Edward Conard

Top Ten New York Times Bestselling Author

  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…reminds us that inequality sends a signal of what society lacks most, in America’s case, entrepreneurship and risk taking.” - Lawrence Lindsey, CEO, The Lindsey Group, former Director of the National Economic Council
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “Unintended Consequences is far smarter and more thought-provoking than most economics written for the general public” - Greg Mankiw, Harvard University, Former Chairman of the Council of Economic Advisors
  • “Unintended Consequences is full of substance, it is one of the must-read books of the year, and once I finish it I will be giving it a second read through right away.” - Tyler Cowen, Professor, George Mason University
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…reminds us that inequality sends a signal of what society lacks most, in America’s case, entrepreneurship and risk taking.” - Lawrence Lindsey, CEO, The Lindsey Group, former Director of the National Economic Council
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “Unintended Consequences offers deep and well-argued analyses on almost every issue.” - The New York Times
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Friday, November 21, 2025

AI Summary. Survey data show that among renters with net worth below $300,000, 4–6% say always giving their best effort at work is "not important," roughly twice the 2–3% rate among homeowners at all net worth levels; renters with net worth above $300,000 resemble homeowners. The authors argue that abandoning the aspiration of homeownership may reduce the value of additional work effort.

Seung Hyeong Lee and Younggeun Yoo Social Science Research Network
Date Posted:

Renters with net worth < $300,000 are 2x as likely to report that “always giving my best effort at work” is “not important” than either renters with > $300,000 or homeowners.

Do renters with lower net worth report putting less effort into work?

Core argument: Renters with net worth below $300,000 report low work effort at 4–6% in MRI-Simmons psychographic data, nearly twice the 2–3% rate among homeowners, which stays flat across all net worth levels.

Once that aspiration [homeownership] is abandoned, the marginal utility of additional labor may decline, leading to a re-optimization of work effort by renters. Figure 6 reports the share of respondents who chose “not important” for the MRI-Simmons psychographic questions that ask the importance of “always giving my best effort” at work. While standard labor market outcomes such as hours worked or employment status may fail to capture effort margins (for instance, a full-time worker who is still formally employed may reduce discretionary effort - commonly referred to as “quiet quitting” - without any observable change in hours), this measure serves as a useful proxy for changing attitudes toward work. Among homeowners, the proportion reporting low work effort remains low and relatively flat at 2–3% across all net worth levels. Among renters with net worth above $300,000 - those still plausibly on track to purchase a home - the proportions are similar or even slightly lower, possibly reflecting stronger motivation to work toward eventual homeownership. However, among renters with net worth below $300,000, the share reporting low work effort increases to 4-6%, nearly twice the rate observed among homeowners.

Takeaways by Macro Roundup® AI

  1. Renters with net worth below $300,000 report low work effort at 4–6% in MRI-Simmons psychographic data, nearly twice the 2–3% rate among homeowners, which stays flat across all net worth levels.
  2. Renters with net worth above $300,000, those still plausibly on track to buy a home, report low work effort at rates similar to or slightly below homeowners, possibly reflecting stronger motivation to work toward eventual homeownership.
  3. Once the aspiration of homeownership is abandoned, the marginal utility of additional labor may decline, prompting renters to re-optimize work effort through margins such as "quiet quitting" that hours worked and employment status may fail to capture.

Related Articles:

  • US Median First-Time Homebuyer Age Now At Record-High of 40 — The median age of first-time homebuyers in the US has reached a record 40, up from 33 in 2021 and 29 when the National Association of Realtors survey began in 1981. High home prices and mortgage rates are delaying homeownership for millions of Americans.
  • Why Are Young Adults in the English-Speaking World So Unhappy? — The share of young adults regularly experiencing stress and anger has risen sharply in the US, Canada, UK, Ireland, Australia and New Zealand while remaining largely stable elsewhere in the West. Real house prices since 1995 are up 85% in the US and over 200% in the other five countries.
  • Housing and the American Worker — Wages net of housing costs were once substantially higher in high-density areas than low-density areas for workers both with and without college education, but over the past half-century that urban wage premium has persisted only for college-educated workers, not for those with a high school degree or less.

AI Summary. Since the second half of 2023, four-quarter US productivity growth has averaged nearly 2.5%, well above the pre-pandemic trend of about 1% to 1.25%, supporting real GDP growth. It remains unclear whether this reflects a higher productivity regime fostered by rising AI use or a temporary post-pandemic reorganization surge.

