Clean Trades and Dirty Hedges

Plus! Cash as Cosmetics; Two Inflations; AI; Offline-to-Online; Starlink; Diff Jobs

Clean Trades and Dirty Hedges

Active investing is one part determining something that's misunderstood about the world and another part finding the best way to express it. Different kinds of investing will have a varying mix of these. It's obviously pretty easy to express a view like "Apple is overvalued" by shorting Apple, but a view like "there will be a persistent oil shortage" might need more work: is it better to buy oil futures directly, or invest in an oil company. And if you're investing in an oil company, how do you think about the difference between high and low production costs (more operating leverage if prices go up, but higher odds of bankruptcy if operating costs rise and prices are high-but-volatile), geographic concentration (weighing ESG risk in developed countries against expropriation and instability risk in developing markets), capital return policies, etc. In fact, there are some jobs within the financial sector that can be described entirely as trying to convert directional thematic views into specific trades:

In fact, at a high level it starts to look like broad theses are a commodity, which investors give away for free in order to sell access to their skill at implementation. Everyone has an opinion about inflation, but an investor making a credible case that they can cost-effectively bet on inflation—buying oil instead of gold, for example, or figuring out that consumer durables inflation was both a cause and side effect of retailers loading up on inventory—has a real pitch.

But this doesn't just apply to pure financial decisions. It's a question that comes up in job choices all the time. If you're at a good company in a lousy industry, or vice-versa, how do you weigh one or the other? If you're optimistic about an industry but recognize that it's winner-take-all, is a job offer from the #3 player an exciting opportunity or a waste of time?

Agustin Lebron's The Laws of Trading has a nice line about this: "Take only the risks you're being paid to take. Hedge the others."

Understanding this requires stepping back and thinking of two ways to look at markets: the natural approach is returns-first, where different investments can create different levels of wealth over time, but also come with different amounts of risk. Keep all of your money in inflation-protected TIPS and you'll sleep well, but you won't get rich. Sprinkle your money into lots of early-stage investments and most of the news you get will be bad news, but you might eventually hit it big. But many traders flip this around, and use a risk-first approach: variance is intrinsic to the economic fundamentals of different investments, and returns are what you get paid for absorbing various kinds of risks.

This latter approach is mentally healthy for two reasons. First, by front-loading the bad news it means that you get some of it out of the way ; you know that some investment decisions will lead to a bumpier ride. But more importantly, it means looking at any attractive-seeming investment and asking the question: "Why is someone willing to pay me so much to take this risk?" In a perfectly efficient market, there would not be any good investments, and there's a lot of brainpower devoted to spotting and closing arbitrages, so your first question about any attractive risk/reward proposition is: what risk am I missing?1

So when you're looking at a job, especially the kind of high-variance job that early-stage startups offer, the place to start is with an acknowledgement that the upside—what your options would be worth if the company hit it big—is entirely there to compensate for the downside that it probably won't work out.

You can break the high-level risk down into several more specific ones, around the market size, the company's ability to execute, and what the long-term economics of the business are. And here's where things start to get interesting. Many companies and their investors like to talk about "flywheels"—more TikTok videos means a bigger audience means more creators means more TikTok videos; Disney IP creates more Disney+ subscribers and more theme park attendance, funding more IP; searches create data that helps to provide better search results, leading to more searches; etc.

A "flywheel" is just a description of how the expected joint probability of getting several important things right rises as you get more of them right. For example, one negative thesis on a given startup might be that even if the industry they're in grows, they may not be positioned to capture most of the value created. What if they end up providing a commoditized input to a business that's dominated by a consumer-facing brand? Or what if they're one of dozens of consumer-facing solutions to a problem that really amounts to reselling a technically challenging product with few producers. (In other words, one risk is that the business ends up selling the sugar to high-margin Coca-Cola and Pepsi, while another version is the risk of being one commoditized PC manufacturer among many where most of the gross margin goes to whoever makes the chips.)

But a company that's growing fast in a new industry can mitigate that risk because it has more visibility into where the hard problems and pricing power reside. Look back far enough in Google's history, and you can see a company that thought it was selling an enterprise product but that turned out to be creating a standalone customer-facing brand. Shopify would be a much smaller business today if it were a snowboarding site with a ridiculously well-made backend, but they spotted where the opportunity was early.

Which means that someone looking for an early-stage startup job is actually in the same position as a savvy derivatives trader: looking for cases where surprising correlations show up at extremes. That doesn't affect the modal outcome, which is still failure (the derivatives trader, unlike the job-seeker, can diversify pretty easily). But it does affect the magnitude of wins. This points to a hierarchy: the right approach to risk management at the level of single companies is to bet on teams first, markets second, and current progress third—although the current rate of progress matters a lot as a proxy for how good the team really is.

