Edward Conard

Top Ten New York Times Bestselling Author

  • “Unintended Consequences provides a provocative interpretation of the causes of the global financial crisis and the policies needed to return to rapid growth. Whether you agree or not, this analysis is well worth reading.” - Nouriel Roubini, New York University; Chairman, Roubini Global Economics
  • “…a comprehensive explanation of the modern economy.” - Julian Robertson, Founder, Tiger Management
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “Unintended Consequences offers deep and well-argued analyses on almost every issue.” - The New York Times
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “A full-throated defense of economic dynamism.” - The Wall Street Journal
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “…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
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Friday, September 6, 2024

AI Summary. Markets are never perfectly efficient because betting against other investors' errors carries risk, and the author argues social media, investor overconfidence, and gamified 24/7 trading have made markets meaningfully less efficient, with indexing and very low interest rates a smaller contributor. The recommended response is to adopt the longest possible investment time horizon.

Clifford Asness The Journal of Portfolio Management
Date Posted:

.@CliffordAsness notes large and prolonged swings in US stock market valuations since 1990, and contends that the market has become “less informationally efficient.” He advocates having “the longest time horizon possible” to wait out bubbles.

Have markets become less efficient, and how should investors respond?

Core argument: Asness argues markets are never perfectly efficient because betting against other investors' errors, even when correctly identified, is not riskless; efficiency rises as more rational investors take the other side and falls as more irrational traders act more irrationally.

Markets are never perfectly efficient, as taking the other side of other investors’ errors, even if you correctly identify them, is still not riskless. The more rational investors taking the other side, the more efficient. The more minnows doing crazy things, and the crazier they are, the less so. I think the rise of indexing and the super low interest rates of the last 10-15 years may have had exactly this effect (more crazy, and crazier, minnows, fewer rational G&D disciples), but I conjecture that’s a relatively small part of the story. I’m far more certain that social media, the overconfidence that comes when people think all the world’s data is at their fingertips, and gamified, fake-free, instant, 24/7 trading has done so in a significant way. Assuming a valuation change will continue to go on forever is obvious folly. So, pushing yourself (and whoever you report to!) to have the longest time horizon possible is definitely #1 on the list. Quoting myself, “A long-term horizon is the closest thing to an investing superpower.”

Takeaways by Macro Roundup® AI

  1. Asness argues markets are never perfectly efficient because betting against other investors' errors, even when correctly identified, is not riskless; efficiency rises as more rational investors take the other side and falls as more irrational traders act more irrationally.
  2. Asness expresses greater certainty that social media, the overconfidence that comes when people think all the world's data is at their fingertips, and gamified, fee-free, instant, 24/7 trading have made markets significantly less efficient.
  3. Asness conjectures that the rise of indexing and the very low interest rates of the past 10–15 years produced more and crazier small irrational traders and fewer Graham-and-Dodd disciples, but calls this a relatively small part of the story.

Related Articles:

  • Cliff Asness: AI is ‘Still Just Statistics’ — Artificial intelligence in investing is still statistics — data goes in and a forecast comes out — and is not as revolutionary as many claim, the author argues. Questions such as the premium for high- versus low-quality stocks or value versus growth involve a small-data problem, not a big-data one, so more computation cannot answer them.
  • Economist Eugene Fama: ‘Efficient Markets Is A Hypothesis. It’s Not Reality’ — Technology stocks surged and then collapsed in the late 1990s, but the underlying premise that the internet would create hugely profitable new companies proved correct, with some prices too low even as most were too high. Amazon is now worth about a quarter of the Nasdaq's entire market capitalization at its dotcom peak.
  • Which U.S. Stocks Generated the Highest Long-Term Returns? — Of the 29,078 U.S. common stocks in a database spanning nearly a century, 51.6% had negative cumulative returns, while 17 stocks returned more than 5m% — led by Altria at 265m%. Those top performers averaged annualized returns of just 13.47%, underscoring the value of time invested.

