The social construction of value
“In the short run, the market is a voting machine but in the long run, it is a weighing machine.” Benjamin Graham
VC is more art than science. Nowhere is this more evident than when it comes to the dark arts of private valuations. Most seed startups have no revenue, let alone cashflows to discount. There are no fundamentals for your trad value investor to analyze.
Valuations do somewhat reflect technological and commercial progress. But valuations are largely socially constructed, useful milestone signals to highlight the trajectory of a startup. Bill Gurley likes to argue that “all these private valuations are fake. ... It's all on paper, it's all a myth.”
VCs can use a panoply of creative (financial & non-financial) yardsticks to benchmark companies (e.g. valuations based on multiples of the number of engineers or the number of AI PhDs).1 The further a company is from reliable cashflow, the more a valuation is socially constructed. LeadEdge has this helpful hierarchy of information (Figure 1).
Figure 1. LeadEdge’s hierarchy of information
This is not to belittle these proxy valuation yardsticks, as they are more useful than holding one’s finger in the air. But these valuations are not your business school DCF.
Valuations can themselves help a company curate a narrative to create commercial value. Companies with $1bn valuations (which used to attract the social cachet of the label “unicorns”) can attract higher quality talent, customers & investors. But unicorn status does not always translate to actual value: the rotting corpses of the 2021 unicorn vintage are piling up (e.g. Hopin).
What is more, not every $1bn valuation is created equal. Social value is ascribed to being backed by a specific investor (and in many cases to the specific GP). A Series B led by Martin Casado (currently ranked #1 Series B GP on SignalRank models) at a $1bn valuation is ascribed to have greater value than the numerical valuation alone implies. The support of the “right” VC can frame how the market perceives a startup.
Ultimately it is DPI that determines the best VC firms. But venture capital is an opaque industry where it is challenging to find actual cash on cash returns for specific GPs. It also takes a long time for such cash returns to materialize.
It is therefore important in the meantime for GPs and firms to curate an image which projects the firm as having network centrality. Brand building is a critical component of a VC’s role, both at the firm level and personally as a GP (Figure 2). VCs need the next Patrick Collison / Alexandr Wang / Melanie Perkins to consider their fund as the go to investor.
Figure 2. The importance of brand in VC
According to Marc Andreessen in Nate Silver’s new book, On The Edge:
“At the point of contact with an entrepreneur, when we go in and we do our whole sales process and try to win the deal… 90% of that fight is over before it begins because it has to do with reputation that we’ve established, the track record – you know, the brand.” Andreessen argues VC becomes a self-fulfilling prophecy: “there might be one thousand ways to get in the positive feedback loop. But the reality it, you’re either in it or you’re not.”
Missing a deal can materially hurt returns. Andreessen again: “The mistake of omission are much, much bigger mistakes. We almost never kick ourselves for mistakes of commission; we do kick ourselves pretty hard for mistakes of omission.” The combination of oodles of capital, FOMO & a brand which signals support to idiosyncratic founders makes for a heady cocktail. See A16Z’s monstrous $350m into Adam Neumann’s (of WeWork fame) rental real estate business, Flow.
There is an implied hierarchy of investors. With actual cash returns skewed towards the best investors, it is important to be perceived to be “Tier 1” and to ensure that your name glistens on the Midas List. (LPs also pay attention to these lists, so perception of value can translate to actual value by both attracting LP dollars and deploying into the next big thing). In fact, the more interesting and unusual funds, such as Zeev Ventures or Harrison Metal, are those who engage in effectively zero brand building but are picked up by our algorithms as having a canny ability to select power law companies.
On the face of it, SignalRank’s cold quantitative approach appears to take the opposite approach. Our model ranks investors by stage based on a number of factors, including number of unicorns backed, unicorn efficiency (unicorns / all investments) and MOIC for every investment. Yet it is still socially constructed, in that it aggregates human intuition and decisions across multiple rounds (instead of looking purely at the fundamental data itself). Instead of relying on personal networks or narratives, SignalRank paradoxically turns subjective social sentiment (in the form of investors attaching their names to companies which go on to grow) into quantifiable objective signals.
In practice, SignalRank is leveraging data to codify the social construct of tiered investors because we look at almost zero fundamental company data. We purely look at investor patterns & behaviors across multiple rounds for a company. The strength of this approach lies in its potential for scalability and for standardization of subjective decision-making.
Let’s dig into why this quantitative approach can have predictive power.
Art as an analogy for VC
Benedict Evans (former partner at Andreessen Horowitz) flagged in his eponymous newsletter a few years back that his favorite book about venture capital is Talking Prices by Olav Velthuis. Except this book is not about venture capital. It’s a book about art, written by a Dutch sociologist.
Velthuis applies a sociological lens to explore how art markets operate differently from other economic markets, focusing on the symbolic, social, and cultural dimensions of pricing contemporary art.
