EIF Research & Market Analysis

Working Paper 2017/41

The European venture

capital landscape:
an EIF perspective
Volume III:
Liquidity events and returns of
EIF-backed VC investments

Dario Prencipe 
 Dario Prencipe worked at EIF’s Research & Market Analysis team.

Corresponding author: Simone Signore.

Contact: [email protected]
Tel.: +352 2485 81 636

Editor
Helmut Kraemer-Eis,
Head of EIF’s Research & Market Analysis, Chief Economist

Contact:
European Investment Fund
37B, avenue J.F. Kennedy, L-2968 Luxembourg
Tel.: +352 248581 394
http://www.eif.org/news_centre/research/index.htm

Scan above to
obtain a PDF

Luxembourg, April 2017 version of this

working paper
Disclaimer:

This Working Paper should not be referred to as representing the views of the European Investment Fund (EIF)
or of the European Investment Bank Group (EIB Group). Any views expressed herein, including interpretation(s)
of regulations, reflect the current views of the author(s), which do not necessarily correspond to the views of
EIF or of the EIB Group. Views expressed herein may differ from views set out in other documents, including
similar research papers, published by EIF or by the EIB Group. Contents of this Working Paper, including
views expressed, are current at the date of publication set out above, and may change without notice. No
representation or warranty, express or implied, is or will be made and no liability or responsibility is or will be
accepted by EIF or by the EIB Group in respect of the accuracy or completeness of the information contained
herein and any such liability is expressly disclaimed. Nothing in this Working Paper constitutes investment,
legal, or tax advice, nor shall be relied upon as such advice. Specific professional advice should always
be sought separately before taking any action based on this Working Paper. Reproduction, publication and
reprint are subject to prior written authorisation of the author(s).
This page intentionally left blank
 Abstract†

Despite the sometimes intensive media coverage and exuberant storytelling around the industry, ven-
ture capital (VC) investors tend to operate in highly opaque markets. On this premise, this work
contributes to the literature via a hand-collected dataset of about 3,600 EIF-backed VC investments
made in the 1996-2015 period, with the aim to analyse their liquidity events and returns. The paper
finds, inter alia, that VC returns show sensitivity to the economic cycle. At the same time, it discusses
how their heterogeneity leaves room for VC firms to pursue diversification strategies and minimise the
correlation with other asset classes. Moreover, this work provides preliminary evidence in support of
the often claimed heuristic that VC returns follow a power-law. Finally, it employs competing risks
models to analyse time-to-outcome data, observing that VC firm experience only relates positively
to performance when outstanding (e.g. 3rd generation fund or above). However, this may also be
a reflection of EIF’s high-standard screening of first-time VC teams. The paper is structured as fol-
lows. Section 1 introduces the key research motivations, while section 2 discusses the features of the
analysed dataset. Section 3 provides a descriptive overview of the data, while section 4 discusses
the statistical test of power-law behaviour. Section 5 explores exit outcomes against the background
of profitable or unprofitable trade sales. Last, section 6 analyses the determinants of exit outcomes.
Section 7 concludes.

Keywords: EIF; venture capital; performance; fat tails; divestment; IPO; competing risk analysis

JEL codes: G11, G24, L11, M13

†
I thank Simone Signore for his continuous support and the long hours of constructive discussion behind this
work. I would also like to acknowledge the invaluable contribution of my EIF Research & Market Analysis
colleagues Helmut Kraemer-Eis, Frank Lang, Wouter Torfs and Salome Gvetadze. I am also grateful to Uli
Grabenwarter, Laurent Braun, Arnaud Vanbellingen, Stephanie Descoubes, Alicia Bodeau for their useful
remarks, as well as Margherita Biffi and Frank Vancamelbecke for their help with the data. Finally, I thank
prof. Stephen Jenkins, London School of Economics, prof. Xavier Gabaix, Harvard University and prof.
Thomas Hellmann, Oxford University for their insightful comments. All errors are attributable to the author.

3
Non-technical Summary

This work is the third volume of the series of working papers entitled ”The European venture capital
landscape: an EIF perspective”. The series aims at assessing whether EIF’s VC activity positively af-
fected beneficiary start-up companies, contributing to the broader theme of government intervention
in the field of venture capital. With the aim to investigate exit returns in Europe, this paper analyses
data on about 3,600 EIF-supported seed and start-up VC investments from 1996 to 2015.

Throughout 20 years of EIF-supported venture capital (VC) activities, the exit scenery of venture
investments has shown sensitivity to the business cycle. While trade sales steadily increased following
the expansion of EIF’s VC activity, major recession events such as the dot-com bubble (2001–2002)
and the European sovereign debt crisis (2009–2010) were linked to peaks in investment write-offs.
On the upside, profitable trade sales continue the increasing path started in 2010, following the
modest economic recovery, low interest rates and a rekindled confidence in the tech industry.

The data illustrated in section 3.1 provides evidence that start-up valuations are responsive to move-
ments in the NASDAQ Composite index. However, while the European VC ecosystem may be shaped
by common macro-factors, we observe high heterogeneity in the exit trends within Europe, particularly
across geographies and industries. At fund or fund-of-funds level, the diversity and the granularity of
start-up investment opportunities leave room for investors’ diversification strategies that can effectively
lower the correlation with other asset classes. Moreover, recent years display a clear upward trend
in both average and median returns: the weighted average of the multiples on cost (MoCs) at exit
for realised VC investments stands at 1.16x for the entire period, the median being 0.12x. The dis-
tribution of exit MoCs is extremely right-skewed: 70% of exited investments are either written-off or
sold for an amount below cost. Deals in which venture capitalists (VCs) sell at cost account for 8%,
whereas the remaining 20% are profitable liquidity events.

Noteworthy, 4% of the exits have returned more than 5 times the investment. This 4% generates
almost 50% of the total aggregated proceeds. As a result, the performance of VC firms is mainly
driven by the occurrence of tail events in a VC fund. In order to better understand the process leading
to such distribution, section 4 formally tests the standard assumption that returns in the VC industry are
power-law distributed. Power-laws rarely surface works in the field of financial economics. However,
the presence of this empirical regularity delivers concrete implications for VC investment strategies and
risk management. For instance, it might shed light on the changes of portfolio valuations following
a Unicorn exit and/or to the share price collapse of a quoted investee. In light of this, empirical
evidence from our data partially confirms the assumption, in that EIF-backed VC returns comply with
the power-law distribution for multiples over 2.35x. Nevertheless, as alternative distributions (and
theories) fit the data as effectively, only further research will be able to reach conclusive evidence.

Section 5 focuses on exit outcomes, providing further descriptive insights on M&As and IPOs of
EIF-backed start-ups. Among several other findings, we observe that, on average, about 50% of
the performing EIF-backed European investees are acquired by non-European corporations, par-
ticularly from the US. US-based buyers are typically larger in terms of assets and revenues, more

4
innovative and mostly active in the ICT domain. This raises the issue of whether the missing scale-up
phenomenon in Europe could be linked to the lack of serial tech buyers, that is, incumbents in highly
innovative and competitive sectors (and often former successful start-ups). At the crossroads between
scaling-up or being acquired, later-stage funding becomes essential. While both acquisitions or for-
eign buyers are not per se negative, their joint existence may be a signal that European start-ups
lack the growth capital necessary to expand and strengthen their position. As such they may end up
being acquired, unless they have the chance to go public: in section 5, we also map 152 IPOs of
EIF-backed start-ups observed in 1996–2015 and 20 different stock exchanges around the world.

Finally, section 6 concludes our analysis by focusing on the factors affecting the exit performance of
VC investments. The empirical analysis delivers key results that can be summarised as follows:

• Geography-related take-away. While fund managers from the NORDICS region display
higher propensity to write-off their positions, investments performed by UK&IRELAND in-
vestors are strongly associated to a greater incidence of profitable trade sales and an initial
public offerings.
Although causality is never claimed, British and Irish VC funds might play an important
role in shaping performance at fund-of-fund level. Moreover, in some specifications of our
model, a shorter fund-company geographic distance is associated with a higher chance of
a positive trade sales.

• Industry-related take-away. Compared to ICT, Life Sciences investments have a significantly

higher chance of IPO. On the other hand, investees in the Services industry seem related
to a higher probability of profitable trade sales.

• Startups-related take-away. Becoming a Unicorn is related to a large increase in the prob-

ability of being acquired, but not significantly related to the chance of going public. This
dynamics might suggest investors’ caution for IPO exit strategies when the company private
valuation is very high.

• Investors-related take-away. More recent vintage years are associated with less likely write-
offs and IPOs. Moreover, the first investment amount, i.e. the size of the initial investor’s
bet on a start-up, is negatively associated with the probability of unprofitable trade sales
and positively associated with IPO likelihood. Hence, investors appear able to recognize
and cherry-pick successful companies at the time of the first check.
In some specifications, a larger number of investments made by a VC fund is related to a
higher probability of experiencing a write-off, but also to a greater chance of IPO. This last
result is non-trivial, because it advocates for the idea that enlarging the number of portfolio
companies in a fund increases the chance of getting an outlier on board for an investor.
Last, venture capitalist’s experience is strongly correlated with a lower probability of write-
off and higher probability of profitable sale. However, first-time VC teams are no significant
predictors of any exit outcome, a finding that defies the expectation of under-performance.
This might hint that VC experience is relevant only when it is markedly high. At the same
time, it might also suggest the strong selection effect of EIF to choose only high-potential
first-time VC teams.

5
 Table of Contents

Abstract 3

Non-technical Summary 4

Table of Contents 6

List of Figures 8

1 Introduction 9

2 The Data 10

3 Trends and Returns 13

3.1 The European exit environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2 VC investment returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 On the empirical features of venture capital returns 22

4.1 Rationale: from data to theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.2 Testing the power-law distribution of VC investment returns . . . . . . . . . . . . . . 24

5 Exit Outcomes 27

5.1 Acquisitions: who are the active buyers? . . . . . . . . . . . . . . . . . . . . . . . . 27

5.2 IPOs: going public after VC in Europe . . . . . . . . . . . . . . . . . . . . . . . . . 30

6 Understanding exit outcomes: a competing risks analysis 32

6.1 Model set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

7 Conclusions 39

References 41

6
Appendices 46

Appendix A Granger-Causality Test . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Appendix B Additional evidence on the distribution of exit returns . . . . . . . . . . . . 47

Appendix C Additional evidence on the power-law assumption in VC returns . . . . . . 50

Appendix D Additional evidence on exit outcomes . . . . . . . . . . . . . . . . . . . . 51

Appendix E List of covariates in the competing risks model . . . . . . . . . . . . . . . 54

Appendix F Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Appendix G Current EIF-supported Unicorns . . . . . . . . . . . . . . . . . . . . . . . 57

About 58

EIF Working Papers 59

7
List of Figures

1 Exit trends of EIF-backed VC investments by exit type . . . . . . . . . . . . . . . . . 13

2 VC investees quarterly real valuation over time . . . . . . . . . . . . . . . . . . . . 15

3 Median VC investee valuation and NASDAQ Composite Index . . . . . . . . . . . . 16

4 Distribution of the exit return class, unweighted MoCs . . . . . . . . . . . . . . . . . 17

5 Distribution of the exit return class over time, unweighted MoCs . . . . . . . . . . . 18

6 Exit Returns of EIF-backed VC investments over time, weighted MoCs . . . . . . . . 19

7 Exit Returns of EIF-backed VC investments exited with MoC ≥ 1, weighted MoCs . . 20

8 Exit Returns of EIF-backed VC investments by vintage year, weighted MoCs . . . . . 20

9 Exit Returns of EIF-backed VC investments over time, by macro-sector, weighted MoCs 21

10 Exit Returns of EIF-backed VC investments over time, by macro-region, weighted MoCs 22

11 CCDF log-log plot of VC investment exit returns . . . . . . . . . . . . . . . . . . . 25

12 CCDF log-log plot for empirical data, power law and log-normal fit . . . . . . . . . 26

13 Private acquisitions of European EIF-backed VC investees, by buyer geography . . . 28

14 Sector type of acquisitions of European EIF-backed VC investees, by buyer’s geography 30

15 EIF-supported tech IPOs over time . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

16 Country and sector distribution of EIF-backed tech IPOs . . . . . . . . . . . . . . . 32

17 Geographic distribution of the EIF-supported tech IPOs . . . . . . . . . . . . . . . . 33

18 Changes in the cumulative incidence function for acquisition . . . . . . . . . . . . . 39

8
1 Introduction

There are numerous reasons in support of a study on the return perspectives of venture capital (VC)
investments in Europe. First, it would increase transparency among market players. The importance
of such first point cannot be understated: despite the sometimes intensive media coverage and exu-
berant storytelling around the industry, VC investors tend to operate in highly opaque markets (Da Rin
et al., 2013).1 Against this backdrop, prospective venture capitalists (VCs) may lack the expertise
and in-depth knowledge that is crucial to mitigate information gaps and to deliver good performance
(Sorensen, 2007; Sorensen, 2008). The provision of new comprehensive data on European VC re-
turns may thus advance investors’ current knowledge of the market and potentially enhance their
investment strategy (Chaplinsky and Gupta-Mukherjee, 2016; Masters and Thiel, 2014).

Second, the lower presence of venture funding in Europe has been repeatedly ascribed to the chronic
lack of attractive and liquid markets for VC exits (Black and Gilson, 1998; Hege et al., 2009; Tykvová
et al., 2012). As such, evidence on exit dynamics throughout Europe becomes of paramount impor-
tance for policy-makers in charge of designing public intervention in the European VC ecosystem.2

Third and last, there is an inherent need to enlarge the body of economic research on the performance
of venture capital investments in Europe. Currently, the evidence produced for European VC is
dwarfed by the volume of publications focusing on the US VC market. A key challenge in this area
is obtaining data to compute investment returns, which typically restricts the focus — particularly in
Europe — to exit-type data.

In this paper, we employ a novel micro-level dataset of 3,600 VC investments performed via EIF-
backed venture capital funds from 1996 to 2015 to address the topic of VC returns. Data is sourced
from EIF internal records, i.e. the quarterly reports submitted by private equity funds in which EIF
is Limited Partner (LP). Departing from the commonly used data sources, we avoid some of their
well-known limitations.3 On the other hand, as EIF only supports a limited portion of European VC
fundraising,4 an important disclaimer relates to the potential non-representativeness of our sample
with respect to the overall population of European VC investments. However, the breadth of EIF’s
geographic and sectoral focus, coupled with the diversity of EIF-supported VC teams in terms of
experience and past performance, guarantee sufficient heterogeneity across the analysed sample.

1
The case for severe information asymmetries in venture capital has often been discussed, e.g. by Berge-
mann and Hege (2005), Casamatta (2003), Schmidt (2003), Hellmann and Stiglitz (2000). Lack of infor-
mation on expected returns and the existence of double moral hazard shape some of the industry’s defining
features, such as, inter alia, investors’ activism, usage of convertible securities and equity rationing.
2
Arguments supporting the need for governments to tackle such ”European VC gap” are discussed in
Tykvová et al. (2012) and Kraemer-Eis et al. (2016b). For a critical view of public intervention in the
VC market, see Lerner (2009).
3
Most empirical research on VC exits is grounded on either data from ad-hoc surveys or commercial
databases, such as VentureXpert and Venture Source (see Kaplan and Lerner 2016 and Da Rin et al. 2013
for an overview.). The limitations of such data sources are well-known: reporting bias, bias towards US
investments, lack of complete time series on financing rounds, under-representation of the Biotech sector
and the difficulty in knowing the status of investments whose outcome is other than IPO or acquisition.
4
In 2007 (2014), 27% (30%) of VC funds’ final closings in Europe have been backed by EIF. For the
complete time series, see Kraemer-Eis et al. (2016b).

9
The aim of this paper is threefold. First, it provides a descriptive overview of venture capital returns in
Europe: while section 2 describes the main features of our sample, section 3 looks at historical exit
trends. In this first part, we discuss investee valuations over time, comparing them to stock market
behaviour. In addition, we provide a descriptive analysis of VC investment returns.5

The second objective, based on the insights of the opening sections, is to investigate the empirical
regularities of VC investment returns. To this end, section 4 dives deep into the statistical properties
of the returns distribution with the aim to identify and discuss the physical process that underlies
European venture capital returns.

