VC and the Systematic Search for Outliers
It's fun to look at investmentment strategies from the perspective of the portfolio manager, but it's always helpful to take a step back and look at the perspective of their backers.1 For limited partners in venture funds, they're making a pair of paradoxical bets:
- That there will be more unpredictable outlier companies that produce extreme returns, and
- The identity of the fund most likely to back such a company can be predicted in advance.
The skewness of venture returns is extreme, even at the level of funds rather than venture-backed companies; The Power Law cites data from fund-of-funds Horsley Bridge: the VC funds they invested in from 1985 through 2014 backed a total of 7,000 startups, and 5% of the capital deployed accounted for 60% of total returns. That kind of skewed return can show up in other domains, sometimes in nonintuitive ways.2 Finding the companies in that 5% subset that makes the entire industry worthwhile is hard to optimize for in advance, since the companies that produce high returns are necessarily the ones that do something no one else was able to do successfully at the same time; to know what those companies are going to be like in advance is, to have an edge at predicting things that are fundamentally unpredictable.
You can think of venture firms as managing a portfolio, where the key differentiators are dealflow and convexity. (The linked piece goes into more detail, but the basic idea is that the two things a fund needs to do are to get into the best companies and then to ensure that they maximize the dollars they deploy to the best of those. Unless a startup is very capital-efficient, most of the dollar returns will be made by investors who get in fairly late3.)
But to go for persistent returns from one fund to the next, you have to think of VC firms as managing an institution, ensuring that they continue to get access to the best dealflow and the best partners. The Power Law highlights three good models of this:
- The Accel Partners dictum that "Every deal should lead to the next deal." This is a wise approach on many levels. For one thing, it's a reminder that institutional investing is an iterated game: doing something that makes one investment produce a marginally higher return while also constricting future deal flow doesn't pay off long-term. But it's also useful because closely-monitored investments produce a continuous flow of information: one company's constraint is another company's business opportunity (which is a sort of distributed version of classic pivots; Shopify started selling e-commerce software because their original plan, selling snowboards online, was so hard to implement; when a business is a bit more successful than an online snowboard store, but still running into scaling problems, that's an opportunity for a follow-up investment that breaks down those constraints). This strategy can go in the other direction, too: if one part of scaling is easier than it looks, that indicates that there's an un-consolidated part of the market. Investors who backed Wayfair, TripAdvisor, Yelp, and other SEO plays might have noticed that these companies all grew fast in their early days because Google didn't fully monetize its search results, so some of the alpha from being a VC in Yelp came in the form of knowing that the ceiling on Google's search revenue growth was far higher than it looked.
- The "prepared-mind" approach, originally attributed to Accel but now more widespread. This shows up in deals like Sequoia's investment in YouTube, where the fund tracked what happened to digital media consumption patterns when countries reached a given threshold of broadband penetration. Keeping a prepared mind means knowing about important investment themes, knowing what will matter to those themes, and thus being in a position to quickly evaluate new companies that stand to benefit from them or catalyze them. Since time has a high opportunity cost—and time spent on research is time not spent writing checks, and it's just as possible to fall behind on execution as on research—this is not a zero-risk strategy. In fact, it's a form of investment: a fairly fixed initial outlay of time that creates the opportunity to add cash and more time in order to produce returns.
- Managing Allocating financial and reputational credit in a way that makes sense for the firm rather than for individual partners. The Power Law attributes the decline in Kleiner Perkins' relevance in the 2000s to this problem: within the firm, senior partners had enough political capital to get credit for investments that worked (even if they didn't source them) and to shift blame for the ones that failed (even if they were at fault). It's naturally very hard to manage attribution even with full information, so this is a judgment call, but it seems directionally true, especially compared to more egalitarian setups where new partners can get board experience early and share in some of the credit. What's especially pernicious about this practice is that in the short term, hoarding credit actually helps the firm: having high-profile senior partners is good for fundraising, so in the short term letting them take more credit for big wins actually leads to bigger funds. But it also means that talent leaks out of the company over time, so they end up with the catastrophic combination of more money and fewer good investment opportunities. That's a decent recipe for canceling out decades of above-average returns by getting subpar results on a larger capital base.
Investment returns can persist over time, but that's not the default state. There's an element of luck in any given fund's returns, which means there's natural mean reversion. On the other hand, the sample size of companies that could plausibly contribute to outlier returns is small enough that in principle, a single firm with a small investment staff could commit to at least getting a look at all of them. Returns get evaluated most closely at the fund level, the best performing VC firms seem to be the ones willing to lose out on some opportunities for investing and fundraising in order to maximize the expected return of the funds they've yet to raise.
The Paradox of Decentralization and NFTs
This Wired profile of NFT marketplace OpenSea is a good illustration of the paradox of decentralization: anyone can mint an NFT of any digital asset they own, but anyone can also mint an NFT of a digital asset they don't own but wouldn't mind selling. OpenSea won in part by being less curated than other exchanges, but it's also produced a backlash by acting as a centralized arbiter of what constitutes an acceptable NFT. That's happened through delisting of spammy offerings and of trollish ones (even when the trolls were making a point about NFTs). It's not enough moderation to make normal users feel completely comfortable, but it's also more moderation than purists want.