A core theme is the social construction of prices. Art prices are not solely determined by supply and demand but are influenced by social relations, status, and symbolic meanings. The value of art is often tied to reputation, gallery affiliations, and the artist's prestige rather than purely economic factors:
“Prices in the art world are socially constructed: they do not reflect intrinsic qualities of artworks, but rather result from the interplay of social factors such as reputation, networks, and the narratives that surround both the artist and the work.”
The most interesting parallels here are to compare the role of galleries/dealers with VCs. Galleries serve as gatekeepers, helping benchmark prices and establish reputations. They are useful signalers. The perceived quality of a gallery can influence the perceived quality of an artist and their work:
"Artists are dependent on galleries not just for their livelihood but for their reputation and the narrative surrounding their work. The gallery invests in this narrative, promoting it to collectors and critics alike."
Value is constructed, signaled, and maintained through social and symbolic means.
Algorithm aversion
One way to think about SignalRank is that our investor ranking model codifies the social hierarchy of VCs, and looks for patterns around how this hierarchy plays out over multiple rounds. The name “SignalRank” itself is a nod to Google’s PageRank, their core initial algorithm which ranked web pages based on importance and relevance. PageRank measured the significance of a page by analyzing the links that point to it, effectively treating a hyperlink as a "vote" for the importance of that page. Note how SignalRank & PageRank both weigh human decisions.
At present, it appears that society prefers algorithms which weigh human behaviors rather than applying abstracted unstructured algorithms, even if the unexplainable algorithm is more accurate than the human weighted algorithm. Researchers coined this term “algorithm aversion”.
This idea that today we do not fully trust algorithms is particularly pertinent as we ask AI to do ever more stuff. A new paper by Gertjan Verdickt and Francesco Stradi looked at whether investors trust an AI-based analyst forecast. They found that, “although investors update their return beliefs in response to the forecast, they are less responsive when an analyst incorporates AI. This reduced trust stems from a lower perceived credibility in AI-generated forecasts.” In other words, AI-written or aided reports are today perceived to be less influential than those authored by humans; people became more distrustful the more complex the method sounded. This could change as AI models become more commonplace.
At SignalRank, we are of the view that it could be some time before LPs commit hard-earned pensioner capital to pure machine-learning quant funds because these models are not currently explainable and therefore not currently trusted. Our model today is codifying socially created transparent heuristics. We can fully explain the algorithm step by step. Plus we only invest when a world class human is leading the round (ie we are price takers). In due course, AI models could come to be the norm.
Standing on the shoulders of giants
SignalRank is not the first to codify investor decisions and behaviors into a model. Alfred Winslow Jones, credited as the inventor of the original “hedged fund” pioneered the strategy in the early 1950s.
According to Sebastian Mallaby in More Money than God:
“Jones invited brokers to run ‘model portfolios’ for his fund: Each man would select his favorite shorts and longs, and phone in changes as though he were running real money. Jones used these paper portfolios as a source of stock-picking ideas. His statistical methods, which separated the fruits of stock selection from the effect of market moves, allowed him to pinpoint each manager’s results precisely. Jones then compensated the brokers according to how well their suggestions worked. It was a marvelous technique for getting brokers to phone in hot ideas before they gave them to others.”
More recently, Marshall Wace, one of Europe’s largest hedge funds, has built an “alpha capture system” which analyzes investor patterns & data. TOPS (Trade Optimised Portfolio System) is the flagship strategy of Marshall Wace, representing $30bn+ of their $65bn AUM. In essence, TOPS captures the best “sell-side” strategies electronically and trades on them.
In 2001, Marshall Wace hired a new Oxford graduate, Anthony Clake, to design a database that would assimilate sell-side broker recommendations. TOPS integrates real-time insights from a network of 5,000+ brokers and analysts with advanced algorithmic models to make informed investment decisions.
The TOPS Opportunistic portfolio returned 24% gross of fees in its first full year, versus a market benchmark down 21%. TOPS has continued to perform well, showing how a hybrid approach of human intuition and machine precision can create a highly diversified & active portfolio that performs at scale.
SignalRank is to venture capital what TOPS is to public markets.
How does SignalRank combine social constructs with machine learning?
SignalRank ranks investors by round stage and then looks for patterns across multiple rounds. Our algorithm considers a number of data points over a rolling five year period, including number of unicorn investments per stage, unicorn efficiency (unicorns divided by all investments at that stage) and MOIC for every investment at that stage. We then look for patterns for individual rounds and across multiple rounds; we convert investor scores into round scores which are then converted into company scores. We benchmark these company scores relative to similar rounds in the same vintage (to create percentile scores for rounds & companies). Our algorithm has increased the probability of identifying a company which can deliver 5.0x MOIC in five years from Series B to above 30% (compared to 10% for the market average).
SignalRank does not take into account fundamental company data (founder background, revenue, unit economics, etc). For a quantitative investment strategy in venture, it is our contention that there is much more predictive power looking exclusively at investors rather than company level data. Markets change, business models change, and technology changes. This leads to noisy company level data. Exceptional companies are by definition exceptional, so the data may not pick this up at the company level. And entrepreneurs are playing a single hand of cards. VCs on the other hand are playing multiple hands across multiple tables with somewhat publicly available track records. A specific investment decision might be based on intangible factors like intuition or social networks, but, when aggregated across many instances, these decisions can reveal consistent patterns. Backing multiple successful companies might reflect good judgement over time.
