Welcome back to The Diff. Here are the subscribers-only posts you missed this week:
- Fracking in the Age of Oil Runoff looks at how the US oil industry is adapting to a world where consumption may have peaked—and the paradox that whether the US is the world's main oil importer or the world's swing producer, its role remains the same: to provide a buffer against the fundamental instability of oil prices.
- In Learning from the Death of the Afternoon Paper, I look at a largely forgotten media transition. In 1950, 60% of newspaper subscriptions were for afternoon, not morning, newspapers. The transition to the morning paper made fortunes and led to a newspaper golden age, but it wasn't without its difficulties.
- The Gun Company Payoff Matrix explores the incredibly morbid game theory of investing in firearms stocks.
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 19,088 subscribers, up 134 since the last edition.
In this issue:
- The Road Not Taken: Stripe, Ant, PayTM & Defi
- Facebook's Outsourcing Moderation
- Amazon's Outsourced SEO
- Scarce and Free Resources
- Moderating Social (Pt. 3)
- Business Travel and Veblen
- Intel's Semi-Hack
The Road Not Taken: Stripe, Ant, PayTM & Defi
Four paths diverged: Stripe, Ant, PayTM & Defi
The evolution of the financial systems in different geographies has taken some improbable twists. To illustrate why, let’s play a matching game:
Governments in the 1990s:
1) Democratic superpower at the forefront of innovation and the commercialization of the internet—championing privatization and liberal values across the globe
2) Communist government striving for a delicate balance between market reforms and single-party, technocratic political control
3) Messy, bureaucratic democracy—known for poor infrastructure investment, dysfunctional politics, and a challenging business environment
Financial systems today
A) Private tech juggernauts capitalizing on the mobile revolution and (until-recently) light regulatory scrutiny to blitzscale closed-loop payment monopolies—largely outside the purview of the legacy system
B) State-of-the-art inter-bank transfer infrastructure produced by a government-led coalition to upgrade the legacy tech within the existing, regulated system
C) A hodgepodge of legacy infrastructure and regulation colliding with private company middleware to make 1960s technology (barely) usable for 21st century commerce
From the perspective of the 1990s, the logical matching exercise would have probably rendered: 1 -> A, 2 -> B, and 3 -> C. But clearly, history had other plans. The world’s three largest countries have diverged markedly in their financial architecture—in unexpected ways.
The evolution of payments is central. Evolution means adapting to fill a niche, so each payment system was built around the deficiencies of the environment in which it operated. Payment processing is a business where earning modest amounts of revenue requires touching large amounts of value, which means that payment companies accumulate data much faster than similar-sized businesses. Winning payments is one of the most powerful ways to create barriers to entry and network effects in a sea of otherwise commodified offerings. As software eats the world, the evolution of payments often foreshadows where value will accrue within a system.
Through the lens of key payments innovators, we can get a glimpse of the different paths taken in the US, China, and India to ready their financial systems for the 21st century. For this, we have chosen Stripe, Ant Financial and PayTM to help tell our story.
In each case, the maturity of economic development, consumer/merchant behaviors, the regulatory environment, and the smartphone revolution played key roles in shaping very different outcomes.
Stripe - the middleware growing the GDP of the Internet
The paradox of the US payments system is that because it automated early, it has dated-but-mission-critical legacy code scattered throughout the system. Bank of America started its first computer project in 1950, the Electronic Recording Method of Accounting. It started a pattern: ERMA sped up the processing of checks, sorting them 134x faster than a clerk could by hand—which made BofA more locked in to using checks. (Check payments in the US rose from ~8bn annually when ERMA was commissioned to a peak of 50bn in the late 90s, before declining.)
The space of entities that offer some form of payment—bank-to-bank, card-to-merchant, various peer-to-peer mechanisms, some run by centralized entities and some completely decentralized—is vast. Some of these entities have modern, up-to-date ways to handle payments. Many do not. Stripe is a convenience layer that allows businesses to treat “the payments system” as an abstraction, the same way you treat electricity and plumbing as things that Just Work. There was a time when starting a car, using electric illumination, or removing sewage with an intricate system of pipes and vats would have required specialized experience, and still had a high failure rate. Eventually, transferring money between arbitrary parties will work the same way, with the exception of currency controls and sanctions, and Stripe is making that happen.
