Month: August 2024
Sometimes people overrate “marginal value” and underrate “average value”
When people criticize cars, you often hear it mentioned that many people die in road accidents, many more are injured, and so on. All that is true, dangerous driving does represent a negative externality on others, and it implies that roadway reforms should be undertaken.
Still, the point remains that, for those who choose to drive, the average return to driving is higher than the average return to whatever the alternative was. The network of roads and driving still is making those individuals better off.
If anything, noting the negative externality may imply that especially large improvements are possible for the driving experience.
You also can note that moving into a large city may benefit other urban residents, through density externalities. The opera scene will get better, and so on.
Still, the net American flow is toward the suburbs. That implies for most movers the average returns of suburban life are higher.
Many urbanists focus on various externalities from urban and suburban life, positive and negative accordingly. They are a bit loathe to admit where the average returns are highest for a growing number of Americans.
At the macro scale, perhaps we should be favoring the suburbs. If, in some far-off part of the state, you could open up a new city, or some new suburbs, the suburban option might boost welfare more. Even if there are positive net externalities to urban density, and some negative externalitlies from suburban life.
Yes, the marginal revolution blah blah blah. But do not forget the averages!
The Less-Efficient Market Hypothesis
Market efficiency is a central issue in asset pricing and investment management, but while the level of efficiency is often debated, changes in that level are relatively absent from the discussion. I argue that over the past 30+ years markets have become less informationally efficient in the relative pricing of common stocks, particularly over medium horizons. I offer three hypotheses for why this has occurred, arguing that technologies such as social media are likely the biggest culprit. Looking ahead, investors willing to take the other side of these inefficiencies should rationally be rewarded with higher expected returns, but also greater risks. I conclude with some ideas to make rational, diversifying strategies easier to stick with amid a less-efficient market.
Saturday assorted links
2. Alexa will be powered by Claude. And ChatGPT goes to church?
3. In praise of reading reference books.
4. AI analyzes interviewing styles, in this case with CWT with Nate Silver.
5. Five productive years from Stephen Wolfram.
6. The microfoundations of Long Covid.
7. New documentary on home schooling.
8. The new Meta glasses might be called “Puffin.” (The Information)
9. But what do the AIs think of you? (NYT)
The Daylight Computer
I am pleased and also honored to have been sent an advance copy of The Daylight Computer.
It performs functions similar to those of an iPad and a Kindle, but with improvements.
My first surprise is that I proved capable of operating the thing. It requires no expertise above and beyond what you need to use your current devices, arguably less.
Here is the review of Dwarkesh, and here is the review of @patio11. Both are consistent with my impressions, but Patrick McKenzie’s uses are closer to mine. I’ve been looking for a Kindle improvement for a long time, and this is it. Kindle Fire was not.
This seems to be the best general reading device humans ever have invented. Compared to a Kindle, the page is much larger, the color choice is excellent, scrolling is easy, it is far better for showing maps, and it captures far more of “does this feel like reading a book?” impression than a Kindle ever did. It also can handle all sorts of glare and sunlight issues.
It can connect to a wireless system more easily and effectively than a Kindle — ever have that problem in your hotel room? The hotel makes you fill in extra fields, and the Kindle interface is not well suited for that.
The Daylight Computer just seems very generally well thought out.
I am also told that an AI function will make it possible to query reading passages at will, and easily, yet without leaving your reading window. This is not yet up and running on my demo version, but it will be a major advance.
So I will continue to use this device and also will travel with it.
There is some other set of associated benefits, something about being able to use some iPad-like functions, but without the full distractions of the internet (see the Dwarkesh review). That is not relevant for my own planned consumption habits, but it may be a significant benefit to many.
You can pre-order yours here.
Neo, and the Norwegian Century
Neo, and the Norwegian Century? (It’s happening) And more here. And yet more.
It’s not just Magnus Carlsen, people!
Friday assorted links
1. NYC is starting a crackdown to ensure payment of bus fares (NYT).
3. 100 million token context window?
4. Egypt sending troops to Somalia?
5. “Japan plans cash incentive for women to marry and leave Tokyo.” I think they will find it hard to achieve the desired ends here.
6. Does America have enough states to support statistical inference?
8. Brazil’s military relies on Starlink for operations and security. Solve for the equilibrium. And the very latest development.
9. Italy considering a tourist tax on expensive hotel rooms (FT).
10. For the first time, more than half the entering class at Caltech will be women.
Capitalism debate at MIT
September 16, 7-8.30, 2-190: “Is Capitalism Defensible?”, with Tyler Cowen (GMU Economics), and Alexander Gourevitch (Brown Political Science). Please register here.
Here is the link.
Mpox Vaccines Stuck in Limbo: WHO is at Fault
In 2022, Mpox, a viral disease endemic to parts of Africa and primarily transmitted through close contact—especially sexual contact between men—spread to developed countries, including the United States. The U.S. saw over 30,000 cases and approximately 58 deaths. Despite two available vaccines there was not nearly enough supply to vaccinate even the high-risk populations. Fortunately, health authorities adopted vaccination strategies my colleagues and I had recommended for COVID such as first doses first and fractional dosing. For example, several small studies (e.g. here and here) suggested that 1/5 doses delivered intradermally could be effective and the FDA, EMA, and the UK all recommended this fractional dosing strategy. As result, the US was able to vaccinate around 800,000 people and the epidemic ended (natural immunity and other preventive measures also played a role).