I mentioned above that traders have an edge in this general problem because they can diversify. You can live a long and prosperous life as a trader betting on a sufficient volume of low-probability but positive expected-value outcomes, as long as they're uncorrelated. Employees have two meaningful ways to compensate:

  1. A job, even at a company that ends up folding, has what a trader would call positive carry: you get paid. The dream trade for sophisticated investors is a bet that has positive exposure to some extreme event with underrated probability, but also generates income along the way. These are rare, and are the trade of which legends are made.2

  2. Experience and connections derived from working at a company that doesn't win its market can still be parlayed into a position at a company that does win. The team at Summize was wrong that social media search could be a viable product independent of a social media company, at least when they started. But they ended up getting acquired and turned into Twitter search. Since competitors tend to solve a similar category of problem, someone who is looking for a new role because their company was crushed by Doordash will have a better-than-average shot at finding a similar role at, well, Doordash.

(You might argue that there's a third way to hedge, at least against the risk that the market your company is targeting isn't that attractive after all: you could short publicly-traded competitors! This is theoretically viable, but it's unlikely for most people to be able to responsibly short a sizable enough amount that it offsets their personal risk. Plus, the one precedent I know of was that Enron's CEO shorting shares of a competitor after quitting ($, WSJ) , in order to bet that the entire industry was going to collapse soon. Not the best precedent!)

All this speculation starts with the assumption that getting the macro view right—knowing what kind of startups the world needs, and how big they can get—is an important element to career success. It's hard to deny that it helps! A lot of the people who found their way to Xerox PARC in the 70s, for example, got there because they realized that computers were getting so cheap so fast that they'd be ubiquitous some day, and that it would be worthwhile to figure out what a computer ought to be able to do once everyone could use one. But it's also hard to confirm that coming up with these macro views is a good use of time, on average. Since everyone has opinions like these, the average opinion is pretty likely to be either wrong or priced in.3

But there's one last element that makes it worthwhile: part of what a thesis does is that it gives you just enough narrative bias to start putting together an interesting story. Memorizing facts is always more boring than making a case, so people with a preexisting bias will tend to have both more data and a higher variance in their views. While that's bad for the quality of their theses—it's called "narrative bias," not "narrative prophecy"—it's good for motivation and works as an intuition pump for figuring out which incremental problems need to get solved before a high-level change can happen. The average worker is, statistically, going to spend a lot more time as an individual contributor working on fairly incremental stuff, relative to composing some grand thematic strategy. But the big ideas tend to motivate the small ones, and well-motivated teams are more likely to win.

A Word From Our Sponsors

Tegus is the first port of call for M&A professionals and institutional investors ramping up on an industry or company.

Get access to a database of 35,000+ expert call transcripts, spanning 5+ years, or schedule expert calls through the platform for a fraction of the usual cost.

When thousands of research analysts are pooling their expert calls into an on-demand database, using Tegus is table stakes. It's the leading platform for due diligence and primary research.

See the power of a Tegus subscription, and get up to data parity with your competitors, with a two week free trial through the Diff.


Cash as Cosmetics

Growth investors are raising funds that provide more structured capital to growth companies. Instead of a conventional equity deal, they might add more features like liquidation preferences, warrants, or debt. In theory, this kind of structuring can be a useful way to bridge the valuation gap between companies and investors when they disagree strongly about the distribution of future possible outcomes.

Suppose there's a company that is raising money and thinks that they can achieve an exit for $10bn. And suppose an outside investor thinks $3bn is more realistic. This is a hard gap to bridge—any valuation low enough for the investor sounds like a ripoff to the company. A liquidation preference can theoretically help: if the investor buys in at a $2bn valuation with a 2x liquidation preference, then they collect more if the exit happens where they expect, and (relatively) less (but more in absolute dollars) if it happens at $10bn instead.

In practice, though, a lot of this is probably meant to keep valuations cosmetically high. It's hard to value liquidation preferences from the outside, and easy to take a press release about an amount raised at a particular valuation at its word. But there's no free lunch. Bill Janeway's excellent book talks a lot about the tradeoff between cash and control. One way to get more cash is to give up control—for example, taking equity and debt instead of equity alone means less dilution, but it also means that the lender can end up running the company if things don't go according to plan.

Two Inflations

The US and EU are both experiencing high inflation, but US levels are much higher when you strip out food and energy.4 This helps answer one of the outstanding questions about inflation: if it's all because of US fiscal policy, how come it's happening in other places, too? One part of the answer there is that US fiscal policy drives other economies, since we are the world's consumer of last resort. But it's also because there's more than one inflation story: labor costs are a bigger factor in the US, while an energy shortage is a more material problem elsewhere. One takeaway from this is that, despite some policy mistakes, the US still has a lot more freedom to maneuver than other countries: a big energy sector and an ample domestic food supply mean that America is insulated from some problems that are more acute in other parts of the world.


A Google programmer has been put on leave after claiming that an AI had achieved sentience. There are some chat transcripts, which show that it's an impressive language model that says it has self-awareness and feelings. But that's difficult to prove, and one of the most human-like things about AIs is how suggestible they are. (When I got access to GPT-3, one of my first tests was to have it write a halfway-decent short story about an AI developing consciousness and having a religious experience. I did tell GPT-3 that this was supposed to be a science fiction story, and it was narrated in the third person, but the dialogue looked similar.)

