for
Masters in Finance: Corporate Finance
1st July 2018
University of Amsterdam, Amsterdam Business School
Student Name: Jola Danaj
Student Number: 11866608
Student Email: j.danaj@live.com
Thesis Supervisor: Tolga Caskurlu
Program Faculty: Business and Economics
Statement of originality
This document is written by Student Jola Danaj who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no
sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of
Aknowledgements
I would first like to thank my thesis advisor, Dr. Tolga Caskurlu of the Amsterdam Business School at the University of Amsterdam for his guidance and expertise throughout the making of this research. I further, want to express my profound gratitude to my parents Rezarta Danaj and Dhimitraq Danaj and my grandparents Olimbi and Viktor Rrapushi for their everlasting unconditional support: even from miles away, they are true role models. A warm thank you goes to Bryn Watkins, not only for his love and continuous encouragement, but also for his economist expertise throughout this study year. Finally, I want to thank my good friend Coen Binnerts, whose valuable comments and library coffee breaks were very much appreciated throughout this research. This accomplishment would not have been possible without these people. Thank you.
ABSTRACT
In this research paper I study the relationship between technological innovation and Corporate Venture Capitals. In more detail, this paper focuses on the investment intensity of tech-centric CVCs on new ventures as well as the probability of success of the incumbent firms during periods of high technological growth. To test this, I use a difference-in-differences regression model, a Tobit regression model and also a combination of both. To correct for the implied endogeneity in the model, the regression estimation incorporates two different exogenous shocks as identifiers: The DotCom bubble burst and the launch of Amazon Web Services. It is found that during periods of high technological growth, incumbent are less likely to receive funding from tech-centric CVCs, after the launch of AWS and the DotCom bubble burst. Second, it is found that there is a lower probability that an incumbent firm will receive a follow up funding from tech-centric CVCs after the launch of Amazon Web Services. To conclude, this research proposes that the DotCom bubble and Amazon Web Services can be used as relevant exogenous shock, to identify endogeneity when studying CVCs and technological innovation. Further, it is also proposed to use the CVC follow up investment as a relevant variable proxy for the success of firms’ incumbent to a CVC fund in future research relevant to CVC investment.
Key words: Corporate Venture Capital, Technological innovation, DotCom Bubble, Amazon Web Services
Table of Contents
1. INTRODUCTION ... 6
2. LITERATURE REVIEW AND RESEARCH BACKGROUND ... 9
a. Corporate Venture Capital ... 10
b. The DotCom bubble ... 14
c. Amazon Web Services – “a breakthrough innovation” ... 15
d. Hypotheses... 17
3. SAMPLE AND DATA ... 18
a. Data sample ... 18
b. Variables ... 19
i. Dependent Variable ... 19
ii. Independent Variable ... 20
iii. DID dummies ... 20
c. Data Analysis ... 21
4. ENDOGENEITY AND ROBUSTNESS ... 21
5. METHODOLOGY ... 22
a. Empirical Methodology ... 22
i. Difference-in-differences ... 22
ii. Tobit ... 24
b. Estimation Model ... 24
6. RESULTS AND ANALYSIS ... 25
a. Results ... 25
b. Analysis ... 27
i. First Hypothesis... 27
ii. Second Hypothesis ... 29
7. DISCUSSION ... 31
8. CONCLUSION ... 32
9. TABLES AND GRAPHS ... 34
10. APPENDIX ... 40
INTRODUCTION
A recurring theme that has recently resurfaced in Corporate Finance is the effects and implications of technology and innovation on the investment framework. Corporate Venture Capitals (from here on referred to as “CVC”, “CVCs”), are a type of an investment vehicle derived from Venture Capital ideologies. CVCs are an “old novelty”: although they have been around since the early 60s, this type of investment vehicle has not been in the spotlight until recent developments in the tech field. The first corporate venture fund was founded in the US around the 60s and proved to be very successful. This was followed by several CVC waves, starting from the early 1970s, where more than 25 percent of the Fortune 500 firms attempted corporate venture programs (Gomper and Lerner, 2000). As such novice in the field, CVCs are also highly volatile. However, although pioneering studies have found that CVCs are considered to be volatile, they reap just as much, if not more returns and benefits as Venture Capitals do, ceteris paribus. Past research supports the theory that even when start-ups incumbent to a CVC portfolio historically do not show signs of success1 they still appear to
bring value to the parent CVC (Gompers and Lerner, 2000; Ma 2016). Although CVC have come to the scene relatively late, their innovative nature has led to high number of deals every year, with the highest number of active CVCs in the second quarter of 2017 (CB Insights, 2018). Over the last few decades, CVCs have become increasingly important. According to CB Insights (2018), “186 new corporate VC units globally made their first investment in 2017.” This number represents a 66% increase in CVCs from 2016. The existing literature has already touched upon the determinants of success for a CVC, and therefore in order to offer a new perspective on CVCs, I plan to use industry shocks and technological shocks as a factor of reference to determine their relationship with technological advancement as well as estimate the probability of their success in the face of such occurrences.