Third and last, this paper looks at the outcomes of EIF-backed investments. Section 5 matches re-
alised investment data with Bureau Van Dijk’s (BvD) M&A database Zephyr. This enables the retrieval
of further insights on acquisitions and IPOs experienced by EIF-backed VC investees. It also allows
to tackle our final research question, related to the determinants of VC exit performance: to this end,
section 6 follows the approach in Giot and Schwienbacher (2007) to estimate a series of competing
risks models. These are used to assess which fund- and investee-level factors can significantly predict
the probability of experiencing a write-off, liquidation, acquisition or IPO.6 Section 7 concludes.

2 The Data

This work builds on multiple data sources. The starting point is a dataset of 2,951 start-ups invested
by 355 EIF-backed funds between 1996 and 2014.7 For this analysis, we focus on investment-level
data, comprising 3,592 fund-investee records. Each record has been augmented with information
on the status of investments throughout their lifespan, producing quarterly-updated panel data.8 As
EIF typically acts as LP in funds it invests into, data on investment statuses is considered unbiased, i.e.
not suffering from the reporting bias of voluntary self-disclosure. Instead, the information analysed
results from the formal reporting requirement fulfilled by general partners (GPs) since the date of final
closing up until the fund’s liquidation. This guarantees that our data is complete up to a reasonable
extent and that it allows following all portfolio companies from vintage to exit, independently of
the economic, financial and legal outcome of their investments. Finally, data is standardized, i.e.
EIF-backed VC firms comply with IPEV Reporting Guidelines.9

5
In this paper, the expression returns points to the gross returns of VC investments. These focus on the
performance of the underlying companies, thus meant to quantify the ability of VCs to generate returns.
Net returns gauge instead the ability of LPs to pick successful VC fund managers (Da Rin et al., 2013).
6
In the listed order: a write-off is a divestment where the investor assigns zero to the value of its equity hold-
ings (or a symbolic amount), thereby withstanding a full or partial investment loss; a liquidation indicates
every non-write-off liquidity event whose proceeds for the VC investor are less than her original investment,
i.e. an unprofitable trade sale; acquisition is used as a synonymim for profitable trade sales; an Initial
public offerings, or IPO, marks the first time that shares of a company can be publicly traded.
7
Details of its construction were discussed in Kraemer-Eis et al. (2016b) and Signore (2016). Consistent
with these studies, this work only focuses on EIF-backed investments in seed and start-up stage companies.
8
The panel is unbalanced because of different investment periods and sample attrition, i.e. investments
dropping out of the sample when exited.
9
In a small number of cases investments are carried out by legal structures other than ”standard” VC funds,
e.g. co-investments with business angels. Nevertheless, their reporting requirements mostly hold the same.

10
Choosing a well-suited metric for VC returns is far from trivial. Even in the case of granular and
unbiased data, standard asset-pricing models (i.e., CAPM) deliver estimated measures of risk-return
that can be challenging to interpret. Indeed, the assumptions of liquid and transparent markets
underlying the CAPM are far from being satisfied in private equity investments (Ang and Sorensen,
2012).10 At the same time, alternative metrics such as the Public Market Equivalent (PME) or the
Internal Rate of Return (IRR) present some relevant drawbacks when applied to a company-level
framework. Despite being a robust relative performance indicator, the PME, introduced by Kaplan
and Schoar (2005), has been used in the literature only to evaluate net fund returns for LPs. The
IRR introduces computational issues and manipulation risk with the unrealistic assumption that the
investment proceeds can be reinvested by the VC fund manager at the IRR rate.

Therefore, as our return measure, we use the Total Value to Paid-In multiple (TVPI), or Multiple on
Cost (MoC, henceforth) measured at exit. Despite the MoC being rougher than the IRR, in that it
does not embed time value of money in its calculation, it is the most popular performance measure
among investors and entrepreneurs, mostly because of its intuitiveness and computational simplicity.
On the other hand, analyses based on the MoC — which assumes a zero-rate reinvestment of cash
flows — tend to be more conservative than those based on the IRR (Ang and Sorensen, 2012).

Define t0 = 0 as the vintage quarter and Ti as the exit quarter for investment i = 1, . . . , 3592. In this
work, we define the MoC for investment i at quarter t ∈ [t0 , . . . , T] as follows:

valuationi,t + cumulated_proceedsi,t
MoCi,t = (1)
total_invested_amounti,t

where total_invested_amounti,t is the overall cost of the investment and cumulated_proceedsi,t is

the total amount of cash realised from sale or liquidation at quarter t. The term valuationi,t is the
end-of-quarter valuation of the start-up shares held by the VC firms. This valuation is typically equal
to cost at t0 and it reaches zero when the investment is fully exited, i.e. at Ti . As funds report their
equity stakes on a quarterly basis, we are also able to track the start-up valuation over time.11 To
ensure the comparability of time series, all monetary amounts in Equation 1 are expressed in real
terms, by using GDP deflators of the country hosting the main headquarters of each VC firm.

Quarterly reports submitted to EIF also provide data on two main types of exit route: trade sales and
write-offs. By combining this information with the MoC at exit date, we further define ”unprofitable
sales” as trade sales with MoCi,Ti < 1, and ”profitable sales” as trade sales with MoCi,Ti ≥ 1.
Profitable trade sales indicate that start-up shares were sold, in either private or public markets, for

10
Building on Cochrane (2005), Korteweg and Sorensen (2010) use US company-level data to estimate
CAPM alphas and betas for VC investments in start-ups. By using a dynamic selection model to correct for
endogenous selection of the observed returns, they find an average beta of 2.8 and negative alphas after
2000. However, due to specificities of their continuous-time model, interpreting the economic magnitude
of the excess return is not clear-cut.
11
Comparing valuations of VC-backed start-ups on the basis of this approach entails a number of stronger
assumptions. Most critically, we must assume that VC firms’ equity stakes are always reported based on
the outstanding fully diluted shares. If instead the reported equity stake refers to the basic outstanding
shares, we expect a potential downward bias in the computed valuations, due to the lack of consideration
for equity-related instruments such as employee stock options, warrants and convertible notes.

11
an amount no less than the investment cost. Instead, unprofitable trade sales mark an exit event
also referred to as liquidation, i.e. the investor’s attempt to minimise the loss of a low-performing
investment. Table 1 shows the sample split by exit outcome and provides some descriptive statistics.

Table 1: EIF-backed VC investments by exit status12

Exit Status N Perc. Start-up Age Investment Age Tot. Invested (EUR m)
Company in portfolio 1527 42.5% n.a. n.a. 3.0
Write-Off 709 19.7% 7.2 4.9 2.6
Unprofitable Sale 829 23.1% 8.5 6.0 3.0
Profitable Sale 527 14.7% 7.9 5.1 3.3

Note: based on a sample of 3,592 investments in 1996–2015. Columns 4 and 5 contain the exit age for companies and
investments respectively. Column 6 provides the average of total (i.e. first and follow-on) investments from vintage to exit.

Besides some well-expected data features, e.g. more profitable investments receiving additional
financing along the way, a robust finding immediately emerges by looking at the percentages. That
is, in venture capital, gains come from the outliers. Moreover, investments resulting in a write-off
are, on average, the quickest to be terminated, whereas liquidations take the longest to materialise.

An obvious limitation with investment performance figures is that, despite the insights on whether
returns were realised, not much is known about where such realisations stem from. We therefore
complete our dataset on exits with data on equity deals sourced from BvD’s Zephyr.13 While write-offs
and non-exited investments are typically not traceable in the Zephyr database, the exercise still allows
to retrieve deal information on 591 exits, out of 1,356 profitable and unprofitable trade sales. Not
surprisingly, Table 2 shows that most exit types not matched in Zephyr tend to be the ones generating
the lowest returns for VC firms.14 A case-by-case inspection of quarterly reports reveals that untraced
exits mostly correspond to liquidation-related share buybacks or management buy-outs (MBOs).

Table 2: Exit MoCs for VC exits by deal type

Deal Type Avg MoC Median MoC Min MoC Max MoC N
Acquisition 2.1x 1.0x 0.0x 105.4x 447
IPO 4.3x 1.5x 0.0x 139.0x 111
Other 2.1x 1.2x 0.0x 12.5x 33
No Zephyr Deal Data 1.0x 0.3x 0.0x 28.5x 765
Total 1.7x 0.6x 0.0x 139.0x 1356

Note: based on 1,356 EIF-backed trade sales. Write-offs and non-exited investments are excluded. Deal type ”Other”
includes IBOs, MBOs, MBIs, joint ventures and buybacks. Source: BvD Zephyr (2016).

Equity deals data will be extensively used in section 5.1 and section 5.2 to provide descriptive evidence
on acquisitions and IPOs experience by EIF-backed investees, with the specific aim to pinpoint start-
up buyers and the stock exchanges used to go public. Moreover, the deal-level information allows
a finer identification of the different exit outcomes used in the competing risks models of section 6.

12
Unless otherwise stated, all figures in this research are an elaboration of the author, based on EIF data.
13
BvD Zephyr is an information solution containing merger and acquisition (M&A), IPO, private equity and
venture capital deals. As of April 2017, Zephyr has information on over 1.5 million deals and deal rumours.
14
Note that the number of IPOs reported in Table 2 does not show all the EIF-backed IPOs, but only the
divestments through IPO. Indeed, it might well be that an IPO happens after the VC has sold its shares.

12
 3 Trends and Returns

3.1 The European exit environment

Throughout 20 years of EIF-supported VC activities, the exit scenery of venture investments has shown
sensitivity to the business cycle. While both profitable and unprofitable trade sales steadily increased
following the expansion of EIF’s VC activity, major recession events such as the dot-com bubble
(2001–2002) and the European sovereign debt crisis (2009–2010) were linked to peaks in write-offs.
On the upside, successful trade sales have been firmly increasing since 2010, alongside an economic
recovery, low interest rates and a rekindled confidence in the tech industry, evidenced by the upturns
of NASDAQ and venture capital valuations.15 Figure 1 shows the trends for EIF-backed VC exits.

Figure 1: Exit trends of EIF-backed VC investments by exit type

100

80

60
Nr of Exits

40

20

0
‘96 ‘97 ‘98 ‘99 ‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15
Year

Write-Off Unprofitable Sales Profitable Sales

Note: based on 2,065 EIF-backed early-stage VC investments exited between 1996 and 2015.

The EIF-backed subset of the European VC ecosystem appears to be shaped by common macroeco-
nomic and institutional factors.16 However, there is also significant heterogeneity in the exit trends,
particularly across geographies and industries.17 Following the dot-com crash, for instance, the

15
Nevertheless, some caution is necessary when looking at the last years of the sample period, as most of
the recent investments are likely to be still held by funds’ portfolios.
16
Remarkably, the observed trends are in line with European-level statistics based on data regarding the
entire European VC market. See Kraemer-Eis et al. (2016a) for details.
17
Eight geographical regions were defined as follows: DACH: AT, CH, DE; NORDICS: DK, FI, NO, SE;
FR&BENELUX: BE, FR, LU, NL; SOUTH: GR, ES, IT, MT, PT; UK&IRELAND: IE, UK; CESEE: BG, CZ, EE,
LT, LV, PL, RO, SK, TR, CY; US; ROW (Rest Of the World): AR, AU, CA, CN, CR, HK, IL, IN, MX, PH, RU,
SG, UY. See Signore (2016) for a breakdown of the EIF activity by regions. The industrial nomenclature
follows Invest Europe’s classification. For details, see Signore (2016).
13
SOUTH macro-region endured the highest relative incidence of write-offs,18 while at the same time
showing greater resilience to the sovereign debt crisis — despite the economy of the region being
impacted the most — which only caused a short-lived reduction in trade sales. Conversely, the region
UK&IRELAND seems to have been highly sensitive to the worsening of the macroeconomic condi-
tions. During both crises, investments in UK accounted for 30% of the overall write-offs. As expected,
investments in Life Science companies were almost unaffected by the dot-com crash, contrary to ICT.

The sensitivity of aggregate divestments to the economic cycle may be due to multiple circumstances.
Write-off waves might follow VC firms’ negative growth expectations of portfolio companies in the
aftermath of a recession, while M&A activity might be affected by tech-buyers’ appetite to invest and
expand their business. This appetite, in turn, is impacted by expectations, uncertainty and volatility.
Notwithstanding, the main driver of these patterns seems to be the stock market, as documented by a
well established literature. According to Gompers et al. (2008), fluctuations in VC activity are related
to changes in the stock market, so that public market shifts can provide useful insights to investors.
Black and Gilson (1998) stress how the stock market affects VC exit strategies, while Michelacci and
Suarez (2004) develop a model in which stock markets trigger start-up creation. Pastor and Veronesi
(2005) provide evidence that IPO figures vary in response to changes in public market conditions.

Further responsiveness to macroeconomic conditions emerges when looking at start-up valuations.

The outstanding volatility of start-up valuations is certainly a well-known fact (Gompers et al., 2008).
Phenomena like winner-takes-all competition may give raise to natural outliers that heavily affect the
average start-up valuation. Figure 2 shows the evolution of average and median valuations over
time. Median valuations tend to be more stable than averages, although both of them follow the
cyclical pattern suggested by Figure 1. Interestingly, there is a clear upward trend in both series over
recent years, with the median start-up valuation reaching its historical peak.

To understand whether these trends are in line with the public markets behaviour, in Figure 3 we
plot the median VC investee valuation against the real value of the NASDAQ Composite Index, an
index of all common stocks and similar securities listed on the NASDAQ stock market. The use of
this index is motivated by its significant bias towards information technology and biotech companies
and, in turn, by its suitability to represent VC-backed start-ups that go public.19 Apart from the
2009–2012 period, where the flat pattern of European valuations might have been a reflection of
the sovereign debt crisis, the valuation trend seems to be anticipated by the NASDAQ shifts. We
formally test the assumption that NASDAQ movements help predicting median start-up valuations by
means of a Granger-Causality test (Granger, 1969).20 The Granger-Causality test is a statistical test
to determine whether one time series is useful in forecasting another. Results suggest that variations
in the NASDAQ Composite Index contain useful information on future short-term fluctuations of the
median valuation of start-ups. Further details on this finding are discussed in Appendix A.

18
For instance, Italy accounted for 15% of all the write-offs between 2000 and 2002.
19
For a discussion on the role of stock exchanges in providing exit routes to European venture capitalists,
see Da Rin et al. (2006) and Bottazzi and Da Rin (2003).
20
We choose median valuations by quarter over average, as the latter can be significantly influenced by VC
investments in start-ups that undergo a public listing.

14
Figure 2: VC investees quarterly real valuation over time
80 8

60 6

40 4

20 2

0 0
20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1

20 q1
q1
99

00

01

02

03

04

05

06

07

08

09

10

11

12

13

14

15

16
19

(lhs) Average Start-up Valuation (EUR m)

(rhs) Median Start-up Valuation (EUR m)

Note: valuation statistics prior to 1999Q1 not considered due to negligible sample size. Amounts refer to
company valuations (see section 2), as opposed to investment valuations, used instead to compute the MoC.
Average and median values only consider start-ups whose shares are held by at least one EIF-backed fund at
each respective quarter. Monetary values expressed in constant EUR prices, with 2005 as the base year.

In principle, venture capitalists aim at maximising their profit by realising their investments when
valuations are at their highest. The challenge of choosing the right time to exit a VC investment is not
trivial, and it may have a significant impact on the return on investment of VC firms.21 As time-to-exit
is a key parameter shaping the general exit environment, we might be interested in looking at how the
average holding period evolves for our sample of investments. However, the time series we analyse
are not long enough to establish a significant trend, as many of the investments are still held in the
VCs’ portfolios. Also, short holding periods in the first years of the sample are intrinsic to the first
EIF-backed investments being carried out no earlier than those years. Longer time series are needed
in order to establish a clearer holding period pattern.

Overall, data so far shows that the exit environment for EIF-backed VC investments is not only linked
to technological cycles,22 but also to economic cycles. Accordingly, venture capital is likely to show

21
The VC investment in Tiscali, an Italian telco company founded in 1998, is a textbook example: the
company, one of the pioneers of flat-rate solutions for consumer Internet services, sailed on the enthusiasm
of the dot-com boom and faced an IPO in October 1999. The share price at IPO was EUR 3.75 and
peaked at EUR 61.59 in March 2000, +1540% in less than 6 months. In June 2002, share price went
below 3.15 and steadily declined ever since, where the current price fluctuates around EUR 0.04 (Source:
Thomson Reuters Eikon). While the investment generated more than 100x MoC for an early seller, a
different investor obtained around 2.5x by exiting the investment a mere 18 months later.
22
By technological cycle, we mean the rise and fall in the adoption of new technologies and its effects on the
economy. For instance, several different industries nowadays face new business opportunities due to the
advent of technologies such as AR/VR, blockchain, driverless cars, etc.

