Some simple economics of AI and macro cycles

by  Tyler Cowen in

Has AI been propping up the American economy?  For instance “the Bureau of Economic Analysis’s category for investment in information processing equipment and software accounts for over 90 percent of economic growth in the first half of 2025.”

The key question is what would have been done with those resources otherwise.  Regardless of their specific allocation, it is reasonable to assume they would have been allocated with considerable less urgency than the AI resources.  That means more resources sitting around, with their owners exercising a bit more option value.  It is probably also the case that the alternate allocations would have, on the whole, been less risky and less correlated than the AI allocations.  They would have been bets on a variety of different technologies, rather than a single technology.  After all, what else on the table could have been as “big” as these AI bets?

So without the AI boom yes we would have had a lower gdp growth number, but by no means do those resources just disappear.  We also would have lower expected returns from the alternate resource allocations and lower risk.

Since these resource allocations seem large enough “to matter” and to create systemic risk, the American economy would have had lower returns and also lower risk.  But we, collectively, opted for the scenario with higher expected returns and higher risk.  (Do not think you had no role in this!  You could have bid ridiculously high dollars to have the economy experiment with some new breakfast cereals instead.)

This is risk-based business cycle theory, people, much of it derived originally from Fischer Black.  I wish us luck people!

The Economic Geography of American Slavery

by  Tyler Cowen in

What would the antebellum American economy have looked like without slavery? Using new micro-data on the U.S. economy in 1860, we document that where free and enslaved workers live and how much they earn correlates strongly—but differently—with geographic proxies for agricultural productivity, disease, and ease of slave escape. To explain these patterns, we build a quantitative spatial model of slavery, where slaveholders coerce enslaved workers into supplying more labor, capture the proceeds of their labor, and assign them to sectors and occupations that maximize owner profits rather than worker welfare. Combining theory and data, we then quantify how dismantling the institution of slavery affected the spatial economy. We find that the economic impacts of emancipation are substantial, generating welfare gains for the enslaved of roughly 1,200%, while reducing welfare of free workers by 0.7% and eliminating slaveholder profit. Aggregate GDP rises by 9.1%, with a contraction in agricultural productivity counteracted by an expansion in manufacturing and services driven by an exodus of formerly enslaved workers out of agriculture and into the U.S. North.

That is from a new NBER working paper by Treb Allen, Winston Chen & Suresh Naidu.

Friday assorted links

by  Tyler Cowen

1. More on US pedestrian deaths.

2. Helen Andrews on the Great Feminization.

3. How the swap with Argentina will work.

4. Adda in NYC is a very good Indian restaurant.  And SRV in Boston is very good, get the Venetian dishes such as the pork and beef meatball and then the crab pasta with chile and breadcrumb.

5. Koyama on Mokyr.

6. The striking liberalism on the chatbots.  Are they the number one force for liberalism in the world today?

7. We are living through a literary golden age.

8. Kevin Bryan on the Laureates.

Predicting Job Loss?

by  Alex Tabarrok in

Hardly a day goes by without a new prediction of job growth or destruction from AI and other new technologies. Predicting job growth is a growing industry. But how good are these predictions? For 80 years the US Bureau of Labor Statistics has forecasted job growth by occupation in its Occupational Outlook series. The forecasts were generally quite sophisticated albeit often not quantitative.

In 1974, for example, the BLS said one downward force for truck drivers was that “[T]he trend to large shopping centers rather than many small stores will reduce the number of deliveries required.” In 1963, however, they weren’t quite so accurate about about pilots writing “Over the longer run, the rate of airline employment growth is likely to slow down because the introduction of a supersonic transport plane will enable the airlines to fly more traffic without corresponding expansion in the number of airline planes and workers…”. Sad!

In a new paper, Maxim Massenkoff collects all this data and makes it quantifiable with LLM assistance. What he finds is that the Occupational Outlook performed reasonably well, occupations that were forecast to grow strongly did grow significantly more than those forecast to grow slowly or decline. But was there alpha? A little but not much.

…these predictions were not that much better than a naive forecast based only on growth over the previous decade. One implication is that, in general, jobs go away slowly: over decades rather than years. Historically, job seekers have been able to get a good sense of the future growth of a job by looking at what’s been growing in the past.

If past predictions were only marginally better than simple extrapolations it’s hard to believe that future predictions will perform much better. At least, that is my prediction.

Those new service sector jobs

by  Tyler Cowen in

Yes — as of late 2025, several robotics and AI startups are literally paying people to fold their laundry (or perform similar chores) while recording themselves, in order to train robots in dexterous, human-like task performance.

Companies such as Encord, Micro1, and Scale AI have launched paid “data collection” programs aimed at generating real-world video datasets for robotic learning. Participants are compensated to film themselves carrying out everyday household activities — folding laundry, loading dishwashers, making coffee, or tidying up. The footage is then annotated to help AI systems learn how to manipulate deformable objects, coordinate finger movements, and complete multi-step domestic tasks.