Part of this research will focus on studying the investment intensity and success of these firms, comparably to the Gompers and Lerner (2000) study but whilst incorporating data on CVCs for a longer period of time (after year 2000). The differing effect of these estimations when tech-centric CVCs are exposed to different shocks will be noted. Furthermore, investigations will be carried out to determine whether technological shocks make a difference on the likelihood of success that an incumbent firm sees in times of high technological innovation.
1 An example of these types of exist are an Initial Public Offering type-exit or an outside acquisition (Gompers and Lerner, 2000)
This study builds on from the Gompers and Lerner (2000) by addressing one of their suggestions on further research. One of the questions that they highlight from the implications of their findings is that although there are a lot of companies that are still conducting a large amount of investments even when there is no clear strategic reasoning behind it, what is this prompted from? Their study was conducted in the late nineties, shortly before the DotCom bubble burst (also known as the Internet bubble), hence I believe that technology, along with investor behaviour might be a factor of interest. I want to estimate whether the technological advancement, while controlled for exogenous shocks, has an impact on the investment decision making nature of a CVC. To test whether technology and innovation indeed have an effect on the amount of CVC investment I initially test the following research question throughout my study:
Does technological innovation lead to higher overall CVC investment?
I further want to test the implications of technological innovation on the success of these incumbent firms2. Which leads me to the second research question that I will be addressing in
this study:
Does technological innovation lead to higher success for the CVC investments?
A possible weakness of this research is the endogeneity problem of technology and investment that arises within the model. I address this issue of identifying a more correct estimate of the nature of such investment growth by identifying two exogenous shocks through a randomized experiment: the Dotcom bubble and the launch of Amazon Web Services. This will be performed with a difference-in-differences model and later the robustness will be checked with a Tobit and difference-in-differences combination statistical model.
To estimate these hypotheses, I will initially look at the linear relationship between CVC investment and technological advancement, through an Ordinary Least Squares and a Two Stage Least Squares, for which I expect a positive correlation as technological advancement is always a positively linear variable.
The results of this thesis can be divided into two parts. In a difference-in-differences analysis, the relationship of technological advancement and CVC investment intensity is tested while using the tech-centric CVCs as a treatment. I use two time dummies, namely the DotCom
2 From here on after, since in its essence, CVC investment is the investment in the fund’s portfolio companies, I use the term “incumbent firms” interchangeably with “CVC investment”
Bubble crash and the Amazon Web Services launch as the time treatment. For my difference-in-differences analysis I use two dependent variables to count for CVC investment, Round Number and Equity Investment - for which Equity Investment indicates a much higher statistical and economical relevance. This analysis shows more statistically significant results when using the DotCom crash as the time dummy where, after the crash, in the presence of technological innovation incumbents of tech-centric CVCs, are shown to receive less equity investment funding.
Second, I use a combination of a Tobit regression with a difference-in-differences analysis to assess the probability of success that the firms incumbent to tech-centric CVCs face in the presence of technological advancement. Similarly, to the first part of my analysis, for the difference-in-differences analysis I use both the DotCom Bubble crash and the Amazon Web Services launch as the time dummies and treat them for tech-centric CVCs. This analysis found that for incumbent firms that receive funding from tech-centric CVCs after the launch of AWS and the DotCom crash, the probability of success is lower when technological innovation is higher. However, although this estimation was statistically significant when using AWS as time dummies, it showed very low economical relevance.
