Venture Capital Growth in Portfolio-level Returns November 2021 Richard Nigel Inciong, CFA Contents We analyze data from over then thousand deals Crunchbase deals to determine the impact of several factors drivers to portfolio-level returns of early stage venture investments. These factors are: 1) 2) 3) 4) Introduction ......................................................................................... 3 Background ........................................................................................ 4 Preliminary Definitions ........................................................... 4 Introduction to Power Laws (Othman, 2019) .... 5 Capital Asset Pricing Model (CAPM)....................... 6 Number of investments within the portfolio Portfolio Age, and Loss due to startup failure rate. Investment Horizon This allows us to examine the magnitude of impact of these factors to a hypothetical early stage venture capital portfolio. Our results suggest that there is indeed an empirically measurable positive impact to portfolio-level returns from the increasing number of investments BETA Coefficient ................................................................... 6 Crunchbase Data ....................................................................... 7 Individual Investment Returns ......................................... 7 Portfolio Simulation Methodology ................................ 9 Assumptions on Loss............................................................10 Correlation & Beta to Overall Market Returns ..........11 Results .................................................................................................. 12 Effect of Different Portfolio Sizes ................................ 12 Impact of Portfolio Age ....................................................... 15 Impact of Investment Horizon........................................16 Closing Thoughts .......................................................................... 17 References ........................................................................................ 18 KEY TAKEAWAYS Our analysis based on investing in a randomized VC portfolio suggests that : ο· Expected portfolio returns increase with the number of companies giving approx. 5bps higher IRR for each company made ο· Expected portfolio returns tend to stabilize at 3+ years with approx. 50% chance of at least 5% IRR ο· Majority of companies do not see positive returns but are made up for by outsized gains from a small portion of portfolio ο· Timing of Investments does play in impact to portfolio returns supported by demonstrating a negative -0.66 beta to S&P returns Portfolio IRR vs. Portfolio Size (n=1000/series) 12.0% 10.0% 8.0% 6.0% 4.0% 2.0% 0.0% 25 50 Mean IRR 75 Median IRR 100 Introduction In this paper, we use Crunchbase data of early-stage investments to determine whether there is an impact on portfolio-level returns from several factors including: 1) 2) 3) 4) Number of investments within the portfolio, Portfolio Age Loss due to startup failure rate (which in our paper is based on our assumptions) Investment Horizon Based on our analysis, our results suggest that making an increasing number of seed-level VC investments can increase the overall level of portfolio returns. This differs from conventional equity investments where increasing the number of investments generally only serves to decrease return volatility. We approximate the impact of this to be 5bps to portfolio-level returns from each additional seed-level VC investment made. While the number of investments made make the most significant & measurable impact, other factor do impact VC portfolios. We see that portfolio returns between 1-3 years generally do not see appreciable returns due to the irregular nature of VC funding. Failures in the VC space also make a large impact to overall returns. Given the lack of information on failures, assumptions were used in the analysis which also shows that differing levels of failure rates can have outsized impacts on the portfolio returns. Also, the investment horizon and/or time period unsurprisingly makes a difference. Another interesting result of our analysis shows that overall correlation to the market is low (-0.11) and has a beta lower than 1 (-0.66). This shows that including VC investments can be shown to have a positive impact as part of a diversified investor portfolio. Venture Capital Growth in Portfolio-level Returns | Page 3 Background In this section we provide definitions