This is the once-a-week free edition of The Diff, the newsletter about inflections in finance and technology. The free edition goes out to 21,320 subscribers, up 58 since the last edition.

In this issue:

• Partial Deregulation as an Antipattern

• China's Inflation Inflection

• The Next Amazon IPO

• The First Shopify Pivot

• The Other Two Supply Chains

• Non-Monolithic China

Partial Deregulation as an Antipattern

In software development and project management, there's a concept of the "antipattern," popularized by the book of the same name. It's the inverse of the design pattern, a generalized way to solve particular classes of problems. Antipatterns are a generalized way to fail at solving particular classes of problems. Outside of software, they're often ideas preceded by "Why don't we just..?"

One common antipattern that leads to repeated economic disasters is the partial deregulation: taking an industry that has some kind of public policy impact (it provides an important good or service, it's a natural monopoly, it can cause a depression), strictly controlling one aspect of its behavior, and leaving adjacent parts totally flexible. The pattern arises in many places:

Savings and Loans: Type "s&l crisis" into Google and you get two autocomplete suggestions: 1980, and 1989. This makes it sound like a wild industry, which should be something that couldn’t be further from the truth. The S&L industry was supposed to be the most mundane and local part of the US financial system: small banks that took local deposits, and mostly made mostly mortgage loans. In Liar's Poker, Michael Lewis describes S&Ls as following the 3-6-3 model: pay 3% on deposits, lend at 6%, and be on the golf course by 3pm.

It's a very good model as long as deposits consistently pay 3% and mortgages reliably earn 6%. It was a great one in the 60s, when the 10-year yielded 4% and people were buying plenty of housing. It was not such a great model over the course of the 70s, despite elevated household formation, because long-term rates rose from 7% in 1970 to 11% by 1980, peaking at almost 16%. If you make a 30-year loan with a 6% yield, and interest rates rise to 16%, your loan is now worth 40 cents on the dollar, at least if it's marked-to-market. Of course, the thrifts would generally collect these loans eventually, so as long as they had enough liquidity to pay higher rates on deposits, they'd survive. They were technically insolvent, but in practical terms they could keep operating.

Normally, an insolvent borrower has trouble borrowing more money. Lenders like to get their money back. But bank depositors are a special kind of lender, and bank deposits are a special kind of loan; they're insured against default up to a set amount. So, depositors didn't need to perform any kind of credit analysis.

At this point, policymakers had three choices:

1. Recapitalize and re-regulate the banks; limit the mismatch between their long-term loans and their demand deposits, or at least incorporate that gap into the rules about how much they can grow.

2. Ignore them, and hope the problem goes away.

3. Let them grow faster; if they make more loans, those loans start at 100 cents on the dollar, so a) the average state of the loan book improves, and b) the banks may be able to claw back their losses.

Congress, in the Depository Institutions Deregulation and Monetary Control Act of 1980 and Garn–St. Germain Act of 1982, chose the last option: they raised the amount of deposit insurance from $40,000 to $100,000, let banks raise more money from wholesale deposit brokers instead of traditional deposits, and loosened the restrictions on who or what savings and loans could lend to. Meanwhile, interest rates finally did start to decline, so the S&Ls were bailed out of some of their most precarious positions.

But giving them the opportunity to grow out of one bad debt problem let them grow into another. In 1980, the S&Ls had a maturity mismatch problem, funding long-term loans with short-term deposits at a time when those deposits got more expensive. By the late 80s, they'd switched to a credit problem: lending too much to speculative real estate projects. By the late 80s, S&Ls were failing again, and at large scale. Eventually, about a third of the industry went under.

It's hard to regulate banks because the business looks best when it's getting worse: when credit is more abundant, the biggest beneficiaries are the marginal borrowers, who are able to roll over loans they otherwise couldn't pay. So typically the point when banks are making the worst loans is also when default rates look their best.

In retrospect, it's actually unclear whether recapitalizing the industry in 1980 would have been cheaper than the bailout that happened a decade later, which ended up costing around $130bn. But on a risk-adjusted basis, it was a very poor decision: when an entire industry is heading towards insolvency, giving them more leverage solves some problems, but magnifies others.

Energy deregulation is another case study. It usually involves creating a hybrid of a market economy and a command economy. Power companies, for example, may be able to charge fixed rates to customers, but pay floating wholesale rates for the power they sell to customers. Or, as in the case of Texas' infamous Griddy, they may even sell electricity to customers at whatever the market rate is.

