Diversification in Startup Investing: The Autopilot Thesis

Photo by Denys Nevozhai on Unsplash

In the past couple of months, we have been busy structuring our co-investment fund, Autopilot. Launching a first-time fund is never an easy job. Amongst the key challenges that need to be addressed is to have a clear and differentiated investment proposition. And for us one of the main differentiators is diversification. We are building what we believe to be one of the most – if not the most- diversified portfolio of early-stage tech assets.

There were two fundamental points we believe to be true:

  • Extreme diversification statistically increases the chances of landing home runs and unicorns – if there is a mechanism to ensure the quality of startups don’t fall with increasing the size of the universe
  • With extreme diversification, we are still able to maintain the target returns albeit with much less volatility!

Bold statements, but are they supported by evidence?? We set out to examine the benefits of diversification in order to test our hypothesis, and the results surprised even us. But first, a bit of background.

So what is diversification?

Common wisdom suggests that you should not put all your eggs in one basket. And that is what diversification is about. The goal is to reduce risk. The effect of diversification is to offset the negative performance of some assets by the positive performance of others.

This is particularly important in early-stage investments which are inherently high risk (and high return) assets. But what is the optimum level of diversification in such portfolios? How many investments do you need to make to increase your chances of delivering your target returns? As a general rule, only 1 out of 10 investment makes it big, with 2 out of 10 just about break even and 7 out of 10 becoming zombies. Traditional VC funds typically invest in 15-20 portfolios, bar some larger funds who can afford to invest in more startups and a new wave of VCs, primarily passive and data-driven, who target a much higher level of diversification to capture more of the better performers in their portfolios. So where does Autopilot stand in this spectrum?

This question led us to conduct a comprehensive analysis to model Autopilot diversification. The key objective of this exercise was to understand how risk and return profile of a portfolio changes with varying levels of diversification i.e. portfolio size, with the ultimate aim to prove our case for having a massively diversified portfolio.

We started off investigating historical return distributions among VC funds. We found a study by Correlation Ventures to be a good foundation to base our analysis on. Their study was conducted on more than 21,000 funding rounds in the US between 2004 and 2013. The table below summarizes their findings:

We used this distribution – buckets of realised multiples- as a starting point to base our theoretical models on. Next step was to define how to calculate risk and return. IRR is the most common measure of return in VC, and that is what we used as well. Standard deviation, on the other hand, is a measure of volatility and we opted to calculate that as a measure of risk. We then used a number of assumptions based on some empirical evidence. We assumed an 8 year time to exit for all companies and an identical start and end date for simplicity i.e. all companies being funded at the same time and exit at the same time. We also assumed that all portfolios will follow the same distribution of returns as the one exhibited in the table above.

Next we defined different portfolio target sizes i.e. portfolios comprising of 15, 25, 50, 100, 150, 250, 500, 1000 companies. For each portfolio size, we calculated the approximate number of companies within each return bracket.

Next, we generated random variables within each bracket and for different portfolio sizes to model the return of a fund which follows the same distribution. We then calculated the IRR and Standard Deviation for a number of different scenarios for each portfolio. We ran this both with normally distributed random variables as well as log-normally distributed random variable as VC returns tend to follow a power law distribution – a small number of companies account for the majority of the return at the portfolio level…

Ok, we are getting a little bit geeky here so I will leave that for another post. Here are our findings and tweaks:

  • The first and most common sense finding is that extreme diversification significantly enhances chances to land more of the best performing startups. That is if there is a mechanism in place to ensure quality of startups are not compromised as the quantity increases. Not many investors can claim to do that. This is where a passive investment strategy with automated and standardised vetting can pay dividends!
  • As the portfolio size increased, the IRR diminished slightly but the standard deviation as a whole increased! This was not what we expected! We actually wanted to prove the opposite to make a case for extreme diversification! But, here is what we had ignored: At large portfolio sizes, we are statistically increasing the chances of having extreme positive returns i.e. a company with let’s say a 100x return which in turn skews the standard deviation numbers – a property of power-law distribution
  • As a next step, we decided to exclude the extreme scenarios i.e. any return above 20x. A more conservative and realistic scenario. The outcome was much more in line with our expectations as an increase in portfolio size then resulted in a decrease – on average – of standard deviation and hence the volatility of returns.
  • To model this even more accurately we then segregated the upside volatility from downside volatility and focussed on the downside. Why ignoring upside? Well, at the end of the day who doesn’t like to have astronomical positive returns in their portfolio? So that is not a contributor to risk. This way we could see more clearly that an increase in portfolio size reduces the downside volatility around a set average.
  • So far all findings seem to be common sense but does all this mean that we have to compromise on returns i.e. IRR at larger portfolio sizes? Well in all of the scenarios, we did see a decrease in IRR which some might use as an argument against extreme diversification, but at the same time the target return of 20% IRR could still be maintained more or less in all cases. So a comparatively small compromise in IRR could yield much safer and smoother returns with much lower volatility over time. And considering only a tiny portion of the VC funds do actually meet their target returns, isn’t it a great proposition to investors?
  • Bonus point: Probably the most astonishing outcome that we observed, which we would like to explore further more systematically in our future studies, was to show that a large number of decent companies with decent returns of 3x-5x could on aggregate could yield comparable returns to other typical VC funds, even in the absence of any home runs and unicorns! Could it fundamentally change the early stage VC model if proven true?

We acknowledge all the limitations of our model. Yes, we have made a number of simplistic assumptions that although were based on empirical evidence, may not be an accurate reflection of reality. So there is so much more we can do to improve the model however the key findings out of this exercise were quite significant and did, in fact, prove the benefits of the sort of diversification levels we are targeting with Autopilot. The findings may seem common sense but nevertheless great to be proven scientifically.

This is just the beginning. We will continue improving our models and fine-tune assumptions and scenarios but even at this level, we are quite excited about the findings! Stay tuned for more to come…

Leave a Comment