There is an analogy data-driven VCs use of VC being a horse race. Investors need to back either the horse (market / company) or the jockey (entrepreneur). At SignalRank, we believe that there is value in looking an abstraction higher by evaluating the horse trainer (VC).
A critical component of our algorithm is that we are looking for behaviors across multiple rounds. We exclusively invest at Series B; we require a company to have decent scores in the seed & Series A rounds, as well as the Series B. There is an interesting sociological side to our business in that seed investors’ key skillset is reading people while Series B investors’ key skillset is reading businesses with growth potential. We like the pattern when a top 100 seed investor invests 2x in a company prior to the round. It is very hard to trick a world class seed investor twice (especially when that investor also has insider information on the company prior to the second investment). Some of the largest frauds in the last few years (FTX, Theranos, Frank) raised very impressive Series Bs. But none was backed by world-class seed investors. They didn’t pass the sniff test. Our model would have eliminated these companies from our candidate list.
Our model also works because of persistence in venture capital (unlike in public markets or in private equity). If you made a good investment 10 years ago, you are more likely to make a good investment today. VCs are effective brand builders and storytellers which is necessary to demonstrate network centrality. You can ride the tale of a hit investment for a number of years, which then itself projects your credibility which leads to superior dealflow. This can create a self-perpetuating cycle. See Jason Calacanis’s career post his Uber investment. Indeed, this persistence is even more powerful at the GP level than at the firm level. Correlation Ventures has shown that value of a VC fund is no more than sum of its partners. The persistence in performance by partners is six times more important at explaining fund returns than firm attributes.
Our model does use paper mark-ups but ultimately relies on actual realizations. Instead of relying on stale primary valuations, we are in the process of plugging in real-time daily valuations from secondary markets. 90% of secondary transactions are in $1bn companies, so the data is largely focused on unicorns. But given that all fund returners are unicorns (even if all unicorns are not fund returners), this is the data set to focus on. To rank highly on our models at any stage, you need to be in a number of unicorns. In other words, our model balances leveraging social signals across rounds while also calibrating for real-time cash on cash returns.
Blowing bubbles
A quick aside on bubbles. You can’t write a piece on social value without mentioning bubbles, the “mortal enemy of the Efficient Market Hypothesis”.
Bubbles are the ultimate expression of social exuberance impacting value. Value decouples entirely from fundamentals and simply reflects collective behaviour. This is driven by fear, greed, uncertainty, and the human tendency to follow the crowd. It happens to the best of us. Even the great Sir Isaac Newton lost much of his fortune in The South Sea Bubble.
The word bubble gets bounded around a fair amount. There have been at least two clear bubbles in the US financial market in the post war period. By looking at the value spread (the ratio of expensive stocks to cheap stocks) in US public markets, we clearly see the Tech / Telecom Bubble and the Covid / ZIRP Bubble (Figure 3).
Figure 3. Fama/French value spread, 1950-2024
Despite all the posturing about contrarian thinking, VCs are particularly susceptible to herd mentality given the FOMO dynamics discussed above. Sebastian Mallaby in The Power Law talks about how “in some ways, VCs are the ultimate herders. You go to Sand Hill Road, and you see that they all have offices on the same road… There is one characteristic in particular which encourages groupthink – you can’t go short. Because you’re syndicating and you want dealflow, you can’t even speak short, you can’t say negative things about other people’s deals.” Hence the Patagonia vest memes, etc. But more importantly hence the likely subpar returns from the 2020/21 vintages when the herd moved as one.
We are frequently asked about how SignalRank deals with bubbles in our algorithm. It feels intuitive that there is wisdom in the (VC) crowd, which is how our algorithm is built. But what happens when the crowd is not making independent decisions, and acts in unison as a herd? How does your algorithm deal with that?
We have developed a dynamic thresholding system which takes into account the “heat” of the market using 90 day averages (see Figure 4). In order for any company to unlock our capital, it must be score in the top 5% of any Series B in the last 90 days, as well as be backed by top ranked investors across multiple rounds. We can also tweak this threshold in real-time if we see that the market is experiencing one of its moments of madness.
Figure 4. Series B thresholds (publicly available at signalrank.ai)
Conclusion: Combining the Best of Human and Machine
At the end of the day, venture capital remains an art. But it’s an art that can benefit from the precision of data. By paradoxically codifying subjective social constructs into a model with objective insights, SignalRank allows investors to navigate the opaque world of venture capital with greater confidence, combining the best of human intuition with the rigor of machine learning.
SignalRank’s model can help its investors ensure that they are exclusively investing in companies backed by the best investors in the world at scale.
Lightspeed’s Nnamdi Iregbulem has this great post on how seed valuations are not valuations but a proxy for tech wages.