Typically, the original idea behind deregulation is to make every price float, and then politicians quickly realize that their constituents have the same sort of loss aversion as everyone else: paying $20/month less is slightly noticeable, but paying $240 extra one month is very memorable. So they adjust rules to tamp down some of the volatility, and try to architect rules that encourage a steady supply of electricity.

This wonderful classic Matt Levine post goes into some of the hijinks that ensue: one set of rules says that power plants offering a competitive price should get to sell at that price; another rule says that power plants shouldn't have to turn on and off all the time, regardless of price. Combine the two, and you can create a system for running a plant as a very temporary loss-leader, and then selling the rest of its output at higher prices. There were many, many such strategies; similar ones bankrupted PG&E in 2001. It’s like trying to hack a system that can’t get patched; once you find a vulnerability, you can just keep on exploiting it.

The very broad argument for energy deregulation is that the rules are complicated, and some kinds of service are simply not allowed. The argument against deregulation is that prices are volatile, and it's a lot better for large, well-capitalized utilities to be the economic shock absorbers, rather than households. (And that, in practice, deregulated energy markets don’t seem to offer consumers much in the way of savings.) Some of that volatility, of course, is induced by the rules; if the system is limited in its ability to match supply and demand, then the parts of the system with market prices will have very volatile ones. And some of this just boils down to the complexities of the electricity market in general. It's a nontrivial problem. And it ends up being a value judgment: if it's intrinsically hard to guarantee both a certain level of service and a certain price, then someone will have to absorb the volatility. If it's individuals, they'll be upset, sometimes rightly so—if you're economically precarious already, more variance in your bills means a meaningful chance of being unable to pay them. But companies will also want compensation if they have to bear risks, and for regulated utilities, that compensation comes in the form of the returns they'll willingly accept on new projects.

A last example is the Soviet pseudo-financial system. As described in The Oligarchs, Soviet Russia had two kinds of currency, one of which could be used for salaries and retail purchases, and another of which could be used for transactions between companies. The latter currency was nominally worth just as much as the former, but in practice was valued at about a tenth as much. If a factory owner needed something, it was much easier to get it with real money than with company-only monopoly money. The Soviet central bank could exchange one form of money for the other, but the rest of the economy had to treat them as separate.

And, in 1987, Mikhail Khodorkovsky, a former deputy head of the Communist Youth League, figured out a way to exchange one for the other, at a 1:1 ratio. He could accept the nearly-worthless factory-rubles, and turn them into actual rubles. This was very lucrative for Khodorkovsky (peak net worth: $15bn), but also made him some enemies (his companies were confiscated early in the Putin years, and he's currently living in exile in London).

The parallel currency system makes a certain amount of sense: currency is a good way to ration goods and services, but you can never be sure that it will get to where it's supposed to go. Subsidies that are supposed to go to building a steel mill might buy the head of the steel company a new vacation home and car, for example. By essentially offering the money in pre-marked bills, Soviet central planners were at least able to ensure that less money would leak out. But, as it turns out, the fungibility of money is a feature, not a bug. And because it's such a desirable feature, anything sufficiently moneylike will eventually develop fungibility, and if the financial system rests on the assumption that that can't happen, the result is an unpleasant surprise. There were many weak points in the Soviet system, and arguably the fake-ruble-to-real-ruble arbitrage made it work better, since it meant that a recipient of fake-ruble subsidies could actually use them to make purchases. But the system was designed around the assumption that this couldn’t happen, and it didn’t adapt in time.

Markets abhor persistent arbitrage opportunities, and any time a fixed system bumps into a more dynamic one, the dynamic system's drift naturally creates these opportunities. Building a system that's resilient to this is tricky; in each case, one of the main goals was to increase certainty and reduce volatility. But some level of uncertainty and volatility is a fixed trait of reality—rules can mitigate it, but they can't squash it. And if the rules manage to suppress it most of the time, they're just pushing it out to the tails. Deposit insurance, for example, was a good policy in that it allowed savers to sleep at night. But when it became a subsidy for insolvent banks to expand as fast as possible, and make the most speculative loans possible, it turned into a wealth transfer from taxpayers to the property developers who got loans. As it turns out, some systems are hard to perfect, and getting them 95% reliable just makes the last 5% worse.

