovered and are settling in. As the final word of the title rolls off their tongue, they are about to click to the introduction slide when a hand shoots up in the audience…

“I read the paper linking the new technology to crop production, but…[dramatic pause]…where the heck are the prices? Those would seem to be really important variables to be included in the analysis.”

The speaker is rattled. The questioner seems genuinely upset, and this was not expected so early in the presentation. They stammer a bit but aren’t panicking yet. They decide to go with a succinct reply in hopes of moving on. Remember, be polite…

“Great point, thank you. We agree that prices are important drivers of production but we do not need them in order to estimate the effect of interest for reasons that I will discuss later.”

The questioner doesn’t like this answer, they are a little offended…

“Well you want a job here and I asked you a question. I would appreciate an answer now.”

The speaker is again surprised and fidgets a bit. They had everything perfectly planned out. Number of slides. Length of presentation. Leave enough room for questions along the way, sure, but this soon in the presentation? Not sure what exactly to do, they summarize their strategy succinctly in the hopes that they can move on…

“One of the really cool aspects of this project is that we spent a large amount of time setting up the experimental design in which the technology was randomly assigned to different producers and thus plausibly exogeneous to both input and output prices.”

Experimental design??? Sensing shenanigans, they dig back in…

“Okay…hmmm…I don’t know about all that but I do know that you’re trying to explain production outcomes and prices are really important in that context. We are economists after all, not agronomists”.

It’s clear the questioner is well practiced in this slight against agronomists, putting just the perfect amount of snark into the comment to elicit a chuckle from the audience. Damn! After an internal sigh, the speaker wades back into the madness….

“I appreciate your concern and we have thought a lot about it, I suppose there might be some incidental in-sample correlation of prices and technology adoption so to check that this isn’t indeed a major concern we did consider a robustness check where we controlled for prices with annual fixed effects and the estimates remained stable.”

“A what effect?”

“A fixed effect. Like a dummy variable for each year in the sample.”

“You controlled for prices with a dummy variable?”

“Yes, a different one for each year.”

“Like a zero-one variable? Prices are continuous variables how can that possibly work?”

“I know right, its super cool. As long as the producers all face the same prices it is observationally equivalent to putting prices directly in the model.”

“Aha! But the economic environment is different by regions so that can’t possibly work.”

“Agreed! That’s why we also used region-by-year fixed effects!”

“What the what???”

“Yeah, our results actually seem to be pretty robust to a wide range of possible confounders.”

Baffled, and sensing they might be a bit behind on current empirical approaches as some of their colleagues in the audience seem to be nodding along in agreement with the speaker, they make one last attempt….

“But prices are a really important drivers of production and I want to see what their effect is in this study”

“That’s not our focus here”

“Well…it should be!”

“Ummmm…okay…”

OK, to appreciate the humor, perhaps you have to have been in this kind of seminar room. For those who were not, it’s probably useful background to know that for a long time, agricultural economists had a heavy focus on “production functions”–that is, a function where the inputs would be land, labor, seeds, irrigation, fertilizer, farm machinery and the like, and the question was how changing various inputs would affect output.

The goal seems straightforward, but as has been well-recognized for a long time, drawing inferences in this approach can be messy. For example, if one observes a bunch of farmers using more farm equipment, and producing higher output, is the farm equipment causing the higher output? Or is the greater use of farm equipment a reaction to some factor not included in the analysis, but also affecting output? For example, perhaps some land is just much better-suited to farm equipment, and if you don’t have that kind of land, buying farm equipment won’t benefit you as much. What if certain inputs are not well-measured in the analysis? For exmaple, perhaps some farmers may be more entrepreneurial, risk-taking, and knowledgeable than others, but you can’t just take the inputs those farmers are using, apply them to farmers with different managerial characteristics, and expect the same result.

The question of how to get more plausible causal estimates is at the center of lots of economic empirical work in the last few decades, including agricultural economics. As a straightforward example, imagine a study of several hundred or several thousand farms. Half of the farms are randomly chosen to receive a certain intervention: perhaps seeds that produce more if given particular care, or loans to buy additional fertilizer before planting, or insurance against future fluctuations in crop prices, or conservation payments, or a sophisticated weather app. Later, one can then compare output of farms that got randomly got the technology and those that didn’t.