15
Figure 3: Median VC investee valuation and NASDAQ Composite Index
10 6000

8 5000

6 4000

4 3000

2 2000

0 1000
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
20 q1
q1
99
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
19

(lhs) Median Start-up Valuation (EUR m)

(rhs) NASDAQ Composite Index

Note: the NASDAQ Composite Index is a market capitalization weighted index with more than 3,000 common
equities listed on the NASDAQ Stock Market. The index includes all NASDAQ listed stocks that are not
derivatives, preferred shares, funds, exchange-traded funds (ETFs) or debentures. Series expressed in real
terms (2005 = 100). Source: FRED, Federal Reserve Bank of St. Louis.

correlation with other asset classes. However, the geographic and sectoral heterogeneity of venture
capital investments leaves room for VC firms to pursue diversification strategies. From a portfolio
management perspective, finding the tail events remains a key condition to generate excess return.
On the other hand, exploiting the granularity of the start-up ecosystem via diversified investment
strategies may allow VC firms to hedge against the business cycle and the stock market.

3.2 VC investment returns

With the aim of providing a descriptive analysis of the EIF-backed VC exits, we start by inspecting
exit returns for 2,065 realised investments (58% of the overall 3,592 EIF-backed investments in the
sample, see Table 1). To ease the visualisation of highly skewed data, Figure 4 breaks down the
distribution of exit multiples in five discrete classes. About 70% of the exited investments have been
either written-off or sold for an amount below cost. Deals where the VC sells at cost account for 8%,
whereas the remaining 20% of the liquidity events are profitable. Only 4% of exits have returned
more than 5 times the investment.

The reader acquainted with VC dynamics will not find Figure 4 much surprising. Venture capital
is mostly a losing game, with investments falling on the right-hand tail of the distribution destined
to return the amounts necessary for the fund to break-even and make profits. To this purpose,
Table 3 reports, for each exit class, the average share of the investor’s fund returned by the exit.

16
 Figure 4: Distribution of the exit return class, unweighted MoCs
75%

57%

50%
Percentage of total exits

25%
17%
13%
8%
4%

0%
<0.25 0.25 to Cost At Cost Cost up to 5 >5

Note: based on 2,065 early-stage VC investments, exited, made between 1996 and 2015 by EIF-backed VC
funds. The figures include all the exit types, i.e. write-off, liquidations and successful sales. Exit MoCs are not
weighted. The ”At Cost” bucket includes all the MoC values such that 0.8 ≤ MoC < 1.2.

The percentages in Table 3 can be interpreted as follows: suppose a VC investor realises an invest-
ment with an exit multiple between 1 and 5. Then, as per our findings, (s)he should expect that exit
to return, on average, 10% of the fund.

Table 3: Average share of VC fund returned by an exit, by exit class

Exit Class Share of the fund returned (Mean) Share of the fund returned (Median)
<0.25 0.1% 0.0%
0.25 to Cost 1.7% 1.0%
At Cost 3.7% 2.5%
Cost up to 5 10.0% 7.3%
>5 43.9% 25.3%

Note: based on 2,057 EIF-backed exited investments, including write-offs. Exit classes defined as per the MoC at exit.

Although the right-skewdness of venture returns is intrinsic to the venture capital business model,
some heterogeneity in the distribution can be observed across time, space and domain (see Figures
B1 — B3 in Appendix B). Investments in Life Sciences and Services display less under-performers (and
more out-performers) than investments in the broader ICT space. At the geographical level, start-ups
in the FR&BENELUX and US macro-regions have returned more than the average. The vintage year
plays an important role in shaping the returns distribution. While the share of profitable exits was
16% of all the investments made between 1996 and 2001, the proportion of successful investments
made between 2007 and 2015 was higher than 30%.

17
The question of whether this phenomenon is cyclical rather than structural is certainly relevant. In
order to shed more light on the time dimension of the investments performance, Figure 5 depicts
how the distribution of unweighted exit returns fluctuated in recent years.

Figure 5: Distribution of the exit return class over time, unweighted MoCs

100%

80%
Percentage of total exits

60%

40%

20%

0%
99
00
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
19
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
Exit Year

<0.25 0.25 to Cost At Cost Cost to 5 >5

Note: based on early-stage VC investments exited between 1999 and 2015 by EIF-backed VC funds. The
figure includes all the exit types, i.e. write-off, liquidations and successful sales. Exit MoCs are not weighted.
The ”At Cost” bucket includes all the MoC values such that 0.8 ≤ MoC < 1.2.

There seems to be cyclicality in the returns distribution, with the proportion of non-performing exits
increasing during recessions and shrinking over growth periods, in line with section 3.1 as regards to
the number of exit deals. Remarkably, 2010 can be marked as the starting year for an expansionary
trend of profitable EIF-backed investments. The extent to which this points to an improvement in VCs’
ability, or it simply reflects the positive momentum in tech company valuations, is not as easy to say.

On the one hand, the representation of the unweighted exit multiples gives information about the
”biological” properties of start-ups, i.e. how much they return given their business performance and
growth. On the other hand, from an investor perspective, there is value in considering these figures
taking into account how much has been invested in each of these companies. That is, studying
returns weighted by total investment cost (the denominator of the fraction in Equation 1).

Against this backdrop, given two VC investments with different exit returns, the exit MoC of the start-up
receiving more financing will outweigh the exit MoC of the lesser financed start-up in the calculation
of average returns. Figure 6 shows weighted average and weighted median exit returns over time.

18
Figure 6: Exit Returns of EIF-backed VC investments over time, weighted MoCs
5x 200

4x 160
Multiple on Cost

3x 120

Nr of Exits
2x 80

1x 40

0x 0
‘97 ‘98 ‘99 ‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15
Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Note: based on early-stage VC investments exited between 1997 and 2015 by EIF-backed VC funds. The
figure includes all the exit types, i.e. write-off, liquidations and successful sales. Exit MoCs are weighted. Grey
bars report the number of exits in each year (right-hand axis).

The weighted average of the MoC at exit for realised VC investments over the entire sample period is
1.16x, whereas the median is 0.12x. As shown by the grey bars, the statistics computed for the first
years of the sample period are based on a small number of exits, therefore they should be interpreted
with caution. Returns show sensitivity to the cycle. When the average return peaks during pre-crises
periods,23 the median return follows, but almost never exceeds the 1x threshold. Recent years display
a clear upward trend in both average and median VC investment returns.

Given the asymmetric returns pattern of VC, some insights can be also derived from the evolution of
MoCs linked to purely profitable investments. To this end, Figure 7 computes the weighted average
and median returns for investments with exit MoC ≥ 1. Against the undoubtedly lower number of
exits covered in Figure 7, the data offers a snapshot of the upper-tail of the returns distribution. And it
may actually prove as representative as Figure 6, e.g. in the case venture capitalists were to employ
liquidation preferences provisions to protect themselves against hypothetical losses.

Figure 8 shows instead the returns by vintage year. The vintage of investments is an important element
in the analysis of the cyclicality of returns. Indeed, vintages from downturn years should have lower
entry prices that, ceteris paribus, lead to a higher expected exit MoC. Conversely, in ”boom” periods,
when valuations may become inflated, higher entry prices could affect exit returns. This pattern is
not so clear-cut when looking at Figure 8, with vintages 2000 and 2001 under-performing and
vintage year 2007 scoring rather positively. Figure 8 discards vintage years with less than 10% of
total investments realised (2012 being the last observed vintage year). However, both Figure 6 and
Figure 8 bring evidence of generally increasing returns for EIF-backed start-up investments.

23
The peaks in Figure 6 are driven by two outlying exits in 2001 (Tiscali) and 2007 (Skype). Their omission
leads to an average return of 1.15x and 1.37x for 2001 and 2007 respectively.
19
Figure 7: Exit Returns of EIF-backed VC investments exited with MoC ≥ 1, weighted MoCs
8x 80

6x 60
Multiple on Cost

Nr of Exits
4x 40

2x 20

0x 0
‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15
Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Note: based on 520 early-stage VC investments, exited with MoC ≥ 1, made between 2000 and 2015 by
EIF-backed VC funds. Exit MoCs are weighted. Grey bars report the number of exits with MoC ≥ 1 in each
year (right-hand axis).

Figure 8: Exit Returns of EIF-backed VC investments by vintage year, weighted MoCs

5x 100%

4x 80%
Share of exited investments
Multiple on Cost

3x 60%

2x 40%

1x 20%

0x 0%
‘97 ‘98 ‘99 ‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12
Vintage Year

% Realised Average Exit MoC Median Exit MoC

Note: based on early-stage VC investments taking place between 1997 and 2012 by EIF-backed VC funds.
The figure includes write-offs, liquidations and profitable trade sales. Exit MoCs are weighted. Grey bars
report the portion of each vintage year that has been realised to date (right-hand axis).

20
 In order to investigate the return profile for specific subsets of VC investees, we create ”synthetic”
portfolios by grouping different groups of start-ups (e.g. Biotechnology for sectors DACH for regions).
Results are displayed in Figure 9 and Figure 10.

Figure 9: Exit Returns of EIF-backed VC investments over time, by macro-sector, weighted MoCs

ICT Life Sciences

5x 125
4x 100
3x 75
2x 50
1x 25
Multiple on Cost

0x 0

Nr of exits
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

Manufacturing Services
5x 125
4x 100
3x 75
2x 50
1x 25
0x 0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Note: based on early-stage VC investments exited between 2003 and 2015 by EIF-backed VC funds. The
figure includes write-offs, liquidations and profitable trade sales. Exit MoCs are weighted. Grey bars report
the number of exits in each year (right hand axis).

Among ICT investments, those in Computer Related and Electronics/Automation have performed
better than investments in Communications (see Figure B4 in Appendix B). Life Sciences investment
returns have significantly soared in the recent years, with average exit MoCs floating over 2x between
2013 and 2015. Despite the relatively low number of exits achieved so far, start-ups in the Services
domain (mainly Consumer Related and Financial Services) have shown above-average returns over
the last years. Noteworthy, despite low performance in the past, exit multiples in the NORDICS region
have thrived from 2011 onward.

Interestingly, there seems to be no significant difference between returns from start-ups invested by
first-time VC teams and VC firms with one or more past raised funds (See Figure B5 in Appendix B).
However, investments carried out by first-time teams perform worse during market downturns (see
2008-2010 period in Figure B5), where experience turns out to be key in shaping the investment
outcome.

21
 Figure 10: Exit Returns of EIF-backed VC investments over time, by macro-region, weighted MoCs

DACH NORDICS FR+BENELUX

5x 50

4x 40

3x 30

2x 20

1x 10
Multiple on Cost

Nr of exits
0x 0
‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15 ‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15 ‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15

SOUTH UK+IRELAND CESEE

5x 50

4x 40

3x 30

2x 20

1x 10

0x 0
‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15 ‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15 ‘03‘04‘05‘06‘07‘08‘09‘10‘11‘12‘13‘14‘15

Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Note: based on early-stage VC investments exited between 2003 and 2015 by EIF-backed VC funds. The
figure includes write-offs, liquidations and profitable trade sales. Exit MoCs are weighted. Grey bars report
the number of exits in each region-year (right-hand axis). To avoid over-inflating trends for CESEE, the exit
return of LiveRail (Romanian start-up acquired by Facebook in 2014) was omitted from the calculation.

4 On the empirical features of venture capital returns

This section shows evidence on an empirical regularity that is often postulated in the economics of
venture capital. That is, returns follow a power-law.24 Before diving into the topic, we may first
question its relevance. In other words, how useful can it be to know how VC returns are distributed?

4.1 Rationale: from data to theory

As Clauset et al. (2009) argue, for many purposes it may be enough to know that a quantity follows
a heavy-tailed distribution. In our case, observing Figure 4 would be sufficient for investors to derive
basic implications of VC economics. However, if the aim is to infer plausible mechanisms that might
lie behind the formation and evolution of exceptionally successful start-ups (or an entire ecosystem

24
Practitioners and investors are acquainted with the topic, although it has hardly been subject to rigorous
research, at least with regards to venture capital. For instance, according to Marc Andreessen from An-
dreessen Horowitz (a Silicon Valley-based VC firm) ”The key characteristic of venture capital is that returns
are a power-law distribution”. The VC firm itself published data showing the skewness of the returns distribu-
tion (see https://a16z.com/tag/power-law/). Sandeep Bhadra, Principal at Menlo Ventures, discusses this
in a lecture at UC Berkeley called ”Power-Law Returns in Venture Capital: Strategies for Building and Work-
ing with Great Companies”. VC investor and entrepreneur Peter Thiel builds further on power laws in VC
and their implications for investors in Masters and Thiel (2014). For a comprehensive investor-perspective
review on power laws in VC, see Neumann (2015).
22
of these), then it is of utmost importance to comprehend the mechanism — the physical process —
that is the cause of such empirical regularities. Why are 70% of start-up investments unprofitable
while a few of these reach companies like Skyscanner and are sold for amounts greater than the VC
fund itself?25 Assessing whether VC returns follow a power law — or Paretian — distribution is thus
the first step towards an understanding of the broader economic phenomenon.

Quoting Scherer (2000), ”the potential variability of economic outcomes with Paretian distributions is
so great that large portfolio draws from year to year can have consequences for the macroeconomy”.
Indeed, if the size of companies26 in a certain industry (e.g., ICT sectors that are typically affected by
network externalities and winner-takes-all effects) is power law distributed, then a random shock to
the largest companies, or to the narrowly defined sector, can generate aggregate fluctuations in the
entire economy (Gabaix, 2011).27 This finding yields concrete implications to be factored in when
designing, for instance, policies in support of ”scale-ups” (Hellmann et al., 2016).

Perhaps more fascinating, the topic of power-law distributed returns entails concrete suggestions for
portfolio management in venture capital. First, on the risk management side, the exact characterisa-
tion of the distribution would allow to better predict shocks from ”extreme returns” in a VC portfolio, in
particular with respect to listed portfolio companies — more exposed to market risk — and in general
to sudden shifts in valuations due to unexpected factors. Second, as discussed in Sornette (2002),28
the power-law scale parameter α dictates whether VC financing is characterised by economies of
scale (specifically, if α < 2). That is, whether the expected return of a VC fund increases with its
portfolio size, i.e. the number of investments carried out. Although subject to the existence of some
additional conditions, the power of such result lies in the implication that, ceteris paribus, a higher
number of small bets at very early stage is a better strategy than a smaller number of larger bets.

As venture capital is about building companies and relationships, money is only one part of the story.
Starting from the assumption of power law-distributed returns, Masters and Thiel (2014) argue that
the VC should focus time and resources only on those companies that have the potential to be in the
”upper-tail” of the returns distribution. The heuristic the authors derive by combining years of literature
on VC power-laws reads as follows: in the first round, invest small amounts in as many promising
companies as you can. Then, once the few fast-growing portfolio investments start emerging (e.g.
via positive feedback mechanisms), mostly revert your focus to these few winners.

25
In the VC jargon, these companies are called Dragons. A Dragon is a VC investee that returns no less than
the entire VC fund when exited. This means that, for a VC fund, the proceeds from selling the equity stake
of that company (in a trade sale or IPO) are bigger than or equal to the fund size measured at closing
date of fund-raising. Although nowadays Unicorns are by far the focal core of the tech community talks, a
Dragon is the first item in each venture capitalist’s wish-list.
26
Hence returns: as generally assumed, VC returns from a start-up investment follow both past and prospec-
tive growth of the underlying firm. Further details based on EIF data were discussed in Signore (2016).
27
The tumble in the Nokia-led Finnish economy’s outlook following the telco company’s decline (partly due
to the rise of Apple) is an example (see http://www.economist.com/node/21560867). Also, Pomeroy
(2016) claims that the rise of VC-backed ”giants” (such as Amazon, Uber, Spotify) might have exerted an
important downward pressure on global prices and inflation.
28
Sornette (2002) discusses the case of returns from a portfolio of R&D projects. Without loss of generality,
the framework can be extended to returns from a portfolio of start-up investments.