I expect to see a positive relationship between technological growth and investment in CVC, however when controlled for the DotCom bubble I expect that tech-centric firms CVCs would be more reluctant to invest in companies due to a fall in confidence. My estimation results confirmed this expectation. As for when I use the AWS as a time dummy, I expected to see an increase in investment from tech-centric CVCs. The results to my estimation show the opposite reaction from these measures. I had the same expectations concerning the success of CVC firms: I expected a lower probability of success for tech-centric CVCs investments after the DotCom bubble and a higher probability of success for tech-centric CVCs investments after the launch of Amazon Web Services. Similar to the first hypothesis, my results confirm the relationship sign for my theory concerning the DotCom bubble, but not that of Amazon Web Services. I expect that this occurrence is due to an increase of the competition in the market, which pushes tech-centric CVCs to invest in smaller increments on a higher number of companies, thus seeming to jump ship a lot.
So far as is known, this topic has not had any research dedicated to it yet. Implications of this research could lead to more insight on the investment intensity and dependency on technology on the area of investment focus concerning CVCs. I aim to find results that show if these shocks
push for more investment in CVC as well as see if they have an impact on the success of the portfolio companies. I include data from 1990 until 2015. Furthermore I am proposing two new identifiers to correct for the endogeneity in the framework of studying the implications of technology on investment: the DotCom Bubble crash and the launch of Amazon Web Services that can prevail as exogenous shocks for empirical studies. It is also proposed to use the follow up in CVC investment as a proxy variable for the success of CVC investments. Lastly, this thesis provides an extension to the literature on corporate venture capital investments and its relationship with technology.
The paper is structured as follows. After examining the appropriate literature on CVCs and the main theories around this innovation in LITERATURE REVIEW AND RESEARCH BACKGROUND, I will also focus on previous studies and past theories of investment intensity for such deals as well as observe their implications towards this study. DATA SAMPLE AND SELECTION section describes where the data is retrieved from and how it is constructed, as well as address my primary measures which will serve as dependent and independent variables in my model. In METHODOLOGY, I will explore the research method and the empirical model that I use to get my estimates. Here I will also explore identification strategies to control for endogeneity and sample selection issues. In ROBUSTNESS CHECKS, I will address the relevant robustness checks I used for my model. RESULTS presents and analyses the regression outputs and the relevant coefficients of interest. DISCUSSION discusses the results and compares my findings to that of existing literature, as well as presents an outlook based on that outcome, and CONCLUSION concludes my research and suggests new ideas for research.
LITERATURE REVIEW AND RESEARCH BACKGROUND
In this segment, the relevant literature will be reviewed throughout several sections. First, I will present the existing background for Corporate Venture Capital and lay out its definitions as they are widely accepted by the general literature. Then I will discuss the recurring theories that scholars have studied on CVCs. Further, I will then review and compare the relevance of the Gompers and Lerner (2000) study with my research as well as tackle the issue of the underlying endogeneity in my model. Following this theory I then present an identification strategy through the Internet Bubble and the launch of Amazon Web Services as well as expand on the reasons and theory behind using these shocks and elaborate on their respective effects on the economy and consequently my empirical model. Lastly the segment will present the hypothesis relevant to this research.
Corporate Venture Capital
Corporate Venture Capital is a type of investment vehicle in which a corporation creates a fund (or several funds) in order to directly invest in external start-up companies (Chesbrough, 2002). This usually takes the form of a minority equity investment (Gompers and Lerner, 2000). In his definition of a CVC, Chesbrough (2002) excludes investments made through an external fund, if they are ultimately managed by a third party, even if they are prompted by a single parent corporation. So in general, a parent corporation launches a CVC fund which is then in charge of managing investments and deciding which external start-up should receive funding and capital. Another way to distinguish CVCs from other forms of Independent Venture Capitals is by looking at the underlying dimensions with which a CVC operates with. A CVCs investment objectives can be categorized as twofold: (1) strategic centric investments, which are primarily pursued to enhance the parent corporation’s profits and exploiting potential vertical and horizontal synergies between them, and (2) financially centric investments, where the CVC is primarily seeking to make a return on the investment, similar to or higher than a normal VC would (Chesbrough, 2002).
Although the concept of Corporate Venture Capital stems from Venture Capitals (“VC”), one of the main differences between a regular Venture Capital, or an Independent Venture Capital (“IVC”) and a Corporate Venture Capital is that VCs are interested in a liquidity event which means that they have an underlying interest to push for an IPO or other forms of exit and are more concerned with financial returns, whereas corporate investors in this industry are primarily interested in the underlying technology and its future value (Ceccagnoli, Higgins & Kang, 2017). CVCs or investment arms, provide the founders of the incumbent firms with knowledge, know-how, connections to the industry and a more secure exit timetable, thereby creating a better symbiotic relationship between the two. These parent companies gain in return a “window” on new technologies as well as the ability to recognize and seize market trends in advance.