Stripe is keenly aware of this. While the plumbing comparison is funnier—when payments break, it’s very messy!—we’ll stick with the electricity one. Which they do. For example, in one of their blog posts, on future-proofing the API, they say: “Just like a power company shouldn’t change its voltage every two years, we believe that our users should be able to trust that a web API will be as stable as possible.”
Even simplification requires a relentless fight against complexity. If an API is designed to make different payment options interchangeable for the user and the merchant, it needs to somehow encapsulate all the ways they are not, in fact interchangeable. As a result, the Stripe Payments API has a ratio of careful thought required per character written that approaches that of a mathematical theorem or the Lunar Plaque.
The natural suspicion about any company like Stripe—a company that earns a small cut of a large volume of commerce it touches—is that Stripe will move up and down the stack. Proprietary purchase data is, for example, a great way to seed an ad business with unique targeting data, as many brick-and-mortar retailers have realized( $).
Stripe does some of this. Stripe Capital, for example, provides working capital loans to merchants who use Stripe. That’s a direct way to apply the information from payments in a way that adds value both for Stripe, and for the merchants. On the Stripe side, there’s a capital structure arbitrage:
- Stripe’s cost of capital is low, and the effective rates on these loans are high. (The two ways alternative lending companies die are either a) running out of good ways to acquire customers, or b) making bad loans. Starting with payments makes both problems much easier.
- That cost of capital is partly a function of growth, and capital-constrained merchants are merchants who will grow faster when they’re provided with capital. Numbers are hard to come by, but at a sufficiently optimistic price/sales ratio, Stripe’s valuation may ratchet up every time they invest more through Stripe Capital.
But Stripe has other opportunities closer to the core product. Fraud detection, for example, is a direct complement to payment processing. Reducing payment failures qualifies, too. They’re both tasks that thrives on data collection and work best when many incremental signals can be applied at scale. As Patrick McKenzie put it:
I was employee #650 or so. A common worry of folks joining around the time I joined was whether there was anything left to do. Have we solved all the challenges we’re going to solve? Have we built all the things we’re going to build? Has all the fun work already been done?
This was a serious worry in 2016. It is laughable now. We have almost 3,000 employees and it feels like too few to do all the work. We have solidified a lot of our operations, security posture, resiliency, etc, and it still feels like we have huge opportunities for improvement.
And, in another post:
An example which is just a boggling fact about the world: what’s your finger-to-a-wind guesstimate about what percentage of credit card payments fail with error code I Don’t Know Sometimes Things Fail In Credit Card Land? Hint: it’s higher than you think. Those failed payments cost conversions at the margin. When Stripe fights that number down by a basis point, that creates value across our entire portfolio, forever.
A competitive moat measured in basis points doesn’t sound durable, but piling up these tiny improvements is very hard to compete with. Every time Stripe reduces one incremental bit of fraud, or onboards one more set of potential purchasers, it’s a slightly easier default for merchants. One compounding advantage leads to another. In fact, the model of very picky hiring, zillions of tiny sources of incremental profits, and better data does have one good real-world analogue; Stripe is formed from the same Platonic mold as Renaissance Technologies.
In one sense, payments remains an unsolved problem because it can be solved, reasonably well, for certain sets of transactors. Because a workable solution can be invented, the world is full of mutually-incompatible systems. But the default expectation of every business is that they should be able to accept money from customers, and the default expectation of customers is that they should spend their money, so Stripe is gradually, but thoroughly, adding a general solution on top of a lumpy, heterogeneous class of problems.
Ant—the aggregator that swallowed the system
Perhaps the antithesis of Stripe as the “platform of platforms” - removing friction within the legacy financial system - is Ant. Ant primarily removes friction not from within China’s system, but by pulling as much activity out of the legacy system as possible. Ant is China’s US$17t financial aggregator.
