Tag: venture capital

  • The Elasticity of Risk

    The Elasticity of Risk

    As described in a previous article, risk is the product of venture capital. If you do not understand risk, you do not understand venture capital.

    In theory, the greater the risk an investment is seen to have, the greater the payoff if it works out. This is the skill of the venture capitalist; recognising asymmetry of risk and payoff.

    In this calculation, there are essentially three categories of risk:

    1. Idiosyncratic Risk
      The core of what is traditionally seen as venture capital, associated with investing in unproven ideas.
    2. Systematic Risk
      The market-related risk in venture capital. The influence of shifts in sentiment or cost of capital.
    3. Execution Risk
      The risk associated with building and scaling a particular solution, and dealing with competitive forces.

    Any startup may have minimal or maximal levels of all three, although there are some relationships to explore.

    • As an example of idiosyncratic risk, consider SpaceX. There was no space industry. There was no market map, playbook, or competition. It was like nothing else.
    • As an example of systematic risk, consider OpenAI. The company’s success is heavily reliant on excitement about AI keeping the cost of capital low for large model providers.
    • As an example of execution risk, consider Uber. The business clearly worked at scale, but reaching scale (dealing with regulators, unions, etc) was the central challenge.

    Each risk category is a spectrum and all startups will have some degree of each. You’d have to go to extremes (e.g. a laundromat) to find a business close to zero risk.

    e.g. for all that SpaceX is a champion for idiosyncratic risk, clearly there was also execution pressure. Similarly, OpenAI was highly idiosyncratic at inception, although that has changed over time as the AI market expanded.

    Risk Relationships

    While each risk exists on a separate axis, there are some risk relationships worth understanding:

    Idiosyncratic risk is generally opposed to systematic risk

    If a company is doing something truly original, it is unlikely to be relying on category hype to fund their development. Thus, a highly idiosyncratic company usually has low systematic risk.

    That said, painfully novel ideas with an unclear path to profitability are less likely to be funded in cold markets. Investors will retreat to the safety of more knowable investments.

    To understand this, we have to separate hype (AI, dotcom) from hot markets (early 80s, ZIRP), although often the two overlap. Each introduces a different flavour of systematic risk, one related to technology and the other to market conditions.

    Idiosyncratic risk is generally opposed to execution risk

    Execution risk is typically a product of highly competitive environments. If building a great company wasn’t hard enough, there is now a clock ticking on market saturation.

    Implicitly, this is less of an issue for highly idiosyncratic startups who may be pioneering a whole new category with much more limited competition.

    That said, some companies are so painfully idiosyncratic that there is also significant execution pressure. To court investors, they must keep smashing through development milestones.

    In summary, there is a U-shaped relationship between idiosyncratic risk, and the two other risks. As covered above, the highest theoretical payoff comes when all three are maxed out.

    Scaling Risk

    Time and time again, in hot markets, incumbent firms have attempted to “scale” venture capital. That is, to find elasticity in the relationship between fund-sizes and risk, allowing individual firms to dominate large chunks of the market.

    Each time, they have failed, because the risks outlined above do not scale linearly with fund size.

    • Idiosyncratic risk is inelastic. It is the art of making good judgements about the future, and you cannot improve outcomes by throwing more people at the problem.
    • Systematic risk is exponential. As you inflate a category or market, you herd-in capital which accelerates growth and raises the stakes dramatically.
    • Execution risk is an unknown. There is a belief that patterns might reveal founders who excel at execution, but research largely indicates these are misleading.

    If risk cannot be made elastic, then neither can fund sizes.

    The largest firms cannot bear this reality, because a world where venture capital doesn’t scale eliminates any incumbent advantage; all firms must continually prove their right to exist.

    Instead, they throw their resources at financialising the industry, creating a hyperreality where these risks are treated differently:

    • “Consensus is good, actually“
      You don’t need to worry about scaling idiosyncracy if it’s not highly valued. In a world where consensus rules, the largest allocators will naturally rise to the top.
    • “Escaping the cycle“
      Systemic risk already scales well, and a large firm is in the perfect position to amplify that inflation and ride-out the consequences of the subsequent crash.
    • “The fundable founder“
      If a founder can keep raising money and generating signal, there’s a chance they’ll persist even with subpar execution. Especially with support from a large platform team.

    Ultimately, this produces a “smart beta” strategy, reliant on market momentum and performance that isn’t directly connected to the quality of outcomes.

    e.g. research finds that large, well-networked firms are more likely to inflate valuations and pump capital into consensus opportunities.

    This drags venture capital away from its core purpose: the entropic distribution of capital to opportunity. Instead, it rewards concentration and accumulation, in firms and companies.

    As a consequence, innovation suffers and returns slip. As hot markets have driven fund sizes and management fees up, with fewer firms in more dominant positions, technological progress has increasingly stagnated.

    A less competitive environment is fundamentally less meritocratic, and the focus on consensus themes and pattern-matched founders excludes swathes of opportunity and potential outliers.

    The future of venture capital, and whether financialisation deepens, depends on narrative dominance. If the market continues to reward consensus seeking, concentration and dogmatic belief, the trend will persist.

    It’s the kind of existential question which would lead large firms to fund social media bot farms, groom tech influencers, and play a growing role in the media side of venture.

    (top image: “Untitled 2”, by Julie Mehretu)

  • Simulcra and Success

    Simulcra and Success

    Venture Capital’s Search for Scale

    A recurring question in each techno-financial cycle: “Can venture capital scale?”

    Each time the market is primed with enthusiasm, and the macro stars align, fund sizes swell and managers promise that this time it will make sense. Inevitably, returns fall and the market contracts, yet it’s the greatest accumulators that survive.

    Thus, the goalposts have gradually shifted towards scale, and the market has nudged its way towards “hyperreality” — celebrating the signals of success.

    The Liquidation of Referentials

    French philosopher, Jean Baudrillard, argued that society has moved from a “real” economy (based on production and utility) to a “hyperreal” economy (based on simulation and sign value), when signals offer disproportionate reward.

    For example, a wealthy individual may buy a Ferrari as a reward for their hard work. Another person might buy a Ferrari on credit to access the status of perceived wealth.

    Alternatively, a successful VC will end up accumulating “unicorns” in their portfolio. Alternatively, a VC that wishes to appear successful might offer “unicorn” terms to their portfolio companies.1

    This behaviour reflects the “liquidation of all referentials”, where the signal itself becomes desirable, rather than the underlying reality.

    Thus, as explored below, venture capitalists (particularly in hot markets) less frequently fund “reality” (idiosyncratic, novel innovation based on practical demand), and more frequently fund “simulacra” (scalable software models that simulate value through rapid growth detached from physical constraints).

    A History of Hyperreality

    It’s in our nature to want to leave a legacy, and typically that’s either through quality or scale. Which of these is easier, and more appealing, is a reflection of where we’re at in the grand boom-and-bust cycle of financialisation.

    “Financialisation” itself essentially refers to something like hyperreality in the narrow context of financial markets.

    In recent years, it appears that scale and accumulation have become the primary objective. While it has become more obvious (and more widely discussed) in the last five years, the origin goes back a decade further. Indeed, much credit goes to Seth Levine for calling “bifurcation” as far back as 2010.

    Implicitly, we have then moved in the direction of hyperreality and the pursuit of signals — but can it be demonstrated empirically rather than just theoretically and philosophically?

    To assemble a more complete picture, we can compare studies on various periods of venture capital activity and the “revealed preference” in the data:

    From the 1960s up until 2020, there is a surprisingly clear shift of investor priorities which aligns with Baudrillard’s theory of “hyperreality”.