23
4.2 Testing the power-law distribution of VC investment returns

Several phenomena observed and measured in the real world tend to have a size or ”scale”, i.e. a
typical value around which individual measurements cluster (Newman, 2005). Among several classic
examples, the height and weight of human beings, the speed of cars on a motorway, the temperature
in Luxembourg at noon on vernal equinox.29 However, there are also natural processes that do not
show a central tendency, due to few ”extreme” observations or events affecting the shape of the entire
distribution and of course the central tendency itself. For instance, in our data 4% of investments
realised with MoC greater or equal than 5x account for about 50% of the total aggregated exit
proceeds. When certain mathematical conditions are satisfied, these processes driven by ”extreme
events” are called Paretian, or power laws.30

The study of natural power laws is an increasingly fertile research field. One reason being that they
occur in many different areas: the population of cities, earthquakes size, the frequency of use of
words in any human language, the sales of books, music records and other branded commodities,
people’s income, stock market returns and many others.31 To our purpose, a nonnegative random
variable X follows a power law distribution if:

Pr(X = x) = Cx−α (2)

for all the x s.t. x ≥ xmin > 0,32 where C > 0 is a normalization constant and α > 0 is the
scale parameter. In a power law, the tails fall according to the scale parameter α. The lower
the α, the higher the amount of probability gathered in the tails. Roughly speaking, a power law
distribution foresees a much higher probability of ”extreme events” than the one admitted by other
common distributions used in financial economics. Remarkably, when 2 ≤ α < 3, the mean exists
but the variance and higher-order moments of the underlying population diverge as the number
of observations tends to infinity. When α < 2 the situation becomes even more extreme, as the
population mean becomes undefined, i.e. infinite, and the sample average is dominated by extreme
events. Accordingly, as the sample size or time series enlarges, ”standard” summary statistics are
dominated by ”black swan” phenomena (Taleb, 2007), or ”meaningful outliers” (Sornette, 2009).33

29
This regularity is widespread — ”normal” — because of the Central Limit Theorem: when independent
random variables are added, their sum tends toward a normal distribution, even if the underlying variables
are not themselves normally distributed.
30
The formal treatment of power laws goes well beyond the scope of this section. Newman (2005) and
Clauset et al. (2009) provide an in-depth review of both theory and empirics of power laws. For a thorough
discussion of power laws in economics and finance, see Gabaix (2009). Seminal works in the power law
literature are also Zipf (1949) and Mandelbrot (1997).
31
See Newman (2005), Clauset et al. (2009) and Gabaix (2016) for a comprehensive list.
32
As Equation 2 shows, this probability diverges as x → 0. Therefore, there must be a lower bound xmin such
that the power law holds from that point onward.
33
An on-going research debate is concerned with extreme events originating from power law distributions in
stock markets. On the one hand, Taleb (2007) argues that extreme events are intrinsically unpredictable,
due to scale-invariance and self-similarity in power laws. That is, a stock market crash starts as a small
ordinary drop (that keeps building up), thus not distinguishable ex ante from other standard small-scale
fluctuations. On the other hand, according to Sornette (2009), positive feedback phenomena that give
rise to power laws implies a phase transition that can be identified, e.g. a market bubble before its burst.

24
To investigate the presence of a power law we first revert to visualisation, e.g. by looking at the
empirical distribution of exit multiples. However, the extreme right-skewness and kurtosis of the VC
returns data make standard histograms hardly readable. For this reason, Figure 11 displays the
complementary cumulative distribution function (CCDF). That is, Pr(X ≥ x), the probability that the
exit multiple is greater than a certain value x. For data to follow a power law, we expect that the higher
the x, the lower Pr(X ≥ x), i.e. the chance of being observed. The use of log-values in Figure 11 is
not accidental: the log-transform of Equation 2 implies that a necessary (but not sufficient) condition
for power-law behaviour is that such log-log plot approximates a straight line, with slope −α.

Figure 11: CCDF log-log plot of VC investment exit returns

0
(log) Pr{Exit multiple ≥ x}

-2

-4

-6
1 2 3 4 5
(log) Exit multiple

Note: based on the MoC data for 2,065 early-stage VC investments, exited, made between 1996 and 2015
by EIF-backed VC funds. The chart plots the logarithm of the empirical complementary CDF against the
logarithm of the exit MoCs for the upper-tail of the distribution.

Despite Figure 11 showing an approximately straight line in the tail of the distribution, this cannot be
taken as conclusive evidence that VC exit multiples follow a power-law, as documented by Clauset
et al. (2009) and Cirillo (2013). In this respect, a more rigorous procedure is necessary. We thus
follow the quantitative approach formalised in Clauset et al. (2009) in order to test whether the tail
of the distribution is consistent with a power law.34 Table C1 in Appendix C presents the key esti-
mated quantities obtained by implementing the Clauset et al. (2009) procedure for continuous data.

34
The procedure involves a Monte Carlo goodness-of-fit test based on measurements of the ”distance”
between the distribution of the empirical data and the hypothesized model. The distance measurement,
based on the Kolmogorov-Smirnov statistic, is derived by comparing data and theoretical model. The
quantity is further confronted with distance measurements for comparable synthetic datasets drawn from
the original hypothesized model. The p-value is defined as the fraction of the synthetic distances that are
larger than the empirical distance. If p is small, i.e. less than 0.1, the model is not a plausible fit to the
data. For further details, see Clauset et al. (2009).

25
The method delivers an estimated scaling parameter α = 2.45 (±0.09). The statistical test also yields
a p-value of 0.915, implying that the null hypothesis of the data being consistent with a power law
distribution cannot be rejected at most typical confidence levels.

Overall, the empirical evidence does not contradict the claim that VC returns are approximately
Paretian. However, not much is said about whether the power law fits the data better than any
other distribution. Indeed, there are several distributions that behave similarly to power laws. For
instance, despite its finite moments, the log-normal distribution is very similar in shape to the Paretian
distribution (Mitzenmacher, 2003). The point of which distribution fits the data better is important in
order to understand the underlying economic mechanism that generates the distribution, as different
models yield significantly diverse predictions and implications.

We test the goodness of fit of the power-law against other plausible distributions, following Clauset
et al. (2009) and adopting the methodology of Vuong (1989). Figure 12 provides a graphical rep-
resentation of fitting a power law and a log-normal distribution against our data, while Appendix C
discusses the implementation details. Findings point to the sample size being too small to unequivo-
cally assert the supremacy of power law in fitting VC exit returns. Virkar and Clauset (2014) mention
a ”rule of thumb” threshold for ntail — the minimum size of the tail-end sample to identify a best fit —
i.e. ntail > 300, which is not matched by the number of observations currently at hand (details in
Appendix C). Thus, further research with larger datasets may lead to future conclusive evidence.

Figure 12: CCDF log-log plot for empirical data, power law and log-normal fit
0

-2
(log) Pr{Exit multiple ≥ x}

-4

-6

-8
1 2 3 4 5
(log) Exit multiple

Lognormal Fit Power Law Fit Empirical Data

Note: the plot represents the upper tail of the distribution only, as in Figure 11. The distribution fits are
obtained with the Maximum Likelihood method of Clauset et al. (2009). Software implementation follows
Ginsburg (2012) and Alstott et al. (2014).

26
Against this background, some insights can be derived from a theoretical discussion of these two
well-fitting distributions. Power laws and log-normal distributions are intrinsically connected, for
similar generative models can lead to either distribution on the basis of seemingly trivial variations
(Mitzenmacher, 2003). A log-normal distribution arises from a multiplicative process. To see why,
suppose X0 to be the MoC at investment date: if in every subsequent quarter the MoC grows by a
random i.i.d.35 factor, then the MoC at exit can be shown to be log-normally distributed.36 Instead,
Reed and Hughes (2002) show that a process whose finite-time expected growth is exponential (as
is the case for most start-ups) follows a double Pareto distribution, i.e. a distribution with a log-
normal body and a Pareto tail.37 An additional mechanism leading to power laws in venture returns
is preferential attachment, or positive feedback: in markets characterised by strong network effects,
growth can generate additional growth. That is, initially successful start-ups can raise more capital,
attract more talents and extend their user-base, leading to a virtuous cycle of sustained growth.

In conclusion, this section brings evidence that EIF-backed VC investment returns are consistent with
a power law behaviour in the upper-tail of their distribution. This has first-hand implications for the
analytical tools used by risk and portfolio managers in VC. When the investment returns are power
law distributed, standard tools used in asset allocation that rely on the variance, such as Sharpe
ratios, become unreliable. It may be argued that this only affects direct VC investments, in that
the distribution becomes more ”normal” when collecting returns at fund or at fund-of-funds level
(Weidig et al., 2005). However, the mathematical properties of power laws imply that a combination
of power-law distributed investments (e.g. a VC fund or fund-of-funds) also follows a power-law
(Gabaix, 2009). As a result, alternative statistical approaches (e.g., based on extreme value theory)
could be considered more appropriate in the context of VC returns. All in all, additional work may
shed further light on this certainly promising field of research.

5 Exit Outcomes

5.1 Acquisitions: who are the active buyers?

In this section, we focus on the acquisitions experienced by EIF-backed VC investees, with the goal
to identify profiles of start-up buyers in Europe. We thus focus on 447 VC investments exited through
acquisition (regardless of exit MoCs). These are defined as equity deals in which at least the majority
stake of the VC investee38 is purchased by an external company — either industrial or financial —
and contextually a major liquidity event is reported by the VC firm. As introduced in section 2,
the analysis leverages on the retrieval of data from the BvD Zephyr database, allowing for detailed
information not only on investee companies, but also on buyers.

35
Independently and identically distributed.
36
A standard reference for this type of processes in financial economics is Hull (2006), that builds on Black
and Scholes (1973) to model the value of a security that moves in discrete time steps.
37
There are several theories of why power laws arise in natural and men-made phenomena. For a review,
see Newman (2005) and Gabaix (2009).
38
It should be noted that more than 90% of these deals report a 100% stake purchase, i.e. full acquisition.

27
In light of the above, we first split acquisitions according to the geographic location of the buyer.
We restrict the analysis to exits with Europe-based VC investees as the deal target.39 We further
distinguish between:

• Same European Country: the buyer shares the same European country40 with the acquired
start-up;
• Other European Country: the buyer is from an European country but from a different
country than the start-up headquarters;
• US: the buyer is from the US;
• Other Non-European Country: the buyer is from any non-European country but the US.

Figure 13: Private acquisitions of European EIF-backed VC investees, by buyer geography

40

30
# Deals

20

10

0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

Same European Country Other European Country

US Other Non-European Country

Note: based on private acquisitions of EIF-backed early-stage VC investees occurred between 2003 and
2015. Source: BvD Zephyr (2016).

Figure 13 shows the distribution of the buyer’s geographic origin over time. From 2003 to 2015, an
average of 44% of exited EIF-backed VC investees was acquired by non-European buyers, particularly
from the US. Deals in which a US buyer acquires a European start-up represent 1/3 of all acquisitions

39
We thus discard 58 observations for which either we do not have geographic data or the start-up location
is outside Europe. Moreover, we observe that some companies report multiple headquarters located in
different countries. In such cases, we rely on the country of the company’s Global Ultimate Owner (GUO)
as reported by Bureau Van Dijk.
40
All EU28 countries plus Norway, Switzerland and Turkey.

28
(For additional details, see Figure D1 in Appendix D). 41 As Table 4 shows, US buyers targeting EIF-
backed investees are typically the largest in terms of assets and turnover, are more innovative (i.e.
more patents registered). They also are more technology-focused42 (see Figure D2 in Appendix D)
and mostly active in the ICT space (see Figure D3 in Appendix D), while European buyers seem
generally more specialised in Life Sciences.

Figure 14 shows the extent to which these start-up acquisitions represent more vertical- or horizontal-
type integrations. Moreover, we distinguish between financial and industrial acquisitions. Financial
acquisitions are defined as deals in which the buyer is an independent private equity firm (e.g. pri-
vate equity or hedge funds). Remarkably, most US buyers acquire start-ups from the same industry
in which they operate (vertical integration), while European buyers largely populate more traditional
sectors. However, the latter are more willing to integrate and/or expand horizontally towards inno-
vative technologies. Furthermore, financial buyers are often from the same European country of the
acquired start-up, suggesting that geographic proximity — perhaps signalling knowledge of local
markets and regulations — plays a relevant role in these transactions. The most frequent corporate
buyers of EIF-backed start-ups in our sample are GlaxoSmithKline, Broadcom Corp., Alcatel, eBay,
Microsoft and Apple (see Table D1 in Appendix D for further details).

Table 4: Economic and financial profile of EIF-backed start-ups buyers, by buyer geography
Buyer Geography Total Turnover* Nr of Size of the Nr of Buyers
Assets* (EUR m) Patents* Corporate
(EUR m) Group*
Same European Country 168 126 0 24.5 105
Other European Country 763 231 5 166 106
US 1182 618 132 86 125
Other Non-European Country 1459 626 38 65 53

Note: *expressed in median terms. Based on 389 private acquisitions of early-stage VC investees, invested between
1996 and 2015 by EIF-backed VC funds. The ”Size of the Corporate Group” reports the number of all ultimately owned
subsidiaries by the Global Ultimate Owner of the subject company. Source: BvD Orbis (2016), BvD Zephyr (2016).

Overall, these figures provide preliminary evidence in support of the missing “scale-up” opportunity
for start-ups in Europe, which is often claimed by many experts as one of the primary reasons behind
the US vs EU venture capital gap (Mind the Bridge and CrunchBase, 2016; Hellmann et al., 2016;
Duruflé et al., 2017). Duruflé et al. (2017) discuss a conceptual framework for the ”scale-up” cross-
roads. During the lifetime of a successful start-up, there comes a moment when the company faces
three options: going public, staying private or being acquired. While the first route is often jeopar-
dised by the lack of European stock exchanges suited to host scale-ups (Duruflé et al., 2017), the

41
As a reference, Mind the Bridge and CrunchBase (2016) exhibit M&A data regarding 1,271 European start-
ups acquired from 2012 onward by US and European companies. They find 44% of start-up acquisitions
to be performed by US companies. However, their data does not track acquisitions performed by non-
European, non-US buyers. Thus, while our data may perhaps be less representative, the time span of our
figures is larger and we account for non-US (e.g. Asian) buyers as well.
42
Shown by the higher proportion of US buyers listed on tech-focused stock markets, specifically the NASDAQ
stock exchange. It could be reasonably argued that US companies benefit from an over-representation
in the NASDAQ — an American stock exchange — but Pagano et al. (2002) show how also European
high-tech firms have a strong tendency to cross-list on US stock exchanges.

29
second option requires an important pre-condition. That is, additional capital to make the necessary
investments to foster growth (e.g. acquire competitors, expand internationally). The lack of follow-on
growth funding might thus force companies to their last option, i.e. being acquired. It should be
noted that the existence of acquisitions from foreign buyers is not negative per se. In fact, this may be
perfectly in line with market supply and demand dynamics. Moreover, this phenomenon may reflect
the often discussed higher risk-aversion of European entrepreneurs and investors, compared to US.
However, we should not rule out the presence of a deeper structural problem, i.e. a lack of growth
capital driving the scale-up gap. We leave it to further research to shed more light on this key issue.

Figure 14: Sector type of acquisitions of European EIF-backed VC investees, by buyer’s geography

Same European Country Other European Country

20

15

10

0
# Deals

‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

US Other Non-European Country

20

15

10

0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

Same sector Same macro-sector

Different macro-sector (financial acquisition) Different macro-sector (industrial acquisition)

Note: based on private acquisitions of EIF-backed early-stage VC investees occurred between 2003 and
2015. Source: BvD Zephyr (2016).

5.2 IPOs: going public after VC in Europe

Back in section 2, Table 2 showed how the Initial Public Offering is the type of VC exit that generates
the highest average and median return for the venture capitalist. The finding is in line with a well-
established literature that documents the positive correlation between IPOs and investor returns as
well as the IPOs’ out-performance and their relative rarity with respect to other forms of exit (e.g.
acquisitions, management buyouts and buybacks, see Phalippou and Gottschalg, 2008; Amit et al.,
1998). A total of 111 EIF-backed VC investments exited through IPO. That exclusively accounts
for liquidity events in which the VC firm sold its shares following the company’s IPO. Moreover, the
average IPO return is more than 100% higher than the average VC return from acquisitions.

30
However, not all IPOs of EIF-backed start-ups are quickly followed by a divestment of the VC firm. In
fact, there are several cases where the stocks of newly listed companies continue being held by the
investor. By adding these to the former, we reach an all-time 152 EIF-backed start-ups to have gone
public in 20 different stock exchanges worldwide. Figure 15 shows the trend of EIF-backed IPOs
over time, alongside the total number of start-ups supported by EIF. Although 2014 sees the highest
number of IPOs to date, there is a decrease in the IPO rate over time, considering the rise in EIF VC
activity in the period. In general, IPOs show extreme sensitivity not only to stock market conditions
but also to changes in technology-related sentiment and uncertainty about the future profitability of
innovative firms.43
Figure 15: EIF-supported tech IPOs over time
25 3000

20 2400

EIF VC Portfolio Size

15 1800
Nr of IPO

10 1200

5 600

0 0
‘98 ‘99 ‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15
Year of IPO

IPOs (lhs) Nr of EIF-backed start-ups (rhs)

Note: based on IPOs of EIF-backed start-ups occurring between 1998 and 2015. Grey bars indicate the
historical cumulative number of EIF VC-supported companies. Source: BvD Zephyr (2016), Thomson Reuters
Eikon (2016).