As opposed to the US, where digital card-based networks proliferated before the commercialization of the internet, China’s own network, UnionPay, ran into adoption issues on the merchant side. Despite a highly banked population, merchants were not keen to stomach even 1% processing fees or upfront terminal costs. Cash was king, but still inconvenient as the largest notes were just ~100 yuan, or about $16. Credit bureaus also remained immature, limiting financing value-add from the network.
The door was left open for alternatives.
The geopolitical backdrop during the rise of Alibaba and Tencent is important to note. After the economic turmoil of the 20th century’s civil war, the Great Leap Forward, and the Cultural Revolution, China’s reforms were still in progress and economic development well behind the US & Europe in the early 2000s. Much of offline retail was (and still is) micro-sellers with ~1/10th the formal offline retail footprint of the U.S. Furthermore, the leadership remained deeply suspicious of U.S. tech giants.
This was the vacuum that Alibaba stepped into: an under-developed offline retail footprint, with 1b+ rising consumers, a favorable regulatory climate, global capital inflows, and a seemingly unlimited TAM of whitespace for digital offerings. To say the team capitalized is an understatement.
With this backdrop, eCommerce erupted. The marketplace commerce model matching the fragmented merchants and consumers online quickly reached escape velocity in a world barred from outside competition. And with the arrival of the smartphone, eCommerce went parabolic.
If history has proven anything, horizontal eCommerce marketplaces are by far the best use case from which to launch a thriving payments ecosystem (eBay -> PayPal, Alibaba -> Ant, MercadoLibre -> MercadoPago, Shopee -> SEAmoney). The AliPay mobile app launched in 2009, just in time to ride the mobile revolution. Leveraging its massive (and rapidly growing) network of consumers and merchants, the company blitzscaled a novel closed-loop payments network to >1b users within a decade.
Slower-moving SOE banks struggled to keep up. SME merchants were hostile to the network fees. Consumer behaviors unsolidified. Smart-phones proliferating. Regulators keen to build local champions. Dominant, well-funded internet giants with unprecedented distribution. It all happened so fast. With abundant capital, the digital sphere became a land grab.
The “super-app” was born.
But central to every super-app is the customers’ wallet. Thus began the massive campaign to build out an offline presence. Through aggressive marketing campaigns, no upfront hardware costs, and breakeven merchant discount rates, Ant and Tencent pushed payments beyond their sprawling ecosystems to offline use cases. Not only were Alibaba and Tencent’s own offerings expanding, but the wallets would be accepted at more 3rd-party locations meaning less need to transfer money out of the ecosystem.
The coup of the legacy financial system was nearly complete. With more and more transactions in the ecosystem, Ant was better positioned to underwrite credit or facilitate other financial services - such as loans, insurance, and wealth management.
And that’s the gist; The financial system quietly wrestled away from the most powerful communist technocracy on earth by a private entity. A tech company blitzscaling a closed-loop payments network on the back of the mobile revolution to empower US$17 trillion in payments across 1 billion users, 80 million merchants, and 2k+ financial institutions—while taking a rake on an increasing number of financial services. The “capillaries” of the Chinese financial system appear to be more of a 21st century replacement.
Ant’s IPO was hotly anticipated, but just a few weeks before it was priced, Jack Ma gave a speech in which he argued that legacy banks had a “pawnshop mentality.” This was, in context, less of a damning claim about their business than it was a bluntly-worded claim that lending based on collateral rather than ability to pay was an antiquated model.
To be fair, Jack Ma may have been correct in his speech. Ant has capable data scientists and, more importantly, an unprecedented quantity of transactions and consumer behavior data under one roof. The combination means Ant can potentially underwrite risk more effectively than any prior financial intermediary in history.
However, the amount of control ceded to a private entity largely outside the bounds of the existing system is being questioned. While the shelving of the Ant IPO was a major blow, regulators have been trying to pry open the payment network for years - first announcing designs for “Wanglian” (loosely translated: internet payments union) in 2016.