    In the first paper, covering 1965–1992, the authors found that venture capital is broadly less efficient in hot markets. Capital overshoots the opportunity, duplicate companies are funded, and the per-dollar return of the strategy slides. This is essentially the origin of “venture capital doesn’t scale”, and a hypothetical ceiling on appropriate allocation.

    In the second paper, covering 1985–2004, the authors found that venture capitalists tend to invest in companies with greater risk during hot markets. The distribution of outcomes widened, with a greater rate of failure but higher valuations for successful exits, alongside more patents and patent citations.

    “The flood of capital in hot markets also plays a causal role in shifting investments to more novel startups – by lowering the cost of experimentation for early stage investors and allowing them to make riskier, more novel, investments.”

    Investment Cycles and Startup Innovation

    Here is where the divergence begins, with the Great Financial Crisis and an extended period of low interest-rates coinciding with the rise of scalable cloud-based software with subscription revenue. There is marked change in the risk appetite and preferences of venture capital, shifting towards the hyperreal.

    In the third paper, covering 2010–2019, the authors found that venture capitalists had narrowed their focus on software companies in recent years, leaving other (more “real”) sectors bereft of funding. Particularly, that this did not appear to be connected with greater opportunity in increasingly crowded software markets.

    “While venture funding is very efficacious in stimulating a certain kind of innovative business, the scope is increasingly limited. This concentration may indeed be privately optimal from the perspective of the venture funds and those who provide them with capital. It is natural to worry, however, about the social implications of these shifts.”

    Venture Capital’s Role in Financing Innovation: What We Know and How Much We Still Need to Learn

    In the fourth paper, covering 2010–2019, the authors found that venture capitalists have increasingly pursued scalability during hot markets — rather than greater novelty, as described in earlier research. Instead of a greater appetite for idiosyncratic risk (associated with innovation), it was execution risk which grew (associated with experimental scaling strategies).

    Significantly, these investors choose to keep these companies private in order to absorb greater levels of private capital, rather than seeking an exit — and a reckoning with reality. As long as value appeared to grow at the desired rate, there were no unplesant questions about the reality of it.

    In the fifth paper, covering 2005–2019, the authors found that “centrally located” (connected, influential) investors will syndicate to inflate a company to “unicorn” status for the purpose of grandstanding. The newly minted unicorn will then more easily attract subsequent investment from “lower quality” investors, like moths to a flame.

    The tyre-marks of hyperreality are clear in the post-2009 data, where a period of capital abundance did not result in more ambitious companies being built, but rather a more ambitious approach to scale. Rather than bigger ideas, venture capitalists pursued ideas of bigness.

    We have since seen that this behavior (predictably) left more than 600 “zombie unicorns” stranded when the market contracted. Companies that were only really “unicorns” in a hyperreal sense; frail simulcra of billion-dollar businesses.

    Further research shows that such “well connected” investors tend to perform better in hot markets, and worse in cooler markets, which illustrates the damage done when a market correction shatters the illusion of performance.

    The Hyperreal Market Mirage

    Putting all of this together, we can establish the following:

    Investors found that success does not scale in venture capital, as a greater volume of capital simply widens the distribution of outcomes and erodes the aggregate returns.

    However, signals of success do scale. If you treat “unicorns” as a measure of success, where success brings you more capital, you can create as many unicorns as you wish.

    Thus, after a period of immense accumulation during the post-GFC low-interest-rate period, venture capital embarked on a spree of unicorn creation with the goal of compounding success into durable competitive advantage.

    During this time, influential investors led syndicates that anointed unicorns, encouraging founders to stay private in order to absorb more capital, print more growth, and post increasingly impressive virtual performance.

    In Q2 2022, the market was forced to reckon with reality as inflation set in and prompted a serious escalation of interest rates. Suddenly, LPs were keen to know exactly how “real” the reported performance was.

    Today, the market is not particularly conducive to “hyperreal” investing, as skepticism of venture capital behavior remains. However, the most influential firms aren’t going to let this go; scalable venture capital is the promised land of compounding advantage and ultimate oligopoly.

    So, they manifest the hyperreal by wrapping certain parts of the industry in narrative. They invest heavily in both creating and manipulating media to influence the perception of opportunity and performance.

    By bending their apparatus toward a particular narrow band of influence, they are able to continue achieving scaled “success” in the simulation.

    Secondary transactions, and the “democratisation of private markets”, are your ticket to participate in hyperreality, and the opportunity to become deluded exit liquidity for these accumulators.

    Or you could join the smaller (and less visibly successful) pool of investors who stubbornly grapple with reality.

    (top image: “The Matrix”)

    1. Or a group of VCs may organise around systematically assigning unicorn status to companies, rather than investing in companies that will earn it. [↩]
  • Ghost in the Shell

    Ghost in the Shell

    The danger of inherited biases in VC LLM applications

    There’s a fascinating history of research comparing venture capital decisions to algorithmic outcomes. In a basic sense, these studies pitch human judgement against machine scoring systems.

    What’s particularly interesting about these studies (detailed below) is the clarity they provide about our vulnerability to cognitive biases. Where investors are able to reason objectively, they comfortably outperform algorithms. Where they can’t, they are soundly beaten.

    This topic is increasingly relevant, as more studies are released looking at the use of LLMs in the venture capital decision making process (also detailed below).

    The main point I will explore here is that humans represent one end of a spectrum (hyper qualitative), and algorithms represent the other (hyper quantitative), and LLMs land somewhere else entirely: they offer powerful automation, but have inherited our very human biases.

    There are also important points related to the GP:LP interface and the relationship-based nature of venture capital, which are both connected to the question of why venture capitalists have resisted these findings so far.

    Not Better, Not Worse, But Different

    Fittingly titled, Predictably Bad Investments by Diag Davenport covers the basic sin of pattern matching. Looking at a sample of 16,000 startups, and comparing real-world outcomes versus the output of an algorithm, the study finds that bad investment decisions are typically a product of over-indexing on founder attributes (particularly “star power” and charisma) and under-indexing on the idea itself.

    This theme appears again in Machine Learning About Venture Capital Choices by Victor Lyonnet and Léa H. Stern, where again venture capital outcomes are compared to algorithmic choices. This study goes beyond venture-backed companies, using government data for a perspective on missed opportunities. The conclusion is that not only do venture capitalists make predictably bad investments, they also miss predictably good ones by overweighting “great founder”stereotypes.

    These papers bear important consequences for venture capital, but are ignored for an uncomfortable truth they reveal: the common industry wisdom about picking founders doesn’t hold up to scrutiny.

    The real answer is a little more complicated. To help complete the picture, we can turn to Do Algorithms Make Better — and Fairer — Investments Than Angel Investors? by Torben Antretter, Ivo Blohm, Charlotta Sirén and Dietmar Grichnik. Here, the authors compare the performance of human investors against an algorithm once again. Importantly, they don’t simply look for a winner, but an understanding of who wins, and why.

    Once again, the study found that an algorithm broadly outperformed human investors, producing an IRR of 7.26%, versus an average of 2.56% from the 255 angel investors. However, humans still came out on top by a wide margin, with experienced investors who showed negligible signs of cognitive bias producing an average IRR of 22.75%. Crucially, it was the lack of bias rather than the experience which drove that outperformance; experienced investors with clear signs of bias produced a fairly pathetic 2.87% IRR.

    Combining the three papers, we can drill further into the conclusion: algorithms outperform most human investors because an algorithm cannot be swayed by charisma, it does not care if you went to the same university, if you live in the same area, have similar facial features or matched political beliefs. Investors who wish to fall into the high-performing category must master these cognitive biases, to reduce the drag they produce on investment performance.

    Another perspective on this question can be found in And the Children Shall Lead: Gender Diversity and Performance in Venture Capital by Paul Gompers and Sophie Wang. This study examined the make-up of venture capital investment teams, finding that diversity is positively related to returns when it is the result of the firm recognising and controlling the same cognitive biases. If there’s institutional clarity about objectively recognising talent, it will reflect through the whole system.