Some of this is the usual evolution that user-generated content platforms go through: when they start out, having loose rules is a competitive advantage, but once they're big enough to be worth spamming and suing, they start making rules. NFTs add a layer of complexity, since they're meant to be decentralized, but a decentralized system can still have centralized onramps. (Email is pretty unusable without spam filters, and the web in general is worse without search, for example.) The uncertain dynamic here is that a big proportion of NFT buyers are people who are cashing in appreciated crypto, and the ones who were buying crypto early tend to be true believers in the vision. So OpenSea will probably follow a more open approach than other companies its size would.
Pigouvian Benefits Update
Companies' Pigouvian responses to the pandemic (first mentioned in The Diff last year) continue: Amazon is no longer allowing unvaccinated employees to take unpaid leave when they're sick with Covid ($, WSJ) among other tweaks to their rules. There's a sense in which Dave Clark, the CEO of Worldwide Consumer at Amazon, is the US's acting Secretary of Labor; Amazon's 1.6 million employees and high turnover mean that working at Amazon is at least an option for a large swathe of the workforce, at all levels but most concentrated among unskilled workers. So when Amazon decides how to split the difference on vaccination's costs and benefits, it tends to become something close to a universal standard.
Two Kinds of Growth
The FT has an interesting look at Super Bowl ads ($), with the implicit takeaway that pricing can provide growth even after audiences peak. Omitting the 2021 outlier from Covid, the cost of Super Bowl ads has risen over 3x since 2000 even though the game's audience share was roughly flat over that period. Another takeaway from the data is that after a blowout ad year, it takes a long time to reach a new record; Super Bowl ad prices were flat from 2000 through the financial crisis, before rising again in the 2010s.
Since Super Bowl ads are so expensive and poorly-targeted, it's worth thinking about why companies buy them and whether or not that's a strong indicator (if this year's advertisers are anything to go by, the Super Bowl Indicator is bearish for crypto). One useful framework is that some companies can grow by tapping marketing channels in which they have a competitive advantage, but once they've done that to its limit, a broadly-targeted ad is a good way to measure what their returns are on more general advertising. On the other hand, Super Bowl ads are prestigious, or at least get talked about a lot. Tech companies' brand advertising sometimes targets customers but often seems to target employees, especially employees trying to explain why they have a real job at a legitimate company.
The International Energy Agency has published revised oil consumption numbers showing that world consumption was 2.9 billion barrels higher than previously thought over the last fifteen years, with much of the excess being attributed to petrochemicals companies in China and Saudi Arabia. This is partly a helpful way to resolve a gap between measured consumption, production, and storage, but also indicates that long-term oil demand is higher than it looks because relatively more of it comes from the chemicals industry rather than transportation; if less oil consumption can be substituted with silicon for solar panels and lithium for batteries, then the world will be using oil for longer than expected.
As tensions with #1 global natural gas exporter Russia mount, the EU has mysteriously lost interest in an antitrust investigation of #2 natural gas exporter Qatar ($, FT). Importing raw materials often forces countries to hold their noses about human rights and legal abuses (in countries that developed their resources before they developed a long history with property rights, the natural resources tend to be a government monopoly, and to behave in a monopolistic way). Even if a given option is bad, there's a need in the short term to choose a least-bad option. Or, if nothing else, it's a useful negotiating ploy to ensure that natural gas will be available from somewhere.
This is a helpful, albeit boring, rejoinder to endless discussions about hedge fund underperformance of the S&P, since funds are targeting lower volatility and lower returns, which their investors are well aware of. The more interesting meta conclusion on that topic is that hedge funds beat their risk-adjusted benchmarks before fees, but probably match those benchmarks after fees, which means that on average a hedge fund LP can assume a broadly efficient market and look for the exploitable inefficiency of a manager who is a) bad at sales, but b) still in business. ↩
For diversified systematic strategies, of either the variety that makes numerous intraday trades or the sort that's long 1,000 stocks and short 1,000 more, you can still see this sort of skewness in which signals produce more of their total returns. Though for the high-turnover strategies, it's typically more muted: high turnover means a short time to reach statistical significance, so strategies have a short half-life before someone else figures them out. ↩
That's especially true in recent years, for a few reasons. First, as employee compensation rises, it gets more costly to build a company of a given scale. Raising money is easier than it used to be, but also more essential than it used to be. The second big reason is that there are fewer inefficiently-priced marketing channels. Robinhood, with its "tweet a link to our site to move up the waitlist" growth hack, may be the last of the ultra-low-CAC large-scale growth stories. There will still be companies that find some kind of edge in distribution, but ad pricing has gotten more efficient on Facebook and Google, while Gmail's "promotions" tab makes email marketing harder. Twitter was a better marketing channel when the ad load was lower and the feed was mostly chronological; an algorithmic feed is a good way to demote unpaid commercial messages and require them to be replaced by ads. ↩