Amazon Sees Like a State

Plus! Pensions, Podcasts, Safe Havens, CPI, Toilet Paper, Boomers, Zoom, and more...

This is the once-a-week free edition of The Diff. Paywalled newsletters this week include: a useful economic law that’s always wrong, a review of a timely Edward Luttwak book (from 1968), and an overview of the game theory of crises.

Amazon Sees Like A State


Thirty years after the end of the Cold War, America’s government  enjoys Brezhnevian sclerosis—which septuagenarian will be in charge next  year!? Our biggest hyper-capitalist companies run internal cradle to grave, with free, high-quality public transportation and socialized food.   They’ve delivered the “what” of utopian socialism (albeit with monster  doses of Marxist labor alienation—sure, at one level you’re Organizing  the World’s Information, but in practice that means a tiny optimization  that makes pages render 1ms faster). But they’ve also delivered the  “how”: big tech companies get big when they achieve a monopoly on  resolving an information gap. Economic profits ultimately rely on  information gaps.[1] There are a few short descriptions for how you  aggregate messy user-level information into economically valuable data.  Depending on whether you want to be valued at 5x revenue, 10x, 20x, or  50x, you can call it statistics, data science, machine learning, or  AI.[2] All of these basically involve converting a question into a giant  matrix, and solving it. This is basically what the USSR tried to do.

There are many practical critiques of socialism—body counts, corruption, darkness everywhere. But even before that, there was a damning theoretical critique: the calculation problem.  A price system sets an exchange rate between any one thing and every  other thing, at any point in time. Stalin didn’t know the exchange rate between a tractor and a sock and a poem. He just had to guess.

(This is one reason Russia embarked on such a breakneck industrialization program.  The sooner you achieve post-scarcity, the sooner you stop worrying  about tradeoffs. But to get there, you still need to know the exchange  rate between an arc furnace, a truck chassis, and a ton of cement…)

In economic terms, the trouble with command economies is that supply and demand are curves,  and a price is an intersection between those curves. The price is the  most informative point on the curve, but price still projects  multidimensional curves onto a one-dimensional number line. The Soviets  tried mightily to solve the problem, building “IBM-compatible” machines and ever-more-complex models to meet demand, control prices, and avoid shortages.

It failed: insufficient data and bad incentives. But now Google,  Facebook, and especially Amazon do the same thing: they analyze reams of  data to implicitly estimate supply and demand curves. Amazon has to explicitly  estimate them, since it provides the logistical backend for actually getting products to customers, and treats reliably fast shipping as a  key differentiator. When Amazon makes a pricing decision, it’s making a  decision about warehouse space, packing efficiency, last-mile shipping  capacity, and working capital. And since so many goods it sells are  complements and substitutes, it’s making that decision, incrementally,  for countless other products besides. This is the sort of complex  tradeoff that, in theory, should be intractable for a single  institution. In theory, the way to handle complex tradeoffs  between products in a catalogue and the entire supply chain required to  deliver them is to let prices and incentives do the calculation. In practice, Amazon seems to just do it.

We have reached a weird point in history where we can say that true  communism has never been tried, because they didn’t have enough RAM.

When I look at decisions like Amazon only selling masks to hospitals and governments, but using them in warehouses, I don’t see hypocrisy. I see the world’s first functioning command economy elegantly handling an edge case.

The fact that tech companies are very meta makes them a peculiar kind  of altruistic: they own a percentage of economic growth, and given  their model it’s usually high-margin (and valued at a high multiple).  Once a tech company’s share of economic growth multiplied by its  price/sales ratio equals one, the company becomes selfishly altruistic,  at least for all forms of altruism that can be measured in GDP.

Last year, the US economy grew $850bn in nominal terms, and Amazon’s  North America sales growth was $30bn of that. AMZN trades at 3.45x  sales, so its altruism quotient is ($30bn/$850bn * 3.45), or 12%. i.e.  any time Amazon can spend $1 to make the country $8 better-off, it’s a  wash. Since Amazon is growing faster than the US economy, and its  price/sales ratio has drifted up over time, this will only rise.

This gives Amazon one very good incentive, and one very bad one. The  good incentive is to make the world a better place, as measured by GDP.  More money means more purchases of electronics, clothes, food, books,  on-demand videos, etc. But the bad incentive is to make the economy more  measurable. One way to increase GDP is to invent a fabulously  useful gadget. Another way to increase GDP is to substitute a  home-cooked meal using $5 worth of ingredients for a food delivery order  that costs $30 for something not quite as good.

This tractability argument is straight out of James C. Scott’s Seeing Like A State,  which is all about governments' efforts to give people full names,  fixed addresses, centrally-registered property rights, and other traits  that make them more easily taxed and conscripted. Scott was writing from  a historical perspective, and historically the most ambitious people in  the world went into government. I don’t know who the most ambitious  industrialist in Napoleonic France was, but that industrialist was less  of a go-getter than the median French Army captain at the time. Today,  Scott’s argument applies to tech, because tech is where ambitious people  go. The institutional and legal framework are different—Jeff Bezos is  not going to make you march on Moscow midwinter any time soon—but the  incentives are the same.