    Best Served Cold

    As a fundamentally illiquid strategy, studies on venture capital investment behaviour must address the question of cycles. How do these conclusions change if the market hits a bump, versus periods of sustained growth?

    In the cases of investor biases, and the over-reliance on subjective founder attributes, we can turn to Venture Capitalists’ Decision-Making Under Changing Resource Availability: Exploring the Use of Evaluative Selection Criteria by Noah John Pettit. Unsurprisingly, the study finds that investors in “hot markets” tend to lean further into faster subjective judgements (i.e. more System 1 Thinking) as the possibility of failure seems remote in periods of abundance. Implicitly, this amplifies the issues described above, relating to cognitive biases and founder attributes.

    This finding is reproduced in Venture Capitalists’ Decision-Making in Hot and Cold Markets: The Effect of Signals and Cheap Talk, by Simon Kleinert and Marie Hildebrand. Not only do venture capitalists rely more on “cheap talk” (e.g. promises of rapid growth) from charismatic founders in hot markets, but the Fear of Missing Out leads them to neglect more useful but costly analysis. Mirroring the studies above, venture capitalists end up relying on predictably poor information, and ignoring predictably useful information.

    Amongst a host of other adverse incentives, this FOMO-driven laziness helps drive hot markets into the typically spectacular end-of-cycle crash. It’s a tragedy, as when capital is most readily available for important ideas, it is allocated most recklessly.

    Compromising With Context

    All of this brings us to LLMs, and the central question: are they a “better” algorithm with more context for qualitative reasoning for venture capital decisions? Or, have they simply inhereted the same biases of human investors from their training data?

    There are a few pros and cons to consider.

    For example, in Petit’s paper on decision making with changing resource availability, an identified challenge for investors was “laziness”. The unwillingness, particularly in hot markets, to spend time processing and analysing complex information about an investment opportunity. Here, LLMs clearly offer the benefit of streamlining this process. By making complex signals less costly, it might reduce the influence of “cheap talk” — particularly in hot markets.

    Similarly, Davenport’s paper on predictably bad investments highlighted investors’ vulnerability to founders with extreme charisma. This seems like a problem LLMs should be able to avoid, unless the founder is a well-known figure and features in the training set via past writing or press coverage.

    On the other hand, the problems identified in Stern and Foster’s paper, where venture capitalists built the stereotype of a “great founder”, are likely to be magnified. As these stereotypes are entrenched in the discourse of venture capital, and have influenced past investment activity, they will have been trained into LLMs. This risks exacerbating the pattern-matching problem, including amplifying negative biases like the prejudice against solo founders, minority founders, and startups outside of traditional hubs.

    Biased Echoes: Generative AI Models Reinforce Investment Biases and Increase Portfolio Risks of Private Investors, by Philipp Winder, Christian Hildebrand and Jochen Hartmann, looks specifically at the manifestation of these biases in LLMs. This study found that LLMs exhibit the same cognitive biases as human investors (in some cases, more severely) and increase portfolio risk across all five dimensions measured (geographical cluster risk, sector cluster risk, trend chasing risk, active investment allocation risk, and total expense risk). Additionally, the study found that basic debiasing measures (e.g. prompting the model to ignore certain data) only partially mitigated the issue.

    With all of that in mind, we can turn to two more recent papers on the use of LLMs in venture capital, how they eliminate, modify or magnify our existing cognitive biases, and the implication for their use in future.

    VCBench: Benchmarking LLMs in Venture Capital set LLMs on 9,000 anonymised founder profiles to test their “picking” ability. The study demonstrated that LLMs can more confidently pick winners, beating top-performing venture capitalists. For example, DeepSeek-V3 was correct in 59.1% of its picks. However, that came at the cost of being extremely selective; it only identified 24 out of 810 available winners. In a more realistic environment (~135 winners in the sample) it would end up performing similarly to a good VC firm (picking 4 winners, 18 losers).

    It’s also worth noting that a “winner” in the VCBench paper is an exit of more than $500M (or more than $500M in cumulative fundraising), which doesn’t represent an overwhelming success by today’s metrics.

    Essentially, LLMs were over-reliant on patterns and missed a lot of winners in the process. Implicitly, this approach might be more appropriate for later-stage investments in a more concentrated portfolio, or platform firms generating beta, and would struggle with the idiosyncrasy of early-stage venture capital.

    Moving away from investment decisions, Generative AI-powered venture screening: Can large language models help venture capitalists? examines how quickly an LLM can analyse opportunities by distilling large sets of unstructured data. This study looks at the application of LLMs to tasks that might normally be assigned to an analyst, where it appears to be 537x faster and marginally more accurate.

    While the potential gains in efficiency are immense, it’s worth considering the findings of The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers, and how long-term use of LLMs in investment decisions might impair human ability. In order to learn from past investments, investors must fully understand and appreciate why they were made. There’s a dangerous future where LLMs shape decisions, which shapes training sets, which shapes decisions. We end up stuck in a recursive loop in which venture capital loses touch with reality.

    Do Things That Don’t Scale (and Scale Things That Do)

    LLMs are powerful tools, but they are fundamentally backwards-looking in their reliance on training data, with limited ability to reason toward novel conclusions. Thus, they are quite specifically unsuited to early-stage venture capital investments where extreme idiosyncracy grapples with realities that may be a decade away.

    That’s not to say they aren’t useful for venture capitalists. You might, for example, use an LLM to analyse deal memos and understand how your reasoning compares to real outcomes; whether your concerns are predictive of failure, or whether perceived strengths are predictive of success. A large, general model might play a part in identifying trends across research and industry which inform your thesis on future opportunities. They’re also increasingly capable of analysing financials and cap tables to produce useful insights without the typical burden of work.

    In essence, LLMs can be used to wrangle data that might help shape how the firm operates. However, venture capital investments (done well) are pure alpha, extracted from the negative space. There are no patterns to work from, which precludes much of how LLMs reason. In essence, they give you a better map of what is known but they cannot tell you where to go in the unknown fog of opportunity. Indeed, this is exactly why cognitive biases also produce such a drag on venture capital investments; they are the application of patterns from the past to decisions about the future.

    A Game of Telephone

    Unfortunately the bias problem cannot be solved at the VC level alone. Much warped thinking spreads down through the market from LPs, who also enjoy being sold on “cheap talk” about opportunity and simple patterns. The myth of the rockstar stock picker is seductive, eliminating an awful lot of complex uncertainty from venture capital, for all parties. Far easier than embracing a reality where investors must be technically competent and disciplined.

    Thus, the relationship between founder and VC, or VC and LP, is often boiled down to reductive factoids, like “venture capital is a people business”, or a “craft”. If they agree to this shared delusion that venture capital isn’t scientific, then they can abandon the hard work of scientific thought. Opportunistic founders build heat-seeking ARR printers that cater to the vision that VCs want to market to LPs. Allocation is filled, money is moved, marks go up, and most of the combined objectives are achieved.

    This is the danger of LLMs. If they aren’t properly understood from a perspective of behavioral economics, they can easily be misapplied to produce even worse outcomes (both in fairness and performance) than we see today. It would be relatively trivial, for example, to create some hybrid model that performed well on VCBench, and sell LPs on how it would outperform the market in future — when no such thing is true. Instead, it would offer an accelerated version of the past, fuelling greater accumulation and faster markups, with a zero-G plummet at termination.

    Affirmative Smokescreens

    It’s an interesting era for venture capital, current market dynamics aside.