Unfortunately, a new Mpox variant is now spreading in the Democratic Republic of the Congo and nearby countries. Here’s the crazy part: despite declaring Mpox a public health emergency on August 14, the WHO has not approved any Mpox vaccines. You might think, “Who cares what the WHO authorizes?” After all, the FDA, EMA, and the UK have all granted emergency approval. But here’s the catch: the WHO’s approval is crucial for GAVI, the vaccine alliance that donates vaccines to developing countries. Without WHO approval, GAVI is reluctant to provide vaccines to the Congo. To add insult to injury, the Congo itself has approved the Jynneos and LC16 vaccines. Yet, the WHO refuses to authorize and GAVI to donate these vaccines, citing vague concerns about safety and efficacy.
Stephanie Nolen at the NYTimes has a very good piece on this mess:
Three years after the last worldwide mpox outbreak, the W.H.O. still has neither officially approved the vaccines — although the United States and Europe have — nor has it issued an emergency use license that would speed access.
One of these two approvals is necessary for UNICEF and Gavi, the organization that helps facilitate immunizations in developing nations, to buy and distribute mpox vaccines in low-income countries like Congo.
While high-income nations rely on their own drug regulators, such as the Food and Drug Administration in the United States, many low- and middle-income countries depend on the W.H.O. to judge what vaccines and treatments are safe and effective, a process called prequalification.
But the organization is painfully risk-averse, concerned with a need to protect its trustworthiness and ill-prepared to act swiftly in emergencies, said Blair Hanewall…
In addition, no one has followed the other practice my colleagues and I recommended for COVID (which Operation Warp Speed did), namely advance market commitments. So the vaccine manufacturers have basically been twiddling their thumbs and not gearing up for greater production. (The Congo can also be faulted for not buying more on their own account.)
All of this means that when the WHO does authorize and the vaccines begin to flow, we will still desperately need strategies like fractional dosing.
Hat tip: Ben H. and special thanks to Witold Wiecek.
It is wonderful to put inefficient firms out of business
The differences between the most and least productive companies can be startlingly high. By one estimate, in the US alone the most productive firms in a sector can be more than two to four times more cost-effective than the least productive ones. Given the size of those discrepancies, any expansion of trade or innovation that makes it possible to replace less efficient producers could help a sector economize a significant part of its production costs.
That is from my latest Bloomberg column. And this:
There is a converse worry about efficiency changing an economy too quickly — such as the current panic in some quarters over the speed of change that AI might bring.
What I see, however, is that a lot of institutions are unwilling or unable to adopt new, AI-intensive methods of doing business. Another year or two of prodding probably will not change that reality. Then, as AI becomes more important as a competitive edge, firms that do not deploy AI effectively will go out of business. This process could take 10 years or more, coming in fits and spurts, as has happened with most previous major technological innovations.
The column also has other points of interest.
John Arnold on economic polarization
As divisive as the political rhetoric is, the policy divide between the two parties seems more narrow today than any time in recent memory. Bipartisan bills in immigration, energy permitting, and the child tax credit have been negotiated and waiting for political window to reopen. Foreign aid and military spending bills both passed in past year with strong bipartisan support. Same with infrastructure bill in 2021 and CHIPS Act in 2022. Both parties are anti-China, favorable to India, and increasingly supportive of industrial policy and tariffs. Both talk about lowering the cost of housing, more funding for the police, and are leery of big tech. Neither party is proposing big changes in health care, K-12 or social security. College loan forgiveness is in the courts. Abortion is now in the states. GOP opposition to the IRA climate provisions are around the edges, like EV subsidies. Dems aren’t proposing any new significant climate policies. Dems have enacted minor policies against oil and gas but production continues to reach record highs. Both say no new taxes for <$400k. Increasing number of Rs have joined Ds supporting increase in corp tax rates. Perhaps the biggest difference is how to pay for TCJA extension: Dems want higher taxes on the wealthy; Trump wants universal tariffs; the rest of GOP hasn’t been specific. There are other differences for sure. Dems want more subsidies for housing and child care. GOP wants more deportations (though logistically difficult). Dems would be more aggressive against consolidation, health care costs, and junk fees. GOP wants to restucture civil service rules. But there just aren’t many major fault lines on policy between the parties today. Maybe this is why so little of this election cycle is about policy.
Here is the full tweet. I would add that “ten percent tariffs” vs. “25% tax on unrealized capital gains” is a big difference, but at least one of those is never going to happen, even if the Republicans do not capture the Senate.
Jon Haidt on causality (from my email)
“Hi Tyler,
i have big news about the debate over social media harming teens.
So much of it hangs on the claim that the evidence is just correlational, not causal.
