Business Model Meta-Models

Plus! Fed Maximalism, CFOs, TikTok, IPOs vs Direct Listings, Death and Base Rates, Rapid testing, Wuhan, Politics in a One-Party State

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 11,558 subscribers, up 456 week-over-week. This week’s other posts focused heavily on a spate of interesting upcoming offerings:

This week’s subscriber call was a success, with lots of great questions and follow-ups on IPOs, monetary policy, reading lists, and more. I’ll be continuing to do a weekly call with paid subscribers. To join next week’s call, access the full archive, and receive all five weekly issues of The Diff:

In this issue:

Business Model Meta-Models

When you talk about companies, you probably use the term “business model.” Everyone knows roughly what it means. It’s the answer to three questions: What does the company do? What value does it produce? Why does it keep on existing?

But there are many ways to frame business models. Every one of them asks how a business reverses economic entropy and creates more value than it consumes (or at least does that for the people who could choose to shut it down). None of the models are wrong, but all of them are limited. And when one model stops working, it’s important to switch to the next.

The godfather of mental models for businesses is return on investment versus cost of capital. This model is a way to ask the existential question every business faces: is now a good time to grow, or a good time to gradually wind down? Some companies establish an ROI vs cost of capital fairly early. A fast food franchise, for example, can figure out unit-level economics at one store, test their concept in a few markets, and have a pretty good sense of what growth will look like over the next decade. Other businesses have this model embedded within their existing business. The modern Amazon grew out of a series of moonshots, but each moonshot gave Amazon the option to make incrementally profitable investments: one more fulfillment center or delivery station ($) reduces delivery costs and accelerates delivery time by a predictable amount, and Amazon’s cost of capital is also easy for the firm to estimate.

Internally, many companies use this kind of analysis to make investments. Oddly, they typically choose a required rate of return well above their cost of capital, and they don’t tend to update this hurdle rate in response to changes in their cost of capital. Hurdle rates were 15-18% in the mid-80s, when ten-year treasuries yielded almost 12%. In the early 2010s, with the 10-year rate below 4%, hurdle rates were still around 15%.

This model works well for mature businesses, but it breaks down for others. If you run an independent software company, what are your “assets,” and what does your capital cost? You could invent hypothetical values for each—your assets are the skills you apply to the business, your capital is the wage income you forgo when you run it—but those are all hypothetical.

Loops: I was first introduced to loops through Kevin Kwok’s blog. Alex Danco has also written about the loop metaphor, using yet another metaphor. It’s metaphors all the way down.

Loops are best for understanding companies that are growing their consumer surplus first and revenue second. The loop approach is to ask how the company a) continues to add value, and b) accelerates the pace at which it adds value. And the convenient thing about them is that they work both for analyzing how a company ramps up its revenue potential and how it extends that into revenue itself. Take Facebook: their initial loop involved getting users to sign up and connect with their friends (Facebook’s biggest source of growth in the early days was getting users to upload their address books.) Once someone has a handful of friends, they’ll keep logging in, and adding content. And once that happens, their friends are more likely to use it, too.

That loop explains usage, but not monetization. But monetizing ads on an auction-based platform has loops, too. When an auction is competitive, prices are set by the margins of the top bidders, but when auctions aren’t competitive, prices are set at whatever the lowest amount the advertiser will accept to display an ad. So Facebook’s advertising loop is: find cases where an advertiser is paying very little money for lots of traffic, then find another advertiser who wants that traffic, so the traffic is priced closer to its value. Then, find out what inventory that advertiser is bidding on, and repeat the process. Every advertiser getting a great deal on Facebook was an opportunity for Facebook’s ad sales team to tell prospects how cheap a certain demographic was. This loop has the added advantage that low bidders on ads are often bottom-of-the-barrel direct-response products—nearly-fake supplements, scammy subscription offerings, spy cameras (in the early 2000s) and “parental control” spyware apps (later on).

While loops explain the direction of a business, and are a helpful way to compare two companies that compete for the same time or budget, they’re a non-quantitative tool. And they’re much better at estimating how many users or how much time a company could capture than how much money. Loops might be the best way to understand a company after it’s achieved product-market fit but before it’s mature, but they won’t get you to a number.

Comps: I include this one for completeness. It’s common because it’s lazy. One way to understand a business model is to say that it’s X for Y and let the listener imagine the details. It’s Facebook for knitters! It’s eBay for space ! It’s Intel minus fabs or Intel minus everything except fabs! It’s OnlyFans for people who own clothes!

