A light aircraft cost about twice what a car cost in the 1970s. It now costs ten times as much, and the engines in it were designed in the 1950s.
Nothing in the engineering failed. John Cochrane's argument is that the regulation worked exactly as written, and that the same mechanism — every attempt needing permission before it is made — is what has taken growth to 2% in the United States and close to zero in Europe.
"One thing you'll notice — I'm a pilot, so I look up at the sky a lot — the skies are empty in Europe."
Cochrane spent 30 years as a professor of finance at the University of Chicago Booth School of Business, is now the Rose-Marie and Jack Anderson senior fellow at the Hoover Institution, and writes The Grumpy Economist. He is also a pilot, which is how he came to notice the empty skies.
I listened to the full interview so you can skip it. 51 minutes of audio, 21 minutes of reading.
Here are the 13 takeaways that matter.
👤 Guest: John H. Cochrane, the Rose-Marie and Jack Anderson senior fellow at the Hoover Institution at Stanford University, who spent 30 years as a professor of finance at the University of Chicago Booth School of Business and writes The Grumpy Economist
🎙️ Host: Thibault Schrepel, associate professor of law at VU Amsterdam and founder of the Network Law Review, which publishes the show
📰 Published: 30 August 2026 on Scaling Theory
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 51 min | ✅ Time saved: 30 min
Key Takeaways
The cost of a light plane is what regulation looks like when it works as written
$80,000 in 1970s real dollars against $800,000 now, on engines designed in the 1950s
The openings that exist — homebuilts, ultralights — took decades to appear
A percentage point of growth is worth more than the whole climate debate
2% against 3.5% doubles the economy over 60 years; he puts the cost of climate change at 5 to 10% of output over 100
Fracking happened in the US because nobody had to be asked
Subsurface minerals belong to the property owner, so the default answer is yes
Cochrane says the US energy establishment opposed it as hard as Europe's did
Antitrust stopped being about consumers and became a way to make companies ask
His test for finance is the same one: find the market failure first, and check whether a regulation caused it
Regulation should have to prove itself after the fact, not before it
Cost-benefit analysis before a rule is both sides guessing; a five-year retrospective is checkable
Universities were not damaged by scaling — the scaling was the good part
The elite universities never grew; the state universities did, and matched smart people to productive jobs
Peer review's problem is not bias toward famous names, it is that nobody reads the paper
He has reviewed papers saying equation one is wrong and watched them publish anyway
AI is already better than a bad referee, and still worse than a good one
A colleague at Chicago destroyed three parts of his paper that Claude had not noticed
1. Growth Beats Distribution
Schrepel opened on an essay Cochrane published in Blueprint for America in 2016, which argues that growth swamps every other policy question, and asked why politics spends its energy on who gets what instead.
Cochrane's framing is that the last few centuries are the exception, not the norm. Lifespan, health and inequality itself are all dramatically better than they were for our grandparents
"This is the best year of all human history, right now, by a long shot."
The problem he has worked on his whole career is the slowdown, down to nearly zero in Europe and 2% in the US
He does not think artificial intelligence changes the question. On the argument that we are about to repeat the industrial revolution in three years, he said "And I go, we should be so lucky." There will be disruptions, but the growth question is unchanged
His answer on politics is definitional rather than cynical. "What government does is take from A and give to B." Regulation and taxation redistribute, to good ends or bad ones, and the system was designed that way
The psychological half is speculation and he labelled it as such. For thousands of years all wealth came from forcible redistribution — Roman emperors took theirs by conquest — so the instinct is old and the market economy is a thousand years old
"So no wonder most of the action is about envy and redistribution."
2. 2% vs 3.5% Over 60 Years
The host quoted Cochrane's own counterfactual back at him, and Cochrane thanked him for excusing him for forgetting his own papers.
The arithmetic is the argument. "Yeah, 2% versus three and a half% — the economy is twice as big in 60 years."
He set that against the number usually treated as the emergency. "The economic cost of climate change is five, maybe 10% of output in 100 years." That does not make climate change unimportant, he said; it makes growth orders of magnitude more important
The prescription is old and he does not claim otherwise. "I'm a reader and interpreter in growth theory, but we know the answer, and we've known the answer since Adam Smith: get the hell out of the way."
