July 2026 links
Agreed. The Obama and Biden stimulus policies were large one-time fiscal shocks, whilst the unfunded tax cuts affect the flow.
Ignore relative poverty statistics. Only absolute poverty measures matter.
Subsiding agriculture uses far more water than all data centres combined.
An antidote to the prevailing left-wing narrative on UK inequality.
That's why we lack AC here. We've basically banned it.
An excellent report on UK nuclear power policy.
Learning-by-doing reduces the optimal labour taxes for any given SWF.
The US shale gas revolution was a boon for domestic consumers and industry, and the world (more substitutes make hedging against geopolitical volatility easier). We're making a major mistake with North Sea oil and gas, and by restricting fracking.
Chinese and Indian takeaways by far dominate the UK takeout industry, so seeing which one of these is more popular here is a good proxy for who has the edge in soft power. Chinese has led in terms of sales since 2009.
According to this, in 2024, AI produced "nearly half of all newly uploaded tracks" on Spotify. Will we get a Pangram for music?
America's shift towards universal healthcare crowded out private insurance coverage, particularly amongst higher income groups. Universal healthcare is inefficient as most recipients aren't poor.
A stagnating or declining labour share, if offset via a rise in capital share, may not imply stagnation in living standards for the median household. Greater lifetime wealth compensates. However, wealth inequality (pre-taxes and transfers) likely increases.
Why is autism therapy booming? To what extent is this related to this? Are we just medicalising personality differences at this stage?
HANK models tend to overstate fiscal multipliers as stimulus moves some households into a lower MPC category.
Introducing time into monopolistic competition.
Fascinating. Integration occurs regardless of the preferences of migrants and natives, via the education (of future generations) channel.
A shift from coercion to consent-based sexual assault laws reduces fertility and reduces unaminous verdicts (implying greater disagreement on the validity of a conviction). Around 1/3 of men fear false accusations. The worst legacy of feminism. How on earth do you define certain consent ex-ante? How do you prove the lack of it? In Britain, being "reckless" is sufficient for conviction, which shifts the onus of proof onto the defendant. One is now guilty until proven innocent. The rule of law and habeas corpus suffers.
The credibility revolution has skipped most of social science, with economics and political science being notable exceptions. Much of biology and medicine also faces issues with robustness, so identification per se is insufficent.
A good primer on how regulatory stasis drives up the cost of clinical trials. Manufacturing requirements stifle many biotech firms such as ultra rare disease companies.
Daraxonrasib is the tip of the iceberg!
On that paper on fertility and smartphones: are these results plausible? Misclassification in reweighting the TWFE design could undermine the robustness of the paper? Arguably the estimator itself overfits pre-treatment trends.
Palantir delivers results. Why won't MidJourney?
If the bold bets of AI firms pay off, and no reason to expect they shouldn't, expect 7% growth. At the very least, AI is highly likely to boost growth by a full percentage point - in itself an amazing feat on par with the industrial revolution.
"Urbanization, multigenerational living, women's & husbands' education, women's employment, & the non-agricultural sector all predicted declines" in fertility, but "women's education & sectoral composition" are the main forces after covariate adjustment. I think what makes this convincing is that levels and changes are well-explained by these factors. In "socially conservative" countries with a high % of female grads (e.g. Italy, China), TFR is very low. In those with low female education (e.g. Islamic world), TFR is high.
AI's politics.
Some fairly robust evidence that the smartphones might be reducing fertility. I'm willing to concede on this one, although I think the effect size is small here.
On average, gays have a smaller anterior hypothalamus than straight men. Suggests that homosexuality does indeed yield a biological explanation invariant to individual choice.
Indivisibility of AI agents is a case for, not against, granting them property rights and some legal concept of personhood. The rule of law relies on this, or else criminal arbitrage becomes the equilibrium.
Reform's rise has nothing to do with immigration, or if it does there are lags (which might be a result of an equilibrium with rational agents individually optimising, given rational inattention). Means-testing the winter fuel allowance seems to be a large driver of Starmer's and Labour's polling decline.
