Welcome back to The Diff. Here are the subscribers-only posts you missed this week:
- Climate Control and Financial Centers argues that the rise of cities like New York and Singapore, with complex financial markets and high-volume exchanges, can be explained as a side effect of the rise of air conditioning.
- In Building an Incomplete Graph I look at how companies handle gaps in the graph—what Netflix does about movies it doesn’t have, how Facebook thinks about non-members, how ride sharing companies prioritized launches.
- Taiwan Semiconductor Manufacturing: A Successful Faith-Based Organization looks at the challenging economics of making chips without designing them, and asks how any company can turn a profit when it’s spending $488 per second on capital expenditures without knowing what that capital will ultimately produce.
Thanks again for reading The Diff. This newsletter is a solo operation, and it’s funded by subscribers. Join them today to get the full experience.
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 17,496 subscribers, up 289 week-over-week.
In this issue:
- Taiwan and Supply Chain Frenmity
- Airline Green Zones
- Short Sellers and the Supply Chain of Information
- Housing and Pull-Forward Demand
- Facebook and Visa Arbitrage
Taiwan and Supply Chain Frenmity
One of the most economically important cities in the world is Hsinchu. Where Hsinchu is located depends on who you ask: if you’re booking a hotel, Marriott will let you narrow it down to “Mainland China and Hong Kong, Macau, Taiwan,” while your United Airlines flight can get you to “Taipei” (no country listed). If you try to ask the World Health Organization about the country Hsinchu is located in, they will encounter unfortunate technical difficulties. Hsinchu is, of course, located in Taiwan, a country whose formal existence is actively questioned and whose practical existence is undeniable.
In yesterday’s post, I wrote about the history of Taiwan Semiconductor Manufacturing from a microeconomic perspective: why a chip company that didn’t design chips was a nonsensical proposition for decades, why it became workable, how it grew essential, and what its model means today. Today’s post looks at the same phenomenon from a different angle: why did the business start in Taiwan, and what does it mean today?
Taiwan’s history as a major economy starts after the end of the Chinese Civil War; the Communists conquered the mainland, and the Nationalists retreated to Taiwan. While Taiwan was certainly not communist, it wasn’t especially capitalist, either. Like other successful East Asian countries, the Taiwanese government broke up large estates and encouraged local farming—in Taiwan’s case, mostly as an effort to a) weaken non-state power, and b) make the country self-sufficient in the event of a blockade. Taiwan did want some exports to generate hard currency, but with a particular purpose in mind: the country had one of the largest armies in Asia in the 50s and 60s, and had active discussions with the US about what to do about China in the early 60s.
Japan and South Korea developed their economies in a state-directed, banking-heavy way: large companies got loans in exchange for promoting exports, and the initial focus was on heavy industry: steel, chemicals, machinery, and later cars. Taiwan’s model was almost the exact opposite, with a large array of tiny companies, mostly targeting labor-intensive exports. Each country’s industrialization was promoted by a few big secular trends—higher consumption and rising wages in the US and Europe, and the growth of the container ship industry, particularly the one-way shipment of military supplies to Vietnam that made rates from Asia to the US low. But the exact timing of their growth led to different development models. (For much more on this, see my piece on the East Asian economic miracle and this one on containerization.) US consumption in particular benefited from an accident of tax policy: Congress passed a law in 1954 offering a tax credit for investing in business, which was meant to support manufacturers. It was written vaguely enough that it applied to malls and shopping centers, and the growth of the Interstate Highway System made those businesses more viable. So a law designed to promote domestic manufacturing ended up creating demand for cheaper foreign manufacturing instead.
Taiwan’s growth coincided with three important trends: first, wages in Japan started rising, which made the country’s lowest value-adding industries, like footwear and electronics assembly, less profitable. Second, American retailers started using bar codes and computers to manage their inventory better. Third, American retailers had to adhere to fair-pricing laws, which prohibited large companies from selling brand name products at low prices. Retailers responded by creating house brands, which they could price any way they wanted. Sears, for example, asked American TV manufacturers to build them an off-brand TV set in 1964. The American companies said no, but Toshiba said yes.