Aside from anti-competitive concerns (~90% market share in digital payments for the two giants), regulators are keen to avoid a consumer credit bubble and want to ensure Ant has proper risk-sharing in loan distribution. With Ant’s IPO shelved, more strict capital requirements looming, and the DCEP in pilot phase - Chinese regulators are looking to reset the balance of power.
Perhaps they will look to India for guidance…
PayTM—the aggregator that never was
Peering across the border at PayTM in 2015, Ant must have felt a tinge of deja vu. Another market of ~1.4 billion consumers ~10 years behind China with rapidly accelerating smartphone adoption and a dysfunctional legacy system. With 23 million users relative to Ant’s 190 million (in 2015), Ant paid US$550m in exchange for a 25% stake in the company set to replicate its playbook next-door.
Without a dominant eCommerce platform to piggy-back on, PayTM relied heavily on promotions & cashbacks to build out its network—luring customers out of the banking system and into its own eWallet. Early traction was promising. The next year, PayTM received an unexpected boon from Modi’s demonetization announcement and digital payments erupted. The Ant playbook on the verge of a second coup.
However, that same year, an unexpected challenger emerged. The government.
PayTM founder Vijay Shekhar Sharma wasn’t alone in realizing payments in India were in need of a serious upgrade. Legacy infrastructure, data silos, deferred settlement, and loose regulation plagued Indian payments and weighed on productivity. In 2007, the Payment and Settlement System Act paved the way for the creation of the National Payments Corporation of India (NPCI) - an entity tasked to tackle this very problem. In 2016, after a concerted effort from regulators to push banking penetration from ~53% up towards 80%, the NPCI orchestrated the launch of the Unified Payments Interface (UPI) in India which would change the balance of power dramatically.
As opposed to standing back and letting eWallets like PayTM burn their way into closed loop monopolies, Indian regulators took a forward looking approach to revamp their payments infrastructure for the mobile era. UPI is an interoperable, real time inter-bank transfer network with open API’s on which 3rd party fintech providers can build. The closed loop eWallets like PayTM were instantly upended by cheaper (initially free), more convenient (no need to load a wallet & less strict KYC requirements), and more ubiquitous payment options (bank network of 207 banks).
The open nature of the network fostered competition as opposed to the walled gardens constructed in China. PayTMs eWallet model was dethroned. Google Pay, PhonePe, Amazon Pay, Whatsapp have joined the fray - with UPI adoption skyrocketing (>2b trxn per month in October 2020). To protect against antitrust concerns, the NCPI has implemented a ~30% cap on market share of UPI volumes by any one 3rd party - suggesting that payments (and the financial services which tend to follow) will remain relatively fragmented.
Because there is no online/offline distinction as all payments are done securely via mobile-enabled inter-bank transfer, UPI avoids many of the complexities of the U.S. offline vs. online payment conundrum which created a market for companies like Stripe.
Increasingly, UPI is held as the gold standard for modern payments—a seamless, secure, digital payments system not co-opted by tech companies, but firmly within the confines of the existing financial system.
Both China and India are being held up as models for what 21st century finance should look like. However, there are tinkerers on the periphery wondering if centralized institutions are needed at all.
Governments and tech companies are not the only entities hoping to build a better financial future. The cryptocurrency movement—kicked off with the launch of Bitcoin in 2009 - is a promising hotbed of open financial innovation which aims to dethrone many of the legacy gatekeepers. Many millennials have grown wary of the traditional banking system and central bank monetary shenanigans. The vision for open, transparent, and trustless exchange resonates with younger cohorts.
If able to overcome scalability impediments, blockchains may hold the key to another round of upgraded financial architecture; correcting the internet’s “original sin” by baking in digital scarcity. Exchange without the baggage of intermediaries: cards, gateways, processors, networks, or even banks sitting between you and the counterparty collecting a fee.
One of the more exciting (very early) spaces today is decentralized finance (Defi) largely built on top of Ethereum. The ecosystem hosts a variety of early decentralized financial applications—exchanges, lending, insurance and even derivatives—parallel to the existing financial system.