Robert Valletta Federal Reserve Bank of San Francisco
Date Posted:

Since 2023:H2, 4-quarter productivity growth has averaged nearly 2.5%, well above the pre-pandemic trend of 1 to 1.25%. The question remains whether this is a new regime or “a temporary surge related to post-pandemic reorganization and consolidation.”

Is faster productivity growth a new regime or a temporary surge?

Core argument: Four-quarter US productivity growth has averaged nearly 2-1/2% since the second half of 2023, well above the pre-pandemic trend of about 1 to 1-1/4%, according to the San Francisco Fed.

[Real GDP] growth has been supported by solid productivity gains, which have been uneven since the end of the pandemic but on average well above the pre-pandemic trend of about 1 to 1-1/4%. Since the second half of 2023, four-quarter productivity growth has averaged nearly 2-1/2%. It is too early to know whether these gains reflect a new, higher productivity growth regime fostered by the rising use of AI and related technologies or a temporary surge related to reorganization and consolidation in the aftermath of the pandemic.

Takeaways by Macro Roundup® AI

  1. Four-quarter US productivity growth has averaged nearly 2-1/2% since the second half of 2023, well above the pre-pandemic trend of about 1 to 1-1/4%, according to the San Francisco Fed.
  2. The San Francisco Fed says real GDP growth has been supported by productivity gains that have been uneven since the end of the pandemic but on average well above the pre-pandemic trend.
  3. The San Francisco Fed says it is too early to know whether the productivity gains reflect a higher-growth regime fostered by rising use of AI and related technologies or a temporary surge tied to post-pandemic reorganization and consolidation.

Related Articles:

  • Concentrated Growth: The Role Of The IT Sector — Since 1988, total factor productivity in the information technology sector has risen about 178%, versus about 12% across all non-IT sectors combined, while aggregate productivity rose about 40%. The IT sector accounts for roughly 45% of cumulative aggregate productivity growth, once sectoral linkages to downstream industries are weighted in.
  • How America’s AI Boom Is Squeezing The Rest Of The Economy — About a sixth of America's 2% real GDP growth over the past year came from investment in computer and communications equipment and data centres, and one estimate puts the artificial intelligence boom's total contribution to growth at 40%. Average electricity bills rose 7%, attributed in part to data-centre demand, while housebuilding and non-AI business investment slumped and real consumption flatlined.
  • Productivity During and Since the Pandemic — Total factor productivity growth since early 2021 has run only slightly above its 2010–19 average, suggesting the 2020 productivity surge reflected cyclical dynamics layered on a continuation of the slow pre-pandemic trend. Productivity has since grown somewhat faster than the authors' analysis predicted, but remains well below the 1995–2004 pace.

AI Summary. Americans with at least a bachelor's degree make up a record 25% of the unemployed, over 1.9m workers aged 25 and up, according to US Bureau of Labor Statistics data going back to 1992. Their unemployment rate rose half a percentage point from a year earlier to 2.8%, while other education levels showed little increase.

Matthew Boesler Bloomberg
Date Posted:

25% of unemployed Americans have a college degree. Overall, degree holders had an unemployment rate of 2.8% in September, up 0.5pp y/y. 9.2% of all Americans 20–24 were unemployed in September, up 2.2pp y/y.

Are college graduates making up a record share of the unemployed?

Core argument: Americans aged 25 and over with at least a bachelor's degree made up 25% of total US unemployment in September, with more than 1.9 million unemployed — a ratio never reached before 2025 in data going back to 1992.

Americans with four-year college degrees now comprise a record 25% of total unemployment, underscoring a sharp slowdown in white-collar hiring this year. Government-shutdown delayed monthly figures published Thursday by the US Bureau of Labor Statistics showed the unemployment rate for bachelor’s degree-holders rose to 2.8% in September, up a half-percentage point from a year earlier. Other levels of education, by contrast, registered little or no increase over the same period. here were more than 1.9 million Americans aged 25 and over with at least a bachelor’s degree who were unemployed in September — one in four of the total number of unemployed. Before 2025, the ratio never reached such a high in data going back to 1992. Thursday’s BLS figures showed young Americans are bearing the brunt of the recent rise in joblessness. The unemployment rate for those between the ages of 20 and 24 was 9.2% in September, up 2.2 percentage points from a year earlier — an increase of a magnitude not seen historically outside recessions.

Takeaways by Macro Roundup® AI

  1. Americans aged 25 and over with at least a bachelor's degree made up 25% of total US unemployment in September, with more than 1.9 million unemployed — a ratio never reached before 2025 in data going back to 1992.
  2. The unemployment rate for bachelor's degree-holders rose to 2.8% in September, up half a percentage point from a year earlier, while other education levels registered little or no increase, according to the US Bureau of Labor Statistics.
  3. Unemployment among Americans aged 20 to 24 was 9.2% in September, up 2.2 percentage points from a year earlier, an increase of a magnitude not seen historically outside recessions.

