Portfolio management in power law environments, or how we (might) make money

Reflections on portfolio management in venture capital, in particular misconceptions about power law returns and diversification.

Creating a portfolio of teams (or how we’re going to make money)

Garry Elevator aspires to manage its portfolio in ways that traditional early-stage portfolio management might deem irrational. We concentrate the portfolio behind relatively few investments. We encourage the longevity of “underperforming relationships” through additional capital and effort. Doing so recognizes the important realities of compounding in power law environments, as well as realities of relationship. While traditional portfolio management spreads bets widely, to a certain extent treating people as “return fulfillment machines”, our approach hopes to have multiple shots on goal with teams we know well and trust. This approach also aligns with emerging fields of economics, which tell us that it’s the best way to make money assuming we identify leaders with a strong risk tolerance and good judgment.

What follows is an attempt to answer the question “how will you generate positive returns?”. It should be answered directly by all venture capitalists. I can’t believe that I’m including a Buffett quote in this memo, but I take seriously his comment that “when a management with a reputation for brilliance tackles a business with a reputation for poor fundamental economics, it is the reputation of the business that remains intact”. 

Early-stage investment is a hard way to make a living; accepted wisdom holds that returns follow power law distributions with unattractive medians, tolerable means, and a vast majority of gains accruing to a relatively small percentage of total opportunities. Recent data from AngelList – as close as the venture ecosystem comes to an index – backs that theory with evidence. 

Chart, histogramDescription automatically generated

Figure 1: AngelList Deal Returns [MOIC > 1x, 2011 - 2019]

The distribution above is truncated for the purposes of visibility on a linear scale. Were it to be displayed logarithmically (with losses removed) it would clearly display power law dynamics. 

Chart, line chartDescription automatically generated

Figure 2: AngelList Deal Returns [MOIC > 1x, 2011 - 2019], losses removed for legibility

Understandably, it’s hard to generate consistent returns in an environment that rewards the discovery of a rare winner among frequent losers. Current portfolio management wisdom advocates for significant diversification across company, stage, and vintage dimensions to maximize changes of catching a winner. Firms invest widely, cut losses quickly, and deploy as much capital as possible into successful teams (at ever increasing prices). 

This intuitive approach is valid, but it tends to deliver index-like returns with upside optionality. Most people are waiting to get hit by the lucky truck.

At the start of a new effort, it is worth questioning accepted norms. An analysis of the drivers of return in non-ergodic ecosystems (which produce power law distributions) brings us not a refutation of venture portfolio management, so much as a slight revision. Our portfolio will be more concentrated. Our investment size will be determined by how many experiments it will take a team to learn and monetize. We will take actions (like follow-on investment) to make sure a team's risk tolerance stays high. We will hold successful investments longer than our peers. These small differences, compounded over time, can deliver returns well outside of normal bounds.

Two caveats: 1) that I am not an expert in ergodicity economics, so this analysis will necessarily be less bold and less detailed than it otherwise might be 2) statistical theory on random games can only take us so far in a space that is not completely random. It is a strong place to start, however, and will usefully expose some assumptions in our portfolio management framework (on which I have slightly more bold opinions).

From a statistical perspective, however, the appearance of power law distributions indicate that we’re dealing with non-ergodic processes. A stochastic process is said to be ergodic if its statistical properties can be deduced from a single, sufficiently long, random sample of the process. Non-ergodic processes emerge when compounding probabilities deliver time series experiences that are drastically different than ensemble experiences of similar observations. For example, a game in which a fair coin flip determines a win (+50%) or a loss (-40%). The system as a whole is better off for playing this game…

Chart, line chartDescription automatically generated

Figure 3: Ensemble average of 60M observations [Ole Peters, TEDx Goodenough]

… on an individual time series experience of similar observations, however, the result is markedly different.

Chart, line chartDescription automatically generated

Figure 4: Time series experience of 60M observations [Ole Peteres, TEDx Goodenough, "Time and Chance" 2011]

We see that ensemble experiences are preferable to any one time series because they necessarily contain more flips at or above par value. Given a fair probability, wins at or above par create greater value than wins below par. The percentage nature of wins in this game make it hard to recover from stepping under par as a string of wins is required.

Hard, but not impossible. While it’s less likely to have a hugely positive time series following an initial loss than an initial win, it’s mathematically true that some outlier outcomes experienced difficult beginnings. The ensemble averages do not truncate any of those time series experiences, allowing winners to compound, some losers to recover, and the remainder to move asymptotically towards zero, affecting total value less and less as time goes on.

We come to our first insight. A purely statistical approach would encourage venture capital to allow teams to continue their time series experience as long as they can – a sort of “do no harm” approach to investing more often governed by founder spiritual runway than it is by access to capital. As long as you have the will to keep running this company, keep going.

Prevailing portfolio management wisdom delivers the exact opposite approach, however, exacerbating the power law distribution by truncating some time series in favor of others, typically related to the path of their first few flips. Many investors advocate for a quick acquisition to realize a small gain or avoid an early loss. They do this for understandable reasons such as fund life concerns or constraints on time as well as capital. But the net result is that some venture firms eliminate companies that might go on to contribute positively to the ensemble average. This makes outliers as a percentage of total outcomes more rare. 

