Reflexivity in Tech Returns: Talent Chasing Money Chasing Talent
One inconvenience for anyone who wants to develop a general theory of profitable investing is that two of the best investors in history are Warren Buffett and George Soros. Buffett made his fortune buying cheap and selling dear (or holding forever); Soros made his by buying when things are expensive and selling when they're somehow even more so. Buffett has a theory of investing, which is obvious ("buy it for less than it's worth, and buy things that go up in value over time") but hard to implement, and has been well-described elsewhere. Soros also has a theory, which is non-intuitive and even harder to implement, but more broadly useful.
Soros' theory goes by the name reflexivity. His classic example is mortgage investment trusts: investor enthusiasm about the trusts allowed them to raise cheaper capital, which caused them to grow faster, which made investors even more excited; eventually they reached a point where their underlying profits were being driven more by capital flows than by fundamentals (when yields are declining because prices are rising, early buyers show nice profits when they sell, but they're selling to later buyers). This reversed, sharply, and the sector lost a lot before reaching a new steady state.
Reflexivity is easiest to see when an industry funds growth by borrowing, and has appreciated assets to sell: high stock prices lead to cheaper credit (the bondholders are more confident that the company has alternative sources of capital), and that cheap credit leads to higher growth. But it can show up in other places.
Take the 19th century steel industry. The basic reflexive loop was:
- Cheaper transportation from railroads lowered the cost of inputs like iron and coal, and raised the returns from building larger and more efficient steel mills.
- Larger mills could supply cheaper steel, which was used by railroads—further lowering transportation costs.
So in one sense the entire steel industry was a sort of ponzi scheme, where the supply of steel was growing to meet the demand created by the booming industry of... supplying steel. This was a reflexive feedback loop, since optimism about steel demand created more cheap steel supply, and that enabled the dense transportation network that created even more demand.
Of course, that can't go on forever, and at some point each incremental mile of railroads gets too low a return. (That can take a while, because of network effects: as with airlines, connecting traffic leads to higher travel demand, so for a while adding capacity leads to a net increase in usage.) Building a business that is big enough to handle peak one-time demand means building one too big for steady-state demand.1 As it turns out, the industry was less dependent on that feedback loop than it originally looked. US rail mileage declined from the early part of the 20th century, but US steel production ultimately peaked in the 1970s; as it got cheaper, the number of uses for steel increased.2
It's worth keeping this recursive model in mind because it's an important part of what's going on in the tech sector. I've written recently ($) about high compensation for employees at big tech companies, and in that post I took it as a given that these workers were getting paid more because they were worth more to employers. There are two viewpoints on "worth more," though.
- The techno-optimist argument is that there have never been more people online, and never has so much of the world been at least partly automatable. Your raspberry pi can order you a pizza! You can set up a bot to Slack you when your Zoom webinar has so many attendees you need to upgrade! You can write a script that texts you when SpaceX schedules a new launch. People get access to pocket supercomputers before they have reliable access to running water! There have never been more opportunities to create software, find users, and make a fortune, and while it's a big industry there's still a vast amount of human labor involved in things that computers will be able to do soon.
- The pessimistic view is that that's all well and good, but an increasing amount of this growth is basically fake. Startups raise money because big tech companies are worth a lot, and those companies are worth a lot because they have businesses with high incremental margins. But many of those businesses sell to smaller companies! Facebook revenue gets a boost from TikTok spending its way into being the next Facebook; any business that lets people search for products seems to buy top-of-the-funnel leads from ads on Google; you can't so much as sneeze without incrementing your AWS bill; if you sell through salespeople, then Salesforce sells to you; and so on. It's a pyramid scheme where trying to find the next X is what makes X something investors want to find the next instance of.
There's plenty of evidence for both of these ideas, and it's hard to fully refute either of them. You can argue that many companies are doing frivolous things, but most companies start out looking fairly frivolous, so that's not true. (The easy counterexample is Google, since search existed as a product before them. What Google looked like was an academic project from people who didn't understand online economics; a good search engine moves people off your site and onto their destination, which means it reduces banner ad pageviews. If you make your money from eyeballs, you want to offer a bad search engine.)
One way to avoid this argument is to cite reflexivity as a synthesis of it. The channel through which reflexivity acts in this case isn't leverage, though. It's people. High and generally rising tech valuations mean that tech companies are the high bidder for talent in a way they weren't before, and this expands both the size of the industry and the opportunities for people and companies within it. The more complex an industry is, the more likely it is that any new entrant is a supplier or customer, not a competitor. Or that it's more than one of the above; as Ben Thompson points out, one step of Intel's master plan to compete with TSMC involves giving them a lot of business. The existence of Google enables lots of long-tail media and commerce companies that couldn't otherwise exist; those companies make it more likely that a given Google search will turn up something useful, and those forces combined mean that more questions get automatically translated into searches. Facebook, Shopify, and Alibaba collectively support an enormous long tail of small online shops, with Facebook providing the customers, Alibaba finding the products, and Shopify gluing the two together; the entire system is a mostly-automated way to connect Shenzhen to the American consumer, with some humans in the loop to make decisions. (Even in that market, the amount of human discretion needed is declining, since there are automated tools for identifying which products to pursue next.)