of our core concerts, introduce power-law distributions and an existing academic model of venture capital returns, and discuss the Crunchbase data used in the results. Preliminary Definitions A portfolio’s Internal Rate of Return (IRR) is the rate of growth r that equilibrates between incoming and outgoing cashflows. π£ β (1 + π) = |π£| β (1 + π) In this paper we will quote IRRs on an annualized basis. For overall clarity, we will often abuse language and refer to the quantity 1 +r as the “IRR” of the investment, particularly in the case of draws from a power-law distribution. An portfolio’s return multiple m is the combined sum of the distributed and residual values of the investment divided by the total amount invested. Formally: π ≡ ∑ ∑ π£ |π£| An investment’s effective duration is defined as the amount of time such that the investment’s IRR implies its return multiple. Formally, d is an investment’s effective duration that solves: π = (1 + π) Venture Capital Growth in Portfolio-level Returns | Page 4 INTRODUCTION TO POWER LAWS (OTHMAN, 2019) The seminal quantitative work on modern power laws is Clauset et al. (2009). Taleb (2001, 2007) also discusses power laws at length from a less quantitative perspective, focusing on their relationship to finance, culture, and society. A power-law distribution (with shape parameter α > 1) is distributed according to the probability density function: π(π₯) ≡ πΌ−1 π₯ π₯ π₯ In the case where xmin = 1 power law PDF reduces to simply: π(π₯) = (πΌ − 1)π₯ There are three cases to consider depending on the shape parameter α, each of which can have different implications for portfolio construction and assessment: • • • When α > 3, the distribution has finite mean and finite variance. With finite variance, the Central Limit Theorem holds and so important portfolio theory concepts like the Sharpe Ratio have meaning. If investments draw their returns from a distribution with α > 3 and investors select randomly among them, then the number of investments in a portfolio does not affect that portfolio’s expected mean or expected median return. When 2 < α ≤ 3, the distribution has finite mean but unbounded (infinite) variance; as a result, the Central Limit Theorem does not hold. If investments draw their returns from such a distribution, making more investments at random will increase a portfolio’s expected median, but not expected mean, return. When α ≤ 2, both the mean and variance are unbounded. If investments draw their returns from such a distribution, making more investments at random will increase both a portfolio’s expected median and expected mean return. This last case of an “escaping” distribution is extremely difficult to make intelligible. Such distributions are far outside the realm of our typical experience. In our opinion, the most illuminating description of such a distribution comes from Taleb (2001), who describes a “Refugee Probability Distribution”: for each additional day that a refugee spends outside of their homeland, the number of days that they can expect to wait to return increases by more than one. The heavy tails of power laws are generally thought of as their most unintuitive property. However, another unintuitive property is that the power-law distribution, but unlike a Gaussian or uniform distribution, has qualitative properties (i.e., the existence of moments) that are highly dependent on its shape parameters. Venture Capital Growth in Portfolio-level Returns | Page 5 CAPITAL ASSET PRICING MODEL (CAPM) The Capital Asset Pricing Model (CAPM) describes the relationship between systematic risk and expected return for assets, particularly stocks. CAPM is widely used throughout finance for pricing risky securities and generating expected returns for assets given the risk of those assets and cost of capital. The formula for calculating the expected return of an asset given its risk is (Kenton, 2021): ERi=Rf+βi(ERm−Rf) where: ERi=expected return of investment Rf=risk-free rate βi=beta of the investment (ERm−Rf)=market risk premium BETA COEFFICIENT The Beta coefficient is a measure of sensitivity or correlation of a security or an investment portfolio to movements in the overall market. We can derive a statistical measure of risk by comparing the returns of an individual security/portfolio to the returns of the overall market and identify the proportion of risk that can be attributed to the market. (Beta Coefficient, 2021). We can calculate beta using two formulas: 1. Covariance/Variance