The comps meta-model is implicitly a work-in-progress, because the natural question is “If it’s X for Y, why doesn’t X just do it?” And the answer is usually that X does it, or that X doesn’t do it because it’s a meaningless niche market, or because it’s really not X for Y because Y is so different from what X does. The online travel agency business is great, for example, but there’s a reason you can’t copy-paste it into mortgages, groceries, or data plans. (In fact, Priceline tried, but gave up and doubled down on travel by buying, which was one of the best tech acquisitions of all time.)

A comps-based model can be a good jumping-off point for understanding why two similar-sounding businesses are deeply different. Why isn’t Airbnb just the eBay for space? One reason is that Airbnb’s inventory gets renewed every night, and their demand indirectly does, too. Another difference is in when and how two counterparties to a transaction can disappoint each other: the home might not be as-described, and stay might be a destructive houseparty. On eBay, the most either side can rip the other off for is the cost of the goods sold, but there’s nothing stopping a $200 Airbnb vacation from causing $20,000 in property damage.

Supply chains and alternating commodities: A model I use a lot is to look at supply chains, both literal and figurative. There’s a literal supply chain that starts with raw materials coming out of the ground and ends with consumer goods. There’s also a supply chain of knowledge: someone gets a clever idea, pitches it to their boss, and twenty years later it’s a standard industry practice. Or someone tweets a novel argument, it gets picked up by reddit and Twitter, and 12 hours later it’s on cable news.

The supply chain model is powerful because supply chains, for complicated reasons, tend to have alternating layers of commodity and monopoly. Mobile games are a commodity, the app store is a monopoly. Media production is mostly a commodity, Netflix is close to a monopoly. Starbucks buys coffee beans—a literal commodity with an active futures market—and turns them into a brand-name product. And this means that any change anywhere in a supply chain can ripple through and end up reversing the commodity/monopoly status of individual products. When Microsoft was founded, the theory was that hardware was a commodity, and as it got cheaper it would make software relatively more valuable. Today, Microsoft likes to cut the prices of software products like Github and Visual Studio in order to encourage consumption of Azure—it’s using cheap commodity software to sell more hardware!

This one’s tricky because it relies on the implicit assumption that you, the business analyst, understand the business as well as the people running it. If you can accurately predict the commoditization of a category, you can make a lot of money, but the company that gets commoditized probably spends a lot of its time worrying about the exact problem you’re tracking. Companies occasionally get blindsided, but more often they make a judgment call that turns out to be wrong. Exxon was studying the impact of global warming in 1979, so they probably beat you to it—they just decided that, all things considered, it was a better business decision to keep on emitting carbon.

TAM(?) x Share(??) x Margins(???): There’s a story that Victorian textile magnates imagined how much they’d make if everyone in China wore shirts one inch longer. (The earliest example I can find of this story is from a rather prescient New Yorker piece.) This is part of a general model: estimate the size of a market, estimate how much of a market one company can capture, estimate their margins, and now you know how big the company can get.

This is incredibly tempting, at least as a way to put a ceiling on the upside case, but it omits every important question: how big is “the market”? How do you capture share, and why doesn’t someone else capture share? And what margins are sustainable at what scale? Every market already exists in some kind of steady state, where the incumbents think the marginal profit from adding share isn’t worth the cost. So this high-level analysis skips every important detail. It can help if a company is considering some investment that has a certain cost and a known effect on long-term margins—then it’s a way to see if that cost is plausibly worthwhile. But any time you multiply three numbers together, biases get cubed, too. The companies that make this sort of estimate are typically early-stage, and don’t have other metrics to use, but if they succeed they often end up redefining the market. (When Airbnb put together their seed round deck, Sam Altman told them to change their market size from millions to billions, which turned out to be right.)

Company as an options trade (long calls or short puts): An option is the right to buy or sell a security at some fixed price at a future date. An oil company is a collection of opportunities to convert production expenses (a fairly fixed cost) into oil (with a variable price), so an oil company is really a bundle of options. Other companies get described this way, too: a pharma company has a portfolio of options on drugs, some of which expire worthless and some of which are quite lucrative.

This model is helpful at one thing: measuring how much a company benefits from or suffers from volatility. In oil, for example, a high-cost producer is an out-of-the-money call option; when prices hit extreme highs, they benefit, but at lower levels they’re unaffected. Some companies are actually short options, like legacy companies in a business with a major competitor, or any business that’s just waiting to get banned. Juul, for example, was both long a call option (on the shift from burning tobacco to inhaling vapor) and short a put (against getting banned).

Some companies are more literally an option. For any deeply indebted company, the equity represents a call option on the value of the assets. Hertz equity is probably worthless, but if the company’s fleet somehow appreciates in value enough to pay off their debt, the equity captures all the upside past that point.