Decentralized finance will clearly draw the eyes of regulators as adoption grows which may dampen the “censor-proof” appeal of regulatory arbitrage. However, many qualities - such as open-source development, global liquidity pools, limited intermediaries, and novel financial incentives to bootstrap networks make crypto a real threat to the old guard as it matures.
Decentralized protocols are hard to regulate, but the onramps and offramps are relatively easy to regulate. For now, the de facto legal standard is that all sorts of behavior is permissible in crypto-land—but the proceeds of that behavior can’t be realized in non-crypto form.
Because DeFi is relatively hard to regulate, it tends to be most attractive for things that regulators frown on. There are plenty of DeFi applications that are economically equivalent to margin loans, for example; the demand is there because the traditional financial system doesn’t encourage retail investors to speculate with high leverage, and it discourages that for a reason. But plenty of other innovations start out in a regulatory gray area and then scale into more legitimate ones.
The interesting thing to think about is who will be best positioned for this new wave of financial innovation. Clearly, China appears to be ahead of the curve with its central bank digital currency—the DCEP—already in pilot. However, much of the open, more libertarian ethos of crypto appears to be antithetical to the Chinese government’s vision and pivot towards tighter political control of the private sector. Digital currencies can be anonymized, but they’re also much more traceable by default. India’s new roll-out of UPI is still in its early innings of propelling digital penetration in India - I imagine they will ride those rails for some time before thinking about another upgrade.
The interesting question is if the U.S. system’s frankenstein solution will make regulators (and consumers) more open to a system overhaul in the coming decades. Similar to an ERP system where you don’t rip it out until you absolutely must, perhaps the US financial system is starting to reach the point where a system overhaul makes sense. While companies like Stripe are doing amazing work to make the current system more frictionless for users, what if fully shedding the old system is what the U.S. really needs to retake the mantle of financial innovation. Replacing legacy systems with perfect fidelity often means replicating every single flaw, flawlessly (the Unix convention of starting configurations files with a dot was something between a hack and a bug, for example). Like China and India with card networks, perhaps the U.S. will “leapfrog” the centralized mobile payment revolution and have a head start in the coming decentralized future.
As the last 30 years have shown us, the intuitive predictions are usually wrong. Much of the economy is mean-reverting, but some kinds of entrepreneurs are the opposite—they fill in gaps, or build systems that end up copying what works in other places, even if it’s designed independently from scratch and then evolves due to market forces. Finance is an industry that makes abstractions tangible, and sometimes a long chain of deductions from a commonsense set of abstract principles leads to some very weird results.
Note from Byrne: I have an abundant list of qualitative and quantitative conflicts of interest here, since I a) do business with Stripe—if you subscribe to this newsletter now, they’ll get a cut—and b) am friendly with a number of people at Stripe, some of whom do business with me by subscribing. I did not discuss this piece with anyone at Stripe, though.
Note from Pondering: Damn. I wish I had friends at Stripe.
A Word From Our Sponsors
As economic activity comes online, new data trails are left behind. Over 200 innovative investment firms leverage this data for alpha-generating insights. Thinknum’s alternative datasets power investment decisions across more than 500,000 public and private companies around the world.
For example, track job listings to identify which companies are expanding in real time. Or track the individual tenants by commercial mall to analyze churn on a daily basis.
Interested? Come learn about all of our 35 datasets here.
Facebook's Outsourcing Moderation
Facebook executives have said, implicitly and explicitly, that they absolutely do not want to be in the business of deciding what kind of content should be allowed online. (Given the size of Facebook's moderation team, and its corpus of banned content, Facebook may be very excited about being in the business of helping the rest of the Internet enforce codified rules.) One tool Facebook has, ahead of formal regulation, is their Oversight Board, a body that Facebook defers to for difficult moderation decisions. They're asking the board to decide what to do about Trump. As with other political processes, part of the point is not just to have a procedure for making difficult calls, but to have a procedure that's hard to argue with. Facebook can plausibly tell a potential boycotter that they've left the decision completely up to the board, and that their hands are tied.