In his study, Keil (2000) organizes the corporate venture framework as indicated in Figure 1, from around seven case studies within the tech industry. He initially differentiates between the internal and external corporate venturing; the latter is where the CVC is placed as well, and marks out three categories based on the risk that an incumbent firm bears on the market. In another study, Sharma and Chrisman (1999) describe the former as corporate venturing that results in the creation of organizational entities that reside within the organizational boundaries and the latter as the creation of external, (semi-) independent organizational entities.
Figure 1. (Keil 2000)
To summarize, while the sole objective of a VC revolves around financial return, CVCs have more profound strategic objectives. This can take several forms such as leveraging externa l sources of innovation, bringing new ideas and technologies into the company, or taking an option on a future technology by creating a wide venture portfolio or allowing the incumbent firm to acquaint itself with different future business directions that the firm cannot pursue by itself (MacMillan et al., 2008). Although new ventures might be able to help the incumbent firm with innovation, in reality there are often problems with acquiring these firms, let alone implementing them successfully into the overall corporate strategy.
Throughout this research I refrain from consulting literature that applies exclusively to CVC. This was because CVC is a relatively new subject and therefore there is not an extensive amount of literature that studies the investment behaviour of such ventures although several studies on such ventures have already addressed the influence and several functions of the CVC as well as the value creation of these investment bodies (Dushnitsky, 2006; Maula, 2007; Chemmanur and Fulghieri, 2014).Thus I also refer to Independent/ Venture Capital (IVC/VC) literature. I do this because there are several similarities between CVCs and IVC/VC as previous literature and history has shown. For instance, just like IVCs, CVCs primarily provide capital to early-stage, high-potential and growth companies expecting a return through an exit opportunity (i.e a public offering or a trade sale). These venture capital investments are normally made as cash in exchange for shares in the invested company and are considered a type of private equity capital. The major difference lies however, in the symbiotic relationship
between a CVC and the incumbent firms: the CVC will additionally gain a window on new technologies and diverse channels in new markets, whereas the incumbent firm will additionally have access to complementary companies that may become strategic partners in the future.
This raises another point in understanding an additional occurrence in the CVC framework. One of the most defining characteristics of a CVC is the connecting degree of the operational capabilities of the fund with that of the incumbent firm (Chesbrough, 2002). Furthermore in his article, Chesbrough (2002) formulates a simple framework for mapping the investment types of a CVC to have a better understanding of the incentives of a CVC. His research aligns with previous studies from Gompers and Lerner (1998), which state that a higher strategic fit and operational alignment lead to higher chances of success.
As mentioned before, my study is heavily inspired by the Gompers and Lerner (2000) paper on the determinants of success for a CVC which indirectly prompted the hypotheses of this research. Although their study concludes that investments within the same strategic fit perform better, they notice an increase in investments of a CVC even when strategic fit was not present during the late ‘90s. CVCs are characterized by a particular type of uncertainty which is embedded in the business structure as the model is based on the transfer of innovation. In practice, many problems tend to arise when such companies are acquired, especially when one needs to implement such strategies corporate wide.
Previous research has shown that investments by incumbent firms that are active in the same industry as the new ventures are able to add more value (Coles, Hertzel and Santhanakrishnan, 2002) but Gompers et al (2000) mention that although the strategic fit is what insinuates more profit in the future, there were plenty of investments made when there were no signs of it. Other studies have also noted that during periods of growth investors tend to ignore underlying issues, hence falling prey to overinvestment behaviour during revolutionary economic moments (Pastor and Veronesi, 2009; Qiang and Yong, 2009). For this reason, financial markets, as they are heavily influenced by the new developments in technology, display an irrational behaviour during times of high growth, which is characterized by unjustified optimism.
In its simplest form, the framework setup for the corporate venture capital can be broken down into a supply-demand curve. The demand for corporate capital from these ventures is determined by the number of firms in the market that are seeking capital. Consequently, this supply and demand is not fixed in real life scenarios (Lerner, 2001). Lerner further argues that
major market changes and technological breakthroughs (such as genetic engineering) lead to an increase in Venture Capital investment. As much as the adjustment process is slow and uneven, when the required amount of capital is reciprocated and reacts to the demand, chances are they will overinvest. An important issue that Lerner (2001) addresses in this article is the difference between the supply-demand curve of Venture Capitals versus those of other types of investment. He states that unlike in a general business model, the long run and short run effect of a positive shock such as the emergence of the Internet, for any given amount of investors and their demands on investment return for a VC, is that there is an abundance of investment candidates.