    The optimistic case is that investors will run sensible experiments with what LLMs can add to their process, yielding a marginal uplift. The pessimistic case is that models will be thrown blindly and arrogantly at investment decisions, producing a decay which may not be obvious for years afterwards.

    That fork in the road will be determined by whether or not the industry can develop a more sophisticated view on behavioral economics, and resist current incentives to embrace bias. This is part of a sustained battle to professionalise the strategy and squash rent-seeking grifters who achieved huge success in the ZIRP bull run.

    Essentially, if investors are handed tools that can dramatically accelerate activity, it will end up amplifying the dominant philosophy. That could result in the more efficient allocation of capital to opportunity, or it could enable ever-greater AUM fee extraction. The worry is that incumbents already use “craft” as a smokescreen for accumulation, and LLMs offer a new and more exciting mirage of competence.

    Projecting Forward

    The highest performing venture capitalists in any (rational) environment are likely to have a particularly raw and tactile interface with the world. They are maximally open-minded, energetically curious, and have mastered the reflective act of developing opinions about the future. Most importantly, they have escaped the cognitive prisons of insecurity and ego — the harbingers of bias.

    In the future, LLMs will empower these individuals by handling everything else. The grind of fund admin, portfolio maths, contracts, reporting and communications. Busy-work will be handled by agents, leaving the gifted to chase creativity, opportunity and serendipity. The missing heart of the AI Tin Man.

    How quickly we can manifest this reality will depend on how soon we can exit the current trajectory, which itself is an artifact of broad economic financialisation.

    References:

    (top image: “Ghost in the Shell”, by Production I.G)

  • Venture Capital is Not Competitive

    Venture Capital is Not Competitive

    Early-stage VC can be one of two things:

    • A complex, collaborative and positive-sum discipline
    • A simple, adversarial and zero-sum competition

    The difference between the two is whether or not you pretend there’s a legitimate consensus on “good deals”, and therefore whether competition and access are meaningful forces.

    If you embrace this fantasy, a few convenient things happen:

    1. You can win at VC by brand building
    2. You can win at VC by exploiting relationships
    3. Venture capital becomes “scalable”
    4. Capital and influence concentrates

    Thus, venture capital morphs from a loose collection of boutique investors into an oligopolistic mob.

    The rent-seeking behavior this enables is so seductive that it has been able to resist all contrary evidence:

    What’s left is an industry with few leaders and many followers. For the weak-willed majority, conviction atrophies and cements their position as vulnerable minor parties in a commensal relationship.

    While this is problematic in a few obvious ways (the failure of fiduciary duty to LPs, the failure to founders), there is one critical issue: it completely undermines the positive-sum attitude that has been central to venture capital’s success.

    Finite Games and Infinite Games

    You might hear investors saying that venture capital feels more competitive than ever. What they really mean is that venture capital is more adversarial than ever.

    As power and capital concentrates, and the focus is increasingly on winning a finite number of the “best” deals in each vintage, venture capital is necessarily less competitive.

    Truly competitive games are infinite games. You don’t just care about winning, you also care about the scale of that win.

    To use basketball as an analogy:

    A finite game player might simply care whether or not their team wins the next game, and the next championship. As long as they win, that’s all that matters.

    An infinite game player cares about stats like the ~42,000 US city parks with basketball hoops. They care that ~40% of 13-17 year olds regularly play pick-up basketball, and the ~18,000 US high schools that sponsor basketball programs.

    Many competitive pursuits die as a finite games because they are dominated by an early elite who are hostile to newcomers and casuals. They choose winning smaller victories over competing for a larger outcome.

    Venture capital, the success of Silicon Valley, and much of today’s technological progress, was built on an infinite game: The pursuit of abundance through innovation, rather than simply dominating the status quo.

    When an investor (either GP or LP) says indiscriminately that VC does not need more capital, they are implicitly using a finite game lens. They prefer that competition is restricted, and cannot see how that also limits the scale of opportunity for the category as a whole.

    In the US, this attitude is becoming a problem as boutique firms are displaced by agglomerators. While the two aren’t in direct competition for LP dollars, the environment is increasingly defined by the participants with the most influence.

    In Europe, this attitude is an entrenched problem. European venture capital did not emerge from an infinite game mindset as it did in the US. For all the anecdotes about Europe’s merchant banks, venture capital is an import that has quickly been seized by opportunists. This is why Europe’s venture capital scene has remained relatively small, at a cost to growth and prosperity. There has never been a vision for abundance.

    City Parks for Entrepreneurship

    If you want to take an infinite game approach to venture capital, you need to care that entrepreneurship is accessible, and that the interface between capital and talent is healthy.

    To the agglomerators, this interface is addressed by scout programs where ambitious young adults seek startups that fit a pattern — as the incentives are to optimise for partner approval. So, more capital flows into SF/NY, Stanford and Harvard Grads, ex-Mag7 employees, AI tools, etc. The pool of opportunity narrows, but you aim to win more of it.

    To the boutique investors, this interface is addressed by scouring college campuses and builder communities. They are looking for outliers; the people with ideas so outlandish that it may not have even occured to them to seek investment yet. The pool of opportunity grows, and you may win more of it.

    The former is obviously a finite game. The latter is obviously an infinite game.

    If the goal is (as commonly stated) the pursuit of abundance through progress, venture capital is implicitly an infinite game and we should care deeply about the entrepreneurial equivalent of city park basketball courts; distributing awareness, access and opportunity as widely as possible.

    The fact that so much of the industry is focused on gatekeeping and exclusivity is an obvious signal of distress.

    Rothenberg’s Paradox

    A paradox in this story is that the scale of capital going to the finite game is much larger, and is growing more quickly.

    How then is that the finite game?

    This reflects what we might call Rothenberg’s Paradox: any successful infinite game creates opportunity for rent-seeking via finite game players. The scale of capital becomes the central opportunity, like a snake eating its own tail.

    For now, venture capital maintains the illusion of a booming industry with growing opportunity — while it’s really just getting better at financial engineering and wealth extraction.

    (top image: “Pollice Verso”, by Jean-Léon Gérôme)

  • Venture Industrial Policy

    Venture Industrial Policy

    The state’s role in creating positive-sum games

    Many countries have developed policy aimed at supporting the growth of domestic venture capital ecosystems — hoping to close the gap with the US.

    In the EU, a key premise is that that underdeveloped private markets leave a “‘growth-capital gap” for scaling companies, limiting the potential of EU-based tech startups.

    This problem is one of the key targets in Mario Draghi’s report on EU competitiveness.

    Bluntly, this is what happens when bureaucrats are left to design industrial policy: Solutions are designed to patch-over outcomes, ignoring a more problematic diagnosis.

    In truth:

    • There is less growth investment in the EU because there are simply fewer attractive investments.
    • The most obvious opportunities are snapped up by faster-moving US investors with stronger brands.
    • Thus, there is even less market impetus for the development of the EU’s growth capital environment.

    So, pouring money into this “growth-capital gap” has predictably negative outcomes:

    Instead of addressing a lack of investment by injecting more capital, the EU must tackle the root cause: aiming policy at generating more opportunity.

    Instead of allocating capital to a known quantity of growth-stage companies, policy must drive capital into raw potential, at the earliest stages, with faith that it will electrify entrepreneurship.

    The “inception capital gap”

    To frame this with a few obvious but important facts:

    • The EU has a significantly larger population than the US (450M vs 340M).
    • Pre-seed funding in the US was approximately 13x the scale of pre-seed funding in the EU in 2023.1
    • Angel funding in the US was approximately 23x the scale of angel funding in the EU in 2023.2

    While the growth-scale capital gap might appear more obvious in sheer scale (the total US VC environment is ~$130B larger than the EU / about 4x the scale), it is important to remember that all growth activity is downstream of early investment.