Related Articles:

  • What the Graduate Unemployment Story Gets Wrong — Comparing recent graduates with others newly entering the labor market, rather than with all non-graduates of similar age, shows those without a degree are faring worse: US unemployment among recent college graduates is up 1.3 percentage points from its low, versus 2.4 points for recent entrants without a degree.
  • Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — Using payroll records covering millions of US workers, the authors find employment for workers aged 22–25 in occupations most exposed to artificial intelligence, such as software developers and customer service representatives, declined 6%, while employment for older workers in those occupations rose 6–9%. Employment in less-exposed occupations grew comparably across age groups.
  • Military Officer Quality in the All-Volunteer Force — The authors find a steady decline in U.S. Marine Corps officers' General Classification Test scores, which is statistically associated with the expansion of college attendance and explains 96% of the measured change. They also find a positive association between African American officers and mean test scores, arguing affirmative action has not lowered officer quality.

AI Summary. Consumer sentiment in the long-running University of Michigan survey is worse than during periods of roughly 9% inflation, double-digit unemployment after the financial crisis, and the 1980 recession, even though unemployment stands at 4.3% and inflation at 3%. The article argues people credit themselves for wage gains while blaming inflation for taking those gains away.

Paul Krugman Krugman Wonks Out
Date Posted:

Krugman blames anomalously low consumer sentiment partly on post-traumatic stress from the Covid inflation. “Even when wages rise faster than prices, people believe that they earned their wage increases, only to have their gains taken away by inflation.”

Why is consumer sentiment worse than during far weaker economies?

Core argument: The last pre-shutdown data showed U.S. unemployment at 4.3% and inflation at 3%, both decent by historical standards, yet consumer sentiment in the University of Michigan survey sits at a record low.

By conventional measures [the economy] isn’t doing badly enough to justify the extremely negative views Americans currently hold. The last economic numbers available (delivered before the shutdown) showed unemployment at 4.3% and inflation at 3%. These are both decent numbers from a historical perspective. Yet according to the long-running University of Michigan survey of consumer sentiment [see Chart], Americans have a worse view of the economy now than in mid-2022, when inflation hit around 9%; in the immediate aftermath of the financial crisis, when unemployment was in double digits; and in 1980, when the economy was in recession and had double-digit inflation. [After 1991] we all became accustomed to a low-inflation economy, with the 1970s a distant memory. But for a period of about two years in 2021-2023 the world saw a return to high inflation. Why do people hate inflation? The Biden years were marked by a period of unusually large increases in prices, but they were also marked by unusually large wage gains. Even when wages rise faster than prices, people believe that they earned their wage increases, only to have their gains taken away by inflation. Consumer sentiment is at a record low, and voters are angry at whoever holds the White House.

Takeaways by Macro Roundup® AI

  1. The last pre-shutdown data showed U.S. unemployment at 4.3% and inflation at 3%, both decent by historical standards, yet consumer sentiment in the University of Michigan survey sits at a record low.
  2. Americans hold a worse view of the economy than in mid-2022, when inflation hit around 9%, than in the immediate aftermath of the financial crisis, when unemployment was in double digits, and than in the 1980 recession with double-digit inflation.
  3. The Biden years combined unusually large price increases with unusually large wage gains, and even when wages rise faster than prices, people believe they earned their raises only to have the gains taken away by inflation.

Related Articles:

  • The Cost of Money is Part of the Cost of Living: New Evidence on the Consumer Sentiment Anomaly — Variation in the University of Michigan consumer sentiment index that inflation and unemployment cannot explain is strongly correlated with growth in consumer borrowing costs, such as mortgage and car loan rates. The authors estimate that including interest payments and homeownership costs in the consumer price index would raise measured inflation from 3% to 9%.
  • Why Do We Dislike Inflation? — Survey evidence finds 80% of respondents believe prices rise faster than wages, though real weekly earnings for the median worker grew 1.7% between 2019 and 2023. About 40% think their nominal income would be higher had inflation been lower, and half expect wages to take over a year to catch up if inflation doubled.
  • Zero-Sum Thinking and the Roots of U.S. Political Divides — Zero-sum thinking — viewing gains for one group as losses for another — is associated with greater support for liberal economic policies, stronger alignment with the Democratic Party, and weaker alignment with the Republican Party. The authors also find it linked to believing the QAnon conspiracy theory holds some truth and to sympathy for the attack on the U.S. Capitol.