AI Summary. Taiwan Semiconductor Manufacturing Co.'s trial production yields at its first advanced Arizona plant, which uses 4-nanometer process technology, are similar to those at comparable facilities in Tainan, Taiwan, according to a person familiar with the company. The company had delayed full production at the site, citing a lack of skilled workers.

Debby Wu Bloomberg
Date Posted:

Initial production yields at TSMC’s Arizona facility are on par with the yields at comparable Taiwanese fabs. There have been concerns about the efficiency of their US facilities due to a lack of skilled workers.

Can advanced chip plants in the US match yields achieved in Taiwan?

Core argument: TSMC's Arizona plant reached trial-production yields similar to comparable facilities in Tainan, Taiwan, according to a person familiar with the company, an early indicator the US project is on track to meet its targets.

Taiwan Semiconductor Manufacturing Co. achieved production yields at its Arizona facility on par with established plants back home, an early indicator that its marquee US project is on track to achieve its targets. The Taiwanese chipmaker’s yield rate in trial production at its first advanced US plant is similar to comparable facilities in the southern Taiwanese city of Tainan, according to a person familiar with the company, who asked not to be identified discussing private corporate matters. TSMC had said it started engineering wafer production in April with advanced 4-nanometer process technology. TSMC originally planned to have its first Arizona plant start full production in 2024, but pushed back the target to 2025 due to a lack of skilled workers. The delay fueled concerns that the company might not be able to make chips in the US as efficiently as in Taiwan.

Takeaways by Macro Roundup® AI

  1. TSMC's Arizona plant reached trial-production yields similar to comparable facilities in Tainan, Taiwan, according to a person familiar with the company, an early indicator the US project is on track to meet its targets.
  2. TSMC said it began engineering wafer production at its first advanced Arizona plant in April using 4-nanometer process technology.
  3. TSMC pushed its Arizona full-production target from 2024 to 2025 due to a lack of skilled workers, a delay that fueled concerns the company could not make chips in the US as efficiently as in Taiwan.

Related Articles:

  • With US Chips Act Money Mostly Divvied Up, the Real Test Begins — The US Chips Act is a bet that Intel, Micron, Taiwan Semiconductor Manufacturing, and Samsung can return advanced chip production to the US, with an explicit goal of making a fifth of the world's most advanced processors by 2030, up from roughly zero. Nearly 100 companies have pledged about $400bn in US facilities.
  • The Chips Act Has Been Surprisingly Successful So Far — The US government has spent over half of its $39bn in Chips Act semiconductor incentives, and chip companies and supply chain partners have announced $327bn of investments over the next 10 years, according to Semiconductor Industry Association calculations. US data show a 15-fold increase in construction of computing and electronics manufacturing facilities.
  • Business Spending Is Lining Up to Reinforce Growth — Around 70% of sectors are posting positive year-over-year profit growth, with big tech showing the largest acceleration in demand and profits but strength broadening across the economy. Companies are responding with rising capital spending growth and more optimistic investment plans, as firms typically add capacity with a slight lag after real demand picks up.

AI Summary. Polling shows Kamala Harris underperforming in swing states relative to her national numbers, a pattern especially evident in Pennsylvania. Donald Trump has also been gaining ground in most state polling averages, which are not affected by the forecast's convention bounce adjustment.

Nate Silver and Eli McKown-Dawson Silver Bulletin
Date Posted:

Noting “Harris is underperforming in swing states relative to her national numbers” (she is up 3pp nationally). @NateSilver538 gives Trump a 61.7% chance of winning PA and a 61.5% chance of winning the electoral college. The first debate is Tuesday.

Is the Democratic candidate underperforming in swing states relative to national polling?

Core argument: Polls show Kamala Harris underperforming in swing states relative to her national numbers, a recurring pattern in the Silver Bulletin forecast, especially in Pennsylvania.

Another day, another set of polls showing Kamala Harris underperforming in swing states relative to her national numbers. No, it’s not from the world’s best pollster, but this has been a theme lately, especially in Pennsylvania. Yes, Harris’s forecast — remains subject to a convention bounce adjustment, although the impact is decreasing a bit. But Trump has been gaining ground on her in most of our state polling averages, too — and they aren’t impacted by the convention bounce adjustment at all.