    If companies cannot raise angel capital, or institutional pre-seed and seed rounds, they will never reach the point of contributing to the opportunity of growth-stage investment.

    Breaking Zero-Sum Loops

    The problem many young venture capital ecosystems face is that haven’t truly embraced the spirit of (ad)venture: escape from zero-sum thinking.

    The usual loop looks something like this:

    1. A small pool of early-stage capital has a lower risk-appetite which manifests as restrictions on access.
    2. Restricted access means safer investments in companies that are inherently less likely to be outliers.
    3. It’s also reflected in credentialism and pattern-matching, which further erode returns over time.
    4. The outcome is a limited number of great outcomes which cannot sustain significant growth of the ecosytem.

    The US was able to break this loop via a cultural tolerance of risk. Particularly, a general trend toward positive sum thinking and looseness, often described as a “frontier mentality”.

    Outside of the US, this may be harder to achieve. Leverage can be applied via policy to help break the catch-22: finding motive to invest in the potential of future investment opportunities.

    This is particularly true for institutional LPs, who are unfamiliar with the idiosyncracies of venture capital and more likely to think in Private Equity terms — reluctant to back GPs at the earliest stages, who take the greatest risk.

    Manifesting opportunity

    Following this logic, the role of the government should be to fund emerging managers (funds 1-3) who target venture capital investment at the earliest stages.

    This allocation would be reduced over time as emerging managers develop their track record and attract more private LP capital. By fund 4, with roughly a decade in the market, GPs should be able to function independent of government capital.

    By increasing the capital availability for early-stage managers, they will increase risk appetite, diversify what gets investment, and enable a range of more innovative origination strategies. Esssesntially, this promotes the “entropic distribution” of capital, where capital flows through the cracks and crevices of industry to seek opportunity.

    There is already significant data out there to establish that this approach works:

    In addition, the lack of angel activity should be addressed more directly via tax reduction schemes — especially for employee stock compensation. Stock compensation is how Europe can create a flywheel which distributes success into new opportunities.

    Summary: lessons for the EU (and beyond)

    There is a significant delta between US growth-stage investing and EU growth-stage investing, and a lack of EU scale-up success stories.

    At a glance, the obvious policy move is to push capital into growth investing. Indeed, the EU already commits significant capital to private markets — e.g. 37% of 2023’s VC total.

    This is the EU throwing huge sums of money at symptom management. A colossal waste.

    The more pressing concern (which could be addressed at less expense) is the gaping chasm between US angel and pre-seed funding versus the EU. Funding the generation engine of entrepreneurial potential.

    Indeed, the EU is stuck in zero-sum thinking which is reflected in attitudes across the bloc; in competition between hubs, between governments, and even between specific actors within those ecosystems.

    If institutional LPs in the EU are hesitant to solve this problem, by committing to the uncertainty of emerging managers in the earliest stages, the state should consider stepping in.3

    While this is an almost non-existent view in policy today, the data is clear that this is the central opportunity for government investment in venture capital across a number of metrics:

    • The growth of startup ecosystems
    • Enabling technological innovation
    • Driving high-skill job creation
    • Return on government investment
    • Feeding downstream growth ecosystems

    (top image: The Tower of Babel by Pieter Bruegel the Elder)

    1. A rough approximation of Europe minus the UK, and US data from Carta [↩]
    2. Calculated using EBAN per-country data, report also includes US estimate. [↩]
    3. Potentially with asymmetric terms that offer leverage to participating private institutions. [↩]
  • Coordinating Capital

    Coordinating Capital

    The role of startup valuation in staged capital deployment

    In his article VCs should play bridge, Alex Danco described the Capital Coordination problem in venture capital, created by the staged nature of investments.

    I recommend reading the whole piece, but the summary is that investments are de-risked by staging capital over future milestones (with the implied valuation step-ups) rather than investing everything up-front and hoping for the best.

    The heart of this strategy is the signalling game where investors offer affirmation of the investment for their capital partners.

    VCs should play bridge

    “This gives the next investor cover to say, ok, I’ll do the same thing. I’ll invest at $20M, sending a signal: I believe that this price reflects a discount to the next round. This 20 million price, by current convention, means we believe this hand will play out as another X million in GMV run rate”, or whatever it is your signalling for this new round of bidding.”

    Danco’s assessment speaks to the oddly myopic perspective of “venture math“: investors are primarily focused on understanding incremental progress (measured with ARR) rather than the ultimate outcome. This has come at a cost to sectors that are slower to start generating revenue but solve much more important problems.

    He’s right to frame venture capital as collaborative (Bridge) rather than strictly adversarial (Poker), this also highlights issues like enmeshment and collusion which emerge when so much stress is put on relationship-driven outcomes over objective measurement.

    The approach that Danco describes as “convention bidding” is framed as an alternative to the zero-sum attitude of formal valuation — where investors are primarily concerned with securing their own returns rather than participating in affirmative semaphore.

    Indeed, this reflects a revealed preference for market signals over conviction which should trouble anyone who still believes venture capital is responsible for risk capital formation.

    Discounting the Future

    Further to the above, Danco’s article is a good reflection of venture capital’s poor grasp of valuation, characterised by three common misconceptions:

    1. Valuation is transient, and at each stage a new valuation would need to be calculated — creating uncertainty that threatens signal-driven coordination.
    2. Valuation is the same as price, and entry points are more a function of market norms and comps than they are of fundamental value creation.
    3. The practice of valuation is built on too much uncertainty to be practical for early stage companies.

    While the first is inaccurate, the second is actually dangerous. The general practice of market-driven pricing has created huge structural fragility in venture capital, exacerbating the already painfully boom-and-bust nature.

    The third, however, is just silly. All investments are predicated on valuation, at least in theory (the alternative being mindless momentum investing). The only difference is whether it’s written out with explicit assumptions or a mental calculation with implicit assumptions.

    Indeed, valuation is essentially the story of a startup (“What happens if things go right?”) translated into numbers, with some discounting for the cost of capital and risk of failure. You can think of it as the financial source code of a pitch.

    Usefully, by calculating a terminal value based on that story and discounting it using the expected rate of return in venture capital, you can also ascertain a healthy entry point for future rounds — assuming the startup remains broadly on track.

    Consider the below: the valuation of a Seed round, calculated on Equidam at just over $20M. As a part of this calculation, and the future it’s predicated on, you can also see the valuation trajectory for the forecasted period. If the company raised a Series A at ~30 months, the valuation would be around $85M.

    Expected Valuation Growth – Equidam

    Of course, all of the usual caveats: the future is uncertain, startups pivot, markets shift. However, it’s fairly easy to built those adjustments into a model and see how they change both today’s valuation and the future trajectory.

    Assessing valuation with a higher resolution view of performance enables better judgements: Perhaps top-line revenue is growing on track, but costs are surging. Maybe revenue is lagging, but margins are way better than expected.

    The hope is not that forecasts play out precisely, but that they give you a useful relative perspective on performance over time.

    If venture capitalists wanted a way to coordinate staged capital in the future, to finance a company efficiently (with minimal time spent agonising over terms) while ensuring a good risk-adjusted return for all investors, this is the logical approach.

    Valuation is essentially the process of aligning expectations across market participants by transparently exploring data and assumptions. Typically this occurs between buyer and seller (VC and founder) in a specific transaction, but there’s no reason why downstream capital providers couldn’t benefit from that work.

    It also offers the vital benefits of helping investors better understand the value of novel innovation, and limiting exposure to systematic risk of market-based pricing — getting caught in the trap of “money chasing deals“.

    (top image: “The Bulls and Bears in the Market” by William Holbrook Beard)

  • Limbic Capitalism

    Limbic Capitalism

    If the house always wins, build a casino

    “Should we be concerned about a ‘Venture Deep State’? A ‘FOMO Industrial Complex’? Limbic Capitalism?”