AI Summary. US prices for goods and services are 25% above 2020 levels, and 44% of middle-income respondents in the University of Michigan consumer sentiment survey said their finances were worse than a year earlier versus 23% who said better, with most citing higher prices. Retailers report middle-income customers pulling back on discretionary spending.

Rachel Louise Ensign and Rachel Wolfe Wall Street Journal
Date Posted:
Is Database:
Database

In September, 44% of middle-income Americans reported they were financially “worse off” than a year ago, while 23% said they were “better off,” ~ unchanged from last September, when 48% reported being “worse off” versus 24% who said “better off.”

Are higher prices leaving middle-income households feeling worse off?

Core argument: Costs for U.S. goods and services stand 25% above 2020 levels, and 44% of middle-income respondents to the University of Michigan consumer sentiment survey said their finances were worse than a year earlier versus 23% who said better, on a three-month average ending in September.

America’s middle class is weary. Costs for goods and services are 25% above where they were in 2020. The University of Michigan’s consumer sentiment survey showed that 44% of middle-income respondents said their financial situation was worse than it was a year ago, while 23% said it was better, based on a three-month average ending in September. Those who feel worse off overwhelmingly said it was because of higher prices. The frugality of the middle-class customer figures as a recurring theme in recent corporate earnings reports. Fast-food restaurant Wingstop said this month that middle-income diners have now joined lower-income ones in dialing back purchases. Target reported slumping sales and said customers are spending cautiously on discretionary items such as home decor and apparel. Walmart, meanwhile, posted strong sales as consumers from all income groups flocked to the retailer’s value.

Takeaways by Macro Roundup® AI

  1. Costs for U.S. goods and services stand 25% above 2020 levels, and 44% of middle-income respondents to the University of Michigan consumer sentiment survey said their finances were worse than a year earlier versus 23% who said better, on a three-month average ending in September.
  2. Middle-income respondents who felt worse off overwhelmingly attributed it to higher prices, and Wingstop reported that middle-income diners have joined lower-income ones in dialing back purchases.
  3. Target reported slumping sales and said customers are spending cautiously on discretionary items such as home decor and apparel, while Walmart posted strong sales as consumers across all income groups flocked to the retailer's value.

Related Articles:

  • Real Income Growth Shifts Down, Especially For The Young — Inflation-adjusted median income growth for 25–54-year-old workers declined and has hovered near decade lows, with a more marked slowdown among those aged 25–29. The authors suggest this may reflect a reduced pace of job-to-job transitions, seen in lower hiring and quit rates.
  • Income in the United States: 2024 — U.S. median household income was $83,730, not statistically different from the prior year's $82,690, according to Census Bureau estimates. Across most demographic subgroups, median household income was higher or not statistically different from the prior year, with the exception of Black households, which saw a decline.
  • Jobless Growth — Labor demand growth may be too weak to keep pace with even modest labor supply growth: the labor market is somewhat weaker than before the pandemic, job growth has turned negative outside healthcare, and firms are focused on using AI to cut labor costs. Economists estimate most employment losses in routine manual occupations since the 1970s occurred during recessions.

Monday, November 24, 2025

AI Summary. A reanalysis using the full sample of firms, rather than one dropping observations with missing capital or overhead-cost data, finds that the measured rise in average markups at the end of the sample is driven almost entirely by the finance and insurance sector. Excluding that sector, average markups rise only modestly, from just under 1.30 to about 1.35.

C. Lanier Benkard, Nathan Miller and Ali Yurukoglu National Bureau of Economic Research
Date Posted:
Is Database:
Database

De Loecker et al. found that aggregate markups surged from 18% to ~ 67% from 1980–2014. This comment argues that much of this finding is due to DEU’s deletion of observations with missing values for SG&A, and the rest is limited to Finance and Insurance.

Is the measured rise in markups driven by one sector?

Core argument: In the full sample excluding Finance & Insurance, the average markup rises only modestly, from just below 1.30 in the mid-1980s to about 1.35 in 2016.