Takeaways by Macro Roundup® AI

  1. Polls show Kamala Harris underperforming in swing states relative to her national numbers, a recurring pattern in the Silver Bulletin forecast, especially in Pennsylvania.
  2. Donald Trump has been gaining ground on Harris in most of the Silver Bulletin's state polling averages, which are not affected by the convention bounce adjustment.
  3. Harris's Silver Bulletin forecast remains subject to a convention bounce adjustment, though that adjustment's impact is decreasing.

Related Articles:

  • Many Gen Z Men Feel Left Behind. Some See Trump As An Answer — National polling shows a gender gap of roughly 39 to 51 percentage points among voters under 30, with young men favoring the Republican candidate and young women favoring the Democrat. The Republican message resonates particularly with young men without college degrees and young men of color.
  • Disconnected: The Growing Class Divide in American Civic Life — 17% of Americans report having no close friends, and survey data show a wide educational gap: 24% of those with a high school education or less have no close friends versus 10% of college graduates. Three decades ago, Americans with less formal education reported social circles at least as large.
  • An Unsettled Electorate: How Uncertainty and Apathy Are Shaping the 2024 Election — A survey of young registered voters finds 54% backing Biden and 46% backing Trump, with 60% of young women supporting Biden while 52% of young men support Trump. Young men are more likely than young women to say the country has gone too far in giving women equal rights, 19% versus 9%.

AI Summary. In a study where 200 participants watched a crime video and were then questioned with misleading questions, a generative AI chatbot induced over 3x more immediate false memories than the control condition and 1.7x more than a survey, misleading 36.4% of users' responses.

Samantha Chan, Pat Pataranutaporn, Aditya Suri, et al. MIT Media Lab
Date Posted:

Researchers at MIT @medialab found that after viewing a video of a simulated crime, participants who were asked misleading questions by an AI chatbot generated 3x more false memories than a control group.

Do AI chatbots induce more false memories than surveys?

Core argument: In an experiment with 200 participants who watched a crime video, a generative AI chatbot interviewer induced over three times more immediate false memories than the control condition and 1.7 times more than a survey method.

Participants (N=200) watched a crime video, then interacted with their assigned AI interviewer or survey, answering questions including five misleading ones. False memories were assessed immediately and after one week. Results show the generative chatbot condition significantly increased false memory formation, inducing over 3 times more immediate false memories than the control and 1.7 times more than the survey method. 36.4% of users' responses to the generative chatbot were misled through the interaction. After one week, the number of false memories induced by generative chatbots remained constant. However, confidence in these false memories remained higher than the control after one week. Moderating factors were explored: users who were less familiar with chatbots but more familiar with AI technology, and more interested in crime investigations, were more susceptible to false memories. These findings highlight the potential risks of using advanced AI in sensitive contexts, like police interviews, emphasizing the need for ethical considerations.

Takeaways by Macro Roundup® AI

  1. In an experiment with 200 participants who watched a crime video, a generative AI chatbot interviewer induced over three times more immediate false memories than the control condition and 1.7 times more than a survey method.
  2. Some 36.4% of users' responses to the generative chatbot were misled through the interaction, and the number of false memories it induced remained constant after one week, with confidence in those memories staying higher than in the control condition.
  3. Participants less familiar with chatbots but more familiar with AI technology, and those more interested in crime investigations, were more susceptible to false memories, which the authors link to risks of using advanced AI in settings such as police interviews.

Related Articles:

  • Exploring How AI Will Impact the Economy — Artificial intelligence is projected to pass through to productivity faster than past general-purpose technologies, though peak effects remain years away. The argument rests on lower capital spending per unit of labor saved, capital spending concentrated among major software players that can distribute AI tools through platforms businesses already use, and low unemployment with high wage growth raising incentives to automate.
  • Erik Brynjolfsson: ‘This Could Be The Best Decade in History — Or The Worst’ — Studies of large language models across tasks including software coding, consulting, and writing find double-digit productivity gains appearing within months, with less-skilled workers typically benefiting most. Given the large share of modern work these tools can augment, the author projects US productivity growth near 3% annually in the 2020s, roughly double the Congressional Budget Office's 1.4% projection.