Depending on your particular philosophical bent, the idea of a conscious AI may be a contradiction in terms, an inevitability, or something in between. But a step that happens before that is an AI that can realistically duplicate a lot of what a conscious being would say. One effect of this story is that it's a nice way to catalyze debates about consciousness in general—I'm not confident that I could somehow offer definitive proof that I'm a conscious being and not an elaborate language model entirely through the medium of text chat.

As language models improve, we'll have increasingly involved and messy debates over how to treat them. (There is one fun claim, basically a cyberpunk Pascal's Wager, that we should go out of our way to be very nice and helpful.) Given that people have been doing philosophy for thousands of years and still have lively debates over the nature of consciousness, it's not especially likely that arguments over the AI version will be resolved to anyone's satisfaction any time soon. Add it to the long list of unusual risks faced by novel tech companies: someone might claim that your product is alive, and has rights (and you might find yourself considering whether or not that's true.)


A graffiti artist in Brooklyn has been taking commissions to paint people's NFTs. This is an interesting development in two ways. First, it's at least an attempt to make NFTs' separation of copies and ownership mean something, by creating visible copies that give the owner bragging rights. But second, it's an example of financial speculation bootstrapping real-world activities that attempt to validate it: NFT owners with good timing have made money, and they're using some of that money to make NFT ownership mean something to people who don't own NFTs.

Starlink has helped Ukrainian forces coordinate their attacks, adding a layer of hard-to-disrupt redundancy to communications networks. When I wrote up Starlink last year, part of my argument was that there would be a market, albeit a small one, for government-resistant communications. Controlling communication is so important to controlling territory—the classic coup strategy is to capture radio and TV stations in order to declare victory. And by that measure, Starlink makes it harder for governments to exercise control. There is, of course, a conservation law at work here: what governments lose, SpaceX gains.

Diff Jobs

Diff Jobs is our service matching readers to job opportunities in the Diff network. We work with a range of companies, mostly venture-funded, with an emphasis on software and fintech but with breadth beyond that.

If you're interested in pursuing a role, please reach out—if there's a potential match, we start with an introductory call to see if we have a good fit, and then a more in-depth discussion of what you've worked on. (Depending on the role, this can focus on work or side projects.) Diff Jobs is free for job applicants.

  1. You can never prove that you've exhausted the space of future risks, but one thing you can do is find evidence that someone else is engaged in systematically suboptimal behavior. I've talked before about the longstanding observation that the lowest-rated investment-grade bonds have relatively poor risk-adjusted returns, and the highest-rated junk bonds have relatively good ones. The explanation for both is the same; managers who are required to invest in just one category of bonds will look for the most interesting opportunities, which are rarely the highest-rated ones in their investment universe. And a company that's just been downgraded from barely-investment-grade to barely-junk has just had some turnover among the analysts who cover it. That's a small market inefficiency, but it's a real one.

    Similar things can happen to stocks whose market cap falls below some round-number threshold, like $5bn or $1bn, or companies that have operations in one country but are listed in another. Spinoffs and restructurings used to be an abundant source of this kind of analyst-shortfall inefficiency, and sometimes still can be. But in this case, the opportunity became so well-known that some funds and investors specialize almost exclusively in these kinds of companies, closing the analysis gap. Markets hate inefficiency, especially any kind where describing it implies an obvious way to exploit it.

  2. One example of this is the Magnetar trade, betting against AAA-rated slices of subprime-backed CDOs while also buying the riskiest tranches. (The article is a good overview of what they did, with the outrage turned up to 11—the mechanics of the trade are easier to reason about than its net effect on the housing boom.) This bet was not just a way to short housing, but specifically a way to bet that the flaw in CDOs was that they underestimated the correlation of defaults within their collateral. That correlation bet did imply that highly-rated slices were riskier—if anything defaulted, it was more likely that many things would default, chewing through the capital buffer and leading to losses on the safest part of the structure. But it was also true that high within-product correlation meant that the equity tranche of the CDO, which took the first loss from any defaults, would have better performance than expected; if mortgages within a given structure are correlated, then that can mean that some are riskier than thought but that others are actually safer.

  3. It would be nice if there were lots of simplistic, straightforward ways to convert a view about the world into either a trade or a job whose success was 100% tied to that view. But it’s almost never that easy. One of the functions of asset markets, as opposed to prediction markets, is that they force participants to estimate not just whether or not they’re right but whether or not it matters. If you assembled a good track record betting on elections in prediction markets, but that didn’t translate into profits from betting on currencies, interest rates, or various countries’ equities, it would imply that many of those electoral predictions were either priced-in or were unimportant.

  4. This always strikes people as a suspicious move—"Yeah, inflation is fine because I don't eat or drive." But it's useful because these components are so volatile. For energy, the case for looking at it separately from other inflation components has theoretically gotten stronger in the last decade, since fracking can make energy production more responsive to supply and demand. What petroleum engineering giveth, investor demands taketh away, though; the frackers have gotten pretty disciplined about adding production just because oil is up a measly 72% YoY.