Figure 1 plots three estimates of the average markup over time. The solid red line is the estimate provided in [The Rise of Market Power, Jan De Loecker, Jan Eeckhout and Gabriel Unger] (DEU). The dashed green line is what we obtain from the full sample—i.e., without dropping missing values for Capital or SG&A. Finally, the dash dot black line is what we obtain with the full sample excluding the Finance & Insurance sector. Using the DEU sample, the average markup increases rapidly throughout the 1980s and 1990s, pauses in the 2000s, then increases rapidly again in the final few years of the sample. In the full sample, the average markup increases more gradually through 2008. It then sharply increases in the last few years of the sample, ending at a similar level as in the DEU sample. Moreover, unlike the DEU results, which are robust to the exclusion of individual sectors, in the full sample the increase in markups at the end of the sample is driven almost entirely by a single sector: Finance & Insurance. Excluding F&I, the average markup increases only modestly, from just below 1.30 in the mid-1980s to about 1.35 in 2016. We conclude that the DEU methodology and data, as they are described in the article, do not support the conclusion that broad-based increases in market power have occurred in recent decades.

Takeaways by Macro Roundup® AI

  1. In the full sample excluding Finance & Insurance, the average markup rises only modestly, from just below 1.30 in the mid-1980s to about 1.35 in 2016.
  2. In the full sample, which retains observations with missing Capital or SG&A values, the late-sample increase in average markups is driven almost entirely by the Finance & Insurance sector, unlike the De Loecker, Eeckhout and Unger results, which are robust to excluding individual sectors.
  3. The comment concludes that the De Loecker, Eeckhout and Unger methodology and data, as described in their article, do not support the conclusion that broad-based increases in market power have occurred in recent decades.

Related Articles:

  • The Rise of Market Power and the Macroeconomic Implications — The authors estimate that aggregate markups over cost rose from 18% to nearly 67% between 1980 and 2014, concentrated among firms that already had high markups, which they interpret as a rise in market power. They associate this increase with declining labor and capital shares, lower low-skill wages, reduced labor force participation, fewer startups, and wider wage inequality.
  • Rising Markups and Declining Business Dynamism: Evidence From the Industry Cross Section — Across industries, larger increases in markups are not associated with larger declines in business dynamism—measured by new business entry rates and gross job or worker flows—and the correlation may even be positive. The authors conclude that rising markups as commonly measured are not a major explanatory factor behind declining dynamism.
  • The Economics of Inequality in High-Wage Economies — United States Income, Wealth, Consumption, and Inequality Diana Furchtgott-Roth

AI Summary. Barclays estimates that investment in software, computer equipment and data centers added about 1 percentage point to annualized US GDP growth in the first half of 2025, with artificial intelligence spending accounting for 0.8 points of the 1.6% annualized growth after subtracting imported chips.

Konrad Putzier Wall Street Journal
Date Posted:

Barclays estimates that AI-related investment constituted half of America’s 1.6% annualized GDP growth in the first half of 2025.

How much of US GDP growth comes from AI spending?

Core argument: Barclays estimates AI-related spending added 0.8 percentage points annualized to U.S. GDP growth in the first half of 2025, out of 1.6% annualized growth overall, implying growth of 0.8% absent that contribution.

Barclays estimates that investment in software, computer equipment and data centers boosted GDP growth by around 1 percentage point annualized in the first half of 2025. AI explained much, though not all, of that. Chips such as those sold by Nvidia make up the bulk of AI spending, but most are imported and must be subtracted from total investment to arrive at the impact on domestic production. Accounting for that, AI spending still increased output by an annualized 0.8% in the first half of the year, Barclays estimates GDP grew by an annualized 1.6% during the period. In other words, absent the growth in AI-related spending, growth would have been a sluggish 0.8%. Barclays senior U.S. economist Jonathan Millar estimates that a 20% to 30% stock-market decline could reduce GDP growth by 1 to 1.5 percentage points over roughly a year. If AI investment stopped growing, that could knock another 0.5 point off growth, Millar estimates. If it went to zero, that would knock a full percentage point off.

Takeaways by Macro Roundup® AI

  1. Barclays estimates AI-related spending added 0.8 percentage points annualized to U.S. GDP growth in the first half of 2025, out of 1.6% annualized growth overall, implying growth of 0.8% absent that contribution.
  2. Barclays estimates investment in software, computer equipment and data centers boosted GDP growth by about 1 percentage point annualized in the first half of 2025, with AI explaining much though not all of it.
  3. Barclays senior U.S. economist Jonathan Millar estimates a 20% to 30% stock-market decline could cut GDP growth by 1 to 1.5 percentage points over roughly a year, with flat AI investment removing another 0.5 point and a fall to zero removing a full point.