Amazon's Outsourced SEO
One of the most economically important search ranking algorithms in the world is the one Amazon uses to select which products are displayed for generic searches. There is a lot of money to be made in being the #1 result for "TV" or "Laptop" on Google, but Amazon's customers have a higher propensity to spend; their searches may, in the aggregate, be worth even more. Amazon recently updated its ranking algorithm, to give better rankings to products that get more traffic from outside Amazon, including from Google search. This is a very interesting way for Amazon to invest some of its demand-directing ability towards rewarding merchants on Amazon who do marketing for Amazon. It might temporarily hurt their growth (if Amazon is weighting outside traffic more, it's weighting factors like reviews and total purchases proportionately less), but it's an interesting way to passively move up the sales funnel.
Scarce and Free Resources
Parler's decision to work with a somewhat dodgy Russian hosting (Edit: DDoS protection) company has had an interesting side effect: the company, DDoS-Guard, got caught manipulating the government of Belize into giving them IP addresses. (DDoS-Guard denies this.) IP addresses are an interesting category of good. At first, they were worthless, but eventually they were quite scarce. Allocating a resource that is effectively unlimited but will eventually run out is a challenging problem. The United States ran into that problem when the resource in question was the United States itself. The trouble with expensive address space is that it's exclusionary, and the people who come latest are likely to be much poorer than the people who bought early. The trouble with so-cheap-it's-free space is that it leads to more spam and abuse. The middle ground tends to involve privatizing the space and selling or renting it over time. More random allocation methods, as the DDoS-Guard case shows, can lead to perverse outcomes.
Moderating Social (Pt 3)
A generally unpopular politician with an energized base of supporters, who claims the most recent election was fraudulent, has taken to social media to encourage his fans to protest in his support. Naturally, there are calls to ban these incitements from major platforms like TikTok and Vkontakte ($, FT). This is another case where country-level speech restrictions actually represent a devolving of authority, from multinational social networks to individual countries with varying standards. The Russia sitaution is a special case—Navalny's protest comes after he survived an assassination attempt. But every country is going to see its speech curbs as a special case, and will expect platforms to be better at blocking content than the government itself is.
(Meanwhile, in India, there's another special case: a Prime miniseries contains scenes viewed as disrespectful to Hindus ($, FT). Amazon has made edits, but more legal action is promised.)
Business Travel and Veblen
TripActions has raised $155m at a $5bn valuation, a premium to their 2019 valuation, and their lead investor argues that business travel will rebound. United, too, says that travel will come back, and that demand will overshoot supply when it does ($, Globe and Mail). Business travel is one of those defaults that could potentially reset to a lower level after the pandemic. If consultants find that they're almost as productive working offsite, a return to on-site work will put them at a cost disadvantage. But for some categories of travel, the cost advantage is the whole point; as Alex Danco has noted, part of the point of business travel is for a salesperson to say "I'm so confident you'll buy our product that I set a bunch of money on fire just to pitch it to you in person." If the supply of business-class seats is lower, the 2022 equilibrium might be that business travel happens less often, but when it does it's actually more expensive per seat.
Intel released their earnings yesterday afternoon. And, by accident, they also released them earlier in the day. Intel describes this as a hack ($, FT) which gave some people early access to an infographic summarizing their quarterly results. But it was not especially sophisticated: Intel had an infographic for their Q3 earnings, in a file that ended with "Q3_2020_Infographic.pdf" and had a URL with a sequential numbering scheme. Q4's earnings presentation had the same file naming scheme, so it was easy to guess.
This kind of thing happens from time to time, and it's an interesting edge case in US securities law. Technically, the information wasn't misappropriated; no one at Intel violated a duty to keep it confidential in exchange for some consideration from a trader. But in practice, the technicality matters less than appearances. Because it looks like insider trading, and fits the broad definition of hacking, trading based on the possession of this infographic is a poor risk-reward even if it turns out to be legal.
(If this scenario and analysis strikes longtime readers as familiar, it's because I used this exact situation as a hypothetical example last year, in the Alternative Data Primer ($).)