What should I do in Cape Town, South Africa?
I won’t be there for long, but what should I see and what should I eat? How is the general level of safety these days?
I thank you all in advance.
Financial Statement Analysis with Large Language Models
We investigate whether an LLM can successfully perform financial statement analysis in a way similar to a professional human analyst. We provide standardized and anonymous financial statements to GPT4 and instruct the model to analyze them to determine the direction of future earnings. Even without any narrative or industry-specific information, the LLM outperforms financial analysts in its ability to predict earnings changes. The LLM exhibits a relative advantage over human analysts in situations when the analysts tend to struggle. Furthermore, we find that the prediction accuracy of the LLM is on par with the performance of a narrowly trained state-of-the-art ML model. LLM prediction does not stem from its training memory. Instead, we find that the LLM generates useful narrative insights about a company’s future performance. Lastly, our trading strategies based on GPT’s predictions yield a higher Sharpe ratio and alphas than strategies based on other models. Taken together, our results suggest that LLMs may take a central role in decision-making.
That is from a new paper by Alex Kim, Maximilian Muhn, and Valeri V. Nikolaev, all at Chicago Booth. Via William Allen.
*Best Things First*
The author is Bjorn Lomborg, and the subtitle is The 12 most efficient solutions for the world’s poorest and our global SDG promises. I missed this book when it first came out last year. Here is what Lomborg presents as the twelve best global investments, in no particular order:
Tuberculosis
Maternal and newborn health
Malaria
Nutrition
Chronic diseases
Childhood immunization
Education
Agricultural R&D
e-procurement
Land tenure security
Trade
Skilled migration
Monday assorted links
1. “Our work suggests the intriguing possibility that non-equilibrium order can arise more easily than assumed, even before that order is directly functional, with consequences impacting mutation rate evolution and kinetic traps in self-assembly to the origin of life.” Link here.
2. Kabosu, who inspired the dogecoin meme, has passed away.
3. 52-minute video on birding in northern Colombia.
4. Old but renovated Singapore shophouses are now some of the most expensive properties in the world (FT).
A theory of information, by Noah Smith
So here you go: a theory of how totalitarianism might naturally triumph. The basic idea is that when information is costly, liberal democracy wins because it gathers more and better information than closed societies, but when information is cheap, negative-sum information tournaments sap an increasingly large portion of a liberal society’s resources. Remember that I don’t believe this theory; I’m merely trying to formulate it.
Here is the full blog post, gated after some point.
David Friedman on his father Milton
When my parents got married, they decided that there were certain things that were difficult to say and should therefore be replaced by numbers. Only one survived in actual usage. In their family “number two” meant, in my family still means, “You were right and I was wrong.”
One reason is that it is shorter, so easier to say. A second reason is that using the number reminds speaker and audience that admitting error is a difficult and virtuous thing to do, which makes it easier to do it. A third reason is that using a family code reminds the speaker that he is speaking to people who love him, so are unlikely to take advantage of the confession of error to put him down.
My father used to be fond of the phrase “There is no such thing as a free lunch,” sometimes abbreviated TANSTAAFL. He eventually stopped using it on the grounds that it was not true, that both consumer and producer surplus are, in effect, free lunches. He replaced it with “Always look a gift horse in the mouth.”
Phrases he continued to use included “A bad carpenter blames his tools,” “It is a capital mistake to make the best the enemy of the good” and Cromwell’s “I beseech you, in the bowels of Christ, think it possible you may be mistaken.” He referred to my carrying too many logs in from the woodshed to the fireplace in order to do it in fewer trips as a lazy man’s load.
Here is the full Substack post.
What does it look like to build more housing?
…we examine how housing provision has evolved for the largest four metropolitan statistical areas (MSAs) in California and Texas. Despite differences in their topographies and regulatory environments, we find several common dynamics. As these MSAs grow, we see that fewer new net units are built at the periphery and a smaller share of the new units are built as single-family detached houses. As a greater share of new net units are built in infill locations, more units are built using higher-density—and more costly—multifamily housing construction techniques. Interestingly, we see these housing supply patterns in both “pro-growth” MSAs and “highly regulated” MSAs. Among all of our sample MSAs, we also find a declining share of Census tracts that participate in accommodating growth. Our results are consistent with the existence of a convex housing supply curve. We believe that this secular trend will pose genuine challenges to many urban housing policies aimed at improving affordability.
That is from a newly published paper by Anthony W. Orlando and Christian L. Redfearn. Via the excellent Kevin Lewis.
Sunday assorted links
Hedging Star Wars and Close Encounters
An old story but new to me. Lucas and Spielberg swapped 2.5% net points on Star Wars and Close Encounters. Pretty smart bet for both of them.
Spielberg tells the story:
George came back from Star Wars a nervous wreck. He didn’t feel Star Wars came up to the vision he initially had. He felt he had just made this little kids’ movie. He came to Mobile, Alabama where I was shooting Close Encounters on this humongous set and hung out with me for a couple of days.