That is from Perplexity, original cite from Samir Varma.

My excellent Conversation with George Selgin

by  Tyler Cowen in

Here is the audio, video, and transcript.  Here is part of the episode summary:

Tyler and George discuss the surprising lack of fiscal and monetary stimulus in the New Deal, whether revaluing gold was really the best path to economic reflation, how much Glass-Steagall and other individual parts of the New Deal mattered, Keynes’ “very sound” advice to Roosevelt, why Hayek’s analysis fell short, whether America would’ve done better with a more concentrated banking sector, how well the quantity theory of money holds up, his vision for a “night watchman” Fed, how many countries should dollarize, whether stablecoins should be allowed to pay interest, his stake in a fractional-reserve Andalusian donkey ownership scheme, why his Spanish vocabulary is particularly strong on plumbing, his ambivalence about the eurozone, what really got America out of the Great Depression, and more.

Excerpt:

COWEN: But once we revalue gold, as you know, starting in 1933, you have manufacturing-output growth rates of 7 percent to 8 percent until we screw it up later on with some disinflationary pressures. How much better could we have done? Wasn’t that a pretty good performance?

SELGIN: It was pretty good, but it didn’t last very long. In fact, the New Dealers knew that it wouldn’t last very long. There are a couple of reasons why.

First of all, there was a big burst of output that was connected to the expectation that the NRA, the National Recovery Administration, was going to be coming into effect, because it was one of the early New Deal measures. It was going to artificially raise prices through controls. There was a boom that was based only on manufacturers’ desires to jump the gun and buy inputs and produce inventory before their own costs went up. That was part of the story.

Of course, when you’re coming out from the deepest depths of a depression with a banking crisis and all that, you would expect rather rapid growth to follow from the stabilization of the banking situation itself. I don’t want to deny that there was genuine progress during those early months of the New Deal, and I don’t want to deny that the New Dealers deserve credit for much of it, but it didn’t last. Of course, we all know it didn’t last.

Beyond that first period, once the NRA and associated programs for price controls kicked in, things started to slow down very rapidly. What kept the progress going after that — though at a slower rate — was mostly gold starting to rush in from Europe. It was rushing in only initially because of devaluation. After that, it was mostly rushing in because of fears of the consequences of Hitler coming to power and the possibility of war breaking out.

That’s the story of the early phase of the New Deal: a good start that didn’t last that long, except as a result of help from abroad that was quite unintentional help.

COWEN: Was revaluing the gold price the best way of reflating the economy? Because there were many proposals at the time. You shut down the domestic gold market as well. Could it have been done better?

SELGIN: Yes, it could have been done better. I think that what should have happened was immediate devaluation of the dollar. It was clear by the time Roosevelt took office, the gold standard, as it had been, had to be at least suspended because the New York banks had run out of gold essentially. That was not something there was much choice about.

Then the question was, “Okay, what are we going to do going forward?” As I said, what I think they should have done was to just plan on a devaluation of the dollar, get it over with as quickly as possible. You don’t announce that plan before you’ve suspended gold payments because that’s just going to make the run on gold worse. Once you’ve suspended, then you can go ahead and proceed with the devaluation.

What Roosevelt did was to engage in this crazy gold purchase program for quite a few months, based on a harebrained theory by a fellow named George Warren, who was very influential. They toyed with the price of gold. The theory was that if you raise the price of gold, other prices will start going up. Didn’t happen. Eventually, after many months, general prices had hardly risen at all.

Finally, Roosevelt picked a value for the dollar, a proper devaluation. Confirmed it, put it into effect, and at that point, things started to improve. That’s what should have happened.

By the way, this is as good a time as any to mention, this is what Keynes would have recommended and did recommend. He scolded or criticized Roosevelt for following Warren’s theories instead. I think that on this and many other scores, Keynes’s advice about dealing with the Depression was actually very sound. The myth is that Roosevelt was following it when, in fact, most of the time, he wasn’t.

Recommended, informative throughout.  I am happy to recommend George’s new and excellent book False Dawn: The New Deal and the Promise of Recovery, 1933-1947.  Plus George now owns a rather large number of donkeys…