Share

Elsewhere

China's Inflation Inflection

The early 2000s were a paradoxical period for inflation: commodities prices were rising, but consumers experienced fairly mild inflation. One explanation for both was the growth of China: as China exported more, the country consumed more raw materials. At the same time, China's wages were so low that finished goods built there were cheap. Today, China's workforce is shrinking, and wages have risen substantially (the CEIC index of China's real wages has risen by 8% annualized since 2008, for example). As a result of this wage pressure, raw materials price increases are actually causing Chinese exporters to raise prices on finished goods ($, WSJ). Some of the price increases are temporary; Covid introduced plenty of kinks in the supply chain. But the response is more permanent: if developed-world consumers are going to continue to see low inflation in the cost of imported goods, the world will have to find a new source of cheap labor.

The Next Amazon IPO

Thrasio is a company I've written about a few times, most recently here. Their model of adding value through 1) supplying working capital to growing Amazon sellers, 2) knowing the ins and outs of the Amazon platform, and 3) getting the economic free lunch of diversification, seems to be working. They've raised another round and are rumored to be considering an IPO. This won't be the first company to go public with a business built on another company's platform—Zynga was highly reliant on Facebook when it IPOed, and Jamf is a bet on Apple—but it shows that businesses can scale rapidly when they're the complement to dominant company.

The First Shopify Pivot

First1000 is an excellent newsletter that delves into startups' histories to learn how they acquired their first cohort of users. This is always an informative exercise. Acquiring up to your Dunbar Number of users can just mean slightly pressuring friends to at least nominally sign up for your product, but by 1k, people generally have to hear about the product from someone other than the founder of the company. Their latest post, on Shopify, is an interesting example of a company that moved along the supply chain until it found the most valuable place to be. It's well-known that Shopify started as a store and evolved into a platform, but one of the obvious-but-easy-to-miss aspects of the story is that the original Shopify store, which sold snowboards, was able to pivot because it was such a seasonal business. It's another testament to the value of moderate amounts of slack; any organization that can sell out its inventory of snowboards during the winter will find something useful to do for the rest of the year.

The Other Two Supply Chains

A frequent Diff theme in the last year has been the gradual shift from a global supply chain to a series of country- or bloc-specific ones, especially for strategic products. This shift is still in progress. There's another driver, too: the US and Western Europe tend to discourage multinationals from engaging in certain behaviors, but also from abetting those behaviors; other countries are less worried. One example of this is Congolese cobalt ($, Economist):

By value, as much as 90% of cobalt mined in Congo comes from industrial firms such as Glencore, an Anglo-Swiss trader, and China Molybdenum, a Chinese state enterprise, which use modern methods to drill and refine the ore. The rest comes from miners like Mr Mwansa, who dig mostly by hand on land held by co-operatives, or else illegally, on land owned, say, by industrial firms. They mainly sell to Chinese middlemen who operate “depots” that line the main road out of Kolwezi, where they do basic processing and then transport the ore to South Africa or Tanzania, to be shipped to China.

Yet by employment, artisanal mining outweighs the industrial sort. Glencore’s two mines employ around 15,000 people; more than 200,000 may work on artisanal sites.

Non-Monolithic China

Kevin Xu has an interesting piece on Elon Musk's relationship with China. He links to this FT piece on how the relationship between the Chinese government and information-rich tech companies is not as straightforward as it might seem: while the Chinese government can get access to data, companies are very reluctant to share it, for business reasons. They don't want users worrying about the government watching their every move, and they also don't want data they hand over to get stolen or sold. China's somewhat decentralized governance approach also makes it easy to overestimate the country’s state capacity; if local governments pursue a mix of policies, with mixed results, the newsworthy ones may be the most alarming rather than the most representative.

My general sense is that the Chinese government is more likely to get information on broad swathes of tech companies' users than other governments, but this is a good reminder that it's not cut-and-dried. After all, one reason American tech companies are so protective of their user data is that the American national security apparatus was spying on them. Even though the circumstances vary from country to country, the incentives are fairly similar: states want to collect data, and the companies that are good at collecting it in order to serve ads and sell products are reluctant to make decisions that worsen their business. They always meet somewhere in the middle, although the exact location of this middle will vary from place to place and time to time.