Method: π΅ππ‘π = πΆππ£πππππππ (π , π ) ππππππππ (π ) Where: πΆππ£πππππππ = 1 π−1 1 π (π ππππππππ = (π , −π −π , , , ) × (π , −π ) , ) Re = Individual Equity Return Rm = Overall Market Return 2. Correlation Method: π΅ππ‘π = π , × π π Where: π , = correlation between Re and Rm π = Standard Deviation of Re π = Standard Deviation of Rm Venture Capital Growth in Portfolio-level Returns | Page 6 Crunchbase Data The dataset used for the analysis was constructed in several steps to align with the analysis requirements. The analysis is meant to simulate a portfolio of investments into various companies’ seed rounds and tracking the overall value of the portfolio. Initially, Transactional data was pulled from the crunchbase website from companies in Canada & the U.S. as well as included a company valuation. This initial step included 9052 separate funding transactions between 2000 and 2021. Secondly, due to IPO transactions not being part of the Crunchbase transactional data, IPOs of companies between 2000 and 2021 were pulled separately and added to the dataset. This included 1707 IPO rounds. Lastly, to narrow the dataset to those relevant to the analysis, funding rounds were filtered based on listing their first initial seed round between 2010 – 2018. The final dataset includes 1891 companies which meet this criteria and 2528 total funding rounds related to these companies. Individual Investment Returns Figure 1 shows the power law fit for the return multiples of winning investments. As expected, return multiples are consistent with an extreme lower law distribution. This fit was made using the excel trendline feature using the power-type trendline. Return Multiple for individual Winning investments for x>1 Normalized Frequency of Returns > x 1 0.1 0.01 y = 0.9617x-0.633 0.001 1 10 100 1000 10000 100000 1000000 Return Multiple Figure 1 - Return Multiple for individual Winning investments for x>1 Generally in the standard model, investment duration and IRRs are drawn independently, implying that there should be no correlation between the two quantities. However, the Crunchbase data indicated that there is a Venture Capital Growth in Portfolio-level Returns | Page 7 weak negative correlation between the two quantities (-8.6% correlation). This relationship is shown in Figure 2. That said, the majority of investments (76%) only show one round of funding which would require assumptions to be made on losses. These loss assumptions are expected to have an outweighed effect on portfolio level returns with regards to the effect of portfolio age vs. portfolio returns. Annualized Return vs. Duration (>1Y) (n=487) 2000% 1800% 1600% Annualized Return, r 1400% 1200% 1000% 800% 600% 400% 200% 0% 1.000 3.000 -200% 5.000 7.000 9.000 11.000 13.000 Duration Figure 2 - IRR vs. Effective Duration Venture Capital Growth in Portfolio-level Returns | Page 8 Portfolio Simulation Methodology To simulate the return of a hypothetical VC portfolio, a Microsoft excel model was built to measure the Return Multiple of an entire portfolio of investments monthly from in the initial investment. Each investment within the portfolio is selected randomly based on specified criteria. I.e. having their first seed round within the specified time horizon of the simulated run. There were four time horizons used which are specified in Table 1 – Time Horizons used in simulations and number of companies in the set. Table 1 – Time Horizons used in simulations and number of companies in the set Time Horizon (First Seed Round) 2010-2012 2013-2015 2016-2018 2010-2018 Number of Companies in Set 217 709 965 1891 The Return multiple of the portfolio was measured monthly with each companies’ value updated based on Crunchbase funding round data. i.e. if there was a subsequent funding round after the initial seed with a new valuation, the company’s value within the portfolio would be updated to reflect the new valuation. It should also be noted that each run included 100 companies with portfolio values also being recorded for portfolio sizes of 75,50, and 25. Essentially, each smaller size being a subset of the larger. However, given the vast majority of companies only had a single seed round (76%), these companies values were modelled based on specific loss assumptions to reflect the inherent risk in VC portfolios. These loss assumptions are further detailed in the section Assumptions on Loss. Figure 3 – Example simulation run of Portfolio of 100 Investments vs similar portfolio of S&P investments shows an example of a simulated run on a portfolio. 