    Joe Milam, Founder of AngelSpan

    A previous article discussed why venture capital has such a myopic focus on backward-looking metrics connected to ARR.

    The summary is that people struggle with uncertainty, and there are two ways to cope: embrace it scientifically, or build a structure that lets people gamble on it.

    Enter “casino culture”.

    “Sports betting, shitcoins, meme stocks, vibe coding 100m in six hours, etc, are all expressions of the same deep cultural rot. If youth don’t believe there’s legitimate ways to get rich through work, all of culture will become a rotten sports book for the soul.”

    Will Manidis

    If you want to bet on startups, ARR provides a convenient metric to compress complexity and uncertainty into one dimension, reducing venture capital to a simple horse-race.

    Investors make bets on momentum categories, riding the wave of revenue driven by large injections of capital, and hope they can raise another fund or exit before the music stops.

    Combine that with the compounding influence of the “power law” meme, and the incentives to concentrate capital, you’ve got a real lottery on your hands.

    Like an actual lottery, the net cash return is negative.

    Net cash flows in venture capital head south

    “It’s very difficult to invest money well, and I think it’s all but impossible to do time after time after time in venture capital. Some of the deals get so hot, and you have to decide so quickly, that you’re all just sort of gambling.”

    Charlie Munger, Vice Chairman of Berkshire Hathaway

    The beneficiaries of this limbic shift are the agglomerators, who play the role of “the house”; appearing to participate in the risk while harvesting beta from the associated activity.

    For other investors, there are incentives to play:

    Primarily, it gives GPs a simpler story to sell to LPs, inspiring greater confidence. They can enthusiastically shill consensus ideas and relationship-based access — which is exactly what many LPs want to hear, despite the poor returns.

    Secondly, it diversifies away accountability. As more investors jump on-board the momentum train, individual career risk is exchanged for greater systematic fragility. When the market collapses, they can survive on relationships.

    By turning venture capital into a game of chance, abusing the principal-agent problem, their job is greatly simplified: It’s the blue pill of steak dinners and congeniality, not the red pill of grinding hackathons and building portfolios.

    Unfortunately, by abstracting the performance if investments from the underlying asset value with crude proxy metrics, the concept of “innovation” is mostly a mirage. It may be difficult to perceive at the time, when the traction feels real, but few truly important companies emerge from this environment.

    Value creation in the metaverse

    “Consistently surprised by general lack of world changing ambition with seed stage cos and “early stage” VCs. Era of incrementalism still prevalent.”

    Jimmy Yun, Investor at 8VC

    Limbic capitalism produces what Byrne Hobart and Tobias Huber have described as “virtual innovation” in Boom: Bubbles and the End of Stagnation.

    Essentially, a category emerges as the focal point for venture allocation, concentrating capital, and participants begin flipping coins under the long-standing premise that the upside of winners is always much greater than the downside of losses.

    How could they possibly lose?

    Image
    The Jackpot Age

    In the end, everyone eventually loses — except the house.

    It turns out that the very behavior they engage in is destroying returns, and the only winner is the person collecting fees to manage the bank.

    “Risk too much hunting jackpots and the volatility will turn positive expected value into a straight line to zero. In the world of compounded returns, the dose makes the poison.”

    The Jackpot Paradox

    (top image: “The Romans of Decadence , by Thomas Couture)

  • Venture Math

    Venture Math

    Creative ways to hide financial engineering

    A rough example of the logic driving investment decisions amongst the most degenerate venture investors:

    1. You’re evaluating a company with $7.5M ARR
    2. $5M is net new ARR, annual burn is $10M
    3. That’s a 2x burn multiple (BM)
    4. You invest $30M at a market-rate of 20x ARR
    5. Assuming 2x BM, $30M produces $15M in net new ARR
    6. At 20x ARR, it gets marked up from $150M to $450M

    On paper, that checks all the boxes: an investment that was subsequently marked up 3x using logic that could be decoded on an iPhone calculator.

    Indeed, the role of multiples is to produce calculations that are easier and faster; to get deals done and put capital to work.

    There are two major flaws with this approach:

    The first is that multiples are backwards-looking. By applying to current revenue or burn, they rely on past performance. All assumptions about the future are squashed down into the multiple itself. Venture capital relies on making good judgements about the future, not the past.

    The second is that multiples assume that all revenue is created equal. They ignore unit economics and capital expenditure, and nor do they fully appreciate churn. Finally, and particularly relevant today, they encourage founders to engage in creative accounting to boost ARR.

    “We’ve actually come back to saying there’s a real advantage to seeng the GAAP revenue accounting, to make sure all the money is showing up for real.

    There’s a lot of noise in that multiple, and when they were all SaaS recurring revenue businesses — all seat based, all 90% or 80% gross margin with no CapEx, all enterprise sales with low churn — it absolutely made sense. You could compare two companies. Thats’s why by 2019 or 2020 it almost felt like ‘fill in the form to give me the valuation’.

    None of those conditions are true now.”

    Rory O’Driscoll, Partner at Scale Venture Partners

    Consider, for example, the venture capitalist’s typical disdain for discounted cash flow (DCF) style thinking in valuation: There are too many assumptions, the future is too uncertain.

    So, instead, they price with ARR multiples, which include all of the same assumptions about the future, but obfuscates them into simple calculation. Out of sight, out of mind.

    “Some investors swear off the DCF model because of its myriad assumptions. Yet they readily embrace an approach that packs all of those same assumptions, without any transparency, into a single number: the multiple. Multiples are not valuation; they represent shorthand for the valuation process. Like most forms of shorthand, multiples come with blind spots and biases that few investors take the time and care to understand.”

    Michael Mauboussin, Head of Consilient Research at Morgan Stanley

    Indeed, multiples are a popular tool precisely because they hide all of that detail. Remember, venture capital is (unfortunately) a game of building confidence with simple stories, not demonstrating competence with complex truths.

    Financial engineering built on multiples is the bedrock of venture capital’s creeping financialisation.

    Imagine you invest $20 in a company that generates $4 in ARR from each customer, with a CAC of $20. It trades at a multiple of 20x ARR.

    • In practice, that company is losing $16 on every customer in the first year, with payback over 5 years assuming no churn.
    • On paper, every $1 into the company produces $4 in marked value for the investor. It’s a great looking investment.

    This is precisely the mechanic which incentivises insane capital consumption on negative unit economics. Investors trade financial health for faster markups, knowing that it’s a bust if the environment shifts.

    “Guess what happened once Founders realized that VCs were valuing startups using revenue multiples? They started playing a game of Hungy Hungry Hippo with the goal of accumulating as much revenue as they could!”

    Frank Rotman, Co-founder of QEDInvestors

    “When market turns, M&A mostly stops. Nobody will want to buy your cash-incinerating startup. There will be no Plan B. VAPORIZE.”

    Marc Andreessen, Co-founder of Andreessen Horowitz

    To quickly round-out a few points, lazy thinking with multiples…

    • …biases venture capital towards business that generate revenue quickly, which led to the neglect of deep tech.
    • …promotes pushing companies to scale as quickly as possible, which produces worse outcomes for everyone.
    • …contributes to VCs losing the ability to build independent conviction, encouraging herd-behavior.

    In conclusion, multiples are a tool for quick comparison across peers. Not for pricing, and not for understanding performance or potential of a startup. They include far too many important and unchecked assumptions, limiting an investor’s understanding of the specific future of a company.

    Proper valuation, with multiples used as a sanity-check afterwards, is the way.

    “We often pursue this kind of rationalization as a spot check, generally after going through the valuation process. When the multiple is implied, investors will then compare it to others seen in the public and private markets to get more comfortable. Think of it like a gut check, a way to determine if the valuation feels ‘reasonable’.”