As it turns out, these unrelated trends combined in a way that massively benefited Taiwan’s export sector. Japanese trading companies had good relationships with American retailers, and started offshoring their production to Taiwan. (Since Taiwan was a former Japanese colony, there were some cultural and linguistic ties.) The trading companies could move their equipment from Japan to Taiwan, and take advantage of much lower wage scales: in the mid 1960s, US textile workers earned $333 a month, Japanese textile workers made $69/month, and Taiwanese workers made $23/month.
For some products, fragmented production is a major cost. It’s hard to coordinate thousands of ten-person companies to put together a car. But for other products, it’s a benefit, because it means that the lowest value-added parts of the process can be outsourced piecemeal, while the most valuable ones can stay behind. Assembling TVs is a low value-added task, but some of the components require proprietary technology. Japanese TV companies could outsource the easy work to small Taiwanese contract manufacturers, while keeping the most valuable parts domestic. In bicycle manufacturing, Taiwan’s distributed manufacturers were able to respond faster to changes in consumers tastes; Korean chaebol conglomerates tried to compete, and they had unbeatable costs for large product runs—but couldn’t handle the diversity of products consumers wanted. (Japanese companies continued to make components like derailleurs, which were a) harder to get right, and b) much less sensitive to the whims of consumers.) In shoes, Nike contracted its high-volume product lines to Korea, and specialty products to Taiwan.
The PC market came later, but was an even more extreme example: the product cycle for PCs was extraordinarily fast in the 80s and 90s, and the business was scaling rapidly. Some components could be reused across many product lines, while others were product-specific. Taiwan’s original equipment manufacturers—contractors who built generic PCs for brands like Dell and Gateway—responded in their usual fashion, and the country turned into a key supplier for US-based companies.
At this point, a problem developed: Taiwan was reaching the classic middle-income trap of getting richer by competing on price, and ending up too rich to compete on price. Fortunately, a solution presented itself: China slowly opened its economy. It’s a testament to how entrepreneurial these firms were that the beginning of large-scale outsourcing from Taiwan to China predates the legalization of travel and investment there; apparently many of them reached arrangements with Hong Kong’s banking and border control authorities that would allow them and their money to travel from Hong Kong to China without fully informing Taiwanese authorities.
By this point, Taiwanese contract manufacturers had expertise in managing complex supply chains, deep relationships with retailers and distributors, and a keen understanding of how to react to consumers' whims. All they needed was affordable labor. China, with a population in 1990 of 1.14bn people and a GDP per capita at the time of $320, had both. So, industry by industry, Taiwanese manufacturers moved to China. Sometimes, this happened incredibly quickly—Taiwan was 30% of US shoe imports in 1987 and 3% in 1992; the suppliers mostly didn’t change, but the factories moved. In other cases, it took longer.
It’s around this time that Taiwan Semiconductor Manufacturing enters the picture. TSM is in some ways a counterpoint to the general story of independent, small-scale entrepreneurial Taiwanese companies; it was founded as a joint venture with the government. On the other hand, it’s very much in the spirit of the Taiwanese exporter ecosystem; Taiwan didn’t have many globally recognizable brands, but they did have a comparative advantage in meeting the needs of companies that did. TSM, as a chip foundry that didn’t design chips, had exactly the same mercenary agnosticism as other local manufacturers: a shoe factory could work with Adidas or Nike; a bike company could brand its products as Schwinn or Huffy; TSM can fab chips that will go into anyone’s devices. As Taiwan’s economy matured, the country generated enough savings to fund ambitious investments (in fact, Taiwan’s vast savings create substantial distortions in global bond and currency markets). A capital-intensive business like chip fabbing is exactly appropriate for a country like Taiwan, with a rich, high-savings population and no comparative advantage in labor-intensive businesses.