A researcher can also look for events that introduce a random component for agricultural production, so that out of a group of farmers who are similarly situated and growing similar crops, some of thosse farmers randomly face a situation that others do not. Variations in weather conditions can offer such randomization: more specifically, it’s now becoming possible to look at detailed variation in humidity, wind speed, solar radiation, evaporation, and so on. If you want to know, for example, how global warming might affect agricultural output, these sorts of studies offer a place to start.  Farmer may also vary in what technologies are available to them, or by variations in the policy environment.

And yes, it turns out to be fairly straightforward to add effects of prices into these models, as our pestering seminar participant in the story above wanted back in 2005–although the methods for doing so weren’t yet well-developed 20 years ago. The essay is a good example of a pattern that has happened fairly often over the years, in which agricultural economics raises problems that lead to new econometric techniques: for example, the birth of instrumental variable estimation and fixed-effects methods.

Kevin Warsh was nominated by President Trump in January 2026 to replace Jerome Powell as Chair of the Federal Reserve. He was confirmed by the US Senate last week and sworn in as the new Fed chair earlier this week. What are some of his views and priorities as he takes office? For a clue, I looked at a just-published collection of essays by economists and central bankers, Finishing the Inflation Job and New Challenges for Monetary Policy, edited by Michael D. Bordo, John H. Cochrane, and John B. Taylor (Hoover Press, 2026, scroll to the bottom of the link and the advance page proofs of book can be downloaded for free in sections). The book described papers and comments from a conference held a year ago in May 2025–that is, well before Warsh was nominated. Here are some of his comments that struck me:

On central banks and causes of inflation:

Inflation is a choice. The world’s central bankers get to choose the inflation rate. … The central bank establishes the policy rate and steers in the direction of an inflation objective. Central banks … are not victims. … Inflation is not caused by pandemics or autocrats around the world. The inflation level is set by the world’s central bankers. And without going too far afield, in my view, it is principally determined by government spending and printing. …

It would be nice if one could say bygones are bygones. It would be nice if households and businesses believed that past errors had no bearing. But the precondition for stable prices is confidence on the part of households and businesses that central banks will deliver stable prices. And the best way to give them that confidence is to have achieved it. It is up to the central bank to ensure that whatever shocks occur outside are one-off effects. The inflation
rate, that is, the second- and third-order consequence of changes in prices, not the first-order change in the price level, is up to the world’s central banks.

If central banks assert that outsize changes in the price level affect inflation and then drive a set of inflation outcomes, the banks are, in a way, admitting something against their own interest. They’re saying, in a sense, that their credibility has been impaired and that inflation will occur because they don’t have the credibility to stop it.

On the aftermath of earlier decisions to conducting monetary policy with quantitative easing, and its interaction with conducting monetary policy by adjusting the federal funds interest rate:

In a broad sense, we should acknowledge now what was acknowledged at QE’s creation in 2008: The Federal Reserve established a second monetary policy instrument, a supplementary instrument that has an important effect on inflation. … [W]ith a very active, large, and often growing balance sheet, we have two policy instruments that are imperfect substitutes for each other, sometimes working at cross-purposes and at other times working together. But if the printing press could be quieted, we could have lower policy rates, because a $7 trillion balance sheet is affecting inflation. There are many benefits of a small balance sheet, including lower rates. However, a better economic outcome is probably the most important. …

First, a surge in the balance sheet is understandable in periods of great shocks. The Fed was created after a panic early in the 20th century. So nothing I say should be taken as any direct criticism of what happened to the balance sheet in 2008, or what happened in the darkest days of 2020. One should give the benefit of the doubt, I think, to central bankers in harm’s way. Second, I think it’s strange to say in 2008 and 2020 that the balance sheet expansion was monetary policy by other means, but not in more benign times. It’s odd not to have any rhetorical or real symmetry. In my view, the balance sheet can’t be construed as monetary policy in crisis times, but the balance sheet has nothing to do with monetary policy in any other circumstance. That kind of asymmetry goes against the very spirit of policy rules and moves us to a policy of full discretion.

Finally, when I joined the central bank in 2006, we had about an $800 billion balance sheet. If you were to try to scale that to the growth of the economy, or to the growth of financial markets, one might end up with a $2.5 trillion or $3 trillion balance sheet today. As we sit here, the balance sheet is about $7 trillion. The right question was raised earlier about transitions between policy regimes. The transition from a scarce reserve system—in which banks were relying predominantly on each other for liquidity, with the central bank entering the market more rarely in periods of extreme illiquidity—to an excess reserves regime was not sudden. Going back to some status quo ante, or adopting a new, third-way model, will take time. The transition is not something that could or should happen overnight. But banks will grow accustomed to the liquidity regime around them. And if the central bank has a permanently larger role, not just in crises but in normal times, and is in some sense providing liquidity to the banks during all seasons and for all reasons, then one has fundamentally changed the role and responsibility of the central bank.