Figure 16 provides some descriptive evidence of the sector and the original location of the listed
VC investees. French start-ups have been the most prolific in terms of IPOs (49), followed by UK
(27) and US (26). Interestingly, more than 50% of all IPOs have been originated by companies in
the Life Sciences sector. Furthermore, Figure 17 exhibits the geographic distribution of the stock
exchanges hosting EIF-backed tech IPOs, showing where start-ups went public most frequently. The
stock exchange harbouring the highest number of EIF-backed IPOs is the Euronext Paris (50), followed
by the NASDAQ (29), the London Stock Exchange (23) and the Boerse Frankfurt (14). Moreover,
there are some peculiarities in the geographic-sector choice of listing. For instance, Euronext Paris
and LSE have featured more Computer Related listings, whereas NASDAQ and Boerse Frankfurt were
preferred by companies in the Biotech space (see Table D2 in Appendix D.).

43
See Berk et al. 2004; Pastor and Veronesi 2005, 2006, 2009 for a thorough discussion.

31
 Figure 16: Country and sector distribution of EIF-backed tech IPOs
50 49
Other Services 1

Other Manufacturing 1

Other Electronic Related 8

40

Medical/health Related
Medical/health Related 31

Industrial Automation 1 30
27
26

Nr IPOs
Energy 2

Consumer Related 2 20
Computer Related
Computer Related 37 15

Communications 14
10
6 6
Chemicals and Materials 2 5
4 4 4
2
Biotechnology 53 1 1 1 1
0
0 10 20 30 40 50 AT BE CH DE DK FI FR GB IE IL IT NL NO SE US
Nr IPOs Country

(a) IPOs by industry (b) IPOs by start-up location

Note: based on IPOs of start-ups invested between 1996 and 2015 by EIF-backed VC funds. The industrial nomenclature
follows Invest Europe’s classification. For details, see Signore (2016). ”Country” indicates the newly listed company location.
Source: BvD Zephyr (2016), Thomson Reuters Eikon (2016).

6 Understanding exit outcomes: a competing risks analysis

In this section we build on the descriptive findings on exit outcomes of EIF-backed VC investments to
discuss their underlying process. In addition to the pre-divestment valuation, a realised exit multiple
can certainly be influenced by the exit route as well as the exit timing chosen by venture capitalists.
In turn, these could be linked to a multitude of endogenous factors, both at micro- and macro-level.

Against this background, the goal of this section is to analyse the correlation between exit outcomes’
dynamics against a series of company-, fund- and cycle-related factors. While no causal relation
can be claimed throughout the exercise, there is arguably value in assessing how the likelihood of a
certain exit event varies according to, e.g., time, geography, VC teams characteristics, and so forth.

A number of works shed light on how European VC performs in terms of exit events. With a com-
parative focus, Axelson and Martinovic (2013) study the determinants of successful exits in Europe
and US. They find no systematic difference in the success rate of European and US VC investments in
relation to IPOs, but lower probability of exit through trade sale for European start-ups. Regardless of
geographical location, the experience of venture capitalists and founders is shown to be positively re-
lated to higher rates of trade sales. The authors argue that the shortage of serial entrepreneurship in
Europe may be the driving factor behind the VC performance gap between Europe and the US. Closer
to the approach in this work, Bottazzi et al. (2008) use a survey-driven and hand-collected dataset of
European VC investments, augmented with data from other commercially available databases. The
authors observe a strong positive correlation between VC investor’s activism — proxied by investor’s
business experience — and exit performance, measured through IPO and trade sale exit rates.

On the technical side, most studies analyse VC exit dynamics by focusing on binary outcomes mod-
elled via a Logit, Probit or linear probability approach (see e.g., Sorensen, 2007; Bottazzi et al.,
2008; Gompers et al., 2008; Brander et al., 2015). However, these models fail to account for the

32
Figure 17: Geographic distribution of the EIF-supported tech IPOs

New York
34

Oslo Helsinki
Stockholm 2
1
5

Copenaghen
2
Amsterdam Berlin
London
1 1
25 Brussels
6 Frankfurt
14
50
Zurich
Paris 7
Milan
2

Tel Aviv
1

Note: based on 152 IPOs of start-ups invested between 1996 and 2015 by EIF-backed VC funds. Source:
BvD Zephyr (2016), Thomson Reuters Eikon (2016).

time-to-exit dimension of VC investing. Moreover, this strategy only allows to model the occurrence
of broadly defined exit events,44 disregarding the competing nature of different kinds of exit events,
a feature intrinsic to the venture investment life-cycle. That is, not only start-ups can face a multitude
exit outcomes, but these events are also fundamentally different from one another: the factors and
circumstances behind a sensational IPO are typically very different from those leading to a manage-
ment buy-out. On this premise, Axelson and Martinovic (2013) revert to survival analysis techniques,
specifically a competing risks model to account for the time dimension in their data and to disentan-
gle IPOs from trade sales exits. Priorly, Giot and Schwienbacher (2007) estimated a competing risks
model accounting for three main exit routes for US VC investments: IPO, trade sales and liquidations.

44
The approach typically builds on binary ”exit statuses”, indicating investee companies that went public or
were acquired. It does not distinguish between unrealised investments and unprofitable exits (write-offs,
liquidations), mainly for the challenges in obtaining accurate VC returns data (see Da Rin et al., 2013).

33
This work aims at contributing to this literature by further enriching the body of competing risks
analyses. Based on exit deal type and return performance, we classify exited investments into four
main categories.45 These are based on all exit outcomes discussed in section 2 and read as follows:

• Write-off: the VC firm divests by declaring its stake in the company fully unreturned (i.e.
assigning zero or a symbolic amount to its value).;

• Liquidation: the VC firm sells its stake realising less than the investment cost (MoC < 1);

• Acquisition: the VC firm profitably realises its investment through a trade sale (MoC ≥ 1);

• IPO: the VC firm trades its company shares following the company’s listing.

6.1 Model set-up

The venture capital business model provides a suitable framework for the use of survival analysis.
Survival analysis, mostly used in medical research, deals with the (expected) time-to-event, where
an ”event” can be e.g. the death of a biological organisms or the disappearance of a particular
disease for patients under medical treatment.46 Against this background, a VC-backed start-up can
be viewed as the patient and a risk capital investment as the treatment. Hence, the observed lifetime
is nothing but the random time it takes for such investment to reach a liquidity event.47 The virtue of
survival analysis is that it allows modelling data that is inherently right-censored, e.g. for investments,
perhaps too recent, whose realisation has yet to materialise. This is a crucial advantage vis-á-vis
Least Squares estimation, as it allows to factor in evidence from exits that have yet to happen.

However, real-world patients may not be exposed to one exclusive outcome — say the occurrence
of a specific disease — but can experience alternative events, such as disease disappearance, oc-
currence of other diseases or death. In the same way, VC investments may face different, mutu-
ally exclusive outcomes. To address these aspects, scholars extended the standard survival analysis
framework through competing risks theory. Competing risks models appropriately account for mul-
tiple and mutually exclusive exit events. The advantage of this methodology is that it allows each exit
option to feature its own dynamics, together with the variables that can affect such outcome.

In line with Giot and Schwienbacher (2007), we estimate a competing risks model for 3,592 VC
investments backed by EIF. More specifically, the goal is to study how a set of covariates affect the
probability of a specific exit outcome, taking into account that alternative exit types can occur. In this
context, as discussed by Gooley et al. (1999), the standard Kaplan and Meier (1958) approach fails
to deliver correct estimates of the exit probabilities, because it treats competing events as if they were

45
It should be remarked how a wider range of contractual arrangements exists for VC exits (e.g., MBOs,
MBIs, shares buy-backs, replacement, options and warranties, etc.). However, given the main goal to
disentagle the key economic paths of different exit routes, we apply here some reasonable simplifications.
46
See Jenkins (2004) for one of the most comprehensive reviews of survival analysis methods.
47
The reader might note that this random lifetime mechanism is, together with expected exponential growth,
one of the ingredients for the potential rise of power laws in venture capital (see section 4).

34
censored.48 Another approach involves modelling an outcome-specific — or destination-specific —
hazard function for each exit type under the assumption of proportional hazards (Prentice et al.,
1978).49 However, as pointed out in Fine and Gray (1999) and Jenkins (2004), the interpretation of
regression coefficients is not straightforward with this method. As such, the key metric for our analysis
is instead the probability that an exit of type k has occurred by time t, accounting for the fact that
other exit types can occur in the meantime. This is known as the cumulative incidence function (CIF):
∫ t
Ck (t; X) = Pr(exit time T ≤ t, exit type = k | X) = S(u)hk (u | X)du (3)
0

where S(·) is the overall survival function, hk (·) is the destination-specific hazard and X is a vector
of explanatory variables.

To give an example that fits our context, the CIF for an acquisition at 5 years is the probability that
a VC portfolio company is profitably acquired in the first 5 post-investment years, accounting for the
likelihood of other exit events (e.g., the start-up goes bankrupt and is written-off, the founders buy
back their shares following an argument with the investors) and a set of company/investment features.
Fine and Gray (1999) specify a semi-parametric transformation for the CIF, also called subdistribution
hazard — or subhazard — that is rather similar to Cox regression (Cox, 1972), commonly used in
survival models. We use this as our main approach, as the complexity of this model is offset by the
easiness of its estimation and its ability to provide a more direct interpretation between the covariates
and the probabilities of interest.50

The Fine and Gray (1999) method yields estimated coefficients known as subhazard ratios, expressed
in exponential form. Their interpretation is as follows: a ratio above one implies that an increase
in the covariate x1 raises the incidence of exit type k. Conversely, a ratio below one means that an
increase in the covariate x1 lowers the incidence of exit type k.51 The interpretation of subhazard
ratios bears resemblance to that of odds ratios in the context of standard survival analysis.

Table 5 provides some descriptive evidence on the time-to-exit for the four types of exit classes.
These statistics suggest an average pecking order for exit strategies in Europe: most hopeless bets
are divested the earliest through write-off, while profitable trade sales start being monetised shortly
after that. Liquidations further take place with tumbling growth expectations. Finally, the highest-
growers are brought to public markets to capitalise their post-IPO shares.

48
For instance, suppose we were interested in modelling the probability of a profitable vs an unprofitable exit,
say, a write-off. Now assume that a certain number of investments are indeed written-off at time t. When
modelling profitable exits in t + 1, the approach in Kaplan and Meier (1958) treats write-off occurrences
as if they were still unrealised, altering the risk set and producing a bias in the estimated exit probabilities.
49
Any hazard function denotes the likelihood of exiting at time t having survived thus far. The destination-
specific hazard, hk (t), is instead the risk of exiting at time t specifically through exit type k, given that the
exit has not occurred thus far (see Prentice et al., 1978).
50
See Cleves et al. (2010) for the Stata routine used in this work.
51
The Fine and Gray method assumes proportionality of the subhazards and independent risks. The latter
is a strong assumption, implying that the probability of a certain exit type k is unrelated to the probability
of other exit types. Releasing this assumption adds considerable computational complexity and yields
likelihoods which are hardly tractable. While theoretical works have shed light on this issue, there are still
few empirical applications of competing risks models that assume correlated risks (Jenkins, 2004).

35
Table 5: Years from vintage to exit for EIF-backed VC investments, by exit outcome
Exit Outcome Average Time to Exit Median Time to Exit St.Dev. N
In Portfolio 5.7 4.0 4.2 1523
Write-off 4.9 4.2 3.1 705
Liquidation 5.8 5.5 3.3 779
Acquisition 4.9 4.3 3.3 448
IPO 7.1 7.0 3.3 109
Total 5.5 4.5 3.7 3564

Note: 28 investments made by non-European EIF-backed VC funds were dropped. The data was computed by counting
the number of days between first investment and exit date and further converting back to years.

6.2 Results

Appendix E describes all variables used in the model, constituting a set of investment-, company- and
investor-level features that often appear in numerous leading studies in the literature. Table 6 shows
the estimation results of the competing risks model using the Fine and Gray method. Categorical
variables are presented in bold and are to be interpreted with respect to the omitted class, set to
the mode of the distribution. For instance, the coefficients for ”Fund macro-region” dummies are
relative effects with respect to the omitted class of DACH-based VC firms.52

The empirical evidence concerning the four different exit routes can be summarised as follows. For
write-offs, the predicted incidence is lower for more recent vintage years. This may be both driven
by the high rate of write-offs during the dot-com crash as well as the longer-term decline in write-off
rates observed in Figure 1. However, note that the shorter holding period of very recent vintages may
bias later write-off rates. A lower probability of write-off is also linked to a higher VC firm experience
(proxied by the number of funds raised previously), to investees active in the Life Sciences industry,
and to the fund’s investment strategy being expansion-focused.53 Last, investments from VC firms
based in the NORDICS region are linked to a higher incidence of write-offs.

Lower probabilities of liquidations are linked to higher first investment amounts. As in the case of write-
offs, more recent vintage year and Life Sciences investments are also associated to lower liquidation
rates. Interestingly, an increase in the size of the VC fund is linked to a higher probability for the
investee to be unprofitably sold. This seemingly counter-intuitive fact suggests that a large portion
of the money gathered at fund-raising stage is not necessarily correlated with investee companies’
growth. One might argue that, ceteris paribus, a greater fund size would lead to a larger number
of investments, thus a higher incidence of unprofitable sales due to the limited stock of time and
resources that can be allocated across portfolio companies (Masters and Thiel, 2014). However, we
do not find confirmatory evidence of this reasoning since the size of the portfolio is not significantly

52
The set of ”Geographic macro-region” and ”Fund investment focus” dummies are included in the estimated
model. Note that, out of 3,592 VC investments initially in the sample, 2,823 enter the model, mainly due
to a high number of missing values for very few covariates, such as ”Fund Distance”, the distance between
VC firm and investee, that requires the companies to be matched and geolocalised in the BvD Orbis
database. See Kraemer-Eis et al. (2016b) for details.
53
This finding is omitted from Table 6 for the sake of conciseness. It may be linked to a higher propensity —
perhaps experience-driven — of venture capital-specialised funds to quickly write-off their positions.

36
Table 6: Fine & Gray Competing Risks Model Estimates
empty Write-off Liquidation Acquisition IPO
Age at investment 1.004 0.988 1.014 0.959
(0.23) (-0.70) (0.61) (-0.88)
First Investment 0.953 0.840*** 1.069 1.387**
(-1.14) (-4.87) (1.13) (2.38)
Fund Distance 1.011 0.977 0.977 1.032
(0.68) (-1.49) (-1.20) (0.67)
Vintage Year 0.919*** 0.945*** 1.006 0.824***
(-6.05) (-4.49) (0.47) (-4.32)
Unicorn status 0.347 1.160 2.698*** 0.824
(-1.57) (0.42) (2.80) (-0.19)
First-time VC teams 1.033 1.018 1.000 1.233
(0.28) (0.15) (0.00) (0.59)
Fund Size 0.994 1.205*** 0.994 0.999
(-0.09) (2.89) (-0.07) (-0.00)
Portfolio Size 1.011 0.963 0.896 1.483
(0.14) (-0.47) (-1.29) (1.55)
Funds raised by VC firm 0.809*** 1.037 1.109** 0.947
(-4.36) (1.01) (2.53) (-0.64)
Investee macro-sector (omitted: ICT)
Life Sciences 0.617*** 0.707*** 0.853 4.087***
(-4.60) (-3.43) (-1.24) (6.42)
Manufacturing 1.260 0.933 1.097 0.000***
(1.04) (-0.30) (0.34) (-85.89)
Services 0.768 1.082 1.512** 0.000***
(-1.26) (0.46) (2.28) (-83.16)
Green Technologies 0.846 0.301** 1.304 0.000***
(-0.40) (-2.00) (0.70) (-59.65)
Fund macro-region (omitted: DACH)
NORDICS 2.264** 1.630 0.888 0.473
(2.29) (1.54) (-0.38) (-0.71)
FR&BENELUX 1.078 0.909 1.506* 1.203
(0.37) (-0.56) (1.87) (0.47)
SOUTH 1.039 0.956 1.156 0.159*
(0.13) (-0.15) (0.36) (-1.88)
UK&IRELAND 0.933 0.778 1.979*** 2.463**
(-0.30) (-1.14) (3.32) (2.33)
CESEE 0.598 1.468 1.061 0.000***
(-0.67) (0.64) (0.09) (-10.46)
Observations 2823 2823 2823 2823
Log-Likelihood -4059.96 -4512.15 -2839.67 -617.40
LR Chi-Sq. 929.63 1065.82 811.44 39738.44
Chi-Square(p-value) 0.000 0.000 0.000 0.000
Geographic macro-region Yes Yes Yes Yes
Fund investment focus Yes Yes Yes Yes
Nr Subjects 2823 2823 2823 2823
Nr Exits 555 618 380 94
* p<0.10, ** p<0.05, *** p<0.01. Exponentiated coefficients; t statistics in parentheses
Note: based on 2,823 EIF-backed VC investments. Estimated coefficients reported are exponentiated coefficients, i.e.
subhazard ratios. A subhazard ratio above (below) one means that the effect of increasing the covariate is to increase
(decrease) the probability of the exit outcome modelled. Fine and Gray (1999) estimation implemented through the Stata
program in (Cleves et al., 2010).