The first corporate venture funds began in the mid-1960s as a response to the successes of independent venture capital funds. Gompers and Lerner (1998) argue that corporate venture capital since then has mirrored the cyclical nature of the independent venture capital industry over the last thirty years in their extensive venture capital summary. Throughout their literature summary on the corporate venture capital they indicate three different waves of corporate venture capital. Each of these waves was triggered by a different mixture of diversification, strategic and financial goals. As previously stated, corporate venture capital initially began in the late 1960s. During these years, the main objective for these ventures was to gain knowledge and new sources of opportunity on the latest technologies by establishing CVC programs. These CVC programs however, had a short life span of only four years, where typically a CVC would be dissolved as it completed its goals3. This wave of CVC investment saw its decline
with the collapse of the 1973 IPO market. However, the relaxed regulations in the 1980s4 made
a path for a second wave of CVC investments. These investments reached a peak in 1986, where they mirrored the success of IVCs, to end up holding circa 13 percent of all venture capital in the US. This second wave was characterized by the need of firms to diversify their operations. Similar to the first wave, this second wave suffered the same consequences as the first when the stock market crash of 1987 significantly lowered the probability of a successful exit opportunity yet again. From a staggering hold of 13 percent of all venture capital in the US in 1986 CVCs were holding only 5 percent of the total venture capital pool by 1992. The third wave of CVC investments began in the mid-1990s and was significantly bigger than the previous two. Maula (2001) argues that this wave was primarily based on a financial and
3 Furthermore, many of these programs were also halted due to the collapse of the market in 1973 (Gomper and Lerner, 1998) and consequently lower chances of successful exit opportunities.
strategic rationale. This time, CVCs were initiated in order to scan the market for unexploited opportunities and obtain a know-how abuot the new technological age. By the time the Internet had become a world-wide phenomenon many of the well-established firms were facing technological discontinuities in their markets as well as difficulty keeping up with the new competition. This third wave of CVC investments peaked in 2000 and plummeted yet again when the DotCom bubble burst. A while after the bubble burst, CVC investments remained between five and ten percent of the total venture capital in the US (Landström, 2007), exceeding the levels before 1999.
On that account, for the last few decades, history follow with two quintessential events in technological development: The DotCom bubble of the early century and the launch of Amazon Web services in 2006, which I later use as exogenous shocks in this study.
The DotCom bubble
During the late 90s, starting from 1998 and until February 2000, there was an over 1000 percent return on the Internet stocks for the U.S equity market, which subsequently evaporated completely by the end of 2000. As it has been widely accepted by previous studies and literature, a bubble forms from when an amount of investors, incumbent to the market, become strongly optimistic thus pushing the prices up higher than the intrinsic value, whereas the counterpart pessimistic investors, who usually bring the markets to reasonable prices and level them towards the intrinsic value, are not able to do so due to short sale constraints.
Ofek and Richardson (2002) argue that the logic behind such unreasonably high levels of Internet stock prices that weren’t able to move back down lies in the higher short-term interest for Internet stock prices, higher borrowing costs and greater violations of put-call parity for these specific types of firms. They note that DotCom stocks appeared to have had very high short-sale restrictions due to very stringent forms of short-sale constraints such as lock up agreements, which were most prevalent for Internet firms. On the other hand, Brunnermeier and Nagel (2004) argue that short sale constraints are not sufficient in explaining the lack of speculation that could have deflated the bubble. They study the response of hedge fund investments, which are considered to be a very sophisticated investor, faced with the Internet bubble and confirm that some models may shed a different light on the reality of these high Internet firm stock prices. Because some investors might profit by riding the bubble, that means that they are less incentivized in working towards stabilizing it. Their findings further reason
that because these hedge funds reduced their holdings right before the bubble collapsed, not only was this bubble predictable but also rather exploitable.