Monday, September 9, 2024

AI Summary. Labor market earnings are strongly associated with measures of personality traits and preferences, and over the last two decades returns to cognitive skills have been flat or declined in the US, Sweden, and Finland while returns to "soft" skills such as conscientiousness, extraversion, emotional stability, and social skills have grown rapidly.

David Deming and Mikko Silliman National Bureau of Economic Research
Date Posted:
Is Database:
Database

.@ProfDavidDeming and @m_silliman find “the economic return to social skills in the US more than doubled” between the 1980s and the 2000s, and argue for a multi-dimensional concept of human capital.

Are labor market returns shifting from cognitive skills to soft skills?

Core argument: Economic returns to cognitive skills have been flat or declined in the U.S., Sweden, and Finland, while returns to non-cognitive skills such as conscientiousness, extraversion, emotional stability, and social skills have grown rapidly.

We present evidence for the growing importance and economic value of "non-cognitive" skills - a strong association between labor market earnings and various measures of personality traits and preferences. Over the last two decades, the economic returns to cognitive skills have been flat or declined in the U.S., Sweden, and Finland, while the returns to "non-cognitive" or "soft" skills like conscientiousness, extraversion, emotional stability, and social skills have grown rapidly. We argue that the right term for these capacities is "higher order" skills, in part because they are farther up the hierarchy of cognitive processes than the skills measured by traditional tests of academic achievement. Teamwork and decision-making require workers to apply knowledge to solving new problems and to connect one’s own skills and ideas to others. We argue that modeling higher order skills requires us to consider workers as agents who make choices about which jobs or tasks to do, and whether to do them alone or in a team.

Takeaways by Macro Roundup® AI

  1. Economic returns to cognitive skills have been flat or declined in the U.S., Sweden, and Finland, while returns to non-cognitive skills such as conscientiousness, extraversion, emotional stability, and social skills have grown rapidly.
  2. Evidence of a strong association between labor market earnings and measures of personality traits and preferences supports terming these capacities "higher order" skills, farther up the hierarchy of cognitive processes than skills captured by traditional academic achievement tests.
  3. Because teamwork and decision-making require applying knowledge to new problems and connecting one's own skills and ideas to others, modeling higher order skills requires treating workers as agents who choose which jobs or tasks to do and whether to do them alone or in a team.

Related Articles:

  • The Economics of Inequality in High-Wage Economies — Information technology created investment opportunities exceeding the supply of properly trained talent, while trade with low-wage economies, low-skilled immigration, trade deficits, and aging demographics eased constraints on low-skilled labor and risk-averse savings. The author argues this framework, rather than rising cronyism, explains high returns to success, slowing productivity growth, and low interest rates.
  • Why Do Wages Grow Faster for Educated Workers? — College graduates sort into professional, non-routine occupations where wages rise much more with job tenure, and the author estimates this occupational sorting explains at least half of the life-cycle growth in the college wage premium. That premium is estimated at 22% at age 26, 49% at age 40, and 56% at age 60.
  • Multidimensional Human Capital and the Wage Structure — The wage return to cognitive skills has declined, from about 10% higher hourly wages per standard deviation of test score in the 1980s and early 1990s to 4.5% in the 2000s and early 2010s, while returns to a bachelor's degree rose. Social skill-intensive occupations grew as a share of US jobs, with faster real wage growth.

AI Summary. If consumption growth accounted for 70% of China's future GDP growth with nominal GDP growing 6% a year, the author estimates it would take 16 years for consumption to rise from 53.4% of GDP to 63–64%. Even with consumption driving 90% of growth, the shift would take 6 years.