There’s an old Isaac Asimov story, The Evitable Conflict. The story is set in the year 2052, when the world’s economy is run by four  supercomputers. In the book, the world’s economy has moved beyond  communism and capitalism. When computers make the decisions, it doesn’t  really make sense to ask the question. This struck me as perfectly  sensible when I first read it (I was 12), totally crazy when I reread it  (some time in high school, after a megadose of Thomas Sowell), and  intriguingly reasonable now (because it’s 2020).

Which is not to say that the commies were right. They were wrong all  along, with disastrous results. It’s a victory lap of sorts that VCs  looking for a big score and wall street financiers demanding ever-better  quarters have caused big tech companies to casually implement the most  effective form of economic central planning and socialist nanny-statism  ever devised.

The killer critique of socialist economics was not that it was  strictly impossible, just that it was hard enough to be effectively  impossible. “Technology” is a shorthand for the process of making  impossible things possible, so in a sense it was inevitable that the  socialist calculation debate would eventually be resolved in socialism’s  favor, just implemented by capitalists and judged by an  earnings-per-share metric.

Lenin probably never said that when the revolution comes, capitalists  will sell socialists the rope that’s used to hang them. But Huey Long  did say “Sure, we’ll have fascism in this country, and we’ll call it  anti-fascism.” So it’s all too appropriate to see Soviet-style central  planning coming to America, sporting a $955bn market cap.

[1] I’m back on my footnote game. To flesh out this argument: in the  short term, you can earn an economic profit from economies of scale or  other cost advantages, but I tend to view those outcomes as distorted.  Getting to the ideal scale is a tournament,  but “ideal scale” is not a stable quantity. While there are rewards to  timing chunky capex well—semiconductor foundries are probably the best  modern example—there are big costs to being early. Regional monopolies  are also a tournament; if you own the only gas station in a town that  can support 1.5 gas stations, you have a good business, but either a)  you bought it at a fair market price, or b) your earnings include a risk  premium from correctly betting on the size of the market and  correctly betting that you wouldn’t split the market with someone else.  So those economic profits are somewhat illusory: they’re real to current  owners, but that’s survivorship bias. Economic profit from regional and  scale monopolies tends towards zero when you count all the people who  went out of business trying to earn it.

Information asymmetry persists much longer, and it’s a good  description of many other kinds of monopolies. What is a successful  biotech company other than a monopoly on the knowledge required to make a  particular drug? Of course, tournament dynamics are at play here, too,  but it’s an iterated game: the real value of the firm is its ability to  repeat the magic trick of inventing a new product. Consumer Internet  companies are a sort of economic Maxwell’s Demon, creating a barrier  between consumers who are hyper-efficient at expressing demand and  companies that are hyper-efficient at satisfying that demand. The big  Internet companies charge to make the introduction.

[2] A biography of Bill Gates written in the early 90s refers to the early 80s as “a period where ‘AI’ was becoming a marketing tool.”

Elsewhere

I wrote a piece in Medium’s Marker about pensions.  One way to think about pensions, from the recipient’s perspective, is  that they’re basically equivalent to an annuity. If interest rates drop,  that annuity is worth more today. Which means that to the pension fund  itself, lower interest rates imply that their liabilities are higher.  Normally, the way to hedge that risk is to buy long-term treasuries.  Pension funds have, increasingly, not been doing that; they’ve been  investing more and more in equities, including private equity. When the  economy is growing, that pays off nicely. During a downturn, they’re hit  in two directions: the value of their equities drops, but their  liabilities rise.

I also joined the Airways podcast, for an interview covering airlines, bailouts, politics, and geopolitics.

And in Coindesk, a piece on why Bitcoin is not a safe haven for a dollar-shortage, but is in other cases.

Covid Updates

New York Legalizes E-Bikes

No idea how they found the time to do this, but New York has un-banned motorized scooters.  NYC was an early hotspot for car-sharing, the last major disruption in  transportation technology, so it will be be interesting to see which of  the surviving scooter rental companies grab share after this.

Boomer Housing

Greg Mankiw and Davie Weil predicted  in the early 90s that housing prices, driven by Baby Boomer demand,  would peak some time in the mid-90s. That is not exactly what happened.  Part of why their thesis went wrong: as housing prices went up and  consumers levered up, mortgages became the major channel through which  monetary policy stimulated the economy. Meanwhile, high turnout for  older voters meant that the Levered Boomer Demo was a key political  constituency. Bubbles lead to bifurcation, especially when you can’t  short: if the world consists of housing optimists and housing  pessimists, as prices rise the optimists are increasingly the  price-setters.

But you can’t outrun demographics forever. Eventually, people downsize or die. That’s starting to happen,  according to a new paper. A striking stat: controlling for zip code,  prices for 1- and 2-bedroom houses have been rising since 2012, but  prices for 4- and 5-bedroom homes have been falling. And there’s a  negative correlation between housing price changes and average age by  zip code.

(I previously wrote about the economics of Baby Boomers here.)

Zoom Apologizes

Following their negative press (covered in yesterday’s newsletter, in which I argued that this happens to every growth company and has never killed any company), Zoom has prudently paused new features for 90 days  to lock down security. They also casually mention that Zoom DAUs went  from 10m at peak in December to over 200m today. 2,000% growth off a 10m  base in three months is a record that may never be surpassed.