TSM is over half the foundry business, with 3x the market share of its nearest competitor. And chips are, like oil, dollars, and widebody aircraft, one of those products that basically every country wants to consume but that many aren’t able to produce. A paper I once read said that oil is “the most efficient way to turn economic risk into geopolitical risk,” and semiconductors fall into that category, too: a crucial global industry—more than 1/200th of all economic activity, and growing—has its de facto headquarters in a country that is treated, by the country with the world’s largest active-duty military, as a rebellious breakaway province that must be reunited with the mainland by whatever means necessary.
The Golden Arches Theory of Conflict Prevention holds that countries with close economic ties (e.g. McDonald’s) will not fight a war. The Chinese/Taiwanese equivalent is the Local Province Conflict of Interest Theory: even if national-level policy dictates that China and Taiwan should be politically reunited, local officials in China only hit their GDP and employment growth numbers because the two countries are so economically united, and political action is a threat to that. Earlier in China’s development, this was because the factories were literally owned and operated by people from Taiwan, which is still the case, albeit less so than it used to be ($, Economist).
But China’s reliance on the consumer electronics industry, and that industry’s reliance on TSM, makes conflict murderously expensive, at least for now. But it also means that if you’re worried about future conflict, the scariest statistics to look at aren’t troop counts, or the speed of hypersonic missiles, or other indicators of combat capability. The really scary numbers are semiconductor capital equipment billings by country, which show that Chinese manufacturers are now investing substantially more in chip manufacturing capacity than Taiwanese ones are. There is a large gap between spending money on chip manufacturing capacity and actually making competitive bleeding edge chips—if you don’t believe me, ask an Intel shareholder. But it’s a worrisome sign, because Taiwan’s paradoxical existence makes sense in light of commercial ties, and those ties are weakening.
 Stay tuned for a future Diff writeup on one of the other surprising beneficiaries of that law, the notorious Leasco. Leasco has it all: a bit of technology, a lot of financial engineering, rejections of B-school dogma, and class conflict. It closes with one of the all-time greatest quotes from someone who earned millions of dollars, but also millions less than he thought he was about to: “I always knew there was an Establishment. I just used to think I was part of it.”
 These fair trade rules may be familiar to readers from my Costco writeup a few months ago.
 This is a good time to note that The Diff itself operates much like the Taiwanese export sector, with a large number of modular components, each of which sources key inputs from sites like FT.com and SEC.gov, and which link together to deliver the finished product.
A Word From Our Sponsors
Here’s a dirty secret: part of equity research consists of being one of the world’s best-paid data-entry professionals. It’s a pain—and a rite of passage—to build a financial model by painstakingly transcribing information from 10-Qs, 10-Ks, presentations, and transcripts. Or, at least, it was: Daloopa uses machine learning and human validation to automatically parse financial statements and other disclosures, creating a continuously-updated, detailed, and accurate model.
If you’ve ever fired up Excel at 8pm and realized you’ll be doing ctrl-c alt-tab alt-e-es-v until well past midnight, you owe it to yourself to check this out.
Airline Green Zones
Last week, I cited Delta and Alitalia’s plan to implement their own Covid-safe travel corridor through frequent testing. Delta has elaborated on this plan in an internal memo to employees on testing. Airlines have unique motives and opportunities to deploy test-and-trace procedures:
- Unlike many other businesses, all of their employees and customers have to be identified, and have to pass through specific chokepoints. Testing changes the degree of inconvenience for air travel, but not the kind. And since airlines need their passengers' identities and contact information already, they can actually inform them of test results or past exposures.
- Airlines also have a stronger than usual economic incentive to return to normal as quickly as possible. One way to view the airline industry is that it’s a classic finance story: time-weighted returns look better than the dollar-weighted returns, because airlines always deploy capital at the worst possible time, and then have to raise expensive funding to stay in business.
(Via Tyler Cowen.)
Short Sellers and the Supply Chain of Information
Institutional Investor has an alarmist article with some shocking news: apparently, some investors pay for research! And then they trade on it!