Michael Pettis Carnegie Endowment for International Peace
Date Posted:

.@michaelxpettis argues that rebalancing China’s economy to raise the consumption share of GDP from 53% in 2023 to a target of 63–64% may take decades. The alternative is “surging debt” or a “disruptive contraction in GDP.”

How long would raising China's consumption share of GDP take?

Core argument: With consumption growth accounting for 70% of China's future GDP growth and nominal GDP growing 6% a year on average, China's consumption share of GDP would take 16 years to rise from 53.4% in 2023 to 63–64%.

If growth in consumption accounted for 70% of China’s future GDP growth, and if China’s nominal GDP grew by 6% on average, it would take 16 years for the consumption share of China’s GDP to rise [from 53.4% in 2023] to 63–64% of GDP. Even if we allowed for exceptional circumstances, with an annual average GDP growth rate of 6% and the growth in consumption accounting for nearly all of the country’s future GDP growth (90%), it would still take 6 years for the consumption share of China’s GDP to rise to 63–64% of GDP. From such a low base, it is not enough that the consumption share of GDP growth is above average (say 60 to 70% ) for a few years for China to really catch up to a more sustainable consumption share of GDP.

Takeaways by Macro Roundup® AI

  1. With consumption growth accounting for 70% of China's future GDP growth and nominal GDP growing 6% a year on average, China's consumption share of GDP would take 16 years to rise from 53.4% in 2023 to 63–64%.
  2. Even under exceptional circumstances in which consumption growth accounts for 90% of China's GDP growth at a 6% average annual rate, China would still need 6 years to lift consumption to 63–64% of GDP.
  3. From a base of 53.4% of GDP in 2023, a consumption share of GDP growth of 60–70% sustained for only a few years would not be enough for China to reach a more sustainable consumption share of GDP.

Related Articles:

  • Communist Party Priorities Complicate Plans to Revive China’s Economy — China's leadership objects to Western-style consumption-driven growth, viewing it as wasteful and at odds with the goal of making China a world-leading industrial and technological power, according to people familiar with decision-making in Beijing. Chinese officials also emphasized avoiding a current-account deficit, which would signal greater dependence on the outside world.
  • Why Is It So Hard for China to Boost Domestic Demand? — Chinese manufacturing competitiveness was built on transfers that acted as implicit supply-side support, so reversing them to raise household consumption would likely erode that competitiveness and cause a short-term contraction. Manufacturing accounts for 27–28% of China's economy versus 15–16% of global output, making any such hit especially painful.
  • What Will It Take for China’s GDP to Grow at 4–5 Percent Over the Next Decade? — China accounts for 18% of global GDP, 13% of global consumption, and 31% of global manufacturing. The author estimates that if China grew 4–5% annually for a decade while keeping manufacturing's role in its economy, its share of global manufacturing would reach 36–39%, requiring the rest of the world to let manufacturing's share of its output fall by 0.5–1.0pp.

AI Summary. A New York Times/Siena College national poll of likely voters found Trump leading Harris 48% to 47%, within the poll's three-percentage-point margin of error, alongside battleground-state polling averages showing a tied race or slim Harris leads. In the same poll, 47% of likely voters viewed Harris as too liberal, versus 32% who saw Trump as too conservative.

Jonathan Weisman and Ruth Igielnik New York Times
Date Posted:

The latest Times/Siena poll shows the race for the White House is a dead heat with Trump leading 1pp nationally at 48% (margin of error 3pp). Harris is leading by 2pp in WI and MI and by 1pp in PA. The candidates are tied at 48% in GA, NC, AZ and NV.

Is the presidential race essentially tied nationally and in battleground states?

Core argument: A New York Times/Siena College national poll of likely voters found Trump leading Harris 48% to 47%, a gap within the poll's three-percentage-point margin of error.