The transition to what I think is a more prudent system will take time, deliberation, and an excess of communication with the public and the institutions in the banking system itself.

For those of us who live in the world of editing, writing, and publishing, the ability of the newest generations of AI tools to produce rivers of grammatically correct prose is a deep shock. But is there actual evidence on AI and the quality of published work? Imke Reimers and Joel Waldfogel offer a starting point in their research paper, “AI and the Quantity and Quality of Creative Products: Have LLMs Boosted Creation of Valuable Books?” (NBER Working paper 34777, May 2026, also available here).

The obvious questions here are how to measure quantity and especially quality of new books. Quantity is easier. The authors offer some highly suggestive evidence: “Using data on new books offered for sale at Amazon, we document that the number of new titles appearing each month nearly tripled between 2022 and late 2025 and rose by a factor of nearly ten in some categories; and the increase in new titles 2022-2025 coincides with both the diffusion of LLMs and the incidence of detected AI in books.”

The Amazon data also provides information on the number of ratings for any given book, the sales rank, and the number of “stars” for any given book. These kinds of measures offer a data-driven way of getting at the question of quality. In addition, Reimers and Waldfogel describe:

[W]e take a direct approach employing AI detection on over 50,000 randomly selected titles. We document the growth of AI usage and compare the quality of “AI books” and “non-AI books” (books with and without detected AI). We have four findings. First, the timing of AI growth tracks the growth in releases: Detected AI use is roughly zero through 2022, rises to 30 percent in 2023, to 45 percent in 2024, and surpasses 60 percent during 2025. Second, AI books garner substantially less usage per title. Most have very little usage, and a modest share is somewhat used, both when measured by the number of ratings and eventual sales ranks. AI books are also worse in the sense of having lower star ratings. Third, the human-AI usage gap narrows substantially between 2023 and 2025. Finally, the number and quality of human-authored books has remained stable.

Perhaps unsurprisingly, authors with relatively low sales and rankings in the past much more likely to take up the AI tools.

At some level, the results of the study are unsurprising. AI seems to have led to many more books, but the average quality of these AI-assisted or -generated books is lower than the previous average. The authors write: “The effects of this influx [of new books] on consumer welfare depend on the quality of the additional books. The average quality of new books has fallen with the LLM-induced influx, and books with detected AI are substantially worse than human-authored books, so that much of the new work is of little value to consumers. Still, the LLM influx has delivered some books in the middle range of the usage/quality distribution …”

Consumers vary in their tastes, and more books means more to choose from. At least so far, AI-generated fiction is not hitting the top of the measures of quality. But it’s worth remembering that a large proportion of book says are in the middle range of the quality distribution, including categories like romance, SF, and mystery. Personally, I read a lot of mysteries, and while I try to pick high-quality ones, I read a fair number that could be fairly characterized as in the middle range of quality; indeed, some of the mysteries I read could probably be improved with an assist from AI.

This broad pattern probably characterizes a lot of AI-assisted work in creative areas. It can produce vast quantities very quickly, which will mostly be low-quality, but some of it will reach middle-quality. If the low-quality output can be ignored at low cost, a greater choice among middle-quality output is a modest social gain.

The US and other high-income countries have their own set of concerns about China’s role int the world economy–but the concerns for low- and middle-income countries may be even more severe. A standard way of thinking about the process of global economic development is as a sort of ladder. Low-income countries start out with economies that are heavy on agriculture and subsistence farming. However, they gradually take a step up into low-skill manufacturing (textiles is a classic example), and then with additional capital investment they can take another step into higher-skill manufacturing. As manufacturing displaces agriculture, their service economy begins to expand as well, and the service jobs in everything from logistics to health care, from finance to education, and the broad category called “professional and business services,” all expand as well.

But what happens if the bottom step of the development ladder, the move into low-skilled manufacturing, is blocked by the presence of China in the global economy? Shoumitro Chatterjee and Arvind Subramanian argue that China’s ongoing global presence in low-skilled manufacuturing is shutting off the early steps to economic development for low- and middle-income countries around the world. They make the case in “China’s Mercantilist Squeeze on Developing Countries” (Peterson Institute for International Economics, Working Paper 26-7, May 2026).