37
related to a rise in the incidence of liquidation.54 We leave to future research on the topic the task
to shed further light on the potential spurious nature of this finding.

Higher rates of acquisitions are linked to the number of funds previously raised by the VC firm,
to investees operating in the Services industry, and to the VC firm being located in UK&IRELAND.
Interestingly, becoming a unicorn55 strongly increases the chance of being privately acquired, while
it does not significantly affect the incidence of IPOs. This dynamics might suggest investors’ caution
for IPO exit strategies when the company valuation is very high.

Last, the chances of an IPO decreased in recent vintage years. Of course, a similar disclaimer to
the one described above on write-offs applies: this could be driven by IPOs requiring more time to
materialise, as shown in Table 5. Therefore, recent vintages may naturally experience such lower
probabilities. A Life Sciences portfolio company has a significantly higher probability of going public
compared to other industries. The result is perfectly in line with the previous findings in Signore
(2016). Moreover, first investment amount is linked to a higher IPO rates. As this feature can be
interpreted as the size of the first ”bet” in the company, this finding bears an important implication,
suggesting that VC firms were able to cherry-pick successful companies already at the time of their
first check. This finding is also in line with the evidence brought by Sorensen (2007) on sorting being
twice as important as direct VC influence to explain IPO incidence rates. Finally, being invested by
a UK&IRELAND firm is significantly correlated to a higher likelihood of going public. This key result
conveys some indications of where the VC asset class tends to be more profitable across Europe.
Although causality is never evoked, Irish and British VC firms are observed to play an important role
in shaping the performance of returns at fund-of-fund level.

A further advantage of the Fine and Gray (1999) approach lies in the ability to model how the
cumulative incidence function — the exit probability for a given outcome having survived thus far
— evolves throughout the years following an EIF-backed investment. In other words, the model
can provide, for a change in a given variable, the change in the day-by-day predicted exit outcome
probability. For instance, Figure 18 shows how the probability of an acquisition evolves given different
values of, respectively, the first investment amount and the number of funds raised by a VC firm.

Differences in the cumulative incidence curve provide useful insights on the relationship between
investment features and the chances of reaching an acquisition. In the case of Figure 18b, the
probability of being acquired within 10 years after the VC investment is observed to be 5% higher
when the VC firm has already raised 4 funds, compared to first time VC teams. Similarly, in Figure 18a
the cumulative incidence of acquisition by the 10th year is 5% larger when the first investment is around
EUR 3m, compared to an initial investment of EUR 50,000.

54
As an heuristic on the fact that fund size is not necessarily related to underlying portfolio companies’
performance, one can think of Lowercase Capital as example. The US VC firm launched its first Lowercase
Ventures I in 2010, raising only USD 8m. The fund is claimed to be the best-ever performing VC fund,
generating massive returns from the sale of Twitter and Instagram (Uber and Docker still in portfolio)
and investing almost only in seed rounds. Despite the outstanding returns, the firm has kept their fund
sizes consistently under USD 50m to maintain their strategy and focus on early investments in very small
companies with very high growth potential (Source: CBInsights).
55
In VC jargon, a Unicorn is a privately-held VC-backed start-up whose valuation is greater or equal than
USD 1 bn (Aileen, 2013). See Appendix G for a list of EIF-backed unicorns to date.

38
 Figure 18: Changes in the cumulative incidence function for acquisition
25%
20%

20%

Cumulative Incidence function

Cumulative Incidence function

15%

15%

10%
10%

5%
5%

0% 0%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Years after EIF-backed investment (t = 0) Years after EIF-backed investment (t = 0)

EUR 3m EUR 50k 4th Fund 1st Fund

(a) First investment amount (b) Number of funds raised by the VC firm

Note: based on 2,823 EIF-backed VC investments. Fine and Gray (1999) estimation implemented through the software
in (Cleves et al., 2010).

Finally, we find evidence of the investors’ experience playing a crucial role. Interestingly, our ”First
time-team” dummy is never significant and the findings from section 3.2 also hint at no significant
returns differences. This finding points to the convexity of the effect, for the experience of VC firms
matters only when outstanding. However, an alternative, perhaps leaner explanation for this effect
may be the selection effect of EIF, e.g. during the due-diligence process. In other words, the fact that
EIF may select only high-potential first-time teams may be a source of bias for this particular result.

To ensure the robustness of our results, we compare the estimates of the methodology in Fine and
Gray (1999) with results from the destination-specific hazard. Specifically, we seek to test whether
our results are sensitive to the continuous-time assumption of Fine and Gray (1999). To do so, we
revert to a discrete-time setting, following the approach outlined in Jenkins (2004). In particular,
we estimate a multinomial logit model, 56 assuming independent risks and proportional hazards.
Table F1 in appendix F shows the estimation output. Overall, results hold qualitatively similar.

7 Conclusions

This work represents a further step towards an increased understanding of venture capital in Eu-
rope and its financial performance. It uses a sample of about 3,600 EIF-backed venture capital
investments made in the 1996-2015 period to analyse their liquidity events and returns. The key
contribution of this work lies in the analysis of historical VC exits and start-up valuations. The paper
provides evidence that VC returns are related to the economic cycle. At the same time, it emphasises
the heterogeneity that arises from different industries and geographies, which leaves room for VC
firms to pursue diversification strategies and minimise the correlation with other asset classes.

A second key contribution pertains to the analysis of return distribution. The paper discusses the
statistical features of VC returns in Europe, providing preliminary evidence of their power-law be-

56
The approach is justified by the theoretical results discussed in Allison (1982).

39
haviour. While further research will be needed to reach conclusive evidence on whether power-law
is the best-fitting distribution for VC returns, the insights yield practical implications for practitioners
and policy-makers. Namely, it calls for a cautious rethinking of investment strategies and portfolio
management approaches that do not account for the possibility of extreme outcomes. Secondly, it
suggests that policies to address the ”scale-up” gap in Europe need to carefully assess the implica-
tions of start-ups’ exponential growth in their respective markets, as well as their potential impact on
the broader economy.

Third, the paper looks at exit outcomes and finds that the largest share of VC returns is generated by
two different exit types: private trade sales and IPOs. A careful investigation of these two exit events
provides an overview of start-ups buyers, their domain and their geographic location. Moreover, a
total of 152 IPOs experienced by EIF-supported companies are discussed in detail.

Closing the circle, a final contribution of this work relates to the analysis of the determinants of
successful exit outcomes, and is sought to put into perspective the numerous findings discussed thus
far. The correlation analysis finds, inter alia, that the experience of VC firm is positively linked to
lower rates of write-offs and higher rates of profitable trade sales. However, first time teams backed
by EIF are shown to perform not significantly different than more experienced teams, hinting that
the contribution of experience is relevant only when outstanding. Alternatively, an equally plausible
explanation could be the selection bias brought by EIF’s own high-standard screening of first time
VC teams. In turn, this implies that this particular result may not hold for the entire VC ecosystem.

While the focus of this work is purely on gross returns of investments into start-ups, some lessons
can certainly be transferred to the field of fund or fund-of-funds performance. In the spirit of the
broader series of working papers, forthcoming issues will continue to tackle start-up-level dynamics
with the perspective of enriching the body of research on the effects of EIF-supported venture capital
investments. Against this background, forthcoming issues will have a specific look at the innovative
ability of European start-ups.

40
References

Aileen, L. (2013). Welcome To The Unicorn Club: Learning From Billion-Dollar Startups. Techcrunch. URL
https://techcrunch.com/2013/11/02/welcome-to-the-unicorn-club/, [Online; Posted 2 November 2013].

Allison, P.D. (1982). Discrete-time methods for the analysis of event histories. Sociological Methodology,
vol. 13, pp. 61–98.

Alstott, J., Bullmore, E. and Plenz, D. (2014). powerlaw: A Python Package for Analysis of Heavy-Tailed
Distributions. PLOS ONE, vol. 9, no. 1, pp. 1–11.

Amit, R., Brander, J. and Zott, C. (1998). Why do venture capital firms exist? Theory and Canadian evidence.
Journal of Business Venturing, vol. 13, no. 6, pp. 441–466.

Ang, A. and Sorensen, M. (2012). Risks, Returns and Optimal Holdings of Private Equity: A Survey of Existing
Approaches. Quarterly Journal of Finance, vol. 02, no. 03, p. 1250011.

Axelson, U. and Martinovic, M. (2013). European venture capital: Myths and Facts. Working Paper, British
Private Equity & Venture Capital Association.

Bergemann, D. and Hege, U. (2005). The Financing of Innovation: Learning and Stopping. The RAND Journal
of Economics, vol. 36, no. 4, pp. 719–752.

Berk, J.B., Green, R.C. and Naik, V. (2004). Valuation and Return Dynamics of New Ventures. The Review of
Financial Studies, vol. 17, no. 1, p. 1.

Black, B.S. and Gilson, R.J. (1998). Venture capital and the structure of capital markets: banks versus stock
markets. Journal of Financial Economics, vol. 47, no. 3, pp. 243 – 277.

Black, F. and Scholes, M. (1973). The Pricing of Options and Corporate Liabilities. Journal of Political
Economy, vol. 81, no. 3, pp. 637–654.

Bottazzi, L. and Da Rin, M. (2003). Financing Entrepreneurial Firms in Europe: Facts, Issues, and Research
Agenda. CESifo Working Paper Series 958, CESifo Group Munich.

Bottazzi, L., Da Rin, M. and Hellmann, T. (2008). Who are the active investors? Evidence from venture capital.
Journal of Financial Economics, vol. 89, no. 3, pp. 488 – 512.

Brander, J.A., Du, Q. and Hellmann, T. (2015). The Effects of Government-Sponsored Venture Capital:
International Evidence. Review of Finance, vol. 19, no. 2, pp. 571–618.

Casamatta, C. (2003). Financing and Advising: Optimal Financial Contracts with Venture Capitalists. The
Journal of Finance, vol. 58, no. 5, pp. 2059–2085.

Chaplinsky, S. and Gupta-Mukherjee, S. (2016). Investment risk allocation and the venture capital exit market:
Evidence from early stage investing. Journal of Banking & Finance, vol. 73, pp. 38 – 54.

Cirillo, P. (2013). Are your data really Pareto distributed? Physica A: Statistical Mechanics and its Applications,
vol. 392, no. 23, pp. 5947 – 5962.

Clauset, A., Shalizi, C.R. and Newman, M.E.J. (2009). Power-Law Distributions in Empirical Data. SIAM
Review, vol. 51, no. 4, pp. 661–703.

41
Cleves, M., Gould, W.W., Gutierrez, R.G. and Marchenko, Y. (2010). An Introduction to Survival Analysis
Using Stata.

Cochrane, J.H. (2005). The risk and return of venture capital. Journal of Financial Economics, vol. 75, no. 1,
pp. 3 – 52.

Cox, D.R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society Series B (Method-
ological), vol. 34, no. 2, pp. 187–220.

Da Rin, M., Hellmann, T. and Puri, M. (2013). Chapter 8 - A Survey of Venture Capital Research. In M.H.
George M. Constantinides and R.M. Stulz, editors, Handbook of the Economics of Finance, vol. 2, Part A
of Handbook of the Economics of Finance, pp. 573 – 648. Elsevier.

Da Rin, M., Nicodano, G. and Sembenelli, A. (2006). Public policy and the creation of active venture capital
markets. Journal of Public Economics, vol. 90, no. 8–9, pp. 1699 – 1723.

Dickey, D.A. and Fuller, W.A. (1979). Distribution of the Estimators for Autoregressive Time Series With a Unit
Root. Journal of the American Statistical Association, vol. 74, no. 366, pp. 427–431.

Duruflé, G., Hellmann, T.F. and Wilson, K.E. (2017). From Start-Up to Scale-Up: Examining Public Policies
for the Financing of High-Growth Ventures. Saïd Business School WP 2017-05, Saïd Business School.

Fine, J.P. and Gray, R.J. (1999). A Proportional Hazards Model for the Subdistribution of a Competing Risk.
Journal of the American Statistical Association, vol. 94, no. 446, pp. 496–509.

Fulghieri, P. and Sevilir, M. (2009). Size and Focus of a Venture Capitalist’s Portfolio. The Review of Financial
Studies, vol. 22, no. 11, p. 4643.

Gabaix, X. (2009). Power Laws in Economics and Finance. Annual Review of Economics, vol. 1, pp. 255–93.

——— (2011). The Granular Origins of Aggregate Fluctuations. Econometrica, vol. 79, no. 3, pp. 733–772.

——— (2016). Power Laws in Economics: An Introduction. Journal of Economic Perspectives, vol. 30, no. 1,
pp. 185–206.

Ginsburg, A. (2012). plfit: Python implementation of Aaron Clauset’s power-law distribution fitter. Python
Package Index. URL https://pypi.python.org/pypi/plfit.

Giot, P. and Schwienbacher, A. (2007). IPOs, trade sales and liquidations: Modelling venture capital exits
using survival analysis. Journal of Banking & Finance, vol. 31, no. 3, pp. 679 – 702.

Gompers, P., Kovner, A., Lerner, J. and Scharfstein, D. (2008). Venture capital investment cycles: The impact
of public markets. Journal of Financial Economics, vol. 87, no. 1, pp. 1 – 23.

Gooley, T.A., Leisenring, W., Crowley, J. and Storer, B.E. (1999). Estimation of failure probabilities in the
presence of competing risks: new representations of old estimators. Statistics in Medicine, vol. 18, no. 6,
pp. 695–706.

Granger, C.W.J. (1969). Investigating Causal Relations by Econometric Models and Cross-spectral Methods.
Econometrica, vol. 37, no. 3, pp. 424 – 438.

Hamilton, J.D. (1994). Time series analysis, vol. 2. Princeton university press Princeton.

Hege, U., Palomino, F. and Schwienbacher, A. (2009). Venture Capital Performance: The Disparity Between
Europe and the United States. Finance, vol. 30, no. 1, pp. 7–50.

42
Hellmann, T., Frydrych, D., Hicks, C. and Rauch, C. (2016). Financing UK Scale-ups: Challenges and
Recommendations. Chapter Two of: Scale-up UK: Growing Businesses, Growing our Economy, Oxford
University, Said Business School.

Hellmann, T. and Stiglitz, J. (2000). Credit and equity rationing in markets with adverse selection. European
Economic Review, vol. 44, no. 2, pp. 281 – 304.

Hull, J.C. (2006). Options, futures, and other derivatives. Pearson Education India.

Jenkins, S.P. (2004). Survival Analysis. Unpublished manuscript, Institute for Social and Economic Research,
University of Essex, Colchester, UK. URL https://www.iser.essex.ac.uk/files/teaching/stephenj/ec968/pdfs/
ec968lnotesv6.pdf.

Kaplan, S.N. and Lerner, J. (2016). Venture Capital Data: Opportunities and Challenges. Working Paper
22500, National Bureau of Economic Research.

Kaplan, E.L. and Meier, P. (1958). Nonparametric Estimation from Incomplete Observations. Journal of the
American Statistical Association, vol. 53, no. 282, pp. 457–481.

Kaplan, S.N. and Schoar, A. (2005). Private Equity Performance: Returns, Persistence, and Capital Flows.
The Journal of Finance, vol. 60, no. 4, pp. 1791–1823.

Korteweg, A. and Sorensen, M. (2010). Risk and Return Characteristics of Venture Capital-Backed En-
trepreneurial Companies. The Review of Financial Studies, vol. 23, no. 10, p. 3738.