Related Articles:

  • Minsky Moments and AI CapEx — Capital-spending forecasts for the four largest cloud providers rose nearly 50% over the year, and AI data centers are increasingly funded through leveraged and off-balance-sheet structures, including $3.2bn in senior secured notes and a roughly $30bn asset-backed financing. The author argues this marks a speculative-finance phase where borrowers must roll over principal.
  • Why We Are Not In A Bubble….Yet — The 5 largest US technology companies are worth about 16% of the global public equity market, and the top 10 US stocks nearly 25%, almost $25tn. The author argues valuations are high but not bubble-level, with the "Magnificent 7" median forward price-to-earnings ratio at 27x, roughly half that of the late-1990s leaders.
  • SF FedViews: Solid Growth, but Small Cracks Have Emerged — Since the second half of 2023, four-quarter US productivity growth has averaged nearly 2.5%, well above the pre-pandemic trend of about 1% to 1.25%, supporting real GDP growth. It remains unclear whether this reflects a higher productivity regime fostered by rising AI use or a temporary post-pandemic reorganization surge.

AI Summary. Composite benchmarks show Google's Gemini roughly in line with GPT-5-class models, with any gap small, and the author argues it does not lead on strict low-leakage reasoning exams. The pattern reflects growing training costs yielding modest gains concentrated in vision, video, and some coding rather than broad reasoning advances.

Paul Kedrosky Applied Complexity
Date Posted:

Kedrosky argues Google’s Gemini 3 is “strong, but not transformative,” and ~ comparable to ChatGPT5. “The industry remains on the same trajectory: rising costs, slower progress, and diminishing returns on the problems that matter most.”

Are rising AI training costs producing only narrow capability gains?

Core argument: Composite benchmarks drawn from broader, cleaner task sets place Gemini roughly in line with GPT-5-class models, with any overall gap small in either direction.

Composite benchmarks built from broader, cleaner sets show Gemini roughly in line with GPT-5-class models. On the strictest reasoning tasks—low-leakage exams—Gemini almost certainly does not lead. In many cases it likely trails. Either way. whatever gap exists overall, if any, is small. The Gemini release fits the same trend seen across the industry: growing training costs and shrinking capability gains. Progress has slowed on the tasks that matter most for general reasoning. The gains that do appear come from domains still responsive to brute-force scaling—vision, video, and certain coding workflows—not from breakthroughs in abstraction, logic, or planning. If Gemini had made a clear leap in those harder areas, the missing benchmarks would have been on the above list. They were not, almost certainly because the gains are not there. The core pattern remains: large increases in compute and engineering produce modest, targeted improvements rather than broad advances in general capability.

Takeaways by Macro Roundup® AI

  1. Composite benchmarks drawn from broader, cleaner task sets place Gemini roughly in line with GPT-5-class models, with any overall gap small in either direction.
  2. On low-leakage reasoning exams, Gemini almost certainly does not lead and likely trails, and the absence of those benchmarks from its release suggests the gains are not there.
  3. Gemini's release fits an industry trend of rising training costs and shrinking capability gains, with improvements concentrated in vision, video, and certain coding workflows rather than abstraction, logic, or planning.

Related Articles:

  • Is AI Hitting A Wall? — Artificial intelligence gains have come from a formula of more data and more computing power, but companies have exhausted freely available internet training data and are now striking data-sharing deals with publishers and copyright holders. Training and running large models is energy-intensive, requiring vastly more Nvidia chips with each generation.
  • Why AI Capex isn’t a Bubble: A Perez-ian Perspective — Carlota Perez's framework of technological revolutions casts heavy artificial-intelligence infrastructure spending as a late-stage installation surge, in which financial capital overshoots and builds capacity ahead of demand, leaving both losses and infrastructure for later productivity. The author argues this model is descriptive rather than prescriptive, and that executives and policymakers invoking it rationalize losses in advance and erode market discipline.

AI Summary. Pressure on technology company bonds has been uneven: Alphabet, Amazon and Microsoft fund most AI spending from quarterly cash flow, while Meta had to offer yields well above its existing debt on a $30bn bond sale. Oracle, already burning cash as it shifts toward AI cloud computing, carries yields higher than almost any investment-grade technology peer.

Sam Goldfarb Wall Street Journal
Date Posted:

Since September, hyperscalers have issued ~$90B of investment-grade debt at higher yields than expected. Post-issue prices of some of the bonds have declined. The cost of CDS’s on Oracle’s debt have climbed.

Does bond market pressure fall unevenly across AI-spending tech companies?

Core argument: Meta issued $30 billion of bonds at the end of October at yields comfortably above its existing debt, and despite double-A ratings those bonds now yield roughly the same as single-A-rated IBM debt after secondary-market prices edged lower.