For example, consider China’s share of global markets in the kinds of low-skilled production that has in the past often been an early rung on the development ladder. The size of global market is measured by domestic value-added of those exports (after some inputs were imported ). Notice that high-income countries dominated these global markets in the 1960s, but then moved on to other industries. However, despite China’s rapid growth in the last few decades, it continues to dominate these markets.

If China’s share of these low-skilled goods should be in decline, by how much? The authors tackle this question in a few ways. For example, they look at China’s share of low-skill/low-paid workers in the world economy, and how it has declined over time, and they look at the history of how today’s high-income countries reduced their global presence in these industries. No matter how they slice it, China is competing very directly in world markets with the low- and middle-income countries of the world in precisely the industries that have often been an early step toward economic development. Moreover, China imports very little of these goods, so it does not act as a buyer in international markets in these areas. As the authors write:

What matters is that China is occupying a larger share of the global value chain in precisely the sectors where poorer countries would otherwise expect to compete. … Not enough attention is being paid to China’s persistent, even rising, occupation of export space in low skilled goods and the attendant impact on the development prospects of low- and middle income countries—the China Squeeze. This impact—hundreds of billions of dollars in foregone LMIC exports—has been felt most acutely in global markets but also in China’s imports in LMIC markets and low access to China’s markets as well. The magnitudes are sizable enough to stymie their structural transformation and ability to escape from low- and middle-income status.

At some level, it doesn’t matter whether China’s position, blocking ability of developing countries around the world, arises “naturally” from China’s economic growth or “unnaturally” from policies of China’s government that offer its manufacturing and exporting firms a direct or indirect subsidy. But the question of “why” is nonetheless intersting.

As one piece of data, Chatterjee and Subramanian look at wages in the apparel industry across countries. Wages in China are way up–but this has not meant a shift of global apparel exports to countries with lower wage levels, as one might expect.

As another piece of data, the authors look at changes in manufacuturing productivity over time. China’s productivity growth in manufacturing has slowed, while other competing countries have seen sharp rises. But again, contrary to what one might expect, this has not led to a lower share for China in low-skilled manufacturing.

Comparing direct and indirect government subsidies across countries is an enormous task, and this paper doesn’t take it on. But the authors do suggest that, in their judgement, perhaps the most likely reason for China’s ongoing global dominance in these low-skilled manufacturig jobs is that China central bank acts to keep its exchange rate artificially low. The result is that China’s exports to the rest of the world are cheaper than they would otherwise be, while China’s imports are more expensive than they would otherwise be. Of course, this also helps to explain China’s consistent pattern of large trade surpluses.

The go-to source for comparing national levels of defense spending arount the world is SIPRI, the Stockholm International Peace Research Institute, which has just published its data on defense spending in 2025. Here are a few snapshots from “Trends in World Military Expenditure, 2025,” co-authored by Xiao Liang, Nan Tian, Diego Lopes da Silva, Lorenzo Scarazzato,  Zubaida A. Karim, and Jade Guiberteau Ricard (April 2026).

This figure shows the inflation-adjusted level of defense spending for the world going back to 1988. The decline after the fall of the Soviet Union in the early 1990s, the rise in the “Asia and Oceania” share as China’s economy and defense spending have grown, and the rise in European defense spending in the last few years are all visible. Moreover, the “Americas” defense spending–dominated by US spending–has been fairly flat since about 2010, so the growth in defense spending is happening in other regions of the world.

It’s interesting (to me, at least) that the growth rate of defense spending over this time has been slower than the growth of the world economy more broadly. As a rough average, world GDP has been growing at 3.5% annually in recent decades. As compounded growth rates take place, the world economy has grown by about 180% in the last 30 years. If you look back at the trough of defense spending in the mid-1990s at about $1.2 trillion, it has risen by “only” about 120% during that time. Thus, the share of world GDP going to defense spending has been falling.

The US continues to have by far the largest level of defense spending of any country, accounting for about one-third of the total.

Finally, for fans of large tables, here’s a list of the countries of the world ranked by total spending on national defense. You can pick out the data-points that interest you the most. (Numbers in brackets mean that the totals were estimated by SIPRI based on government sources.) For example, the total rise in US defense spending from 2016 to 202 was 11%, compared with a rise of 62% for China, 96% for Russia, 118% for Germany, and 39% for India.