Definitely. You have to double-check its output to see if it's not a rewritten draft with different word choices. The more you use AI, the more you understand its limitations and attractors constraining output (hence the often repetitive dialogue of LLMs). You then converge to a position that's not doomer or accelerationist. You see the institutional frictions that limit the rate of increase in adoption, even with AGI. Yet if the rate of adoption is positive, growth accelerates with its compounding effects.
Nationalisation of utilities would "probably provoke significant legal challenges". Burnham is an unserious figure. Britain's decline into a quasi-socialist nation is almost complete.
The overall effect will be to substantially slow down the pace of growth in capabilities, and with it scientific and technological progress. When has progress ever been a net negative?
Claude on air conditioning. Also, yes, we do ban ACs in practice. How is a de-facto ban substantively different from a de-jure ban?
New results on heritability. Twin studies predict greater heritability estimates for height, BMI, smoking, and EA (with higher confidence) than sib-regressions, and lower relative estimates (though again with higher confidence) for IQ. Seb thinks the missing heritability gap is narrowing, and much of what remains is (in my view) explained via molecular studies perhaps being underpowered. It’s also worth noting that being an undergraduate is barely correlated with IQ anymore, so EA is an increasingly unreliable predictor of IQ. University education has generally become less selective, and (especially given extensive public subsidy in recent decades) is a less costly (hence weaker) signal than it used to be. Of course, graduates still command a high wage premium, so this suggests some positive human capital implications of a degree.
Credible estimates of the obesity penalty.
Yes. Likelihood is not probability. With probability, the distribution is already known and observed; eliminating the very need for MLE.
Incorporating RL dynamics into solving HANK resolves the constraints relating to dimensionality, allowing for more efficient computation of global solutions. This is why LLMs don’t sidestep the Lucas Critique. They complement theory via better calibration.
Many applications of AI agents assume, under uncertainty, they react "in roughly human-like ways, if not more rationally". But they're "typically much more sensitive to ordinary choice architecture cues". Frontier reasoning mitigates this somewhat.
"Officials who look more competent, trustworthy, and less aggressive enjoy significantly better promotion prospects" than peers. Indeed this seems to matter as much as performance or connections.
"There is little robust evidence that Ramadan fasting broadly impairs cognitive performance among expert chess players". The longevity field will benefit by not wasting any more time to this fad.
A very good discussion today on mathematical talent and its relation to genius. It strikes me that the most successful people in any domain tend to have fun doing what they're doing. Remember, Feynman had an IQ of only 125, yet he was undoubtedly one of history's greatest scientists. Just because the R^2 is positive does not mean that all the covariance is explained by IQ. There is a certain (yet unknown and domain-specific) threshold where IQ is a necessary condition, but beyond that other factors have strong predictive validity, and in any case IQ is not sufficient for success. The upper end of the heritability estimates for IQ in the twin-studies tend to be around 60%. Results above that are rarer. So if 40% of the variance in IQ is environment etc then one's intrinsic (dis)like of a subject matters.
AI has identified a new biomarker that better predicts risk of sudden cardiac death. Using this, allocation of defibrillators becomes more efficient. Yet people still won't take the log utility bet, even though millions of lives will probably be saved?
The benefits of land reclamation.
Sticky wages are more 'realistic' than sticky prices (e.g. recent EU oil shock shows it's relative input prices that matter for output, and wages could proxy inputs). Search-theoretic frameworks model the entire wage setting process explicitly, but this isn't tractable for NK, and likely entails ditching a PC.
ChatGPT 5.6. does not seem that much of an improvement from Mythos, which is an incremental improvement from Fable, almost equivalent to ChatGPT 5.5. Government decrees often yield arbitrary enforcement, which is one reason to heavily disfavour them.
One could say that growth trickles down.
This is efficient but painful. Yet shielding individuals and places from the pain makes most worse off.
Wooldridge on OLS vs IV.
Deflationary policies raise real minimum wages so should boost employment and output under monopsony. Yet when we had the post-Covid inflation spike, labour markets were running at almost unprecedented levels of tightness in the US, with strong employment and output figures.
Why rent control is one of the most stupid policies on the planet.