900% Portfolio of 100 Value - VC vs. S&P 500 800% Return Multiple 700% 600% 500% 400% 300% 200% 100% 0% 1-Jun-10 1-Jun-11 1-Jun-12 VC Portfolio 1-Jun-13 Date 1-Jun-14 S&P 500 Figure 3 – Example simulation run of Portfolio of 100 Investments vs similar portfolio of S&P investments Venture Capital Growth in Portfolio-level Returns | Page 9 Assumptions on Loss A significant portion of venture investments do not see a return and a large percentage of startups fail. This is partially reflected in the Crunchbase data where the majority (76%) of companies have only raised a single seed round with no subsequent rounds recorded. Due to the limitations of data within the space, it is difficult to ascertain whether companies have closed down or merely stopped listing. To simulate the loss inherent in VC portfolios, companies which only list a single seed round of funding are modelled to lose value using a straight-line amortization (24 or 48 months) from the initial seed round to a specified loss magnitude (50% or 100%). Other sources indicate that approximately 50% of investments fail after the first 5 years. As such for the results section, the 48 months amortization and 50% loss magnitude are used for analysis. Table 2 – Impact of Loss Assumptions to Power Law Coefficients (90% Interval) Loss Assumptions on Single Round investments SL-Amort Period Loss Magnitude 24 Months 100% 48 Months 100% 24 Months 50% 48 Months 50% Power Law Coefficients (90% Interval) K Exponent 0.5089 .800 0.6704 .690 0.6419 .847 0.8805 0.633 Impact of Loss Assumptions to 5Y Return Multiple 100000 Return Multiple 10000 1000 100 10 1 0% 10% 20% 30% 0.1 40% 50% 60% 70% 80% 90% 100% Normalized Frequency of Portfolios > Return Multiple 24Mo/100%Loss 48/100%Loss 24Mo/50%Loss 48Mo/50%Loss Figure 4 – Impact of Loss assumptions to 5Y Return Multiple Impact of Loss Assumptions to 5Y Return Multiple CDF (90% Interval) Return Multiple 25 5 1 5% 0.2 15% y = 0.5089x-0.8 24Mo/100%Loss Power (24Mo/100%Loss) 25% 35% y = 0.6704x-0.69 45% 55% 65% y = 0.6419x-0.847 Normalized Frequency of Portfolios > Return Multiple 48/100%Loss 24Mo/50%Loss Power (48/100%Loss) Power (24Mo/50%Loss) 75% 85% 95% y = 0.8805x-0.633 48Mo/50%Loss Power (48Mo/50%Loss) Figure 5 - Impact of Loss Assumptions to 5Y Return Multiple (90% Interval) Venture Capital Growth in Portfolio-level Returns | Page 10 Correlation & Beta to Overall Market Returns Diversification can play an important role in an investor’s portfolio to producing stable growth. Venture Capital as an asset class can be shown to have a low beta to the overall market and therefore the argument could be made that it’s inclusion in an investor portfolio of different asset classes can reduce systematic risk. For this analysis, each portfolio of VC investments was compared against a portfolio of S&P 500 investments and this methodology is more summarily explained in the section Portfolio Simulation Methodology. The 1year return of 4000 VC portfolios was compared against 4000 S&P 500 portfolios giving the following distribution: Beta - VC 1Y Return vs. S&P 500 1Y Return 10.00 VC Portfolio Return Multiple 9.00 8.00 7.00 6.00 5.00 4.00 3.00 y = -0.6658x + 1.7608 R² = 0.0121 2.00 1.00 1 1.05 1.1 1.15 1.2 1.25 1.3 S&P 500 Return Multiple Figure 6 – 1Y Returns vs 1Y S&P Returns Based on two methods of calculating beta, VC returns are determined to have a beta of -0.66 to overall market returns, respectively. This is shown in Table 3 – VC Beta calculation. Table 3 – VC Beta calculation Covariance Variance Beta - Covariance Method Value (N=4000) -0.0014041 0.0021094 -0.6656278 Correlation Std, VC Std, Market Beta - Correlation Method -0.1100836 0.2777025 0.0459178 -0.6657664 Venture Capital Growth in Portfolio-level Returns | Page 11 Results This section details the impact of different factors to the portfolio level returns from different factors. Effect of Different Portfolio Sizes If it is assumed that VC investments follow a power-law distribution with a shape parameter of less than α <2, then both the mean and variance are unbounded. Therefore, if investments follow this distribution it is expected that making more investments at random will increase both a portfolio’s expected median and expected mean return. This is seen to be the case with VC returns as seen from Crunchbase data. Table 5 clearly shows as the portfolio size of randomized investments increases then so too does the portfolio’s expected mean return as well as the median return multiple. 