In standard economic theory it is stated that the trajectory of stock market prices is a good reflection of the continuous amounts of information input and news that are interpreted by analysts and traders (Cutler et al., 1989). This leads to a school of thought that large shocks in the market, such as a stock market bubble bursting, are a direct or indirect result of potential surprises and that unlike investor demand patterns, the burst is not driven by drastic shifts in the supply and demand of the market (Griffin, Harri, Shou and Topaloglu, 2018). This implies that there is reasoning behind assigning exogeneity to the market due to the DotCom burst.
There are however, opponents to this approach. Although the burst of a bubble can be seen as an exogenous occurrence due to the bad news that creates strong bursts of volatility (Sornette, Malevergne and Muzy, 2018), what exogenous shocks are mainly equated with are unexpected occurrences like political unrest or terrorist attacks. Several past studies have implied that the burst of a bubble may be the ultimate outcome of an endogenous instability that is channelled through rational or irrational herd behaviour of the incumbent agents (Orlean, 1995; Shiller, 2000). For the aforementioned reasons, the DotCom bubble burst which rose throughout 1999 until 2000 will be used as an exogenous shock. The cut-off date will be March 31st 2000, as it
indicates the highest bubble peak before the crash (Townsend, 2015; Ofek and Richardson, 2003). This method will factor in the categorization of the companies which are heavily invested in Internet centric companies. A way to look at this, is by studying the overreaction of the venture capital firm’s investment levels during the emergence of the Internet. As shown in Graph 1 in the Appendix, the years leading up to the Internet bubble have seen the highest number of deals from Corporate Venture Capitals.
Nonetheless, because there appears to be a certain ambiguity about the literature on the endogeneity of the bubble crashes, another identifier will be taken to create my difference-in-differences time dummy (namely the Amazon Web Services launch in 2006) to correct for the endogeneity in my model, which I will expand on in the following paragraph.
Amazon Web Services – “a breakthrough innovation”
Amazon Web Services (hereby referred to also as “AWS”) is a form of cloud computing which provides a physical and digital infrastructure and tools with which others can build yet more and more complex platforms. AWS was internationally launched in 2006 and free for public wide use. AWS marks an important moment in technological history as it not only made such
type of cloud computing platforms available for widespread use, but it opened doors for radical new forms of economies, such as the platform economy.
“Cloud computing on its own, enables organizations to obtain a flexible, secure, and cost-effective IT infrastructure, in much the same way that national electric grids enable homes and organizations to plug into a centrally managed, efficient, and cost-effective energy source.”5.
In their introductory article, Amazon further mentions that “…cloud computing liberates organizations from devoting precious people and budget to activities that don’t directly contribute to the bottom line while still obtaining IT infrastructure capabilities. These capabilities include compute power, storage, databases, messaging, and other building block services that run business applications. When coupled with a utility-style pricing and business model, cloud computing promises to deliver an enterprise-grade IT infrastructure in a reliable, timely, and cost-effective manner.”, thus in other words giving free access to their platform for use to anyone who desires to create IT systems, white label programming, applications and more6. AWS allows the requisition of computing power, storage and several services. There is
no up-front payment or long-term commitment so the client only pays what they use, thereby offering an efficient and low-cost way of computing. As of 2015 and 2016 AWS took approximately 67 percent and 74 percent of total operating income of Amazon. As of today, Amazon runs a global web platform which serves millions of customers, creators, programmers, and manages billions of dollars’ worth of commerce every year. On the same theory, there are some studies in economics that suggest that the results of technological shocks, even negative ones, may describe higher growths and higher potential in the long run (Hallegatte, 2006).
The uniqueness of AWS lies in how they packaged all these software development tools and made them easy to use in one simple package. Firstly, Amazon Web Services unified APIs: through this method, a programmer only needs one account and (in many cases) one set of credentials to use any of the AWS APIs. Second, AWS made it possible to have a form of unified billing. This way for example, a client doesn’t need to pay the storage provider, the hosting company separately as AWS makes it possible for everything to be sent to one account. Third, AWS simplified scaling. As many large-scale developers know you can’t just scale a website by throwing more hardware at it. They further advanced dashboard management by
5 Amazon Web Services, An Introduction, 2008
making it much easier for clients to monitor their own services without the need for a custom dashboard development or integration. AWS also developed the local access by lowering the overhead costs of bandwidth and its other associated costs7. Lastly AWS offers better pricing
on small scale usage8. While there is a lot of reuse of existing technologies, at the time of the
launch Amazon understood the needs of the market and acted early by building out its cloud platform and thus matching it with an equal amount of innovation and execution. In short, Amazon Web Services fundamentally changed the way programming and development was done. Due to its innovative character, AWS made the high end technologies that were that far only accessible to high-end firms available for wide use.