Growth comes from productivity embodied in new companies that put old ones out to pasture, which means people change jobs and some of them lose
His estimate is that 99% of what policy does is get in the way of that process
The rest of the list is unglamorous. Property rights, rule of law, rights rather than permissions, and escaping what he called "the vetocracy that we have over much public and private investment right now"
He objects to the phrase "policy makers" itself, because it implies benevolent technocrats guiding the economy. What exists instead, he said, is politicians who deliver protections to whoever demands them
He conceded the politics of his own position. Nobody, he said, ever won an election on "I'm going to get out of the way."
3. The Nuclear Own Goal
Cochrane's example of the cost of asking permission is nuclear power, and the host — who is French — pointed out that the US ban was good for France's competitiveness.
Cochrane agreed and then said France has forgotten how to do it. France had, in his words, "this beautiful technocratic regime" whose technocrats decided nuclear power was safe and went ahead
"And France has the greatest amount of carbon-free electricity on the planet as a result."
"Building new nuclear power plants is very difficult in France now — even maintaining the old ones is difficult, and for some reason you still don't have air conditioning."
About 10 years ago every member of the government wanted to shut nuclear power down, he said, and now they deny it: "Unfortunately, we have the recording."
The counterfactual he draws for the US is total. "Though when you look at the immense benefits of growth — for example, not allowing nuclear power for 40 years has been a disaster, and that's what our governments did."
Had the US allowed it, he said, there would be abundant cheap electricity and no climate crisis to argue about
The use he names for nearly free electricity is desalination, and growing things across the desert southwest
He put genetically modified foods and AI in the same category — technologies stopped or slowed by the requirement to ask first, with Europe having, in his reading, already made AI illegal
4. Function, Not Size
The host turned to a 2013 Journal of Economic Perspectives piece of Cochrane's — titled Finance, subtitled Function Matters, Not Size — and asked what a functional test would look like in practice.
Cochrane started with the antipathy rather than the economics. Hostility to middlemen and money lenders goes back thousands of years and is, from an economic point of view, completely wrong. The middlemen get things from A to B; if you had to drive to farms to buy food you would not do well
He drew the same line through antitrust. The original regime treated Standard Oil's size as the offence; 50 years of scholarship then established that what mattered was consumer prices, and Standard Oil was making things cheap
"there really is no such thing as a durable monopoly other than that which is enforced by government"
His examples are the companies that were supposed to have the internet locked up: Yahoo, AOL Time Warner and Netscape
His account of why size-based antitrust came back is about power, not doctrine. "If every big company needs to go to Washington and beg for permission to do a merger, whether you're asking permission from Biden or Trump, both the Bidens and the Trump absolutely love this system"
On fees, his answer is that people pay them voluntarily and can stop. He does not like active managers' fees and has read the work saying active management on average adds nothing, so his own money sits in near-zero-fee passive funds
"If you don't want to pay fees, you don't have to pay fees."
Trading commissions, which were a legal monopoly, fell when competition was allowed
The test he wants applied is the standard one. Identify the market failure — an externality, a public good, asymmetric information — then ask why the usual mechanisms do not fix it, and check whether a government rule is what broke it
His estimate: 90% of the time in an already heavily regulated industry, the regulation is the cause
5. Sunset Every Regulation
Schrepel described a paper of his own, still unpublished, proposing an institutional design that would force regulators to run experiments showing their intervention is likely to produce a better outcome.
Cochrane's endorsement came with the Chicago School history behind it. Economics teaches one week of the free market and five weeks of the ways regulation can fix its failures; George Stigler and James Buchanan went and looked at actual regulation and found something else
What the railroad and utility rules of the 1930s show, on his account, is firms and regulators agreeing that regulation stops competition, and politicians finding that approval power buys quiet and campaign contributions
The failure he wants fixed is that a rule is never checked against its own purpose. Rent controls, he said, make the problem worse and have been there for years
Cost-benefit analysis before the fact is theatre. In a legal proceeding both sides are making it up, because nobody knows what the costs and benefits will be
His own proposal is retrospective and mandatory. "But how about every five years you have to look at how this regulation actually did?" A sunset on every regulation, with the cost-benefit case made after the fact and the unintended consequences on the record
He was direct about why the free-market argument keeps losing. The remaining free-marketers, he said, are "an embittered small crowd" who respond to rent control by saying everybody knows it is dumb — and it is still there
6. Rights, Not Permissions
Part two of the conversation was the balance between regulation and permission, built on Cochrane's 2015 essay on the rule of law in the regulatory state, which describes agencies punishing through delay and discretion rather than through rules.