The economics of short-selling research work like this: finding a good short idea takes lots of time, and it takes a highly variable amount of time. Sometimes, the idea is simple; sometimes, it requires labor-intensive research, or even rolling a car down a hill. Because of this uncertainty, it’s hard for dedicated short-selling funds to make money—they can do it, but the payoff of short selling doesn’t line up well with standard hedge fund structures. Instead, short sellers often work independently and sell their research to funds. This presents a tricky information asymmetry problem: pay upfront, and there’s a market for lemons. Pay only after reviewing the idea, and it’s easy for a fund to get the gist of the thesis and recreate the trade itself without paying. (Allegedly, one of the investors who was pitched John Paulson’s subprime-shorting fund thought the idea was good and ran with it himself. The problem of piracy is not confined to the media business!)
Aligning incentives is hard and imperfect, but in this case it’s socially beneficial. Liquidity makes fraud easier, and fraudulent companies compete with legitimate ones for capital. Even an imperfect system of rewarding people for spotting dubious companies is better than a system that lets those companies get away with it.
Housing and Pull-Forward Demand
The last time an economic recession led to interest rate cuts and a housing price boom, it did not end especially well for the world economy. But one of the upsides to a catastrophic once-a-century pandemic is that we can, in fact, say this time is different and mean it. One difference between today’s housing market and the market circa 2006: it’s much more convenient to evaluate a house and make a purchasing decision. This has a few long-term effects:
- It means that the underlying cause of housing’s outperformance—a decrease in the desirability of cities and an increase in the desirability of extra bedrooms and home offices—gets reflected in the market faster.
- It improves the unit economics of ibuyers (for more on them, see my writeup here). They charge a percentage fee and pay interest as a function of time, so anything that reduces their holding period increases their profits.
- A shorter lag between deciding to make a purchase and closing on a new house shifts the move in housing prices earlier in time relative to the change in demand. So any historical model that predicts housing demand based on external factors like interest rates, inflation, and unemployment will have to adjust its lags.
Facebook and Visa Arbitrage
The DOJ has sued Facebook for favoring H-1B workers over US-based workers. This lawsuit illustrates how tricky it is to write a coherent immigration policy, especially one that balances the desire for skilled workers with the desire not to depress wages:
Rather than conducting a genuine search for qualified and available U.S. workers for permanent positions sought by these temporary visa holders, Facebook reserved the positions for temporary visa holders because of their immigration status, according to the complaint. The complaint also alleges that Facebook sought to channel jobs to temporary visa holders at the expense of U.S. workers by failing to advertise those vacancies on its careers website, requiring applicants to apply by physical mail only, and refusing to consider any U.S. workers who applied for those positions. In contrast, Facebook’s usual hiring process relies on recruitment methods designed to encourage applications by advertising positions on its careers website, accepting electronic applications, and not pre-selecting candidates to be hired based on a candidate’s immigration status, according to the lawsuit.
From Facebook’s perspective, it looks like this: an employee was worth hiring, and works well with the team. Facebook would like to keep employing this person. However, there’s some paperwork, so Facebook does the minimal level of paperwork necessary. if the job were an interchangeable, unskilled position, a policy requiring them to advertise it to US workers would not materially harm the company, but would increase US workers' wages. That’s not the nature of these jobs: the workers in question are already at the company, and presumably already working on projects with existing teammates. Replacing them with another worker who has similar but not identical skills, but isn’t used to working with the team, is close to a pure deadweight loss for both Facebook and the H-1B worker.
A visa for workers who have skills that aren’t available locally is mostly an exercise in artfully parsing definitions. In some fields, the half-life of skills is short enough that, over the course of a six-year visa the skills they’re hired for will be obsolete and the work they’re doing will be completely new. Facebook’s technical compliance illustrates how hard it is to write straightforward policy that makes qualitative judgments about workers' skills and employers' needs; easier to set a quantity, hold an auction (with long-term payment plans, to encourage low-cash-flow/high-potential jobs), and let employers quantify how irreplaceable any potential visa holder is.