Kraemer-Eis, H., Lang, F., Torfs, W. and Gvetadze, S. (2016a). European Small Business Finance Outlook,
December 2016. EIF Working Paper 2016/37, EIF Research & Market Analysis.

Kraemer-Eis, H., Signore, S. and Prencipe, D. (2016b). The European venture capital landscape: an EIF
perspective. Volume I: the impact of EIF on the VC ecosystem. EIF Working Paper 2016/34, EIF Research
& Market Analysis.

Lerner, J. (2009). Boulevard of broken dreams: why public efforts to boost entrepreneurship and venture
capital have failed–and what to do about it.

Mandelbrot, B. (1997). The variation of certain speculative prices. In Fractals and Scaling in Finance, pp.
371–418. Springer New York.

Masters, B. and Thiel, P. (2014). Zero to one: notes on start ups, or how to build the future. Random House.

Michelacci, C. and Suarez, J. (2004). Business Creation and the Stock Market. The Review of Economic
Studies, vol. 71, no. 2, p. 459.

Mind the Bridge and CrunchBase (2016). Startup Transatlantic M&As: US vs. EU. Tech. rep., Startup Europe
Partnership.

Mitzenmacher, M. (2003). A Brief History of Generative Models for Power Law and Lognormal Distributions.
Internet Mathematics, vol. 1, no. 2, pp. 226–251.

Neumann, J. (2015). Power Laws in Venture. reactionwheel.net. URL http://reactionwheel.net/2015/06/

power-laws-in-venture.html, [Online; Posted 25 June 2015].

Newman, M. (2005). Power laws, Pareto distributions and Zipf’s law. Contemporary Physics, vol. 46, no. 5,
pp. 323–351.

43
Pagano, M., Röell, A.A. and Zechner, J. (2002). The Geography of Equity Listing: Why Do Companies List
Abroad? Journal of Finance, vol. 57, no. 6, pp. 2651–2694.

Pastor, L. and Veronesi, P. (2005). Rational IPO Waves. The Journal of Finance, vol. 60, no. 4, pp.
1713–1757.

——— (2006). Was there a Nasdaq bubble in the late 1990s? Journal of Financial Economics, vol. 81,
no. 1, pp. 61 – 100.

——— (2009). Technological Revolutions and Stock Prices. American Economic Review, vol. 99, no. 4, pp.
1451–83.

Phalippou, L. and Gottschalg, O. (2008). The Performance of Private Equity Funds. The Review of Financial
Studies, vol. 22, no. 4, p. 1747.

Pomeroy, J. (2016). Rise of the digital natives. HSBC Insights. URL http://www.hsbc.com/news-and-insight/
insight-archive/2016/rise-of-the-digital-natives, [Online; Posted 12 September 2016].

Prentice, R.L., Kalbfleisch, J.D., Peterson, A.V., Flournoy, N., Farewell, V.T. and Breslow, N.E. (1978). The
Analysis of Failure Times in the Presence of Competing Risks. Biometrics, vol. 34, no. 4, pp. 541–554.

Reed, W.J. and Hughes, B.D. (2002). From gene families and genera to incomes and internet file sizes: Why
power laws are so common in nature. Physical Review E, vol. 66, p. 067103.

Scherer, F.M. (2000). The size distribution of profits from innovation. In The Economics and Econometrics of
Innovation, pp. 473–494. Springer.

Schmidt, K.M. (2003). Convertible Securities and Venture Capital Finance. The Journal of Finance, vol. 58,
no. 3, pp. 1139–1166.

Signore, S. (2016). The European venture capital landscape: an EIF perspective. Volume II: Growth patterns
of EIF-backed startups. EIF Working Paper 2016/38, EIF Research & Market Analysis.

Sorensen, M. (2007). How Smart Is Smart Money? A Two-Sided Matching Model of Venture Capital. The
Journal of Finance, vol. 62, no. 6, pp. 2725–2762.

——— (2008). Learning by Investing: Evidence From Venture Capital. AFA 2008 New Orleans Meetings
Paper, Columbia Business School.

Sornette, D. (2002). Economies of scale in innovations with block-busters. Quantitative Finance, vol. 2, no. 3,
pp. 224–227.

——— (2009). Why stock markets crash: critical events in complex financial systems. Princeton University
Press.

Taleb, N. (2007). The Black Swan: The Impact of the Highly Improbable. Incerto. Random House Publishing
Group.

Tykvová, T., Borell, M. and Kroencke, T. (2012). Potential of venture capital in the european union. Tech.
rep., Directorate-General for Internal Policies, European Parliament. URL http://www.europarl.europa.eu/
studies.

Virkar, Y. and Clauset, A. (2014). Power-law distributions in binned empirical data. The Annals of Applied
Statistics, vol. 8, no. 1, pp. 89–119.

44
Vuong, Q.H. (1989). Likelihood Ratio Tests for Model Selection and Non-Nested Hypotheses. Econometrica,
vol. 57, no. 2, pp. 307–333.

Weidig, T., Kemmerer, A. and Born, B. (2005). The risk profile of private equity funds of funds. Journal of
Alternative Investments, vol. 7, no. 4, pp. 33–41.

Zipf, G.K. (1949). Human Behavior and the Principle of Least Effort: An Introduction to Human Ecology.
Addison–Wesley Press.

45
Appendices

A Granger-Causality Test

This section outlines the details of the Granger-Causality test (Granger, 1969) implemented in sec-
tion 3 to evaluate whether a stock market technology index (the NASDAQ Composite Index) predicts
movements in the median start-up valuation. These are two quarterly time series spanning the period
1999Q1 - 2016Q1.

The intuition behind the method is that, given two time series xt and yt , we use the test to determine
whether one of the two series is useful in forecasting the other. The variable x is said to Granger-
cause variable y if predictions of the future value of y based on past lags of both y and x are better
than predictions of y based on its own past values only.

The test is implemented by estimating a vector autoregressive model (VAR) for the two time series of
interest x and y. The two time series are required to be stationary. Hence, in case of non-stationary
data, one must ”difference” the time series until having them stationary, ensuring that there is no
unit root in the model. Then, a F-test on the joint significance of the other variable’s past lags is
performed on each system equation (Chi-square, LR or Wald tests are used as well).

In our context, we can think of the median start-up valuation as y and the stock market technology
index as x. Before estimating the VAR model, we check for the stationarity of the quarterly time
series by means of an augmented Dickey-Fuller (ADF) test57 . When unit root is not rejected, we first-
difference the series. Noticeably, the median start-up valuation series is differenced twice in order
to get it stationary. Then, we estimate a 4-lags quarterly VAR model, where the number of lags is
selected jointly by lowest AIC, HQIC, SBIC.

Once estimated the VAR model, a Wald test on the joint significance of the coefficients of past
lags of NASDAQ (Median Valuation) in the Median Valuation (NASDAQ) equation is performed. In
particular, for each VAR equation, the null hypothesis is that past lags of the ”excluded” variable are
not significant in predicting values of the dependent variable. As Table A1 shows, the null hypothesis
can be rejected only for the Median Valuation equation with respect to the NASDAQ. Therefore,
there is evidence that a stock market technology index Granger-causes valuations, i.e. it brings
informational content useful to predict movements in the median start-up valuation.

Table A1: Granger-Causality test of NASDAQ Composite Index and median start-up valuation
Equation Excluded F statistic df df-r p-value
NASDAQ Median Valuation .923 4 54 .4575
Median Valuation NASDAQ 2.886 4 54 .0308

Note: the null hypothesis of the first test (first block) is that the median valuation does not Granger-cause the NASDAQ
Composite index. The null hypothesis of the second test (second block) is that the NASDAQ does not Granger-cause the
median valuation.

57
For the technical details of the test procedure, see Dickey and Fuller (1979) and Hamilton (1994).

46
 B Additional evidence on the distribution of exit returns

Figure B1: Distribution of the exit return class, unweighted MoCs, by sector

ICT Life Sciences Manufacturing

.75

0.60 0.61
0.52
.5
.25

0.19
0.16 0.14 0.14
0.12 0.13
0.08 0.08 0.09
0.06
0.04 0.03
0

<0.25 0.25 to CostAt Cost Cost up to 5 >5 <0.25 0.25 to CostAt Cost Cost up to 5 >5 <0.25 0.25 to CostAt Cost Cost up to 5 >5

Services Green Technologies Total

.75

0.58 0.57
0.49
.5

0.26
0.22
.25

0.16 0.17
0.14 0.13
0.11 0.08
0.05 0.04
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

47
Figure B2: Distribution of the exit return class, unweighted MoCs, by vintage year

1994-2001 2002-2006
.75

0.63

0.53
.5
.25

0.20
0.14 0.12 0.13
0.10
0.07
0.04 0.04
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

2007-2016 Total
.75

0.57
0.53
.5

0.23
.25

0.17
0.13
0.09 0.09 0.07 0.08
0.04
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

Figure B3: Distribution of the exit return class, unweighted MoCs, by region

DACH NORDICS FR+BENELUX

.75

0.64
0.55 0.53
.5

0.20
.25

0.18 0.16 0.14

0.09 0.13 0.09 0.09
0.05 0.06 0.06
0.02
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

SOUTH UK+IRELAND CESEE

.75

0.61 0.61 0.60

.5
.25

0.17 0.19
0.13 0.15 0.11
0.09 0.10 0.09 0.09
0.02 0.03 0.02
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

US Total
.75

0.56 0.57
.5
.25

0.19 0.17
0.13 0.13
0.09 0.08 0.04
0.04
0

<0.25 0.25 to Cost At Cost Cost up to 5 >5 <0.25 0.25 to Cost At Cost Cost up to 5 >5

Note: based on 2,065 early-stage VC investments, exited, made between 1996 and 2015 by EIF-backed VC
funds. The figures include all the exit types, i.e. write-off, liquidations and successful sales. Exit MoCs are not
weighted. The ”At Cost” bucket includes all the MoC values such that 0.8 ≤ MoC < 1.2. ”Total” category
reflects Figure 4. ”ROW” excluded for its negligibility of number of observations.

48
Figure B4: Weighted exit MoCs of EIF-backed VC investments, by sectors

Communications Computer Related

6x
60
5x
50
4x
40
3x 30
2x 20
1x 10
0x 0
‘03 ‘06 ‘09 ‘12 ‘15 ‘03 ‘06 ‘09 ‘12 ‘15

Electronics/Automation Biotech
6x
Multiple on Cost

60
5x

Nr of exits
50
4x
40
3x 30
2x 20
1x 10
0x 0
‘03 ‘06 ‘09 ‘12 ‘15 ‘03 ‘06 ‘09 ‘12 ‘15

Medical/health related Services

6x
60
5x
50
4x
40
3x 30
2x 20
1x 10
0x 0
‘03 ‘06 ‘09 ‘12 ‘15 ‘03 ‘06 ‘09 ‘12 ‘15

Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Figure B5: Weighted exit MoCs of EIF-backed VC investments, experienced vs first-time VC teams

Experienced VC teams First-time VC teams

7x 140

6x 120

5x 100
Multiple on Cost

Nr of exits

4x 80

3x 60

2x 40

1x 20

0x 0
‘00 ‘03 ‘06 ‘09 ‘12 ‘15 ‘00 ‘03 ‘06 ‘09 ‘12 ‘15

Exit Year

Nr Exits Average Exit MoC Median Exit MoC

Note: based on 2,065 early-stage VC investments, exited, made between 1996 and 2015 by EIF-backed
VC funds. The figures include all exit types, i.e. write-off, liquidations and successful sales. Exit MoCs are
weighted. Grey bars report the number of exits per year in the sample (right hand axis).

49
C Additional evidence on the power-law assumption in VC returns

This section outlines the key procedure and estimates obtained by implementing the Clauset et al.
(2009) procedure for the validation of power law distributions with continuous data.

Table C1: Clauset et al. (2009) goodness-of-fit test for power law distributions
Obs Mean Max Std n (Tail) alpha sigma xmin MC p-value
1414 1.59x 138.98x 5.34 244 2.45 0.09 2.35x 0.915

Note: based on 1,414 non-zero exit returns observations. The power law parameters are obtained with the ML estimators
proposed in Clauset et al. (2009). Given an estimated lower bound of x̂min = 2.35, ntail = 244 observation have been
used in order to implement the Monte Carlo goodness-of-fit test. The number of MC draws is k = 1, 000. The test p-value
is based on the null hypothesis that the data is consistent with power law behavior. Software implementation based on
Ginsburg (2012) and Alstott et al. (2014).

The method delivers an estimated scaling parameter of 2.45 (± 0.09). More importantly, the p-
value of the test equals 0.915, therefore the hypothesis that the data is consistent with a power
law distribution cannot be rejected. The empirical evidence does not contradict the claim that VC
returns are approximately Paretian. However, this does not tell us whether a power law fits the data
better than any other distribution. Indeed, there are several models that appear to behave like power
laws for some extent. For instance, despite its finite moments, the lognormal distribution is extremely
similar in shape to Paretian distributions (Mitzenmacher, 2003). The point of which distribution fits the
data better is important in order to understand the underlying economic mechanism that generates
the distribution, as different models yield very different predictions and implications.

Hence, following Clauset et al. (2009), we adopt the method of Vuong (1989) and we build a battery
of Likelihood Ratio (LR) tests to compare the power law fit with a series of competing distributions.
When the ratio is positive, the power law is favored against the competing model and vice versa.
Against this backdrop, a p-value < 0.1 indicates that the sign of the LR is statistically significant. One
of the main advantages of this approach is that it does not only state which of the two hypothesized
models is favored, but also when the data is not sufficient to discriminate between the distributions
(Clauset et al., 2009). Table C2 shows the tests’ outcome. The result of the tests indicates that we can
clearly favor the power law against the exponential distribution. At the same time, we cannot claim
that the power law has a better fit than the other competing models, inter alia the lognormal model
(see Figure 12). Indeed, as outlined in Virkar and Clauset (2014), a rule of thumb threshold for
the number of tail observations necessary to discern between Paretian and lognormal is ntail > 300.
Thus, larger samples are needed to produce further evidence on this.

Table C2: Log-Likelihood Ratio tests of power law behavior

Distribution Log-likelihood Ratio p-value Support for Power Law
Lognormal -0.815 0.541 Cannot say
Exponential 62.577 0.037 Yes
Truncated power law -0.320 0.424 Cannot say
Stretched Exponential -0.661 0.699 Cannot say
Positive Lognormal 1.774 0.674 Cannot say

Note: (Vuong, 1989) test based on 244 upper-tail observations. The first column shows the distribution to which the power
law is compared. Positive values of the LR test statistic indicate that the power law model is favored over the alternative.
Statistically significant p-values are indicated in bold. A large p signals that there is insufficient amount of data to discern
among the two alternatives.

50
D Additional evidence on exit outcomes

Figure D1: Geographic trends in the acquisitions of European EIF-backed VC investees

47%
56%
40 49%
42%

66%

30 55% 55% 52% 66%

37%
# Deals

53%
58%
75% 51%
20
44%
63%

64% 48%
45% 45%
34%
34%
10
36% 25%
50%
50%

0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15
Exit Year

European Buyer Non-European Buyer

Figure D2: Buyer Technological Focus, by macro-area

Same European Country Other European Country

20

15

10

0
# Deals

‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

US Other Non-European Country

20

15

10

0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

Non-Tech Tech

Note: based on 389 private acquisitions of early-stage VC investees, invested between 1996 and 2015 by
EIF-backed VC funds. “Tech” if listed on NASDAQ, “Non-Tech” otherwise. Source: BvD Zephyr (2016).

51
Figure D3: Buyer Sector Focus, by macro-area

Same European Country Other European Country

20

15

10

0
# Deals

‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

US Other Non-European Country

20

15

10

0
‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15 ‘03 ‘04 ‘05 ‘06 ‘07 ‘08 ‘09 ‘10 ‘11 ‘12 ‘13 ‘14 ‘15

ICT Life Sciences Manufacturing

Services Financial Services Greentech

Note: based on 389 private acquisitions of early-stage VC investees, invested between 1996 and 2015 by
EIF-backed VC funds. Source: BvD Zephyr (2016).