5 Year Portfolio Return of Different Portfolio Sizes (n=1000/series) 50000 Return Multiple 5000 500 50 5 0.50.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0% 70.0% 80.0% 90.0% 100.0% Normalized Frequency > Return Multiple 25 50 75 100 Figure 7 - 5 Year Portfolio Return of Different Portfolio Sizes (n=1000/series) On the left end of Figure 7, it can seen that there is a small (<5%) probability of outsized returns, which we will define from this point as greater than 1000%. The jump from the rest of the returns is similar to a Bernoulli distribution. It can be seen that the probability of outsized returns increases with portfolio size. This is intuitive as with a greater number of investments in the portfolio, the greater likelihood that one would provide these outsized returns. However, as can be seen in Table 4 – Probability of Outsized Returns, the average return of the portfolio is also lower. Table 4 – Probability of Outsized Returns Probability 1.6% 2.8% 3.5% 4.0% Avg 5Y Return Multiple 12,667 6,489 4,227 3,309 Venture Capital Growth in Portfolio-level Returns | Page 12 However, given the irregular behaviour of the outsized returns compared to the rest of the distribution, it is easier to focus on the 90% interval. This is focusing on the same distribution while excluding 5% of results on both sides which would exclude the asymmetrical pattern seen for the outsized returns. This should prove representative for the majority of returns that an investor might expect from VC portfolios. Normalized Frequency > Return Multiple 5 Year Portfolio Return (n=1000 / series) of different portfolio sizes (90% Interval) 64.0% 0.5 1 2 4 8 16 32.0% 16.0% 8.0% 4.0% 2.0% 25 50 75 100 y = 0.6373x-1.505 y = 0.7491x-1.587 y = 0.8306x-1.609 R² = 0.9511 R² = 0.9557 R² = 0.9539 Return Multiple Power (25) Power (50) Power (75) y = 0.9106x-1.662 R² = 0.9525 Power (100) Figure 8 – Probability Distribution of 5Y portfolio Returns of different portfolio sizes As can be seen in Table 5 and Table 6, as portfolio size increases, so too does the level of expected return. Increasing portfolio size from 25 to 100 increases the average expected return multiple at the 90% interval from 152% to 178%. Table 5 – Shape Parameters and Expected Return Multiples of Portfolios of different sizes (90% Interval) Portfolio Size Shape Parameters 25 50 75 100 K 0.6373 0.7491 0.8306 0.9106 -α 1.505 1.587 1.609 1.662 Expected Return Multiple (90% C.I.) Mean Median 152% 122% 163% 133% 172% 139% 178% 143% With respect to portfolio IRRs, it would also seem there is direct impact to the expected IRR vs size of portfolio with each successive step. This relationship is shown in Table 6 and Figure 9. Averaging these results gives the approximation Figure 10 - 5 Year Portfolio Return of S&P 500 Portfolios of different sizesof a 5bps increase in both mean and median returns per additional investment made. Table 6 - Expected IRRs of Portfolios of different sizes (90% Interval) Portfolio Size Expected IRR (90% C.I.) Mean Median 25 50 75 100 6.5% 8.4% 9.6% 10.5% 4.0% 5.9% 6.8% 7.4% Venture Capital Growth in Portfolio-level Returns | Page 13 5 Year Portfolio IRR (n=1000 / series) 0% 0% 0% 1% 2% 3% 6% Normalized Freq 0% 13% 100.0% 26% 51% 50.0% 25.0% 12.5% 6.3% 3.1% IRR Annualized 25 50 75 100 Figure 9 – Probability distribution of Annualized Return IRR, r It is also noteworthy to show similar portfolios made of investments in S&P have no apparent impact from the number of investments made into the S&P this is shown in Figure 10. 5 Year Portfolio Return CDF (n=4000) Normalized Fequency > m 100.0% 1 2 50.0% 25.0% 12.5% 6.3% 3.1% Return Multiple 25 50 75 100 Figure 10 - 5 Year Portfolio Return of S&P 500 Portfolios of different sizes Venture Capital Growth in Portfolio-level Returns | Page 14 Impact of Portfolio Age Given seed investments in the VC space only change in value when there are subsequent funding rounds and time between funding rounds can be long, it can take time for portfolio to show appreciable returns. As such, the age of the portfolio can play a significant role in the level of annualized returns within the portfolio. As would normally be expected from most investment portfolios, the return multiple increases with portfolio age shown in Figure 11 reflecting the impact of compounding returns. Return Multiple of Various Portfolio Ages (N=4000) Normalized Frequency > x 80.00% 40.00% 20.00% 10.00% 0.5 1 2 5.00% 4 8 Return Multiple 60 48 36 24 12 Figure 11 - Return Multiple of Various Portfolio Ages When looking at annualized returns on the log scale, negative returns cannot be shown and are excluded. What is of note is that as portfolio age increases, the probability of a portfolio producing non-negative returns increases from less than 20% to close to 80% from 1 to 5 years. This is likely due to the fact that loss assumptions are present which cause the majority of investments to decay in value for the first few years. Additionally, many “winning” investments only show returns once there is a subsequent round of funding which can take multiple years. The most interesting fact here is the annualized portfolio returns do tend to stabilize after 3 years, however this is most likely due to the investment horizon for all portfolios being 3 years, at a minimum. IRR of Various Portfolio Ages (N=4000) Normalized Frequency > x 80.00% 0.0% 40.00% 20.00% 0.0% 0.0% 0.0% 0.1% 0.2% 60 0.3% 0.6% 1.3% Annualized Return 48 36 2.6% 5.1% 10.2% 20.5% 10.00% 41.0% 81.9% 5.00% 24 12 Figure 12 - IRR of Various Portfolio Ages Venture Capital Growth in Portfolio-level Returns | Page 15 Impact of Investment Horizon Also present in the analysis results is investments of different time horizons to try to avoid biases in time period selection. As can be seen in Figure 13, each time period does have somewhat of an impact to portfolio returns. Overall, the 3-year investment horizons do show different shapes to overall returns but overall following the same trend. The one exception is the 2010-2018 Seed period which does show lower returns overall due to the averaging effect of spreading out investments. Normalized Frequency > m 5 Year Portfolio Return of VC Portfolio by Horizon(n=1000 / series) 100.0% 50.0% 0.5 1 2 4 8 25.0% 12.5% 6.3% 3.1% Return Multiple 2010-2012 Seed 2010-2018 Seed 2013-2015 Seed 2016-2018 Seed Figure 13 – 5 Year Portfolio Return of VC Portfolio by Horizon When comparing to market returns in Figure 14, returns are much more consistent and tighter in spread with the exception of the 2010-2018 Seed due to similar factors. 5 Year Portfolio Return of S&P Portfolio by Horizon (n=1000 / series) 100.0% Normalized Frequency > m 0.5 1 2 4 8 50.0% 25.0% 12.5% 6.3% 3.1% Return Multiple 2010-2012 Seed 2010-2018 Seed 2013-2015 Seed 2016-2018 Seed Figure 14 - 5 Year Portfolio Return of S&P Portfolio by Horizon Venture Capital Growth in Portfolio-level Returns | Page 16 Closing Thoughts Overall, the VC space is a highly competitive market where the source of returns comes from a small fraction of companies. Based on the results found throughout this paper it would seem prudent that to maximize the return of a portfolio of VC companies the following should be considered: ο· ο· ο· ο· Maximizing the number of companies within the portfolio. Analysis shows that there is a measurable increase in expected returns with the number of investments within the portfolio. Higher returns are not expected within the first few 3 years. This is expected as generally VC investments have low liquidity and longer holding periods. High failure rates are expected. The majority of companies within the VC space do not see positive returns and/or fail. However, with a large & diverse enough portfolio, this is mitigated with a small number of outsized winners Using VC companies can help diversify a portfolio. VC companies do show some negative correlation & beta to the overall market. Using a portfolio of VC investments help diversify an investor’s overall portfolio. Additionally, while the impact of screening companies was not explored, any portfolio would benefit from screening companies that due not fit an individual’s investment thesis. Venture Capital Growth in Portfolio-level Returns | Page 17 References Aaron Clauset, C. R. (2009). Power-law distributions in empirical data. SIAM review, 51(4):661–703. Beta Coefficient. (2021). Retrieved from Corporate Finance Institute: https://corporatefinanceinstitute.com/resources/knowledge/finance/beta-coefficient/ Clauset, A., Shalizi, C. R., & Newman, M. E. (2009). Power-law distributions in empirical data. SIAM review, 51(4):661–703. Kenton, W. (2021, March 31). Capital Asset Pricing Model (CAPM). Retrieved from Investopedia: https://www.investopedia.com/terms/c/capm.asp Othman, A. (2019). Startup Growth and Venture Returns. AngelList. Taleb, N. N. (2007). The Black Swan: The Impact of the Highly Improbable. New York: Random House Group. Venture Capital Growth in Portfolio-level Returns | Page 18