A national poll of likely voters by The New York Times and Siena College found Mr. Trump leading Ms. Harris, 48% to 47%, within the poll’s three-percentage-point margin of error. The national results are in line with polls in the seven battleground states that will decide the presidential election, where Ms. Harris is tied with Mr. Trump or holds slim leads, according to New York Times polling averages. Taken together, they show a tight race that remains either candidate’s to win or lose. 47% of likely voters viewed Ms. Harris as too liberal, compared with 32% who saw Mr. Trump as too conservative.

Takeaways by Macro Roundup® AI

  1. A New York Times/Siena College national poll of likely voters found Trump leading Harris 48% to 47%, a gap within the poll's three-percentage-point margin of error.
  2. New York Times polling averages show Harris tied with Trump or holding slim leads across the seven battleground states that will decide the presidential election.
  3. In the Times/Siena poll, 47% of likely voters viewed Harris as too liberal, compared with 32% who saw Trump as too conservative.

Related Articles:

  • Silver Bulletin 2024 Presidential Election Forecast — Polling shows Kamala Harris underperforming in swing states relative to her national numbers, a pattern especially evident in Pennsylvania. Donald Trump has also been gaining ground in most state polling averages, which are not affected by the forecast's convention bounce adjustment.
  • Many Gen Z Men Feel Left Behind. Some See Trump As An Answer — National polling shows a gender gap of roughly 39 to 51 percentage points among voters under 30, with young men favoring the Republican candidate and young women favoring the Democrat. The Republican message resonates particularly with young men without college degrees and young men of color.
  • An Unsettled Electorate: How Uncertainty and Apathy Are Shaping the 2024 Election — A survey of young registered voters finds 54% backing Biden and 46% backing Trump, with 60% of young women supporting Biden while 52% of young men support Trump. Young men are more likely than young women to say the country has gone too far in giving women equal rights, 19% versus 9%.

AI Summary. A meta-analysis found that polygenic scores, person-specific aggregates of inherited DNA differences, are associated with educational attainment at a correlation of .27 and with educational achievement such as school grades at .24; after accounting for sample ancestry, age at assessment, and education measure, these estimates rose to .29 and .50. Only 15% of estimates came from samples of non-European ancestry.

Kirsty Wilding, Megan Wright, and Sophie von Stumm Educational Psychology Review
Date Posted:
Is Database:
Database

A meta-analysis of international studies finds a 24% correlation between education achievement, as measured by grades, and “polygenic scores that are person-specific aggregates of inherited DNA differences.”

How well do inherited DNA differences predict educational attainment and achievement?

Core argument: A multilevel meta-analysis of 20 estimates covering 314,757 individuals found polygenic scores associated with educational attainment at ρ = .27 (95% CI .22–.32), and with educational achievement at ρ = .24 (95% CI .18–.30) across 19 estimates and 83,788 individuals.

Recent advances in genomics make it possible to predict individual differences in education from polygenic scores that are person-specific aggregates of inherited DNA differences. Here, we systematically reviewed and meta-analyzed the strength of these DNA-based predictions for educational attainment (e.g., years spent in full-time education) and educational achievement (e.g., school grades). For educational attainment (k = 20, n = 16, Ntotal = 314,757), a multilevel meta-analysis showed an association with polygenic scores of ρ = .27 (95% CI from .22 to .32). For educational achievement (k = 19, n = 10, Ntotal = 83,788), the association was ρ = .24 (95% CI from .18 to .30). Eurocentric biases were evident with only 15% of estimates being reported in samples of non-European ancestry. After accounting for sample ancestry, age at assessment, and education measure, the meta-analytic estimates increased to ρ = .29 (95% CI from .24 to .33) for educational attainment and ρ = .50 (95% CI from .39 to .61) for educational achievement, indicative of large effect sizes.

Takeaways by Macro Roundup® AI

  1. A multilevel meta-analysis of 20 estimates covering 314,757 individuals found polygenic scores associated with educational attainment at ρ = .27 (95% CI .22–.32), and with educational achievement at ρ = .24 (95% CI .18–.30) across 19 estimates and 83,788 individuals.
  2. After accounting for sample ancestry, age at assessment, and education measure, the meta-analytic estimates rose to ρ = .29 (95% CI .24–.33) for educational attainment and ρ = .50 (95% CI .39–.61) for educational achievement, described as large effect sizes.
  3. Eurocentric biases were evident in the reviewed literature, with only 15% of estimates reported in samples of non-European ancestry.