    Alex Immerman, Partner at Andreessen Horowitz, and David George, GP at Andreessen Horowitz

    (top image: “The School of Athens“, by Raphael)

  • Systematic risk-management in VC

    Systematic risk-management in VC

    Why the industry loves confidence men

    “Venture capital has thrived in uncertainty: uncertain technologies, uncertain market trends and uncertain capital availability.”

    Michael Eisenberg, Founding Partner at Aleph

    Early-stage venture capital is characterised by uncertainty. Success lies in high-risk, non-consensus ideas that take many years to play out.

    There are two ways that LPs grapple with this uncertainty, as institutional allocators to VC:

    1. Seeking competence: managers that understand how to extract value from uncertainty.
    2. Seeking confidence: managers that inspire the most certainty about future success.

    For reasons best described by Daniel Kahneman, LPs are inclined towards the latter. This manifests as poorly managed risk; under-developed portfolio construction and overconcentration.

    This is despite a large body of research illustrating the following:

    1. Overconfidence is a prevalent bias in venture capital.
    2. Larger portfolios are likely to outperform smaller portfolios.
    3. Concentration creates more downside than upside.
    4. Diversification encourages VCs to take more idiosyncratic risk, which drives outperformance.

    Indeed, this is not without precedent. Some of the greatest early-stage VCs have unusually large portfolios: Boost VC, First Round, BoxGroup and Precursor are obvious examples.

    Despite the data, and the many anecdotal success stories, LPs are notoriously hesistant to back diversified strategies. By believing they can diversify at the LP level, across a portfolio of firms, they miss the systematic benefits of diversification and inhibit the compounding gains from process refinement.


    Portfolio Maths

    Earlier this year I posted a breakdown of expected venture capital returns based on simulating two seed portfolio models: one with 100 small checks, the other with 20 larger checks.

    Portfolio simulations

    The graphic was meant to illustrate a simple point: based on a typical range of VC outcomes, the more diversified portfolio was likely to outperform, delivering a narrower and more attractive band of potential return scenarios.

    “Typical” is the operative word there, as while the graphic implies the diversified portfolio wont exceed a 6x return, that’s only if you deliver average performance. It’s entirely possible for a GP to outperform in either strategy — although research indicates that the diversified portfolio is likely to have more upside.

    The statistical basis for this is simple: given venture capital’s reliance on outliers outcomes, the low probability and low predictability of those outcomes, it is more beneficial to make more investments (with lower ownership), than to have more ownership (with fewer investments).

    There are similar simulations and portfolio calculations from a number of other sources, which all point to the same conclusion: venture capital is systematically and unnecessarily overconcentrated.

    Picking winners chart
    Picking Winners is a Myth

    The coversation that followed (credit to the always-insightful Peter Walker for sharing it with his audience) highlighted a deeper problem with the industry’s understanding of portfolio strategy, risk management and cognitive biases.


    Overconfidence

    A good place to start is Daniel Kahneman’s work on cognitive biases in investing:

    “The confidence we experience as we make a judgment is not a reasoned evaluation of the probability that it is right. Confidence is a feeling, one determined mostly by the coherence of the story and by the ease with which it comes to mind, even when the evidence for the story is sparse and unreliable. The bias toward coherence favors overconfidence. An individual who expresses high confidence probably has a good story, which may or may not be true.”

    Daniel Kahneman, “Don’t Blink! Hazards of Confidence”

    As a crude summary, imagine an LP speaks to two VCs:

    • One says they’ll invest in 100 companies and expect that ~75 of them will be be writen-off. The fund returns will primarily hinge on ~1-5 investments.
    • The second explains that they have a powerful network advantage, and they’ll deliver huge returns from concentrated ownership in just 15 companies.

    The LP is likely to go with the second VC, who builds more confidence by telling a simpler and more coherent story; they simply have access to “better investments”. What’s not to love?

    And few investors demand diversified funds, so GPs don’t offer them. A slow and steady “venture is a numbers game” pitch is much less emotionally compelling than “I am a rock star who can consistently beat the odds.” And GPs need an emotionally appealing pitch to get funded.

    The Pervasive, Head-Scratching, Risk-Exploding Problem With Venture Capital

    LPs want to hear confidence, so that’s what VCs offer. Thus, “overconfidence” is by far the most prevalent and dangerous cognitive bias in VC behaviour.

    Overconfidence and disappointment in venture capital decision making: An empirical examination

    This mode of favouring confidence over competence leads to a number of dogmatic beliefs amongst the LP community:

    1. That relationships drive better investments in VC — which is only vaguely true in hot markets when it’s easy to collect markups from your friends.
    2. That venture capital funds themselves operate with power law outcomes — which is only true because of the unnecessary and toxic levels of concentration.
    3. That VCs are supposed to “pick” their way to success — rather than relying on well designed origination strategies and a systematic approach to capturing outliers.


    Overconcentration

    Overconfidence primarily manifests as overconcentration; portfolios with poorly managed risk. Too much capital, concentrated into too few investments, with high levels of uncertainty. This is typically measured with the Sharpe Ratio in grown-up investment strategies.

    The influence of concentration is obviously double-edged, as it compounds both good and bad investment decisions.

    However, research shows that this amplification is not evenly distributed: overconcentration hurts underperformers more than it helps overperformers.

    Fund Concentration: A Magnifier of Manager Skill

    So, if venture capital were systamatically overconcentrated, you would expect to see a wider distribution of returns and a lower average return, relative to other strategies. As it happens, that’s an accurate description:

    Performance Dispersion in Alternative Asset Classes


    Persistence

    There’s a good argument that the poor persistence of performance in venture capital can be attributed to shallow narratives around “picking” which undermine basic theory around portfolio construction and behavioral economics.

    Allocator Solutions: Evaluating Persistence in Fund Performance

    Not only does this damage persistence, it’s particularly toxic for new entrants who may attempt to piece together a strategy from the “common wisdom” available to them.

    Instead of being encouraged to adopt practices that are optimal for more consistent above-benchmark returns, emerging managers are pushed toward the “rockstar” narrative of outsized promises.

    Thus, they end up taking excessive risk and, more often than not, imploding; only 1/3 managers make it to fund 2, and only 1/10 make it to fund 4 .

    You might raise an eyebrow at this, if you are familiar with the history of venture capital incumbents strategically freezing out new entrants in order to maintain advantageous pricing power.


    Intitutional Insecurity

    “Perhaps the most powerful lesson from Marks is the idea of ‘uncomfortably idiosyncratic’ investing. Citing the late David Swensen of Yale, Marks emphasizes that successful investment management requires taking positions that feel uneasy because they go against the grain.”

    Howard Marks on ‘Behind the Memo’:

    If investment performance is driven by adopting “uncomfortably idiosyncratic” positions, it’s reasonable to assume this is especially true for early stage venture capital — where non-consensus investing drives outperformance.

    “You will know you are doing real venture capital when you aren’t competing with other investors to finance a deal but are instead offering to invest in people, industries and ideas that don’t yet have access to capital. That is where money can be most useful, and also where returns can be the highest.”

    Sam Lessin, GP at Slow

    VCs take a systematic approach to managing idiosyncratic risk through portfolio construction; optimising for larger outcomes and failure rates than other strategies. Indeed, VCs with larger portfolios appear comfortable accomodating more idiosyncratic risk, which ultimately contributes to stronger performance.

    In fact, you could probably summarise the VC strategy as the art and science of systematically extracting value from idiosyncracy.

    The lingering question, given all of this evidence, is why aren’t larger early-stage portfolios the default in venture capital?

    The answer is agglomerator-leak; the “loudest model” of the mega-funds, which ends up influencing practices and perceptions across the whole venture market.