He said, ‘Oh my God, your movie is going to be so much more successful than Star Wars. This is gonna be the biggest hit of all time.’ He said, ‘You want to trade some points? I’ll tell you what, I’ll give you two and a half per cent of Star Wars if you give me two and a half per cent of Close Encounters.’
I said, Sure, I’ll gamble with that, great. And I think I came out on top of that bet. I did a lot better than George!
Both movies were wildly profitable. Close Encounters made so much money and rescued Columbia from bankruptcy. It was the most money I ever made on a movie before, but Close Encounters was a meagre success story. Star Wars was a phenomenon and I was the happy beneficiary of a couple of net points from that movie which I am still seeing money on today!
Blame California?
This paper reexamines the role of social policy in the doubling of divorce rates. We demonstrate that the short-run rise in divorce rates formerly attributed to unilateral divorce solely depends on the state of California. California receives considerable weight in national analyses and adopted several policies simultaneously. When we examine the independent effects of these social policies, we find that legal abortion leads to a clear and immediate rise in divorce rates. However, legal abortion’s impact also hinges on California and may be contaminated by concurrently adopted policies. We then demonstrate that California’s influence extends to the broader unilateral divorce literature. We conclude by describing best practices to confront the challenges of simultaneous policy adoption.
That is from a recent paper by Lauren Hoehn-Velasco, Jacob Penglase, Michael Pesko, and Hasan Shahid. Via tekl.
What Went Wrong with Federal Student Loans?
At a time when the returns to college and graduate school are at historic highs, why do so many students struggle with their student loans? The increase in aggregate student debt and the struggles of today’s student loan borrowers can be traced to changes in federal policies intended to broaden access to federal aid and educational opportunities, and which increased enrollment and borrowing in higher-risk circumstances. Starting in the late 1990s, policymakers weakened regulations that had constrained institutions from enrolling aid-dependent students. This led to rising enrollment of relatively disadvantaged students, but primarily at poor-performing, low-value institutions whose students systematically failed to complete a degree, struggled to repay their loans, defaulted at high rates, and foundered in the job market. As these new borrowers experienced similarly poor outcomes, their loans piled up, loan performance deteriorated, and with it the finances of the federal program. The crisis illustrates the important role that educational institutions play in access to postsecondary education and student outcomes, and difficulty of using broadly-available loans to subsidize investments in education when there is so much heterogeneity in outcomes across institutions and programs and in the ability to repay of students.
That is from a new NBER working paper by Adam Looney and Constantine Yannellis.
Okie-dokie, canonized teen blogger edition
An early-aughts blog is probably not where you’d expect to find the next Mother Teresa, but that is where Carlo Acutis — soon to become the first millennial saint in the Catholic Church — made a name for himself documenting miracles.
The Holy See said Thursday that Pope Francis has recognized a second miracle linked to Acutis, paving the way for his canonization — the final step in a process that can sometimes take decades. It will place the online evangelizer — who died in 2006 of leukemia at age 15 — among thousands of saints recognized by the church…
Born in London in 1991, Acutis has drawn a following for his piety and meticulous research on miracles, which he publicized online. One Catholic publication dubbed him “God’s Influencer,” while another site described him as a teen with a “strong faith and a weakness for Nutella.” Vatican News wrote that he loved soccer, video games and was a “natural jokester.”
Here is the full story.
Saturday assorted links
The decline in Native American wealth
I had not realized how negative were the effects of the 1887 Dawes Act, which broke up many Native American reservations. Before 1912:
There was a nontrivial number of relatively wealthy superintendencies, which runs counter to the common perception of uniform poverty during this period. In 1912, the wealthiest superintendency had total per capital wealth levels above $600,000 in 2019 real terms, while total per capital wealth was just $90 in the least wealthy superintendency…
Our results suggest that, on average, Indigenous Peoples in the early twentieth century had substantial levels of wealth per capita, although there was wide diversity in wealth levels. Between 1912 and 1927, wealth per capita declined by nearly 50 percent.
Per capita indigenous wealth had been above white wealth at the beginning of the period.
Here is the AER version of the piece, by Donn. L. Feir, Maggie E.C. Jones, and Angela Redish, ungated here.
How carbon-emitting is AI?
Our findings reveal that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers, while AI illustration systems emit between 310 and 2900 times less CO2e per image than their human counterparts.
That is from a recent Nature paper by Tomlinson, Black, Patterson, and Torrance on the carbon-friendliness of AI. On one hand this is reassuring, though of course it should be noted that the carbon-emitting potential of AI largely comes from the additional economic activity (and querying activity) it may enable. So perhaps we can think of this as another example of Jevon’s Law, namely that the energy-saving technology in the longer run also increases the demand for energy, thus taking back some or all of the initially earned conservation.
Via Jim Pethokoukis.