John Nye on Joel Mokyr (from my email)

by  Tyler Cowen in

I became an economic historian because I thought then and still believe economic history is more scientific than most of economics. It is/was not afraid to combine rigorous analysis with softer, more verbal theory and analytic narrative.
The Nobel underestimates how important it was that Joel who first promoted the distinction between Smithian growth and Schumperterian growth.
That is, Smithian growth is that growth which emerges from trade and well functioning markets. But societies that experienced Smithian growth usually did not overcome the Malthusian trap enough to provide sustained per capita income growth.
Schumpeterian growth was that based on innovation, which is a combination of creative invention AND the ability to successfully market a viable commercial product that changes the industry. The two do not, and prior to the 18th century only rarely came together.
Furthermore, he later added the idea of knowledge noting that inventing things — such as the imperial Chinese –without a deep understanding of the underlying theory behind the invention’s success limited the extent to which that idea could propagate.
As I tell my students, even when the US was still a minor power (such as the early 19th century) it was already a major innovator with items like the cotton gin, the sewing machine and the American system of manufactures. This idea of systematically making products with interchangeable parts took decades or more to become a true reality, but the Americans were the first to take this seriously at a high level.
Joel understood all this and our discussions on this subject have continued to this day.
Note, that because of the nature of his work, Joel did not get a top 5 pub till the late 2010s. By today’s standards, he would not have gotten tenure at most strong and second tier departments in the world. This shows the limits of the current. The old system where top schools balanced judgment against publication record — which GMU strives to promote — was a better source of truly innovative talent than the mechanical formulas promoted in most of worldwide academia.

AI and the First Amendment

by  Tyler Cowen in

The more that outputs come from generative AI, the more the “free speech” treatment of AIs will matter, as I argue in my latest column for The Free Press.  Here is one excerpt, quite separate from some of my other points:

Another problem is that many current bills, including one already passed in California, require online platforms to disclose which of their content is AI-generated, in the interest of transparency. That mandate has some good features, and in the short run it may be necessary to ease people’s fears about AI. But I am nervous about its longer-run implications.

Let’s say that most content evolves to be jointly produced by humans and AI, and not always in a way where all the lines are clear (GPT-5 did proofread this column, to look for stylistic errors, and check for possible improvements). Does all joint work have to be reported as such? If not, does a single human tweak to AI-generated material mean that no reporting is required?

And if joint work does have to be reported as joint, won’t that level of requirement inevitably soon apply to all output? Who will determine if users accurately report their role in the production of output? And do they have to keep records about this for years? The easier it becomes for individual users to use AI to edit output, the less it will suffice to impose a single, supposedly unambiguous reporting mandate on the AI provider.

I am not comfortable with the notion that the government has the legal right to probe the origin of a work that comes out under your name. In addition to their impracticality, such laws could become yet another vehicle for targeting writers, visual artists, and musicians whom the government opposes. For example, if a president doesn’t like a particular singer, he can ask her to prove that she has properly reported all AI contributions to her recordings.

I suspect this topic will not prove popular with many people.  If you dislike free speech, you may oppose the new speech opportunities opened up by AIs (just build a bot and put it out there to blog, it does not have to be traceable to you).  If you do like free speech, you will be uncomfortable with the much lower marginal cost of producing “license,” resulting from AI systems.  Was the First Amendment really built to handle such technologies?

In my view free speech remains the best constitutional policy, but I do not expect AI systems to make it more popular as a concept.  It is thus all the more important that we fight for free speech rights heading into the immediate future.

What matters for central banks?

by  Tyler Cowen

This study examines the drivers of inflation levels, inflation variability, and growth variability collectively representing long-term central bank performance across 37 advanced economies in the Great Moderation era. A key finding is that central bank performance is consistently linked to the overall quality of institutions, while central bank-specific factors such as independence, exchange rate regimes, or inflation targeting show no significant impact. The analysis is extended to the 2022 inflation resurgence, using pre-2022 country characteristics. The results indicate that reliance on imports from Russia (likely gas) and its interaction with post-COVID GDP growth are the primary determinants, suggesting that the inflation surge was not a reversal of the Great Moderation.

That is from a recent paper by

Democracy and Capitalism are Mutually Reinforcing

by  Alex Tabarrok

Many people argue that democracy is incompatible with capitalism but they differ on whether democracy will kill capitalism or whether capitalism will kill democracy. Peter Thiel, for example, famously said, “I no longer believe that freedom and democracy are compatible.” Thiel’s argument has a long-pedigree. The classical economists from Adam Smith to John Stuart Mill all worried that democracy would kill capitalism. Even Marx and Engels agreed with the analysis arguing that under democracy “The proletariat will use its political supremacy to wrest, by degree, all capital from the bourgeoisie, to centralize all instruments of production in the hands of the State…” they differed only in welcoming such a revolution.

On the other side of the aisle we have the moderns such as Robert Reich and Joseph Stiglitz who argue in Reich’s words that Capitalism is Killing Democracy as “Corporations” and “billionaire capitalists have invested ever greater sums in lobbying, public relations, and even bribes and kickbacks, seeking laws that give them a competitive advantage over their rivals…”

A third argument, consistent with the views of Hayek, Mises, Friedman and others, is that capitalism and democracy are compatible and even mutually reinforcing. Ludwig von Mises, for example, argued that “Liberalism must necessarily demand democracy as its political corollary.”