Hypotheses
As stated above there is a cyclicality to the overall investment market during times of high innovation which remains unexplained by investment incentives, financial objectives or strategic fit. Other literature has concluded that times of high investment in a certain sector and technological advancement, give rise to a high optimistic investor behaviour or are a direct manifestation of herd behaviour. As previous literature has also established (Gompers and Lerner, 2000), CVC investment is neither exempt nor isolated from the markets that suffer from this type of discrepancy in interest alignment. Due to a large increase in the number of Internet deals before the crash of 2000, I expect that technological innovation is one of the drivers of CVC investment, which leads me to my first hypothesis:
H1: Technological innovation positively affects CVC investment
Furthermore, I want to investigate whether these investments lead to successful ventures when they are prompted by periods of high technological innovation. Although corporations and established firms might be highly skilled at selecting and steering CVC incumbents and may be better positioned to provide complementary capabilities (Gompers and Lerner, 1998; Dushnitsky and Lenox, 2005), there are reasons to doubt the viability of these investment vehicles. First, Birkinshaw et al. (2002) shed light on the internal conflict that CVCs face in which corporate parents of a CVC do not compensate their fund managers as a typical VC, so the largest part of the salary is the fixed compensation. This might lead to potentially lower
7 An atypical caveat here is that Amazon has multiple data centers which are across multiple regions. But even though the access may not actually be local, it all resides within Amazon’s network.
8 Before the launch of AWS, an enterprise web server cost more than USD 1 thousand and a database server more than USD 3 thousand and you would have to pay high amounts of up-front costs just to run a simple website. Moreover you couldn’t recover any costs if your systems were under-loaded. A T2 medium instance costs less than $500 a year. You pay as you go and there’s no cost if you need to replace it.
incentives and high management turnover, which affects the performance of the fund (Block and Ornati, 1987). Second, CVC investments are proven to fall prey to a high degree of information asymmetry (Block and Ornati, 1987). Even for a usual VC, the amount of public information is limited due to hesitation to share their innovative ideas, which leads to an increase in the amounts of adverse-selection, where low quality ventures have an opportunity to disguise themselves as better investments. Due to the nature of its investment types, this predicament of adverse selection is even more heightened for a CVC. (Dushnitsky, 2004); Gans and Stern, 2003; Dushnitsky and Lenox, 2005). Building on the above mentioned conflicting schools of thought on the viability of CVCs as an investment vehicle, I want to test and see whether technological innovation leads to a higher amount of these investments (which are made during periods of high technological innovation) and if that in turn leads their incumbent firms to success. This brings me to my second hypothesis:
H2: Investment in CVC during periods of technological innovation leads to higher CVC success
To conclude the literature review, this study will examine the influence that technological innovation has on CVC investments, and expand that to see whether this affects the strategic fit within the fund and the portfolio companies. This research attempts to identify technological innovation as an important driver in decision making for CVC funds, thus overriding the strategic goals of said fund. Because the DotCom bubble crash was what brought the third wave of CVCs to a halt, and the AWS was what encouraged a new type of technological movement, I believe that these exogenous shocks, are events that can easily isolate tech-centric firms, such as Internet specific companies, Internet consumer products etc.
SAMPLE AND DATA
This section will first present the sources of data and selection criteria used. Second, I will elaborate on the primary dependent and independent variable as well as the rest of the control variables used in the model. Lastly, I will analyse the summary of the data and descriptive statistics.
Data sample
To explore the relationship between technological innovation and corporate venture capital investment I constructed a panel dataset of U.S CVC firms and incumbent companies to the funds throughout 1990-2015. The existing literature provides multiple methods for working
with CVC research. The consensus from previous studies (Gompers and Lerner, 2000) lies in using the VentureXpert database from Thomson One, as I do in this research.