Recent pressure on tech company bonds hasn’t been evenly applied across the sector. Alphabet, Amazon and Microsoft have been spared the worst, thanks to their ability to fund most of their AI expenditures with the huge amounts of cash that they generate each quarter. Meta has a slightly less massive cash hoard and is seen as likely needing more debt to realize CEO Zuckerberg’s large ambitions. When the social-media company issued $30 billion of bonds at the end of October, it had to lure investors with yields comfortably above those of its existing debt. Prices of some of those bonds then edged lower in the secondary market, pushing the yields higher. Despite carrying double-A ratings, they now yield roughly the same as IBM bonds, which are rated single-A by the major ratings firms. Oracle is in a more difficult position. It already is burning cash, with plans to burn tens of billions more over the next several years as it tries to transform itself from a leading software company into an AI cloud-computing giant—leasing out the vast clusters of advanced computer chips needed to power applications like OpenAI’s ChatGPT. Rated two notches above speculative-grade territory, Oracle’s bonds now carry yields that are higher than those of almost any of its investment-grade tech peers.

Takeaways by Macro Roundup® AI

  1. Meta issued $30 billion of bonds at the end of October at yields comfortably above its existing debt, and despite double-A ratings those bonds now yield roughly the same as single-A-rated IBM debt after secondary-market prices edged lower.
  2. Oracle, rated two notches above speculative grade and already burning cash with plans to burn tens of billions more over the next several years as it shifts toward AI cloud computing, carries bond yields higher than almost any investment-grade tech peer.
  3. Alphabet, Amazon and Microsoft have been spared the worst of the pressure on tech bonds, owing to their ability to fund most of their AI spending with the cash they generate each quarter.

Related Articles:

  • Minsky Moments and AI CapEx — Capital-spending forecasts for the four largest cloud providers rose nearly 50% over the year, and AI data centers are increasingly funded through leveraged and off-balance-sheet structures, including $3.2bn in senior secured notes and a roughly $30bn asset-backed financing. The author argues this marks a speculative-finance phase where borrowers must roll over principal.
  • Investor Angst Over Big Tech’s AI Spending Spills Into Bond Market — Investors have sold off bonds issued by large US technology companies building data centres, including Alphabet, Meta, Microsoft and Oracle, as concerns over artificial intelligence spending reached the bond market. Bank of America data show the extra yield investors demand to hold this debt over Treasuries rose to 0.78 percentage points from 0.5.
  • The AI Boom Is Looking More and More Fragile — Technology valuations are not extreme by historical standards, with the Nasdaq Composite's average forward price-to-earnings ratio near 29 versus above 32 before the artificial intelligence boom. Concern instead centers on leverage: AI-related companies issued $139bn of corporate bonds this year, about 9% of investment-grade issuance and 23% above last year.

AI Summary. Japan's gross debt is 240% of GDP versus Germany's 65%, yet both carry the same 30-year government bond yield, which the author argues shows Japanese long-term yields are massively mispriced by Bank of Japan bond buying. Letting markets set yields would stabilize the yen but risk a debt crisis; capping yields means further yen depreciation.

Robin Brooks Robin Brooks Substack
Date Posted:

Brooks argues BoJ intervention is still “massively” distorting Japan’s long-term yields, noting that Japan, with debt at 240% of GDP, and Germany, with debt at 65% of GDP, both have ~ the same 30-year government bond yield.

Are Japan's long-term bond yields mispriced by central bank buying?

Core argument: Japan's gross debt stood at 240% of GDP in 2024 versus Germany's 65%, yet both countries carry the same 30-year government bond yield, which the author calls a massive mispricing of Japanese yields.

Japan’s “shadow” long-term yield - the yield that would clear the market in the absence of BoJ bond buying - is much higher than what we observe. The easiest way to see how massively distorted Japan’s long-term yields are is in the chart. The horizontal axis shows gross debt in percent of GDP. The vertical axis shows 30-year government bond yields. Euro zone countries are in red, the rest of the G10 are in blue and Germany (DE) as well as Japan (JP) are in black. German debt in 2024 was 65% of GDP, while that of Japan was 240%. Yet both countries have the same 30-year government bond yield. That’s a sign of a massive mispricing in Japanese yields, which are way below where they should be. Japan has a choice. It can either let markets set yields without BoJ interference. That will stabilize the Yen, but at the cost of far higher long-term yields and a debt crisis. Or it can cap yields, in which case the Yen will continue its slide.

Takeaways by Macro Roundup® AI

  1. Japan's gross debt stood at 240% of GDP in 2024 versus Germany's 65%, yet both countries carry the same 30-year government bond yield, which the author calls a massive mispricing of Japanese yields.
  2. Japan's "shadow" long-term yield — the level that would clear the market without Bank of Japan bond buying — is far above observed yields, according to the author.
  3. The author argues Japan faces a choice between letting markets set yields, which would stabilize the Yen at the cost of far higher long-term yields and a debt crisis, or capping yields and allowing the Yen to keep sliding.