Or in the column on defense spending as a share of national GDP, it’s heartbreaking to see that Ukraine’s share was 3.7% in 2016 and 40% in 2025, even as Russia’s defense spending as a share of GDP has risen from 5.4% in 2016 to 7.5% in 2025. However, the percentage of GDP going to military spending hasn’t changed much for the US, China, or India; to put it another way, the large increases in China’s defense spending in the last decade or so are a reflection of the large increases in the size of China’s overall economy.

Ranking cities by population require some choices: in particular, do you focus only the legal boundaries of the city, or on a metropolitan area? If it’s a metropolitan area, how do you decide on the city limits? According to the World Urbanization Prospects 2025 report from the United Nations Department of Economic and Social Affairs, the most populous city in the world is Jakarta, Indonesia. Their process involves dividing the area into square kilometers,and estimating population density in each square. Here’s their graphic showing the top 10 cities in the world by population in 2000, 2025, and projected for 2050.

It’s interesting to note that Tokyo, the world’s largest city by population in 2000, has grown in population only modestly since then, and isn’t projected to grow further by 2050. However, Jakarta, Dhaka, Shanghai, Karachi,Cairo, and a few others are taking off. I find it hard, with my 20th-century brain, to imagine what it means to have “city” of more than 50 million people.

What got me thinking about Jakarta was an essay by Nithin Coca: “Jakarta’s Remarkable Urban Transit Transformation” (In Development, April 30, 2026). She begins:

For many years, the first word most foreign visitors learned upon moving to Jakarta was macet, traffic jam. Traffic was so bad that transport experts warned in 2013 that if nothing was done, the city could achieve total gridlock, with every part of the city experiencing a traffic jam. In 2014, Jakarta was crowned the world’s most congested city by the Stop-Start Index and a year later was ranked far below other Asian cities on livability by the Economist Intelligence Unit.

Ten years later, Jakarta has the world’s largest and one of the most used bus rapid transit (BRT) systems. The old, crowded diesel commuter trains, famous for allowing passengers to ride on the roofs, are now electrified, air conditioned, and run on regular schedules linking the suburbs to the city center. There are multiple subway and light rail lines crisscrossing the city. The transformation has been remarkable: in 2015, less than 20% of residents were within walking distance of transit. Now, nearly 90% of the city has access to BRT or trains.

How did this happen? One answer is that traffic and pollution concerns had gotten really bad. “Jakarta had also become one of the world’s most polluted cities. By 2011, 58 percent of all illnesses among people living in the city were related to air pollution. … Residents were resigned to spending an average of 16 days stuck in traffic each year.”

Then a former governor of Jakarta was elected president of Indonesia, and used the position to push for a mass transit system. Indonesia then got a low-interest loan from Japan’s development agency. The deal was: “Japan would play a role in basic design, construction, and introduction of transportation systems, including trains, signals, and gate systems, as well as their operation and maintenance. But Japanese contractors were insistent that, while they might build the railway, it was up to Indonesia to run it. Much of the technology would come from Japanese companies like Sumitomo and Nippon Sharyo, but construction, operations, and maintenance would all have to be done by Indonesian companies or the government.” As the system has been built out, a South Korean consortium has become involved as well.

But the story is only a partial success. Jakarta’s population is growing so fast that what has been built so far is far from adequate. Even with expanded mass transit, the number of cars continues to expand as well. “Danny Djarum, an Air Quality Senior Research Lead at WRI Indonesia says that PM 2.5, the measurement of inhalable airborne particulate matter, is now eight to ten times higher than World Health Organization guidelines. `We’re still one of the top 5 most polluted cities in the world, he said.”

Jakarta’s mass transit system continues to expand. But in addition, one intriguing suggestion is to designate certain parts of the city as “low emissions zones,” where access by private vehicles is restricted, green space is expanded and walkability improved.” Another policy suggestion is some form of congestion pricing.

But overall, Jakarta is showing that if you have a local political determination and are also willing to hire and to heed the world’s top experts to manage the project, a big start toward working mass transit system for the world’s largest city can happen in a decade. “In the latest TomTom traffic index, measuring average congestion—the percentage increase in travel time compared to free-flow conditions— the city ranked 24^th, just ahead of the United States’ most famous traffic clogged city, Los Angeles.”