Where this model breaks down is when it applies a simple financial model to dynamic situations. I cheated a bit when I said the price of extracting oil is fixed. Actually, the cost of all of those inputs—roughnecks, drills, trucks, frac sand—goes up right along with energy prices, albeit usually not enough to ruin margins. This doesn’t ruin the model, but it certainly complicates it.

As a general rule, you can make a fortune buying options when they’re too cheap, but you can make a living selling them when they’re too expensive. It depends on whether you have higher tolerance for continuous suffering or occasional bouts of excruciating financial pain.

A multiparty negotiation between labor, capital, management, shareholders, government. This is an especially useful framework when a company is getting in trouble by way of one of the parties above. The model assumes that all businesses create more value than they consume, and then asks the distributive question: who gets what? Some companies seem to be run on behalf of employees, like investment banks in the early 2000s, some dot coms in the 90s, and heavily unionized companies in the 50s through 70s. Outside the anglosphere, companies are run with more input from government or labor—in China, some behavior that does not exactly accord with Maximizing Shareholder Value makes perfect sense if the CCP is ultimately calling the shots.

This model is somewhat similar to the supply chain model, in that it looks at the status quo and asks what will change. If one side in this ongoing negotiation gets more powerful, the next question is to ask who they’ll take advantage of. You could imagine everyone playing nicely, but they face strong internal constraints: if a union has the opportunity to push through a pay raise, or management has a chance to engineer a pay cut, they face immense pressure to do it.

All models are wrong, some models are useful. None of these models are perfectly descriptive. Every one of them breaks down in some situations. But every one of them can answer important questions about a business. These frameworks don’t provide answers, but they do help ask the right questions.


I have a piece in Palladium looking at India’s TikTok ban and digital sovereignty. The fixed and marginal costs of sovereignty affect how centralized the world is. When one cost of sovereignty is building a country-specific social media and telecom infrastructure, some places won’t be able to afford it. This is nothing new; most countries couldn’t afford nuclear weapons programs, either, so they ended up in the sphere of influence of someone who could. And in Marker, a piece on how DraftKings survived without sports. One reason: the gamblers who would have bet on games are betting on DraftKings stock instead.

Fed Maximalism

Yesterday, the Fed announced a sweeping, unprecedented change in policy, dramatically shifting their inflation target from 2% to 2%.

That’s a bit unfair: what the Fed has done is switched from targeting 2% inflation at any one time to targeting 2% over long periods. In the current context, that means they’d be aiming for inflation above 2% until they’ve compensated for the long period of below-target inflation in the past.

One theory on forward guidance like this is that it’s a powerful form of leverage for central banks: if they take an action that’s perceived to be temporary, its response will be weak, so long-term commitments make more sense. But long-term guidance is only credible if it’s carried out even when the optimal choice is to do something different. The way central banks buy credibility for future forward guidance is to pursue suboptimal policies that line up with prior guidance.

The Fed has also changed its attitude on unemployment, essentially saying that there’s no good way to estimate “full employment,” so they’re not going to treat low unemployment as evidence that the economy is overheating.

All of this would have been very meaningful at the beginning of the year, but doesn’t represent a meaningful policy change at the moment. Since unemployment is clearly far from full now, and since recoveries take a long time, it doesn’t imply a different policy over the next few years. Past that point, it might hypothetically mean higher long-term inflation, but forward guidance is hard to trust that far out.

An interesting counterfactual to consider is: what would this approach have implied in the past? During the Long Disinflation after the early 80s, it would have required rates to stay higher for longer. This would have led to an alternate history for the 90s, where tighter monetary policy would have required looser fiscal policy (no more Clinton surpluses), which would have led to a very different late 90s boom.

CFOs Vocally but Passively Worried About Valuations

A new survey from Deloitte says:

84% of Fortune 500 CFOs say the US stock market is overvalued, according to a survey released Thursday by Deloitte. That’s up from the 55% who felt that way a quarter ago. Just 2% of finance chiefs say US stocks are undervalued.

CFOs' opinions matter, but they matter when they’re expressed through market activity. So far this year, IPO proceeds are up 7.8% year to date—which does not exactly comport with a rush to supply public investors with as much equity as possible to exploit overvaluation. Whenever an asset class seems overpriced, the right question to ask is “compared to what?” And in this case, it looks like private companies are overpriced, corporate debt is overpriced, sovereign debt is overpriced, etc. And if everything is overpriced, that’s another way to say that rates have declined. It would be nice, at least for savers, if everything were cheaper. But what the market is saying is that the supply of savings exceeds the demand for investment, so the market price of capital is cheap.