Table D1: Most frequent buyers of EIF-backed VC investees

Buyer Buyer Buyer Sector Nr Acq. EIF Acquired EIF Investees
Country Investees
GLAXOSMITHKLINE PLC GB Medical/Health 4 Domantis Ltd.;
Related Okairos; Glycovaxyn;
Cellzome AG
BROADCOM LTD. US Computer 4 Dune Networks;
Related Element 14 Inc.;
Alphamosaic Ltd;
Siliquent Technologies
Inc.
ALCATEL SA FR Communications 3 Right Vision; Open
Plug; Native Networks
EBAY INC. US Consumer 3 Brands4Friends; Shutl;
Related Skype Technologies
MICROSOFT CORPORATION US Computer 3 Sunrise Atelier Inc;
Related Screen Tonic;
6Wunderkinder
APPLE INC. US Computer 3 Semetric Ltd.; Polar
Related Rose; Acunu Ltd.

Note: Source: BvD Orbis (2016), BvD Zephyr (2016).

52
Table D2: IPO Stock Exchanges of EIF-backed VC investees
IPO Market Total Foreign MktCap At Revenues At Time To Most Frequent
List- List- IPO (EUR IPO (EUR IPO* Sector
ings ings m)* m)*
Euronext Paris 50 2 85.1 11.7 8.0 Computer Related
NASDAQ 29 7 280.5 4.0 7.6 Biotechnology
London Stock Exchange 23 1 172.1 15.4 5.1 Computer Related
(LSE)
Boerse Frankfurt 14 0 155.2 24.2 6.7 Biotechnology
Swiss Exchange (SWX) 7 1 1337.3 88.7 7.5 Biotechnology
Euronext Brussels 5 1 125.0 1.9 5.5 Biotechnology
NASDAQ OMX - Stock- 4 0 126.0 22.1 6.3 Communications
holm
OTC Bulletin Board 3 1 107.5 23.1 9.7 Medical Related
AIM (LSE) 2 0 24.1 8.2 12.3 Computer Related
NASDAQ OMX - Helsinki 2 0 n/a 2.8 2.9 Computer Related
New York Stock Exchange 2 1 190.9 25.6 7.8 Computer Related
(NYSE)
NASDAQ OMX - Copen- 2 0 112.0 0.4 3.6 Biotechnology
hagen
Borsa Italiana (MTA) 2 0 653.5 51.4 5.9 Communications
NASDAQ OMX - Oslo 1 0 n/a 0.0 10.2 Medical Related
Boerse Berlin 1 0 n/a 1.3 4.3 Biotechnology
Tel Aviv Stock Exchange 1 0 n/a 0.4 7.2 Medical Related
AktieTorget 1 0 7.6 n/a 7.5 Medical Related
Australian Securities Ex- 1 1 33.1 1.0 6.0 Computer Related
change
EASDAQ (Brussels) 1 0 n/a n/a 2.2 Electronics
Euronext Amsterdam 1 1 n/a 7.8 8.3 Biotechnology

Note: *average values. Based on 152 EIF-backed start-up IPOs. Source: BvD Zephyr (2016), Thomson Reuters Eikon
(2016).

53
E List of covariates in the competing risks model

Name Label Type Transformation Description

eif_macroreg Geographic Categorical none Set of dummies for each geographic
macro-region macro-region. The dummy switches
on when the investee’s headquarter is
located in that region.
eif_macrosec Macro-sector Categorical none Set of dummies for each macro-sector.
focus The dummy switches on when the
investee’s activity is classified in that
macro-sector.
vintage Vintage Year Continuous none Year of first investment by the VC fund
for the portfolio company.
fund_newteam New Team Indicator none Dummy equal to 1 if the VC fund is the
(dummy) first one raised by the VC firm.
nr_raisedfunds Funds raised by Continuous none Number of funds raised by the VC
VC firm firm. It is a proxy for the VC firm’s
experience.
fund_macroreg Fund Categorical none Set of dummies for each geographic
macro-region macro-region. The dummy switches
on when the VC firm’s operative
headquarter is located in that region.
fund_macrofocus Fund investment Categorical none Set of dummies for each investment
focus focus. The dummy switches on when
the VC fund invests with that focus
(e.g. technology transfer).
logdist Fund Distance Continuous log Log of geodetic distance between VC
firm and investee company.
logfirstinv First Investment Continuous log of real value Log of the first investment amount of
(EUR 2005) the fund into the company.
logfundsize Fund Size Continuous log of real value Log of the VC fund’s total amount
(EUR 2005) raised.
logportsize Portfolio Size Continuous log Log of the total number of investments
made by the VC fund.
unicorn Unicorn Indicator none Dummy equal to 1 if the investee
(dummy) company has reached valuation
greater or equal than EUR 1 bn while
staying private.
age_at_inv Age@investment Continuous none Age (in years) of the company when
invested by the VC fund.

Note: Source: BvD Orbis (2016), EIF internal data (2016).

54
F Robustness checks

To ensure the robustness of our results, we compare the estimates of the methodology in Fine and
Gray (1999) with results from the destination-specific hazard. Specifically, we seek to test whether
our results are sensitive to the continuous-time assumption of Fine and Gray (1999). To do so, we
revert to a discrete-time setting, following the approach outlined in Jenkins (2004). In particular,
we estimate a multinomial logit model, 58 assuming independent risks and proportional hazards.
Table F1 shows the estimation output.

The results are qualitatively the same, let alone for the coefficients of VC firm distance and portfolio
size, which become significant. In particular, under this specification, an increase in the geographic
distance between the fund and the portfolio company is linked to lower probabilities of both liquida-
tion and acquisition, suggesting a common finding in the venture capital literature: longer-distance
investments tend to be performed more conservatively than those with higher proximity (see, for in-
stance, Kraemer-Eis et al., 2016b).

An interesting result pertains to the size of the portfolio. In this second specification, it is linked to a
higher incidence of both write-offs and IPOs. While the obvious caveat relates to its non-significance
in the Fine and Gray (1999) model, it might otherwise indicate that larger portfolios have a greater
propensity to write-off — a finding consistent with faster VC human capital reallocation in large
portfolios (Fulghieri and Sevilir, 2009) — but also a higher chance of exiting through IPO. This last
result is non-trivial, because it promotes the idea that, in the presence of power law return distributions
(see section 4), expanding the portfolio size may increase the likelihood of reaching the outlier that
can potentially return the entire fund.

58
The approach is justified by the theoretical results discussed in Allison (1982).

55
Table F1: Multinomial Logit Competing Risks Model Estimates
empty Write-off Liquidation Acquisition IPO
Age at investment 0.994 0.978 1.007 0.948
(-0.28) (-1.16) (0.29) (-1.05)
First Investment 0.919* 0.823*** 1.032 1.509***
(-1.93) (-5.15) (0.56) (2.60)
Fund Distance 1.003 0.969** 0.964* 1.023
(0.18) (-2.01) (-1.81) (0.51)
Vintage Year 0.948*** 0.984 1.031** 0.879***
(-3.64) (-1.18) (1.98) (-2.90)
Unicorn status 0.000 2.388 15.358*** 8.997**
(-0.00) (1.19) (6.55) (2.11)
First-time VC teams 1.043 0.985 0.933 1.194
(0.36) (-0.12) (-0.45) (0.51)
Fund Size 0.927 1.119* 0.933 0.818
(-1.13) (1.76) (-0.82) (-0.98)
Portfolio Size 1.159* 1.079 0.985 2.198***
(1.77) (0.96) (-0.15) (2.84)
Funds raised by VC firm 0.817*** 1.017 1.095** 0.935
(-4.04) (0.45) (2.09) (-0.72)
Investee macro-sector (omitted: ICT)
Life Sciences 0.553*** 0.608*** 0.738** 3.109***
(-5.37) (-4.87) (-2.35) (4.97)
Manufacturing 1.233 0.971 1.158 0.000
(0.99) (-0.13) (0.53) (-0.01)
Services 0.924 1.211 1.585** 0.000
(-0.39) (1.16) (2.51) (-0.02)
Green Technologies 0.843 0.328* 1.217 0.000
(-0.40) (-1.90) (0.52) (-0.01)
Fund macro-region (omitted: DACH)
NORDICS 2.377** 2.034** 0.952 0.521
(2.46) (2.10) (-0.13) (-0.76)
FR&BENELUX 1.051 0.927 1.373 0.985
(0.22) (-0.41) (1.41) (-0.04)
SOUTH 1.246 1.373 1.353 0.142*
(0.67) (1.00) (0.75) (-1.71)
UK&IRELAND 0.974 0.925 2.073*** 2.478**
(-0.10) (-0.35) (2.95) (2.02)
CESEE 0.426 1.226 0.844 0.004
(-1.02) (0.32) (-0.22) (-0.00)
Observations 63971
Log-Likelihood -9100.96
LR Chi-Sq. 1175.45
Chi-Square (p-value) 0.000
Pseudo-R-squared 0.06
Mc-Fadden R-squared 0.06
Adj. Mc-Fadden R-squared 0.05
AIC 18450
Geographic macro-region Yes
Fund investment focus Yes
* p<0.10, ** p<0.05, *** p<0.01. Exponentiated coefficients; t statistics in parentheses
Note: based on 63,971 quarterly observations from 2,823 EIF-backed VC investments. Estimated coefficients reported
are exponentiated coefficients, i.e. relative risk ratios, which are equivalent to hazard ratios in a destination-specific hazard
framework for competing risks. A hazard ratio above (below) one means that the effect of increasing the covariate is to
increase (decrease) the probability of the exit outcome modelled. Multinomial logit estimation based on Jenkins (2004).

56
G Current EIF-supported Unicorns

The seminal contribution of Aileen (2013) on Techcrunch has introduced for the first time the concept
of Unicorn. In the corporate finance and VC jargon, a Unicorn is a privately held start-up company
with current valuation of USD 1 bn or more. Despite the 1 bn threshold being void of practical mean-
ing, the concept summarizes well the surging phenomenon of VC-backed technology companies
scaling up to unprecedented levels of revenues, innovation and employment creation. Although the
rise of Unicorns has been so far mainly a US fact, more and more Unicorns are emerging in Europe,
alongside the development of its tech start-ups ecosystem.

As of today, CBInsights tracks 186 Unicorn companies worldwide. Collectively, these companies are
worth approximately USD 650 bn and have raised about USD 125 bn. We have used CBInsights
data to identify the European Unicorns and, by matching them with EIF internal records, those upheld
by EIF. As Table G1 shows, out of 21 European Unicorns, almost 50% of them have been financed
by EIF-supported VC funds (marked in bold).

Table G1: List of the current European Unicorns

Company Valuation (USD bn) Country Industry
Spotify 8.53 Sweden Software & Services
Global Switch 6.02 United Kingdom Hardware
Delivery Hero 3.1 Germany eCommerce/Marketplace
Adyen 2.3 Netherlands Fintech
Klarna 2.25 Sweden Fintech
Hellofresh 2.09 Germany eCommerce/Marketplace
CureVac 1.65 Germany Healthcare
BlaBlaCar 1.6 France On-Demand
Oxford Nanopore Tech. 1.55 United Kingdom Healthcare
Farfetch 1.5 United Kingdom eCommerce/Marketplace
ironSource 1.5 Israel Software & Services
Auto1 Group 1.2 Germany eCommerce/Marketplace
Infinidat 1.2 Israel Hardware
Global Fashion Group 1.1 Luxembourg eCommerce/Marketplace
TransferWise 1.1 United Kingdom Fintech
OVH 1.1 France Big Data
Shazam 1 United Kingdom Software & Services
Funding Circle 1 United Kingdom Fintech
AVAST Software 1 Czech Republic Cybersecurity
MindMaze 1 Switzerland VR/AR
benevolent.ai 1 United Kingdom Healthcare

Note: a Unicorn is a privately held technology company worth 1 USD bn or more. EIF-supported Unicorns are outlined
in bold. Source: CBInsights (2017).

57
 About…

…the European Investment Fund

The European Investment Fund (EIF) is Europe’s leading risk finance provider for small and medium-
sized enterprises (SMEs) and midcaps, with a central mission to facilitate their access to finance. As
part of the European Investment Bank (EIB) Group, EIF designs, promotes and implements equity
and debt financial instruments which specifically target the needs of these market segments.

In this role, EIF fosters EU objectives in support of innovation, research and development, en-
trepreneurship, growth, and employment. EIF manages resources on behalf of the EIB, the European
Commission, national and regional authorities and other third parties. EIF support to enterprises is
provided through a wide range of selected financial intermediaries across Europe. Since its inception
in 1994, EIF has supported over 1.8 million SMEs.

EIF is a public-private partnership whose tripartite shareholding structure includes the EIB, the Eu-
ropean Union represented by the European Commission and various public and private financial
institutions from European Union Member States and Turkey. For further information, please visit
www.eif.org.

…EIF’s Research & Market Analysis

Research & Market Analysis (RMA) supports EIF’s strategic decision-making, product development
and mandate management processes through applied research and market analyses. RMA works
as internal advisor, participates in international fora and maintains liaison with many organisations
and institutions.

…EIF Working Papers

The EIF Working Papers are designed to make available to a wider readership selected topics and
studies in relation to EIF’s business. The Working Papers are edited by EIF’s Research & Market
Analysis and are typically authored or co-authored by EIF staff, or written in cooperation with EIF.
The Working Papers are usually available only in English and typically distributed in electronic form
(pdf).

58
EIF Working Papers

2009/001 Microfinance in Europe – A market overview.

November 2009.
2009/002 Financing Technology Transfer.
December 2009.

2010/003 Private Equity Market in Europe – Rise of a new cycle or tail of the recession?
February 2010.
2010/004 Private Equity and Venture Capital Indicators – A research of EU27 Private Equity and
Venture Capital Markets. April 2010.
2010/005 Private Equity Market Outlook.
May 2010.
2010/006 Drivers of Private Equity Investment activity. Are Buyout and Venture investors really so
different? August 2010
2010/007 SME Loan Securitisation – an important tool to support European SME lending.
October 2010.
2010/008 Impact of Legislation on Credit Risk – How different are the U.K. and Germany?
November 2010.

2011/009 The performance and prospects of European Venture Capital.

May 2011.
2011/010 European Small Business Finance Outlook.
June 2011.
2011/011 Business Angels in Germany. EIF’s initiative to support the non-institutional
financing market. November 2011.
2011/012 European Small Business Finance Outlook 2/2011.
December 2011.
2012/013 Progress for microfinance in Europe.
January 2012.
2012/014 European Small Business Finance Outlook.
May 2012.
2012/015 The importance of leasing for SME finance.
August 2012.
2012/016 European Small Business Finance Outlook.
December 2012.

2013/017 Forecasting distress in European SME portfolios.

May 2013.
2013/018 European Small Business Finance Outlook.
June 2013.

59
2013/019 SME loan securitisation 2.0 – Market assessment and policy options.
October 2013.
2013/020 European Small Business Finance Outlook.
December 2013.

2014/021 Financing the mobility of students in European higher education.

January 2014.
2014/022 Guidelines for SME Access to Finance Market Assessments.
April 2014.
2014/023 Pricing Default Risk: the Good, the Bad, and the Anomaly.
June 2014.
2014/024 European Small Business Finance Outlook.
June 2014.
2014/025 Institutional non-bank lending and the role of debt funds.
October 2014.
2014/026 European Small Business Finance Outlook.
December 2014.

2015/027 Bridging the university funding gap: determinants and consequences of

university seed funds and proof-of-concept Programs in Europe. May 2015.
2015/028 European Small Business Finance Outlook.
June 2015.
2015/029 The Economic Impact of EU Guarantees on Credit to SMEs - Evidence from CESEE
Countries. July 2015.
2015/030 Financing patterns of European SMEs: An Empirical Taxonomy
November 2015.
2015/031 SME Securitisation – at a crossroads?
December 2015.
2015/032 European Small Business Finance Outlook.
December 2015.

2016/033 Evaluating the impact of European microfinance. The foundations.

January 2016.
2016/034 The European Venture Capital Landscape: an EIF perspective.
Volume I: the impact of EIF on the VC ecosystem. May 2016.
2016/035 European Small Business Finance Outlook.
June 2016.
2016/036 The role of cooperative banks and smaller institutions for the financing
of SMEs and small midcaps in Europe. July 2016.
2016/037 European Small Business Finance Outlook.
December 2016.
2016/038 The European Venture Capital Landscape: an EIF perspective.
Volume II: Growth patterns of EIF-backed startups. December 2016.

60
2017/039 Guaranteeing Social Enterprises - The EaSI way.
February 2017.
2017/040 Financing Patterns of European SMEs Revisited: An Updated Empirical Taxonomy
and Determinants of SME Financing Clusters. March 2017.
2017/041 The European Venture Capital Landscape: an EIF perspective.
Volume III: Liquidity events and returns of EIF-backed VC investments. April 2017.

61