These complex integrations between companies do have some bottlenecks. Ads have gotten pricier, and that tends to mean less money recirculating through tech and more of it accumulating on Alphabet/Meta's balance sheet.3 But in the business software market it's harder for anyone to control distribution, since so much of it happens through sales teams, so that ecosystem ends up being more open and dynamic.
For the most part, this reflexivity is helpful to everyone involved; investors get better returns and employees get better jobs. But it's not without risks. Netflix had a fearsome 80%+ drawdown in 2011 and 2012. 2012 was also a year where their total revenue growth decelerated to 13% from 48% the year before. Some of that deceleration was obviously exactly what the market was anticipating when the stock dropped, but some fraction of it might have been from temporary talent constraints. A company where employees are taking home lots of equity is one with an extra layer of leverage: employees will be demoralized and tempted to switch jobs at exactly the times when they most need to focus on the task at hand.
Note: a few people have brought up Netflix’s attitude towards equity comp: they let employees choose how much of their pay should be in stock versus cash. They spent $415m on stock-based compensation in 2020, or $44k per employee. This puts their stock based comp per worker number at a lower level than most big tech companies, but higher than Salesforce, and still probably material to how people there thought about their pay.
At a company level, the risk of stock-based comp is that it exacerbates periods of underperformance by leading to higher employee attrition. That doesn't happen the same way on an industry level, but the industry-level version is arguably worse; if tech's growth has partly been the result of so many smart people choosing to work in the industry (thank you, Aaron Sorkin!), what happens if they decide to work somewhere else? It replaces the decade-plus growth cycle with a decade-plus struggle for talent.
This has happened before: one reason for the surge in US state capacity in the mid-20th century was that the private sector slowed its hiring and promotions while the government added lots of headcount and had to promote people fast. When LBJ was 27 he was running the Texas National Youth Administration, which ultimately gave subsidies to 175,000 students and placed 75,000 of them in jobs. There are not a lot of 27-year-olds working for the US government today who are having that big an impact on the world—if you're that age right now and you want to do work that gets 75,000 people new jobs, you're probably working as a product manager at LinkedIn or Indeed.
The financial industry benefited from the same kind of talent flow in the 80s through early 2000s: bankers beget products, hedge funds provide liquidity for them, liquidity subsidizes the creation of new financial products, and the cycle continues. If you'd asked a banker on the Salomon mortgage trading desk in the 70s just how big mortgage-backed products could get, they would almost certainly have undercounted. Similarly, if you'd asked the founding team at KKR how big the leveraged buyout business would be within their lifetimes, they would not have given an answer that ends in "trillion."
Talent swings are more visible on a micro scale, in industries without the same feedback loops. Oil hasn't had a positive feedback cycle for a long time (it had one in the early 20th century, where cheap oil made cars a good idea and the growth of cars made drilling for oil a great idea). But in more recent decades a great way to predict the number of bachelor's degrees in petroleum engineering awarded has been to look at lagged changes in oil prices—because of the time between choosing a major and getting a degree, new petroleum engineers are disproportionately likely to graduate into an industry downturn.
These trends compound for a long time, but they don't compound forever. Timing the reversals is hard; The Alchemy of Finance, which popularized reflexivity, has sold a lot more copies than there are currency-trading billionaires, so it's an argument that's easier to be aware of than to apply. What you can do with it is to say that two things can simultaneously be true:
- Tech profits, valuations, and compensation are all moving higher due to secular trends that have a lot of life left in them—the amount of wealth that can be produced by humans interacting with computers keeps going up. At the same time,
- The rate of growth is influenced by some self-reinforcing factors, and those factors can reverse. An unsustainable trend can sustain itself for a surprisingly long time, but that any trend so durable that people round "a long time" up to "forever" is bound to reverse.
Meanwhile, it provides some explanatory power for now: yes, tech companies and the people who work for them are doing well, and it's hard to explain that purely by looking at non-industry factors. But long-term feedback loops that determine where people focus their attention and what they choose to specialize in can lead to very long-term self-fulfilling effects. The bet to make is that while there is a long trendline, and there are wobbles around that trendline, the wobbles can wobble very far indeed before they revert to the mean.