When I was a young teenager, changing diapers for my baby brother and on babysitting gigs, it was all cloth. I could change a sleepy baby’s cloth diaper in the dark, large safety pins and all. A couple of decades later when I was a parent, it was all disposable. What happened? Virginia Postrel tells the story in “Engineering the disposable diaper: Benjamin Spock told mothers in the mid-twentieth century to buy six dozen cloth diapers and a covered pail. Within a decade, both were obsolete” (Works in Progress, April 24, 2026).

Back in 1957, disposable diapers had about 1% of the diaper market. They were expensive, and mainly used for situation where diapers would need to be changed while travelling. Postrel takes up the story:

After buying Charmin Paper Company in 1957, Procter & Gamble began looking for ideas for new paper products.  Motivated by the less pleasant aspects of spending time with his new grandchild, the company’s director of exploratory development, Victor Mills, suggested disposable diapers. After analyzing existing products and conducting consumer research, P&G created a dedicated diaper research group.

The research this group conducted, like that of its successors and competitors, wasn’t glamorous. It didn’t advance basic science. It wasn’t even an obvious route to profit. (One percent of the market!) It was a high-stakes gamble that required solving difficult engineering problems. How that happened represents the kind of hidden progress that leads to everyday abundance.

P&G’s first design flopped. Tested in the extreme heat of a Dallas summer, the pleated absorbent pad with plastic pants made babies miserable and left them with heat rashes. Starting over, the group had a one piece diaper ready for testing in March 1959. With an improved rayon moisture barrier between the baby and the absorbent tissue wadding, the new diaper was softer and more comfortable. An initial test of 37,000 hand-assembled prototypes went well, with about two thirds of the parents deeming the disposables as good or better than cloth. The next step was mass production.

Designing one well-functioning disposable was hard enough. Turning out hundreds a minute was practically impossible. ‘I think it was the most complex production operation the company had ever faced’, an engineer recalled.

Eventually, the diaper team mastered the process. In December 1961, Pampers went on the market in Peoria, Illinois. Once again, the test failed. This time mothers liked the diapers. But the price was way too high for a single use item: ten cents a diaper, equivalent to about one dollar today. By contrast, diaper delivery services, which served about five percent of the market, charged no more than five cents a diaper. Home laundry costs ran to one or two cents.

Lowering the price of a diaper required much larger volumes. Aiming at about six cents a diaper, P&G engineers spent several years developing what Harvard Business School’s Michael E. Porter described as ‘a highly sophisticated block-long, continuous-process machine that could assemble diapers at speeds of up to a remarkable 400 a minute’. After successfully testing Pampers at 5.5 cents each, P&G began a national rollout in 1966. By 1973, disposables accounted for 42 percent of the US diaper market.

Other firms first entered the diaper market in the 1970s, and then left: Scott Paper, International Paper, Union Carbide, Johnson & Johnson. The competitor that did gain a foothold was Kimberley-Clark, the innovator who had created “Kleenex tissues and Kotex feminine pads … in the 1920s.” After a false start or two, the Huggies diaper, with elastic around the legs and an improved tape closure, swept into the diaper market. It cost 30% more, but for a lot of buyers, the premium price was worth it.

I love a good product development story, and Postrel has lots more details: how new absorbent materials made diapers slimmer over time, reducing logistics costs like storage, handing, and retail shelf space; the environment arguments about disposable vs. cloth diapers; disposable training pants that little ones could pull up by themselves; some well-chosen modern cultural references to disposable diapers; and more. Here, I’ll just offer three takeaways.

First, the creation of a workable disposable diaper, and then improving on that diaper, and being able to take it to mass production at an affordable cost, were all genuinely difficult tasks. The innovations took investment measured in time, money, and varied kinds of expertise. The time-path to disposable diapers having 95% of the diaper market took decades, with a number of failures along the way both within and across companies.

Second, a non-breakthrough innovation like disposable diapers may be nndervalued, in part because Americans take so much for granted that competitive companies will be trying to provide new and improved versions of so many products. But over time, the accumulaton of many such innovations makes day-to-day life so much easier. It’s an enormous benefit of living in a dynamic market-oriented economy that can also be nearly invisible.

Third, when new mothers are asked about the essential needs for their babies, they tend to focus on milk and diapers. But although diapers are affordable in the mass market, they aren’t cheap. I wrote a few years ago in “Some Economics of Diapers” (September 29, 2022) about what an author called a “leaky” part of the US social safety net, along with discussions of diaper “banks” and other methods of assuring that low-income parents and their babies have access to diapers.