The whole distinction is which default applies. Either you must ask permission for everything, or you have the right to act. Property rights and limited government are what contain the regulatory state
His worked example is a neighbor's extension. If the neighbor has the right, a politically connected complainer cannot stop him; if the government can intervene, it usually will
The fracking comparison is the one he keeps returning to. "So why does the US have fracking and Europe doesn't?"
It did not happen because anyone approved it: "This happened over the dead bodies of the entire US energy establishment," a group he says was interchangeable with Europe's around 2010
"But in the US, subsurface minerals are the right of the property owner, and the default is you can't stop me."
"And in Europe, you have to ask permission, so you can just say no — end of story for fracking."
The economics underneath it is about who can trade. If building higher is worth a million dollars to the neighbor and costs $10 of shade on the tomato plants, a market settles it for $20
With no right to be paid, the only move left is the city council, where $10 of harm can block a million dollars of value
His conclusion is that a political process is very coarse at ranking what matters, especially over long horizons
7. The Wright Brothers Test
Schrepel put it to him that the traffic between the two camps runs one way: companies start out wanting permissionless innovation and, as they grow, discover the value of capturing the government.
Cochrane's objection to permission-first rules is that the regulator knows no more than anyone else. "Imagine if Orville and Wilbur Wright had to go to the FAA and ask permission before they fly an airplane."
The imagined response — a commission, a maintenance manual, the inspection frequency on the engine's O-rings — is how you never get anywhere
Safety came from operating, not from foresight. "Flight was dangerous, early flight was dangerous — how did we get to safe flight?" They flew, they crashed, people died, and manufacturers who wanted to stay in business learned alongside a regulator
That, he said, is a good model for AI regulation; requiring every possible danger to be foreseen in advance kills the innovation instead
On why the AI companies are in Washington, he offered the cynical reading and labelled it as one. The industry has spent billions on hardware and software that will be worthless in three years — "a three-year-old AI model is worth nothing, the Chinese are steaming ahead with free open-source models"
To make sense of a hundred-billion-dollar investment you need a government regulating competition and certifying new models, and you need it now
"That's a little bit cynical, but it's certainly not an unreasonable reading of what's going on."
8. The Skies Are Empty
The host's follow-up was whether the two-camp framing is too simple, given that aviation regulation is often held up as regulation that enabled an industry.
Cochrane took the example and turned it over. "Well, aviation's a good example of something that has nearly been killed by regulation, and has been killed in Europe."
"One thing you'll notice — I'm a pilot, so I look up at the sky a lot — the skies are empty in Europe."
The price series is the evidence. "A light plane used to cost about twice as much as a car — a light plane in the 1970s cost 80,000 real dollars, now it's 800,000."
The technology stalled with it. "And technical development of light planes has been completely stymied — we're still flying around in engines from the 1950s that are much less safe than car engines, because car engines have had much more technical development."
What innovation there has been came through gaps in the rules. Homebuilt aircraft were one; ultralights began with people bolting engines onto hang gliders. Both took decades
He would not hand the whole system to the private sector. The accumulation of experience after each crash is real, and expensive. Libertarians say a manufacturers' consortium could do it better, and he thinks they have their priorities wrong: pilots' licensing can wait behind the larger problems
9. Scale Was Not the Problem
Part three turned to universities, and whether the mechanism that let them scale is the same one that selected for conformity.
Cochrane's answer is that the conformity did not come from scale. Universities were about educating an elite until the early 20th century; the research university arrived from Germany in the 1920s and 1930s, and the elite university open to everyone is a concept from the 1950s and 60s
"But Harvard, MIT, Stanford, Yale — they never scaled, they have the same number of students they've always had."