Related Articles:

  • Japan’s Borrowing Costs At Highest In Decades On Fears Of Public Spending Surge — Yields on Japan's 10-year government bonds reached their highest level since 2008 and 30-year yields hit a record intraday high of 3.35%. Traders attribute the sell-off to signs the Japanese government is preparing a supplementary budget of at least ¥25tn ($161bn) to boost the economy and shield households from rising prices.
  • Japan’s Debt Crisis — The Bank of Japan's holdings of Japanese government bonds are so large that maturing bonds exceed gross purchases, making it a net seller equal to -2.6% of GDP over a recent one-year period. Further cuts to gross buying would deepen the decline in its holdings and likely push longer-term Japanese yields higher.
  • Japan’s Debt Puzzle: Sovereign Wealth Fund from Borrowed Money — Japan's public sector operates as a highly leveraged sovereign wealth fund, funding risky assets by borrowing from banks and depositors at below-market rates, with returns sustaining its debt-to-GDP ratio above 200% without primary surpluses. The authors argue the resulting duration and currency mismatches expose Japan to mark-to-market losses that could trigger a fiscal crisis or a run on the yen.

AI Summary. Consumer prices rose about 20% during both the Biden and Reagan presidencies, yet voters punished Biden's party while Reagan won in a landslide. The author argues the difference is that Americans had grown accustomed to high inflation before Reagan, whereas the surge after decades of low inflation came as a shock.

Paul Krugman Krugman Wonks Out
Date Posted:

“Historical Context Matters.” Both Biden’s term and Reagan’s 1st term saw a ~20% rise in the price level, but only Biden’s saw a sharp drop in sentiment. Consumers had gotten used to inflation in the 1970s, while the Covid inflation came as a shock.

Why do voters punish similar price increases differently?

Core argument: Consumer prices in November 2024 stood about 20% above their level when Joe Biden took office, and despite a much lower inflation rate, voters punished his party at the polls.

In November 2024, consumer prices were about 20% higher than they had been when Joe Biden took office. Even though the rate of inflation was way down, Americans were outraged at these higher prices, and punished Biden’s party at the polls. But in November 1984 consumer prices were also about 20% higher than they had been when the president at the time, Ronald Reagan, took office. Yet voters didn’t punish Reagan and his party. On the contrary, he won in a landslide: It was “morning in America.” Why the difference? My answer is that when Reagan took office, America had already been through years of high inflation, which people had come to accept as normal. Nobody expected prices to go back down to their level years in the past, and people were happy simply to see inflation leveling off. Biden, by contrast, took office in a country where inflation had been very low for decades. The sudden surge in prices from 2021 to 2023 came as a huge shock. What will end the vibecession? As I’ve suggested, there has to be some statute of limitations on anger over past inflation. Inflation peaked more than 3 years ago, and eventually we should expect people to accept higher prices as normal.

Takeaways by Macro Roundup® AI

  1. Consumer prices in November 2024 stood about 20% above their level when Joe Biden took office, and despite a much lower inflation rate, voters punished his party at the polls.
  2. Prices in November 1984 were also roughly 20% higher than when Ronald Reagan took office, yet Reagan won re-election in a landslide.
  3. The author argues the difference lies in expectations: Reagan inherited a country accustomed to years of high inflation, while Biden inherited decades of very low inflation, so the 2021–2023 surge came as a shock.

Related Articles:

  • Vibecessions, Part I — Consumer sentiment in the long-running University of Michigan survey is worse than during periods of roughly 9% inflation, double-digit unemployment after the financial crisis, and the 1980 recession, even though unemployment stands at 4.3% and inflation at 3%. The article argues people credit themselves for wage gains while blaming inflation for taking those gains away.
  • Why Do We Dislike Inflation? — Survey evidence finds 80% of respondents believe prices rise faster than wages, though real weekly earnings for the median worker grew 1.7% between 2019 and 2023. About 40% think their nominal income would be higher had inflation been lower, and half expect wages to take over a year to catch up if inflation doubled.
  • Zero-Sum Thinking and the Roots of U.S. Political Divides — Zero-sum thinking — viewing gains for one group as losses for another — is associated with greater support for liberal economic policies, stronger alignment with the Democratic Party, and weaker alignment with the Republican Party. The authors also find it linked to believing the QAnon conspiracy theory holds some truth and to sympathy for the attack on the U.S. Capitol.