Diff Jobs is our recruiting service that matches Diff readers to interesting roles at companies looking to hire them. Some positions we're working on right now:
- A hypergrowth e-commerce platform is looking for a fraud analyst with experience building rules around suspicious transactions. They're also seeking a business analyst with Looker experience. (US, remote)
- A company helping to build decentralized services is looking for someone to join their finance team, working on financial modeling and investor relations. (US, remote)
- A high-growth company in the education space is seeking a chief of staff; it's a great way to get a look at all levels of how a growing company works. (US, remote)
- A firm that helps investors use unique data sources to find and refine investment ideas is looking for analyst/data scientist/engineers who can collect data and analyze it to add insight to trades. (US, remote)
- An insurtech startup helping to automate a very high-dollar part of companies' benefits plans is looking for multiple roles, including security ops, implementation engineers, and integration managers who understand insurance and can help product teams develop features customers need.
Fixed Costs and Airlines
Hong Kong's restrictive Covid rules have made it especially inconvenient for airline employees to operate there, and Cathay Pacific has responded by paying pilots $3,700 to compensate them for needing to quarantine ($, FT). This is a microcosm of general airline economics, where the fixed costs are high so it makes sense to pay up in order to realize a bit more revenue. It's also an interesting case study in how costs filter down: quarantining is a cost to the employees, not the companies, but the economic incidence of a tax is a function of supply and demand elasticity, not just the legal incidence; if there's a two-week-quarantine tax on flying in and out of Hong Kong, it ultimately gets paid by the company that still needs to offer those flights.
Goods Shortages and Worker Shortages
Union Pacific had worse than expected operating metrics because of Covid- and vaccination-related factors ($, WSJ). In one sense it's a sign that the response is well-calibrated if both the pandemic and its countermeasures are making things difficult. It also shows that some worker shortages will, a few weeks later, turn into product shortages. Even though a lot of the current inflation story is about people shifting spending from services to goods, creating a goods shortage, some of it is driven by the difficulty of keeping a company fully-staffed at all. On the other hand, it means that one leading indicator of inflation is, as of the last few days, in decline.
Marketing and Diversification
A few weeks ago I wrote about TikTok's plan to use ghost kitchens to offer foods that were trending on their platform ($), which I noted was not entirely crazy given that the company can use its historical data to infer how long a food fad will last. As it turns out, the idea may in fact have been too crazy after all; their head of marketing has left after that effort and some other odd marketing ploys. It's a good reminder that there is yet another theory of company diversification: sometimes they do things not because they make business sense, but because they're cost-effective PR—and since the impact of PR can barely be measured, even after the fact, those sorts of projects have a very short half-life.
SPACs are trading lower, closing fewer of their deals, and raising less than a third as much money overall as they did during the peak in February and March of last year. One reason the SPAC vogue was so quick was that it's a financial product for which it's unusually easy to manufacture supply when there's a surge in demand. During the real estate boom it took years to build up enough mortgage brokerage infrastructure to supply as many subprime mortgages as investors wanted to buy, and even during the quick-path-to-IPO peak of the dot-com bubble, it took at least a year or two to get a company to the point that it could go public. But a SPAC is just a team, a plan, and some money, and the money gets provided mostly by investors. So a SPAC boom can happen on a compressed timeline.
Tech companies in emerging markets are constantly getting their workers poached by US-based companies offering higher wages. (One defense: "Hernan López Conde, co-founder of Argentine fintech startup Digiventures, and his team have begun to recruit specifically developers who lack English language fluency.") The idea of remote work is getting more refined: it can mean visiting an office every month, being anywhere in the world but sticking with a particular time zone, or operating fully asynchronously, and that diversity of models has made some parts of the labor market more global than they used to be.
As a Peloton shareholder I am acutely aware of this at the moment: Peloton spent $420m in late 2020 buying Precor in order to get manufacturing capacity, and also opened a US-based factory to meet demand last year. As of yesterday they're apparently halting production for a while to let demand catch up (they deny this), and after hours the preannounced disappointing revenue. Figuring out the ideal amount of capacity necessary to supply a one-time buildout is a classic hard problem, and even if long-term demand stabilizes at a favorable level, it's as easy to overshoot as to undershoot, and ambiguous as to which is worse. ↩
That was part of its own feedback loop: cheaper transportation allows higher population density since food can be moved from farther away. If you put enough people in a given number of square miles, you start thinking of space in three dimensions instead of two. So railroads made it possible for millions of people to live in New York City, after which steel skyscrapers made the city livable for them. ↩
Even this has some intriguing reflexive possibilities: the companies that can pay the most are the ones whose economics are more mature, so those are the companies where a) workers are piling up the most savings, and b) where their relative impact on the business is the smallest, even if their absolute impact on the world is high. Marx would have a field day here: attrition from big tech companies is partly driven by labor alienation, but only because it’s coupled with high pay. ↩