What scaled was the state system, and he thinks it was good. There is quantitative work, he said, showing that a good part of growth came not from great ideas but from getting education to smart people who would otherwise have been stuck as bartenders and janitors
Opening the job market to Black people, to women and to gay people belongs in the same matching story
The dysfunction he does attribute to scale runs through the loan program. When the federal government took over student loans it could not bring itself to lend for engineering and not for a subject with no job at the end, so everybody got a loan
The result, on his account, is a generation with a degree, no qualification to do anything, and a belief that learning to fix pipes is beneath them
The larger problem he names is ideological capture, not size. The humanities and social sciences, with a partial exception for economics, have in his view been captured by a far-left ideology that keeps out dissent and is now passing into education, medical and law schools
"So it turned from knowledge to advocacy."
The reputational cost is the part he thinks is measurable: a Harvard degree in sociology used to signal intelligence, and in his telling it no longer does
10. The Minimum University
Asked what the minimum viable university looks like if the model is breaking, Cochrane declined to design one.
His answer was competition rather than a blueprint. Stop subsidizing the old models, and the nonprofit structure that he thinks defends universities from competition, and find out what emerges
He has heard the disruption argument before. The people saying AI makes universities pointless are the people who said it about online classes and, before that, about the printed textbook
Something about showing up appears to be load-bearing. "So there's something in human beings about this experience of going to a place — there's this social force of showing up to class every week that we seem to need as a pre-commitment for us to learn things."
"We still teach the same way Socrates taught 3,000 years ago, and we keep trying different things and it keeps not working."
He would not forecast the technology, only the obstacle. What exists is a system with a great deal of protection against change, and removing the protection is the direction he would push
11. Nobody Reads the Papers
Schrepel put his own thesis on the table: if AI lets top professors publish more, journals will take more of their papers because the citations are safer, and a promising unknown loses out. He named Daron Acemoglu as the example, with the caveat that he has no idea whether Acemoglu uses AI.
Cochrane rejected the premise on the evidence. "So first of all, Daron Acemoglu wasn't always Daron Acemoglu — he was a struggling nobody just like the rest of us."
"And there's been good studies of peer-reviewing journals — it is not true that big names have an easier time publishing; if anything they have a slightly harder time publishing, quality-corrected."
Readers do cite the big names, he said, but that is the readers' choice and not the review process
What he thinks is broken is the incentive. "What matters is where did you publish, not that anybody ever read your article or understood it." and "Articles are getting bloated and completely illegible."
His own reviewing experience is the proof he offers. He has written a report saying equation one is wrong and watched the paper publish anyway, because the other referees wanted appendices D, E and F on robustness and never noticed
This is where he thinks AI helps. He has spent three or four weeks on a theory paper without getting through the algebra: "So now that Refine and Claude are fantastic at finding algebra errors, it may improve the quality"
He also thinks the question itself carries the disease. Asking who gets to publish where and who gets famous, rather than who writes something that changes how we think, is what he called revealing of the rot
The editorial reform he is proud of has since been undone. At the Journal of Political Economy in the 1990s the rule was that the editor reads and understands the paper and referees only advise
Association journals with rotating editors, he said, have gone back to rubber-stamping the referees
Hiring is the last link in the chain. A tenure case that turns on where someone published outsources the judgement to an anonymous referee. At Chicago, he said, people read the papers and argued about whether the t-statistics in table 12 were calculated correctly — and that is very rare
12. Getting Into the Training
The host asked how a scholar should play a world in which most papers are read by AI systems rather than by people.
Cochrane's first answer was to write it yourself. He pointed to his own writing tips for PhD students: write vividly, keep the audience in mind, and remember that a reputation with humans is what counts
His view of machine prose is unsentimental. "AI prose is deadly boring, it's not a very good writer, it's an adequate writer, especially if you're not a native English speaker, but it's not a great writer."
He does expect the mediation to happen anyway, on a 10-to-15-year horizon, and has drawn the obvious conclusion: "Oh boy, I want to make sure that I'm in the training data."
What he has gathered is to keep everything on the web in a publicly visible format, and he says he does not know how to go further
His to-do list has one item on it: ask Tyler Cowen how
On refereeing, his conclusion is a floor rather than a replacement. An AI report is better than a report from a referee who was not paying attention, so it sets a minimum quality standard
Against that: he ran a paper through multiple rounds of Claude and Refine, and then a colleague at the University of Chicago wrote him a letter that destroyed three parts of it the machines had not noticed
"So really good humans are important."
"Overall what I want to get to is a world where people read papers."