At the start of the month, I decided that by process of elimination, Microsoft was the only plausible bidder for TikTok. This was a failure of imagination on my part, and a triumph on the part of investment bankers. Now, bidders include Walmart, which wants to turn it into an e-commerce place, Oracle, which has given more thought to how to pay than what to do with what they’re buying, and assorted consortia. These competing deals will, of course, raise the price of TikTok. But they raise its value, too, by making it more likely that at least one deal will go through. Since advertisers have been pulling back and some creators have defected to other platforms ($, WSJ), anything that makes TikTok more likely to exist in a year makes it more valuable today.

A TikTok deal is allegedly imminent. And in other TikTok news, their head of security gave a very cautious interview. He’s in a difficult position: from the outside, it looks like TikTok wants to be just another global consumer Internet brand, without any geopolitical or espionage-related implications. But the legal environment under which their parent company operates precludes them from guaranteeing that.

Coming Soon: Experimental Data on IPO Unfairness

The NYSE has gotten regulatory clearance to allow direct listings that raise money, so now we’ll be able to test the core empirical claim of IPO critics. The anti-IPO line, most often articulated by Bill Gurley, is that IPO fees and the IPO “pop” transfer money from companies to their bankers and their bankers' favored clients. The counterargument is that price discovery for a newly-public company is an expensive, risky prospect, and the two parties doing the work and taking the risk are getting paid for it. Direct listings with fundraising will help us figure out who is right, although there’s some selection bias. The companies that are relatively easy to understand, either because they’re a straightforward business or because public investors bought the stock when the company was still private, will be more likely to direct list. While more complicated and hard-to-sell companies will probably default to the old process. Still, it’s progress. I’m sure bankers and Gurley would privately agree that there are some companies that benefit from an old-school IPO and some that could do without. Now they both have the option.

Death and Base Rates

My favorite actuarial blog has a post on excess deaths and risk perception. The hook is a recent survey showing that people overestimate the risk of Covid mortality for young people, and underestimate it for old people:

On average, Americans believe that people aged 55 and older account for just over half of total COVID-19 deaths; the actual figure is 92%. Americans believe that people aged 44 and younger account for about 30% of total deaths; the actual figure is 2.7%.

But as it turns out, this is more of a statement about general mortality than about Covid in particular:

While 92% of official COVID deaths are of those age 55 and older, in 2017 (the latest year with completed death data), 87% were of people age 55 or older. Five percentage points aren’t much different. The percentage of deaths of those age 44 and younger in 2017 were 7%. Again, about a four to five percentage point difference with COVID deaths. Did you know how high the first one was, and how low the second? In just a regular year? No, you probably didn’t. I am not going to point and laugh at you Nelson-style, because there’s no reason to expect that you would know that.

This isn’t action-guiding information, but it should calibrate your confidence in understanding Covid mortality. The first step to understanding new data is to understand the baseline.

Rapid Testing

Paul Romer has long argued that extensive nationwide testing would be a near-substitute for a vaccine. It’s taken more time than expected (the title of the article linked above begins with “The only way to get back to normal this summer…”), but cheap, rapid tests are now available, and the government has purchased 150 million of them for $760m. A continuous testing regime has a cost in compliance as well as implementation, and people are resistant to repeated inconveniences, so it’s not as effective as an actual cure. But a 95%-and-slowly-declining solution is a great intermediate step in getting to a ~100% solution.

The Other Wuhan Story

Back in March, I flagged a story about China’s chip manufacturing in Wuhan. Despite the ravages of the pandemic, they were continuing to construct new chip fabs. A global disease didn’t stop the work, but funding problems did. It’s an interesting situation in light of China’s increasing efforts to push speculative capital into technology companies. High savings and skilled workers give China many opportunities to invest in new, strategically important fields, but even when capital is abundant, there’s still competition for it.

Politics in a One-Party State

Scholars Stage selects some highlights from David Shor’s amazing electoral brain-dump, with a novel angle. Shor has talked about how parties benefit when the salient issues are the ones voters trust them on, but this may apply outside of democratic systems:

An interesting question, which I can only speculate on, is whether this same framework applies to authoritarian states where competition between alternating partisan blocs is banned. My first take at the question is that it does. There are some issue sets that the Chinese populace does not have much confidence in the CPC’s ability to manage well (environmental issues and pollution being perhaps the most obvious). Over the last few months American leaders have been careful to draw a distinction between the Party and the people in an attempt to put additional pressure on Chairman Xi and his team. But if they have done this while simultaneously centering Sino-American conflict on issues whose issue salience is favorable for the Party, they may end up increasing Chinese confidence in the Chairman.