My latest paper (WP version) (with Vincent Geloso) is in the new book, Can Democracy and Capitalism be Reconciled? We take the third view and show empirically that capitalism and democracy go hand in hand. We also provide some mechanisms for this correlation which I may discuss in a future post.

The data is very clear that democracy and capitalism go hand in hand. The figure below, for example, uses the Fraser Institute’s Economic Freedom Index to measure capitalism and the Varieties of Democracy Index to measure democracy (we use liberal democracy for convenience but show the correlations are strong with any variety of democracy).

Every major democracy is a capitalist country and virtually every capitalist country is a democracy (Singapore and Hong Kong being the only two exceptions.) Moreover, the upper left region–countries with a lot of democracy and low economic freedom, i.e. state control of the economy–what we might call the “Democratic Socialism” region–is empty.

We show further in the paper that changes in democracy are positively correlated with changes in economic freedom. We can see this very clearly by examining a natural experiment–the fall of the Berlin Wall. The fall of the Berlin Wall created a big positive shock to democracy which was followed by large and sustained increases in economic freedom.

It is sometimes argued that only an authoritarian regime is capable of “imposing” big increases in economic freedom and this is clearly false but it is true that there have been large increases in economic freedom in some authoritarian regimes. In the paper we look at the biggest such cases, Peru, Nicaragua, Uganda and Chile. The case of Peru carries some general lessons.

Peru began in 1970 with an authoritarian regime and only modest economic freedom. Economic freedom declined under an authoritarian regime and to levels well below those of any democracy. Modest increases in democracy brought modest increases in economic freedom. Under the authoritarian Fujimori regime there were large increases in economic freedom which in the 2000s were ratified, solidified and strengthened under democratic governments.

What we learn from this brief history is that authoritarian regimes can decrease as well as increase economic freedom. Indeed, one reason we sometimes see big increases in economic freedom under authoritarian regimes is simply that they are starting from the wreckage left by the previous regime. It’s easy to increase economic freedom a lot when you begin with a base level far below that of any democratic regime. Moreover, the Peru case is representative in that when a democratic regime is established it typically does not reject but instead ratifies and strengthens economic freedom.

What accounts for the correlation in economic freedom and democracy? The paper discusses a number of mechanisms of which I will only mention two here. Consider two ways to get rich, redistribution and growth. Redistribution can make a minority rich at the expense of a majority. A dictator can live in luxury amid national misery. But no redistribution scheme can enrich the majority—only growth can. Broad prosperity comes not from dividing wealth, but from creating it through pro-growth, capitalist policies. As a result, in a democracy, the rulers, the demos, can only get rich through growth and this provides some incentive to think about capitalism and economic freedom. The incentive is not a guarantee, of course, democratic voters can vote for bad policies but if they want to get rich they have to think about growth and that means capitalism.

The second reason for the correlation is a negative one, democratic socialism collapses into authoritarian socialism. As Robert Dahl argued:

It is not the inefficiencies of a centrally planned economy…that are most injurious to democratic prospects. It is the economy’s social and political consequences.

A centrally planned economy puts the resources of the entire economy at the disposal of government leaders. …“power corrupts and absolute power corrupts absolutely.”

A centrally planned economy issues an outright invitation to government leaders, written in bold letters: You are free to use all these economic resources to consolidate and maintain your power!

The bottom line is if you care about economic freedom, democracy is the way to go and if you care about democracy, economic freedom is the way to go.

Read the paper (WP version) for more.

Tanmay Khale on the decline in iconic songs over time (from my email)

by  Tyler Cowen in

https://x.com/wdavidmarx/status/1977162349107900770?s=46

A model that would explain this (which seems plausible to me for how judges generate these lists):

1. Come up with some metric for assessing works which is normalized such that the overall distribution is constant over time, usually by normalizing a metric that isn’t constant over time.

E.g. (a sports example) touchdowns scored by quarterback -> percentile of touchdowns scored by quarterback among currently active quarterbacks.

The metrics are less concrete for art, but I think people try to make similar adjustments in art as they do in sports (to make the distributions constant over time). The motivation that the judges would give for this is that one should assess each contribution based on how exceptional it was for its time.

2. Classing an achievement as “great” when it’s at least a certain percentile compared to whatever preceded it, by one of the metrics above. (“Oh wow, Johnson’s X was far more Y than anything preceding it, what an innovative work!”)

1. and 2. together will basically guarantee that you’ll have (by that definition) fewer great works over time; in the simple case where you’re looking for something that’s better than everything before it by some metric where the distribution is constant over time, the chance of observation n being better than everything preceding it is 1/n…

I don’t claim that judges are doing exactly this, but they only have to be doing some of this (e.g., their assessment criterion is 20% something like this) for it to lead to the behavior highlighted in the Twitter post!