Under pressure from US tariffs and geopolitical conflicts, US imports of most good and services has been fairly flat, but with one big exception: imports related to artificial intelligence. Michael Waugh of the Minneapolis Fed provides evidence in his working paper “Trade in AI-Related Products”(Minneapolis Fed Staff Report 684, April 8, 2026). Jeff Horwich, also from the Minneapolis Fed, provides a shorter and readable overview of the findings in “Much more than microchips: Trade soars in AI-related goods, driving U.S. trade deficit” (May 4, 2026).

As background, it’s useful to know that that all US imports are classified by what is called the Harmonized System, and the most detailed level of classification is called HS10. Waugh uses a large-language model to determine how closely linked different inputs are to the broader notion of what is needed to build AI. He explains:

What are AI-related products? The classification identifies the obvious computer hardware inputs such as data processing units and storage devices. These products account for about half of all AI-related trade and imports of them have grown by triple digits since 2023. But the classification also identifies a broader set of products tied to electrical infrastructure, networking, cooling and HVAC, and specialty materials. These ancillary products account for the other half of AI-related trade and have also experienced strong import growth.Through 2025, AI-related products account for 23 percent of all U.S. imports, up from 15 percent in 2023. Leading into the start of 2024, AI-related products grew no differently than non-AI-related products. But since early 2024, import growth in AI-related products has accelerated sharply. As of January 2026, trade in AI-related products had grown by 73 percent relative to 2023, compared with only 3 percent for non-AI-related products.

Where are these products coming from? Two countries play an important role in the sourcing of AI-related products. Not surprisingly, Taiwan is an important source for computer hardware components (e.g. semiconductors) and Taiwan accounts for about a quarter of imports of AI-related products. The surprising source is Mexico. It accounts for another quarter of AI-related trade and its reach extends to computer hardware and other products related to electrical power, networking, and cooling HVAC. China is less important in AI-related trade and its overall share has diminished over the past two years.

Here’s a figure to illustrate the pattern. The blue line shows AI-related imports to the US. The black line shows all imports. The red line show imports not related to AI. To compare trends, the levels of all three are set equal to 100 for the monthly average in 2023. Looking back to 2022, you can see that the three lines move much the same. But starting in 2024, they separate quite a bit.

Waugh notes that many of the AI-relevant goods are wholly or partially exempt from Trump’s tariffs. He also offers a rough calculation that if AI imports had not risen as they have, the US trade deficit would have been about $200 billion less in 2025. If policies are judged by their outcomes, not their stated rationales, the Trump administration has decided that imports of AI-related items are important enough that the usual rationales for tariffs (cut the trade deficit, it should all be immediately built in the US, and so on) do not apply.

At a global level, it seems plausible that artificial intelligence technologies will cause an increase in world trade, no matter the tariff and geopolitical issues. The World Trade Report 2025 produced by the World Trade Organization (September 2025) is subtitled “Making trade and AI work together to the benefit of all.” The report notes:

WTO simulations suggest that AI could lead to significant increases in global trade and real income. These simulations are based on an extension of the standard WTO Global Trade Model with AI services and incorporate trade cost reductions, a shift in tasks from labour to AI and productivity gains related to this shift. They suggest that AI could lead to significant increases in trade and GDP by 2040, with global trade projected to rise by 34 to 37 per cent across different scenarios. The largest growth occurs in the trade of digitally deliverable services (42 per cent), including AI services. This trade increase reflects (i) reduced operational trade costs, (ii) the strong projected growth of AI services combined with the high tradability of AI services, related to its geographic concentration of production in a few regions, and (iii) the aboveaverage productivity growth in more tradable sectors, in particular digitally deliverable services. The development and deployment of AI are also projected to generate substantial global GDP increases, ranging from 12 to 13 per cent across scenarios.

There have been other times in history when the political winds were leaning against international trade, but underlying economic forces created large increases in trade anyway. As one example that I wrote about here a few months back, the wave of globalization that happened from about 1870 to 1914 happened in the face of tariff levels that were remarkably high by modern standards. However, this was also a time of dramatic improvements in transportatation, communication, and infrastructure, which combined to drive trade higher even in the face of high tariffs. Perhaps AI will become another example in which technological force that drive down the costs of trade and increase the benefits of trade will have bigger global effect than efforts and events that should seem to have a negative effect on trade.