    For example:

    When you are investing billions in each cycle, you must pry your way into hottest companies of each vintage. The future of your firm depends on those bloated private-market darlings, and there is significant career-risk associated with missing them.

    Thus, a number of ideological artefacts are spawned:

    • “You must be in the category winners, at any cost.”
    • “Only a handful of outcomes drive returns for each vintage.”
    • “Concentrated ownership drives outperformance.”

    These artefacts latch onto insecurity like Pinterest self-help quotes. GPs and LPs looking for confidence through coherence, biased toward simple ideas, lap up this accessible wisdom from venture’s most influential characters.

    Unfortunately, it means they apply the same thinking to early-stage investing, and much smaller funds operating a very different strategy, even when all evidence suggests that’s a very silly thing to do.


    Process Alpha

    From a well-diversified base of initial investments, it is possibe for a VC to double-down on winners over subsequent rounds. This is a process that Joe Milam, founder of AngelSpan, has dubbed “process alpha“.

    Staging your capital deployment properly over multiple rounds dramatically improves the IRR for investors, regardless of the MOIC/DPI. And given the natural failure rates of startups (usually within the first 2 or 3 yrs), optimizing on how much you invest in each round improves the risk-adjusted returns available.

    The Impact Of Proper Venture Portfolio Construction-Optimized DPI & IRR

    Essentially, this is how a VC can take the solid foundation of a well diversified initial portfolio and then build on that position with staged deployment into the best opportunities that emerge in subsequent rounds.


    Conclusion

    While it’s difficult to get into specific recommendations, it seems safe to make the case that early-stage venture capital firms should probably be significantly more diversified.

    More importantly, the LP:GP interface is clearly problematic, and LPs need to think carefully about whether they bias towards confidence over competence.

    In order to do so, they must grapple with the economic theory, and the reality of portfolio construction, to recognise why it’s important that there be a good level of diversification at the VC portfolio level.

    “You should not take assertive and confident people at their own evaluation unless you have independent reason to believe that they know what they are talking about. Unfortunately, this advice is difficult to follow: overconfident professionals sincerely believe they have expertise, act as experts and look like experts. You will have to struggle to remind yourself that they may be in the grip of an illusion.”

    Daniel Kahneman, “Don’t Blink! Hazards of Confidence”

    (top image: The Cardsharps, by Michelangelo Merisi da Caravaggio)

  • “VC”: What comes next?

    “VC”: What comes next?

    There is no shortage of think-pieces on the state of venture, bemoaning concentration, slipping returns, and the challenging environment for smaller managers.

    Few have tried to answer the underlying question:

    The perspective that VC consolidation is poisonous for both innovation and returns is slowly but surely becoming mainstream.

    Can’t wait for this to become a given so we can (finally!) move onto the much more interesting topic of “What comes next?”

    Geri Kirilova, September 2024

    First, the most concise formulation of the problem:

    LP allocation to “VC” grew so quickly that there was no opportunity for the strategy to adapt and properly allocate the additional capital. Instead, it was captured by opportunists.

    Thus, a majority of today’s “VC” activity is simple financialisation.

    • Venture capital involves making long-term investments in innovative, high-growth companies, with the goal of capturing outlier returns at exit.
    • “VC” is the process of using financial engineering to optimise fee income. Maximising proxy performance metrics by manifesting herd-like market momentum.

    “VC” isn’t intrinsically bad: the agglomerators are responsible for pulling billions of dollars into technology investment, by developing a product suitable for the largest and wealthiest LPs with the lowest expectations.

    However, there are three points the market needs to grapple with in order for venture capital to move forward:

    1. Agglomerators have no business investing prior to Series C

    The argument for their existence is that technological development often requires vast pools of growth capital, so why do they invest at seed?

    The truth is rooted in financialisation: it’s easier to produce a mirage of proxy metrics if you can pick startups that fit the momentum narrative, and founders that will go along with it.

    Another part of the reason is that these firms are doing very nicely from absorbing capital in the “VC” allocation bucket. Registered RIAs moving further away from early-stage investing might invite some unfavourable comparisons to their PE peers.

    Indeed, exactly whose allocation are they displacing? Is it venture capital? Or is it the PE firms they maybe more closely resemble? Or is it the public markets they have drained of new growth opportunities?

    We also know:

    • The hypercapitalising behavior of these large firms systematically breaks young startups.
    • They poorly built for early-stage investing, as large firm dynamics favour more obvious investments.
    • A decade of deal activity shows these firms are actively concentrating into consensus anyway.

    The early-stage activity of these firms achieves very little of any merit. They bend the market to their consensus narrative, suffocate genuine innovation, and turn startups into volatile commodities in their hunt for “market winners”.

    2. The agglomerator model of “VC” is an experiment

    For all that many of these firms are well-established names in the venture capital ecosystem, their strategy is not.

    In describing this bifurcation of the venture market, back in 2020, Nikhil Basu Trivedi stated “the next decade will be a referendum on agglomerators“. Essentially, until we see how these multi-billion dollar funds play out, it’s unclear what future the strategy has.

    Indeed, we can learn from history: after the dotcom bubble burst, the industry collectively swore off >$1B funds (for a while), with GPs reflecting on the challenges of scaling venture capital.

    There are similar lessons from the history of “mega-buyout funds”, as described by Meghan Reynolds of Altimeter:

    “Mega Fund dynamic in VC mirrors the meteoric rise of Mega Buyout funds in ’05-’08. Post GFC, Mega Buyout went out of favor w/ many LPs. What happened after was an emergence of strong, highly sought after <$2B “middle market” funds that were thought to have greater alpha.”

    Meghan Reynolds, October 2022


    3. Agglomerator practices have corrupted venture capital

    Perhaps the most painful and pernicious aspect of all of this, is that the conflation of these multi-stage agglomerators with venture capital has led to a corruption of standards and practices.

    Let’s be clear: venture capital has very little in the way of standards and practices. What there is, tends to be a mimmicry of “the loudest model“, which is the agglomerator playbook.

    So, when they say things like “entry price doesn’t matter”, or “non-consensus investing is dangerous”, or “the only thing that matters is getting into the best deals”, they are talking their book. It reflects a strategy that simply does not apply to venture capital, where entry price does matter, all alpha is non-consensus, and nobody can predict the best deals.

    The same is true for LPs, who derive much of their understanding of venture capital, and what “good” looks like, from these “loudest models”: When Marc Andreessen says “AI will save the world”, LPs demand that GPs have an AI thesis, even when a smart GP may be looking beyond that horizon, or at overlooked alpha elsewhere.

    So, what comes next?

    Hopefully, a number of things play out over the next five years:

    • We see the limitations of the agglomerator model, and venture capital can reclaim the early stage, yielding more effective and diverse origination with saner pricing and less volatile outcomes.
    • LPs slowly wise-up to this bifurcation of strategies, re-educate themselves about venture capital, make better allocation decisions and feel less inclined to impose upon GP strategy.
    • Agglomerators (re-classified as “venture growth”) become a defined sub-category of LPs private book, like venture capital. Publications like Pitchbook and CA start disaggregating performance into this new bracket.
    • Anyone who wants to play a fee-driven financialised game can go to venture growth, those who are driven by performance and impact can work in venture capital. Two different systems, each better understood and playing to their strengths.
    • Increasingly, venture capital will be able to tap into venture growth for liquidity as their portfolio matures, accelerating liquidity and feedback cycles to produce compounding performance gains.

    Slowly but surely, venture capital may return to being a better-performing and more positive-sum strategy, focused on finding outliers and the discipline of properly managed risk. Less herd behavior, and less cognitive dissonance.

    (top image: Meindert Hobbema’s The Avenue at Middelharnis)