If “do no harm” would be less penalizing to ensemble returns, “active curing” of time series losses would be value-additive. Resetting an entrepreneur’s bankroll after a few initial losses would statistically deliver significant benefit.

This strategy only becomes more relevant if we admit the limits of this imagined game and introduce a few human considerations. For example, statistics imagines that an individual’s risk tolerance and probability of success are static over time. This defies much of our lived experience. Adding additional capital in the event of adverse outcomes would help to maintain an entrepreneurs confidence and risk tolerance. As a result, it may increase their probability of success.

Risk tolerance changes significantly at the long ends of the spectrum – in an endeavor’s earliest days, and at maturity. At inception, risk tolerance is particularly susceptible to influence from outside conditions. If early outcomes affect the persistence of a capital base, investors work hard to have a positive first year, even lowering their risk tolerance early to lock in gains and accepting sub-optimal positive outcomes as opposed to losses. When capital runs low, entrepreneurial risk tolerance can skyrocket well above reasonable levels. They may bet the company’s future on a fundraising process or go all in on one client. 

As demonstrated above, venture capitalists want a high, but reasonable, risk tolerance from leadership in order to deliver power law outcomes. A lack of clarity around follow-on funding dramatically changes entrepreneurial risk tolerance, and it also affects what leaders are willing to ask of and promise to their teams, further lowering the probability of success. Clarity, and variant follow-on strategies can be exceptionally helpful to entrepreneurs that are pursuing high risk endeavors and experience early adverse outcomes.

At the long end, risk tolerance and probability changes materially as well. One shouldn’t wish to play a game with 40% downside forever, as it imperils years of hard-won gains. In recognition, most large companies would assert that they have a significantly lower risk tolerance, but that each win is compounding off a higher base. I find this by and large to be true, particularly for centralized decision-making entities. The best winners, however, are like Dee Hock’s chaordic organism. They effectively push autonomy down the organization so that the individual risk tolerance remains high, but the company itself becomes the ensemble experience – a collection of time series experiments with a few extremely positive outliers that drive value creation at the entity level. Amazon is the classic example of such an organization, with many underlying failures contributing to overall success.

Early winners and continual winners, by virtue of their wins, attract more and more resources to themselves in the form of human, intellectual, and financial capital. As a result, they’re continually restoring or increasing their base above par and maintaining the probability and magnitude of potential wins, a good example of this style of portfolio management.

Venture capital should optimize for holding winning investments as long as possible. This is the kind of diversification that benefits capital partners, as opposed to diversification that comes through ownership of thirty-plus startups per fund. 

The lessons for venture portfolio management are twofold: 1) venture portfolios should reflect a “stop-loss mentality” which regards a first investment as just that, a starting allocation from a larger stack of capital. If it is lost or less than profitable, that is just the first phase of flips or assessment. More capital allows for more flips, and potentially more attractive ensemble experiences. 2) re-allocating a small portion of winnings to underperforming strategies can yield exceptional returns if those strategies have sufficient upside potential. 

Holding winners and supporting underperforming bets is the best way to access outsize returns. 

If a risk taker is making strong upside bets with longevity, an investor’s response shouldn’t be, “not this one, but the next one”, it should be “this one, and the next one”. Recognizing that risk takers open you up to good randomness as much as bad may introduce some uncomfortable decisions, but over time lead to attractive, sustainable returns.

This approach is most difficult because it’s the most relational, not only in the assessment of leadership capabilities, but also in the investment of assets. Relative to the somewhat tight feedback loops of the stock market, venture investors need to be working on sub-scale projects with community leaders to receive feedback on whether they’re making high quality bets regardless of the outcomes. 

Additionally, where other investors allocate only financial capital, venture investors need to invest human and intellectual capital in ways that are not boilerplate across endeavors. Relationship depth and knowledge, even if built in a failed endeavor, lowers the friction to allocating useful resources, an exceptional value that is currently underappreciated by venture portfolio management dynamics.

Investors should take heart that though relationships do not spread, they scale. I cannot create this kind of intimacy with thirty leaders over 3-4 years, as traditional portfolio management would require. The quality of each decision, however, should be more sound, and the amount of capital able to be allocated through any one relationship, significantly higher.

As a corollary, the fewer bets it takes in order to reach a viable business model, the fewer total entities required to deliver sufficient diversification. Given the operating leverage inherent in the technologies our companies build on, we expect that a much higher percentage of them will reach default alive than their venture peers. At that point, we expect them to be flipping coins "above par" and shooting for a generational outcome.

These beliefs coalesce in a portfolio management framework oddly resembling socialism. “From each according to outcome, to each according to need”. The firm will recycle cash and profits from successful endeavors into new or incremental investments. This is not just a kindness. So long as we select and support risk-on leaders who have the longevity to keep building, it is the probabilistically optimal strategy as displayed in this simple video at 

Want to learn with us?

Sign up to get an email every time a new article is published. We share interviews, primary source data, books that provoked us, and reflections on emerging themes.