Zach Rausch and I show that this is not true; the experiments DO show causation, very clearly and consistently.
Here are my 2 tweets about the post:
https://x.com/JonHaidt/status/1829163166066205168
https://x.com/JonHaidt/status/1829165292460859869
A lot of people heard our discussion, and enjoyed how spirited and yet civil it was.
Might you include the link to this post in your daily email:
https://www.afterbabel.com/p/the-case-for-causality-part-1
We have 3 more coming. We think we can prove causality using just the existing experiments.
thanks for considering it.
jon”
TC again: I received this email this morning, and told Jon I would post it on MR without response from me, so here it is.
Thursday assorted links
1. How AI will impact back office support in the Philippines (Bloomberg).
2. Toward a theory of (Finnish) woke. Not directly related to sauna activity.
3. Immigrant cooks as anthropologists.
4. How NYC’s Park Avenue used to look, and maybe will once again.
5. A left market anarchist canon.
6. David Pilling on Benin and T-shifts (FT). The country continues to grow in the six percent range.
7. In many states, having a state-appointed guardian means you cannot vote (NYT).
8. Andreas Backhaus on the recent Jim Crow paper.
9. Just change the step-up basis upon death. The actual proposal is perhaps the worst economic policy idea I have heard since the fall of communism.
Failing Banks
From Sergio Correia, Stephan Luck, and Emil Verner:
Why do banks fail? We create a panel covering most commercial banks from 1865 through 2023 to study the history of failing banks in the United States. Failing banks are characterized by rising asset losses, deteriorating solvency, and an increasing reliance on expensive non-core funding. Commonalities across failing banks imply that failures are highly predictable using simple accounting metrics from publicly available financial statements. Predictability is high even in the absence of deposit insurance, when depositor runs were common. Bank-level fundamentals also forecast aggregate waves of bank failures during systemic banking crises. Altogether, our evidence suggests that the ultimate cause of bank failures and banking crises is almost always and everywhere a deterioration of bank fundamentals. Bank runs can be rejected as a plausible cause of failure for most failures in the history of the U.S. and are most commonly a consequence of imminent failure. Depositors tend to be slow to react to an increased risk of bank failure, even in the absence of deposit insurance.
Theory of bank runs: overrated.
Do consumers hate on-line ads?
Not so much it seems:
Research on the causal effects of online advertising on consumer welfare is limited due to challenges in running large-scale field experiments and tracking effects over extended periods. We analyze a long-running field experiment of online advertising in which a random 0.5% subset of all users are assigned to a group that does not ever see ever ads. We recruit a representative sample of Facebook users in the ads and no-ads groups and estimate their welfare gains from using Facebook using a series of incentive-compatible choice experiments. We find no significant differences in welfare gains from Facebook. Our estimates are relatively precisely estimated reflecting our large sample size (53,166 participants). Specifically, the minimum detectable difference in median valuations at standard thresholds is $3.18/month compared to a baseline valuation of $31.95/month for giving up access to Facebook. That is, we can reject the hypothesis that the median disutility from advertising exceeds 10% of the median baseline valuation. Our findings suggest that either the disutility of ads for consumers is relatively small, or that there are offsetting benefits, such as helping consumers find products and services of interest.
That is from a new NBER working paper by
The wisdom of Gwern, why should you write?
Much of the value of writing done recently or now is simply to get stuff into LLMs. I would, in fact, pay money to ensure Gwern.net is in training corpuses, and I upload source code to Github, heavy with documentation, rationale, and examples, in order to make LLMs more customized to my use-cases. For the trifling cost of some writing, all the worlds’ LLM providers are competing to make their LLMs ever more like, and useful to, me.
And that’s just today! Who knows how important it will be to be represented in the initial seed training datasets…? Especially as they bootstrap with synthetic data & self-generated worlds & AI civilizations, and your text can change the trajectory at the start. When you write online under stable nyms, you may be literally “writing yourself into the future”. (For example, apparently, aside from LLMs being able to identify my anonymous comments or imitate my writing style, there is a “Gwern” mentor persona in current LLMs which is often summoned when discussion goes meta or the LLMs become situated as LLMs, which Janus traces to my early GPT-3 writings and sympathetic qualitative descriptions of LLM outputs, where I was one of the only people genuinely asking “what is it like to be a LLM?” and thinking about the consequences of eg. seeing in BPEs. On the flip side, you have Sydney/Roose as an example of what careless writing can do now.) Humans don’t seem to be too complex, but you can’t squeeze blood from a stone… (“Beta uploading” is such an ugly phrase; I prefer “apotheosis”.)
This is one of my beliefs: there has never been a more vital hinge-y time to write, it’s just that the threats are upfront and the payoff delayed, and so short-sighted or risk-averse people are increasingly opting-out and going dark.
If you write, you should think about what you are writing, and ask yourself, “is this useful for an LLM to learn?” and “if I knew for sure that a LLM could write or do this thing in 4 years, would I still be doing it now?”
Here is the link, or try this link. Of course not many people have the actual purpose of mind to believe such a thing. But a few do.