Power law returns in VC from AngelList data science

Early and late-stage VC are very different asset classes. Returns can be unbounded for the very earliest stage investments, and early-stage investors are better off building a big portfolio by buying “every credible deal.” New evidence for these statements comes from Abraham Othman, Head of Data Science at AngelList, in his paper on Startup Growth and Venture Returns. I’ll skip the math and explain his graphics.

In looking at these graphs, we should remember that they only apply to WINNERS. Othman started with a dataset of 3000 startup investments, and selected “…a database of 684 non-negative investments…”. I assume that most of the other 2000+ investments resulted in losses to investors. These graphs do not make claims about the overall rate of return to investors who have both wins and losses.

Othman focuses on the distribution of winners, modeled as a power-law distribution. He is trying to fit this distribution to some parameters that we can understand:

• Rate of return — how much do we make per year on winners?
• Duration — how long does it take to get our money back?
• Alpha — How much of the returns are in the biggest wins? Lower alpha indicates that more returns are in the biggest wins.

This does create a link to absolute returns, because in situations with low “alpha”, the losses don’t matter. The wins are big enough to wipe out the effect of small gains and 1X losses.

From Othman, Startup Growth and Venture Returns

AngelList startups become winners because they have high annualized returns in their earliest stages. In Othman’s model, they literally slow down as they get older.

What Criteria Do Venture Capitalists Consider When Seeding a Startup? | Data Driven Investor

This supports the Kauffman Foundation’s claim that the “J-curve” of VC returns is created by funds, not startups. If investors are experiencing negative returns in the early years of a VC investment, it’s probably not because the underlying investments take time to show value. It’s either because a lot of investments get written off right away, or because fund managers are charging a lot to get into a deal.

From Othman, Startup Growth and Venture Returns

The chart above introduces the most important concepts of power-law returns. It looks at the “alpha” characteristic of investments made as a company gets older. It divides the world of returns into three categories, red, yellow, and green.

The red category of later-stage investments has an alpha of more than three. It behaves almost like a “normal” investment portfolio, in which you can realize an average return if you have more than 20 or so investments, and you can reduce the variance by having a bigger portfolio.

The yellow category of investments made in years two and three has an alpha of less than three, where the variance becomes unbounded. The more you add to your portfolio, the more big returns you get that throw off the averages. This leads to a problem in which small portfolios are unlikely to earn even the average return, because they miss the big winners. This tendency of typical returns to be less than average returns is called “skew”.

Returns in the green area (alpha < 2) are theoretically unbounded. The more investments you make, the bigger your expected returns. You get more chances to get a big winner, and your returns are concentrated in the big winners. Othman finds that in very early-stage investing, the combination of high skew and increasing returns means that you benefit from buying “every credible deal.” This is the land of spray and pray, or at least large accelerator cohorts.

From Othman, Startup Growth and Venture Returns

The above graph shows how a portfolio can get to an “infinite mean” distribution after five or six years. Public market investments will show a few big wins (high skew and low alpha) after twenty or thirty years, because stocks go up and stocks go down, and this effect compounds. Skew compounds faster with higher returns and higher variance of returns. In early-stage startups, the variance of returns is high, and the average returns for winners is high. This puts some startups on a trajectory to deliver skew-inducing returns within five years.

What about returns?

Can we use these findings to resolve a conflict between two studies of VC returns? A study from Dan Hoogterp of Hatcher+ found that the most lucrative investments are at the very earliest stage, what they call the “accelerator round”. This matches the green area in the AngelList graphs. Only 4.4% of investments at this stage return any money to investors, but the pricing is good due to this risk, and investors with large portfolios get a high density of unicorns (about 1%). Investors in later rounds make less money and struggle to beat the S&P.

Hoogterp shows declining returns for later rounds, where Series A=4

A study from Cameron Stanfill and Bryan Hanson at Pitchbook found the opposite. They acknowledge high returns for winning early-stage deals, but they think that early-stage investors actually lose money because of a high rate of failures. They believe that the very last round, “Series F” has the best returns, at 4.4% annualized.

Pitchbook shows rising failure adjusted returns for later rounds

These studies are different in how they account for companies that just disappear from the statistics. Did they die with losses, or will they have exits later? Pitchbook calculates a failure rate for each round and assumes that a percentage of investments will fail with zero returns. Hoogterp builds a model with machine learning to predict the valuation of missing rounds, including the exit value.

The AngelList findings could support either position. AngelList removed failures from their statistics, so they can be consistent with the Pitchbook finding that if you include failed investments, you get losses. On the other hand, Pitchbook did not include data for the very earliest rounds, where AngelList and Hatcher+ are focused, and where outliers are important.

AngelList shows that early-stage investments are different from later-stage investments. And, they confirm the Hatcher+ findings. I would bet money on the Hatcher+ and AngelList findings of the way that very early stage investments make money.

The Pitchbook results may apply for later stages. I am inclined to be skeptical about Pitchbook’s finding that essentially every VC investment round underperforms public market indexes unless it happens to skew into a “top quartile” fund. This would imply that most VC fund investors are idiots. Maybe bigger returns are still lurking in residual values, in a big win that pops out of power-law returns. However, Pitchbook has good data for later-stage investments and we should use it.

The analysis from Othman is exciting because it opens up the possibility of unbounded returns for very early-stage investments. It also shows why early-stage investors behave differently from later-stage investors. Later stage investors benefit from hands-on attention to their portfolio, where early-stage investors benefit from moving on to expand the portfolio. Pick your category, and multiply your investments.

This is the first in a series of articles on how to deal with the vast scale of Power Law Returns. Please follow me for more.