Related Articles:

  • Standardized Test Scores and Academic Performance at Ivy-Plus Colleges — Using admissions records and first-year grades at multiple Ivy-Plus colleges, the authors find SAT and ACT scores substantially predict college academic success even among students with the same high school grades, while high school grade point average does little to predict it. Conditional on test scores, students from lower- and higher-resourced high schools have virtually identical college grades.
  • Steeper At The Top: Cognitive Ability and Earnings in Finland and Norway — Administrative data on over 350,000 men each in Finland and Norway show a convex relationship between earnings rank and cognitive ability measured by military conscription tests: top-percentile earners score about 1 standard deviation above median earners, who score about 0.5 above bottom-percentile earners. The lowest scores are almost absent among top earners.
  • US PISA Results By Subpopulation — On the international PISA student assessments, American Asian and White students ranked at the top, American Hispanic students outscored Hispanic students elsewhere, and American Black students performed well. In reading, American Asian students placed first and American White students third.

AI Summary. AlphaProteo, an AI system trained on protein structure data, generates novel protein binders for target molecules, including the first successful AI-designed binder for VEGF-A, a protein associated with cancer and diabetes complications. Across seven tested targets it achieved higher experimental success rates and 3x to 300x better binding affinities than the best existing methods.

Vinicius Zambaldi, David La, Alexander Chu, et al. Google
Date Posted:

.@GoogleDeepMind has developed AlphaProteo, an AI system that designs proteins that bind to target molecules at success rates 3 to 300 times better than the best prior methods, potentially accelerating advanced drug design.

Can AI design novel protein binders better than existing methods?

Core argument: AlphaProteo achieves 3–300 times better binding affinities and higher experimental success rates than the best existing methods across seven target proteins tested.

AlphaProteo can generate new protein binders for diverse target proteins, including VEGF-A, which is associated with cancer and complications from diabetes. This is the first time an AI tool has been able to design a successful protein binder for VEGF-A.AlphaProteo also achieves higher experimental success rates and 3 to 300 times better binding affinities than the best existing methods on seven target proteins we tested. Trained on large amounts of protein data from the Protein Data Bank (PDB) and more than 100 million predicted structures from AlphaFold, AlphaProteo has learned the myriad ways molecules bind to each other. Given the structure of a target molecule and a set of preferred binding locations on that molecule, AlphaProteo generates a candidate protein that binds to the target at those locations. AlphaProteo’s performance indicates that it could drastically reduce the time needed for initial experiments involving protein binders for a broad range of applications.

Takeaways by Macro Roundup® AI

  1. AlphaProteo achieves 3–300 times better binding affinities and higher experimental success rates than the best existing methods across seven target proteins tested.
  2. AlphaProteo designed a successful protein binder for VEGF-A, a protein associated with cancer and diabetes complications, the first time an AI tool has done so.
  3. AlphaProteo was trained on protein data from the Protein Data Bank and more than 100 million predicted structures from AlphaFold, and generates a candidate binder given a target structure and preferred binding locations.

Related Articles:

  • Accurate Structure Prediction of Biomolecular Interactions With AlphaFold 3 — AlphaFold 3, a single deep learning model with a diffusion-based architecture, jointly predicts the structures of complexes containing proteins, nucleic acids, small molecules, ions, and modified residues. The authors report accuracy exceeding specialised tools for protein-ligand docking, protein-nucleic acid interactions, and antibody-antigen prediction.
  • Evaluation and Learning in R&D Investment — Pharmaceutical firms favor incremental drugs because they are easier to evaluate, even though novel drugs generate more learning that improves future screening decisions. The authors argue policymakers could strengthen incentives for novel research by making it easier for firms to capture indirect revenues from successor projects, which accrue even when the original project fails.