13. Substack Against the AER
Cochrane moved to Substack at the end of 2023, and the host observed that he now reaches an audience an order of magnitude larger than any paper could.
It has not changed how he writes, because the deeply technical work still goes into journal articles
The trade-off he names has two axes. "There's a trade-off also between depth and breadth, and between durability and accessibility."
Some of his posts get remembered and passed on, but most do not last, in the way op-eds do not last
His example of the other end is Robert Lucas's 1972 paper, which lasts a very long time
The two feed each other. He thinks people read his academic work a little more because he also does popular work
The routing rule is explicit. "I don't send my pontifications about AI and growth to the American Economic Review, because it's not new research, it's the process of digesting."
Deep equation work gets advertised on the Substack but does not travel there
His defence of the digesting is the most quotable line in the section. People think there is new research and then nothing, he said, but "99% of normal science is this process of digesting, simplifying, and thinking."
Bonus Insights
The host's framing was the sharpest statement of the thesis in the episode. In his introduction Schrepel put it as: "Growth, put differently, economic scaling, is what happens when a large number of attempts are made and a few of them work." and "So, anything that raises the cost of an attempt scales into the aggregates."
Cochrane's list of what a government that minded the store would say was delivered as a single sentence of complaints: social programs that pay people not to work, a tax code that leaves too little for savings and investment, billions of dollars per mile to build a road or a train, and "We pay people to be homeless and we get lots of homeless people."
The growth-miracle roll call. Cochrane's evidence that a little getting out of the way goes a long way: "And anytime it's been tried to a greater or lesser extent — even Mao got a little bit out of the way, let a little bit of free market happen, kaboom, the greatest increase in human well-being ever seen is China going from about $200 a person per year to $20,000 a person per year." He added East Asia, Japan and Singapore, and said Singapore's GDP is now almost double that of the US
The construction analogy came from the host, not the guest. Schrepel said he had heard Nvidia's chief executive make the point that no construction worker today looks at a building and thinks it is heavy, because the machine does the lifting, and suggested AI is doing that to academic writing
The joke Cochrane expects to arrive. On AI at both ends of the publishing pipeline: someone turns a set of bullet points into a paper, and then, as he put it, "Oh look, AI turned this paper into bullet points."
The sign-off. Schrepel thanked him for a non-AI conversation and suggested that next time they could throw their two avatars together and let them speak
Cochrane's bottom line is that growth is the only policy question that compounds, that it is produced by attempts nobody had to approve in advance, and that the institutions now deciding which attempts are allowed — regulators, journals and universities alike — are the ones that need a sunset clause.
Products, Companies & Tools Mentioned
Hoover Institution and Chicago Booth (Where he is now and where he spent 30 years as a professor of finance)
Claude and Refine (The tools he says are "fantastic at finding algebra errors" — and that missed three things a colleague at Chicago caught)
Yahoo, AOL and Netscape (His standing example that a durable monopoly needs a government behind it: the companies that were supposed to have the internet locked up)
Standard Oil (The origin of size-based antitrust, and a company he says was making things cheap for consumers)
Nvidia (The host cited its chief executive on machines doing the heavy lifting, as an analogy for AI in academic writing)
Federal Aviation Administration (The agency he says has regulated light-plane safety to the point of nearly killing the business)
Substack (Where he moved at the end of 2023, and where he now reaches far more readers than a journal article does)
Journal of Political Economy and American Economic Review (The journal he edited in the 1990s, and the one he says will not take his digesting work)
Books & Resources Mentioned
Economic Growth – John H. Cochrane (His 2016 chapter in Hoover's Blueprint for America, and the essay the interview opens on)
Finance: Function Matters, Not Size – John H. Cochrane (The 2013 Journal of Economic Perspectives piece behind the whole antitrust and finance section)
Rule of Law in the Regulatory State – John H. Cochrane (The 2015 essay on agencies that punish through delay and discretion)
The Grumpy Economist – John H. Cochrane (The Substack he moved to at the end of 2023)
Writing tips for PhD students – John H. Cochrane (The piece he points readers to when he says the answer is to learn to write well yourself)
The Matthew Effect – Thibault Schrepel (The host's own short piece on peer review, and the prior he discloses before asking the AI question)
Robert Lucas, 1972 (Cochrane's example of work that lasts, against an op-ed that does not)
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: