Look for Boring First

In this issue:

• Look for Boring First—Starting analysis with base rates is a good way to stay realistic, but it turns out that the narrative fallacy is sometimes not such a fallacy after all.
• Moderation—Why Twitch punches above its weight in finding good ways to moderate platforms.
• ex-SBC—Equity-based compensation is a good way to align incentives, but it's still compensation and it's not free.
• Platform Risk—Free software also doesn't mean blanket permission to build whatever you want.
• Strategic Resources—Taking resource constraints seriously.
• Meta and AI—The distribution advantage.

This issue of The Diff is brought to you by our sponsors, Brex.

Look for Boring First

There's a good shorthand explanation for why superforecasters perform so well compared to other predictors, including prestigious experts: the expert's job is to tell a story, or at least to fit a new fact into that story, and the superforecaster's typical approach is to start with base rates and calibrate from there. So if you ask an Asian foreign policy expert about the odds of a Taiwan invasion, they have lots of information at hand about what the CCP has said recently, their naval buildup, the US's statements, etc. A superforecaster without any of that information will probably start with a question like "on average, what's the annual probability that a border dispute between two countries turns into a military conflict?" or even a broader question like "what's the probability of a war starting in any given year?" and then start adjusting that number.

This is a recipe for many different flavors of "nothing ever happens," and lots of boring answers about how some easily-narrativized risk is really unlikely to matter. But that's also an apt description of the real world: we live in an age of technological marvels, vast material abundance, and unprecedented geopolitical turmoil. And yet, most new technologies fail, most growth stories peter out, and most crises simmer for a while and then settle down. This stylized fact is actually reflected in financial markets: systematically writing options produces excess returns over time, but those returns include lots of pain at the worst possible time.^[1] Another way of phrasing this is that most of the time, an option buyer is 1) betting that the world will rapidly change in a surprising way, and 2) betting wrong.

This doesn't mean that such narrative-driven predictions aren't valuable. The incredibly annoying flipside to this argument is that many big events, in markets and in history, do actually fit into a narrative. The big surprises aren't that this is one of the one-in-a-thousand-and-twenty-four times that someone flips a coin ten times in a row and gets heads each time, but that sometimes fifteen or twenty flips in you start to realize it's a biased coin or someone with a weird hobby. The most extreme long-term results have been from finding outcomes that chain together a large number of events that are independently low-probability, but that turn out to be strongly correlated. In financial markets, reflexivity is the usual term to describe this: as a company starts to get more investor attention, it can borrow more cheaply, and that makes its returns look better, starting the feedback loop again. But this shows up in plenty of other places: for startups, for example, a lot starts to click once they hit product-market fit—there are more people they can afford to hire, and more people who want to work for them; it's easier to get meetings with potential partners; customer feedback translates more directly to revenue when they already like the product and are willing to say what would make them love it, etc.

And there are plenty of narrative-driven historical events. Some elections are boring, but some of them represent a preference cascade: if there's some policy that is important, but just slightly beyond the Overton Window, then someone who runs based on that policy can execute a nice maneuver where 1) it's something people are willing to talk about, 2) as a consequence, it's the main thing people talk about, 3) it's synonymous with that particular politician, and 4) they combine underdog status with the kind of momentum that makes victory feel inevitable. (Like any interesting claim about politics, this is not just a vague reference to Trump; Sanders held out as a self-styled socialist for long enough that socialism went out of fashion, waited for a while, and went back into fashion again.) But base-rate dominance is still important; grand narratives about realignment are less predictive, in any given election, than heuristics like "the incumbent usually wins unless the economy is bad."

Base rates are good to keep in mind when evaluating trends in companies and businesses. Specifically:

• Turnarounds usually don't work, because it's hard to identify everything that's broken in a company, and by the time that's obvious, a lot of the best people have left.
• Growth usually slows down over time—in the aggregate, the economy consists of some industries that are actively dying, some that are growing but losing share of incremental GDP, and a few that are supporting the rest of overall growth—and their identities are only obvious in retrospect.^[2]
• Any systematic strategy will tend to get crowded; discretionary strategies can't be rigorously evaluated out-of-sample (if they can, they're systematic) and even though there is such a thing as talent in discretionary strategies, there's also an element of luck.
• New products usually didn't exist for a reason.
• Broad metrics like market-wide valuation ratios, corporate profits as a share of GDP, real interest rates, etc. are fundamentally mean-reverting, but over sufficiently long periods that they aren't worth betting on.^[3]
• When an industry's margins or returns on capital improve, it's probably a result of some temporary supply/demand imbalance rather than a fundamental change. And if it is a fundamental change, there's a decent chance that regulators will still try to reverse it—if people were supplying a product when earning an 8% return on capital, you clearly don't need to reward them with a 20% return on capital to incentivize its production

In a way, comparing studies of superforecasters to personal recollections of high-profile pundits' predictions is unfair. All of the points above can be recast as selection effects: if you're hearing about one of the examples above, it's because there's a convenient broad narrative. If there's such a narrative, there's demand for experts to opine on it, and maybe add context before they make a prediction. They face a tradeoff: provide a deeply data-driven expert-level shrug, and that's probably the last interview they do in a while. (You very rarely see people look back and say things like "Why didn't Zika virus spread everywhere?" or "Whatever happened to that looming recession in late 2018?" so the moderate pundits only get to take a mental victory lap in their own heads.) On the other hand, if they say something apocalyptic, it makes their field more important and makes them seem like more of an expert. Even if they aren't making this cynical calculation, the selection effect is still there: you're more likely to hear expert opinions in a context where there's demand to put a high probability on unlikely events, and the specific experts chosen to expound are more likely to do this.

(Not all pundits, of course. To use a currently salient example, Nate Silver has a good track record of making quantifiable predictions and backing them up with data and methodology. If his answer is kind of uninteresting—something like "in the current election basically everyone has made up their mind, so results will vigorously bounce around 50/50 and you'll always have some datapoint that makes one side look like they're definitely going to win"—then lots of people will complain. But also, he'll end up vindicated. It may not be a coincidence that before Silver made his money predicting elections, he made it playing poker, where he probably learned a thing or two about playing tight when you don't have an edge.)

The curse of base rates is partly a blessing: the historical record has plenty of evidence of bad predictions, and it's good to keep that in mind when hearing them or making them. So they're largely safe to ignore. But the flip side is that interesting events do happen, and some people do correctly predict them. If you want to bet on a story, you can, but since it's easy to invent a story good enough to fool yourself, and very hard to take the approach of a skeptical outsider to your own idea, if you are going to be right, you really have to work for it.

A Word From Our Sponsors

Banking that takes every dollar further.

Runway is life for startups, so why do most banks chip away at it with fees, minimums, and delays? Founders deserve better — that’s why Brex built a banking solution that helps startups take every dollar further. With Brex, you get the best of checking, treasury, and FDIC protection in one account. Send and receive money fast worldwide. Earn industry-leading yield from your first dollar — while being able to access your funds anytime. And protect your cash with up to $6M in FDIC coverage through program banks. Ready to join the 1 in 3 US startups using Brex and take every dollar further?

Elsewhere

Moderation

Social media moderation is both a reflection of the Overton Window and a major contributor to it, but it's also a fairly new problem that hasn't been solved yet. Since the feedback loop for social media interactions is faster than anything more continuous, and more continuous than anything faster, it's probably upstream from most other opinion-forming media. So new developments in the space are worth tracking. A recent one is that video game streaming site Twitch is changing its moderation policies to add a decay factor to small violations, though it hasn't specified exactly what those are. As they point out, a system with a fixed number of strikes implicitly privileges new users over established ones (and a system where you can accumulate more strikes if you're well-established is a system that encourages well-known users to periodically say a lot of egregious things to their now-enlarged audiences). Twitch is certainly not the biggest social site out there, but it's worth watching for another reason: many of its creators are in the business of looking at some system, like a video game, and figuring out all of the ways they could exploit it in surprising ways. So Twitch basically recruits an audience of recreational algorithm-exploiters, and whenever they launch some new moderation policy, they get an early look at what they've now established as the new meta.

ex-SBC

One of the most annoying things about tech companies is the irritating habit of disclosing some adjusted non-GAAP profitability number, and adjusting out stock-based compensation, which is one of their biggest costs and which is definitely a real expense to shareholders who face ongoing dilution. Sophisticated investors will almost never put this directly into a valuation model, because compensation practices vary enough to make companies hard to compare, and because it's almost never useful to know that a given company would be profitable if one of its biggest costs magically disappeared. (The only time in my life that I've used such an adjusted EBITDA number as a valuation input was when I was buying a bond issued by a once-growing company in happier times.) To the extent that such a metric is useful, it's because it's a measure of the company's cash runway: a company whose adjusted EBITDA is positive only when excluding stock-based comp is a company that won't have to do a secondary offering to get to profitability because it's basically doing a mini-secondary every time another tranche of employee stock vests. Zoom has joined other tech companies in trying to dial back equity compensation. Over short periods, the cost of dilution is easy to ignore, but they add up: Zoom had 272.5m shares outstanding in the quarter after its 2019 IPO, and it's up to 307.8m, after spending a cumulative $1.5bn on buybacks. So shareholders own 2.2% less of the company each year. Slowly transferring ownership from investors to employees is a decent way to align incentives, but even if that's directionally true it isn't always worth the cost.

Platform Risk

A common setup in software is that there's an open-source project, and a set of companies in the business of commercializing it. This is true to some extent for basically all of software, since you aren't going to get very far without at least using one of Linux, Python, C, etc., but it's much more true for companies built around commercializing one main open-source product. A very common version of this model is that the software is free, but implementation and hosting are just annoying enough to justify a business. And this looks, from the perspective of the business, like a fantastic deal: someone else underwrites the expensive and uncertain business of deciding on new features and rapidly squashing bugs, and you get to run a company with all the benefits of that R&D and none of the costs.

And sometimes, the founder of the open source product your company commercializes, and is named after, calls your company a 'cancer' and tells your customers to switch. The basic complaints are 1) WP Engine offers hosted Wordpress, but turns off version control for site updates, making it hard to rewind after a mistake, and 2) WP Engine doesn't contribute much to Wordpress. WP Engine hasn't formally responded, but whoever runs their blog was, for whatever reason, inspired to drop this piece on WP Engine's investment in open source at 5:40 pm on a Friday evening.

One way to frame the relationship between open-source projects and the companies that commercialze themp is that as totally parasitic, with private companies identifying something meant to be free in both senses of the word and applying a hasty fix. But there are plenty of symbiotic relationships in this space, and open-source would be a much less fruitful ecosystem if private companies didn't pay people good wages to work on open products. You can also assume that this is the public part of a fight that's been happening in the background for a long time—WP Engine has existed since 2010, and has been PE-backed since 2018. So some of this has been simmering for a long time.

The best way to read it is that a naive view of commercializing open source is that it's an interaction with a piece of software made of ones and zeroes. But it's really an interaction with a community, and that community is made of people who have both incentives and opinions. If Wordpress is run by people who care about Wordpress, then they view negative-sum commercialization as an existential risk—if it's an option, companies will happily convert $10 of Wordpress brand equity into $1 of their own free cash flow until the brand is worthless.

Strategic Resources

In last Monday's Diff, we briefly mentioned China's dominance of the extraction and (especially) processing of certain industrial metals. Now a coalition of 14 Western governments is trying to get financing for projects to extract these metals in friendlier countries ($, FT). China has some contract enforcement mechanisms that aren't as readily available to other countries, both because the financial system is more state-directed and because they can always sweeten the deal by building the head of state a fancy palace. And from an investment standpoint, what that means is that they're cutting the political risk premium, or at least shifting it—from the perspective of a Chinese investor, getting involved in local politics in another country is a tricky and uncertain proposition. While the CCP is also liable to change its mind, that's a risk that they're taking no matter what, so the marginal cost is lower.

Meta and AI

The Information reports that most Meta AI users are using it through WhatsApp, and they tend to be in low-income countries and thus hard to monetize ($). Meta has an immense distribution advantage, which rescues it from the classic AI problem that hyped categories have rapid growth, but also rapid churn as customers jump to the next trendy product ($, Diff). By distributing its AI tools through products that already have organic usage, Meta basically gets as many chances to launch as they need.

Disclosure: Long META.

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• A well funded startup founded by two SpaceX engineers that’s building the software stack for hardware development is looking for a software engineering manager with 7+ years experience developing data-intensive applications. (LA, Hybrid)
• A fast growing start-up enabling SMB focused software businesses to offer accounting functionality to customers with an API call is hiring a founding full-stack engineer. (SF)
• A new data startup that's seeing early traction bringing transparency to a $1.5tr asset class is looking for a frontend-focused generalist engineer. 1-3 years of experience, or compelling internships and side projects. (Hybrid, West Palm Beach/Miami, FL).
• Growing edtech company with successes online and offline is looking for an influencer who can reach the parents of gifted and talented kids. (Remote, full- or part-time).
• A company unlocking transactions in a multi-trillion dollar financial market is looking for a head of actuarial research, with a strong academic background, fluency in Python or R and 3+ years of actuarial experience. (Remote, US East Coast).

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.