To compile a sample for my research I use the data from VentureXpert Thomson One (former SDC Platinum database), USPTO database on monthly patents, which is a patent assignment dataset and WRDS Compustat database for stock returns and US bond yields. For my first specification I will use exclusively a hand matched sample data taken from Thomson One and USPTO. For the data retrieved from Thomson One, I used the equity deals filtered per CVC funds and located in the United States region. The time series spans dates from 1/1/1990 to 1/1/2015 with details on every investment and investment round. I retrieved patent data from the USPTO database, for which there is available data only until 2014. I then eliminate all the missing data. After filtering the data from both databases, the following sample was constructed and presented in Table 1. Table 1 indicates summary statistics for the main variables used forward on this thesis for the first hypothesis. Sample size is similar for all of the variables and all available data is winsorized accordingly at the 1% level.
Due to the nature of the data that I have, it is necessary to make sure that I correct for unobserved heterogeneity across industries and time trends for macroeconomic occurrences. Hence why for this estimation I will use entity fixed effects which apply to the different types of CVC as well as time fixed effects to capture the influence of aggregate (time‐series) trends. Because my data involves every CVC investment made from 1990 until 2015, in many cases this time series variable will be spuriously related simply because of the rising magnitude of aggregate variables (because of inflation, economic growth, population growth, etc.).
Variables
Dependent Variable
The primary dependent variable for this study is the CVC investment. In this model, CVC investment and investment growth is proxied by two different measures, all of which are continuous and I will expand on the next paragraphs. To check for robustness, I will use several methods to calculate for CVC investment. For my first hypothesis the first dependent measure I use is the equity investment on CVC to show an example of the relationship between the amount invested on each incumbent company by the respective CVC and technological innovation. To check for robustness, I further use the round number as the dependent variable. This is the number of rounds from which a company has received funding from a CVC. I use
this measure because I want to see whether there is a difference implied by the presence of technological innovation in a firm’s investment density.
To test for the second hypothesis, I use the success of a CVC as the primary dependent variable. In their study, Gompers and Lerner (2000) use the probability that the firm goes public as a proxy for success for a CVC is the exit opportunities. However due to a lack of data on this type of exit opportunity, I choose to construct a dummy variable which takes the value of 1 if an incumbent firm has received a follow up in investment from the CVC and 0 if otherwise. The number of investment rounds is a relevant unit of observation as it can signal difficulties and poor operational performances thus making it difficult for the incumbent company to receive new funds (Kaplan and Schoar, 2005).
Independent Variable
The purpose of this study is to observe the extent of the impact that technological innovation has on CVC investment. Thus, for the first hypothesis, the explanatory variable is called Technological Innovation and it is represented by the number of patents issued in a month. For the purpose of a seamless regression, I use the patent growth by each year in order to see actual effect that the increasing rate in number of issued patents has on the dependent variables.
DID dummies
In this study I use a difference-in-differences model which requires a treatment and a time dummy. My treatment dummy is Tech which is a dummy variable that takes the value of one if the categorical variable Industry Focus of a CVC corresponds to "Computer Software and Services" or "Internet Specific" category.
The time dummy applicable to the DotCom bubble takes the value of 1 for any date after March 31st 2000 of the variable Time of Investment and the value of 0 if otherwise. The time dummy
specifying the launch of AWS takes the value of 1 for every year after 2006 of the variable Year of Investment and 0 if otherwise.
Other
Moreover, for this sample I use the SP500 return as well as the 5-year U.S. Bond returns to control for the overall state of the economy and its effect on investment. As mentioned before, industry and time fixed effects will be applied to all the regressions where is possible.
Data Analysis
Table 1 shows the summary statistics for the main measures of dependent and independent variables. We see that the average number of deals per month amounts to roughly 136 deals with a minimum of 3 deals up to a maximum of 369 deals per month. On average, CVCs invest USD 19.92 million of equity per deal in incumbent firms with USD 0.3 million per deal being the smallest amount invested. As Figure 2 shows, this measure is heavily skewed on the left as the majority of the investments are small yet in more frequent rounds. Similarly, Figure 3 indicates that the measure for investment rounds is highly skewed to the right and its average lies at 3 per CVC to incumbent company mounting to a very high maximum number of 11 rounds. This issue of such large number of rounds is also mentioned in previous studies (Lerner, 1995; Townsend, 2015) and is theorized that it could stem from staggered disbursements on a single round being misreported. Townsend (2015) corrects this by restricting the sample to companies with rounds no fewer than 30 days and no more than 6 years apart. Following his approach to solving this problem, makes no difference for the data hence I did not apply those criteria to my model.
Table 2 presents the pairwise correlation for my main variables as used in the study. We see that the main measures of interest do not show a high amount of correlation, which leads me to believe that bad control bias shouldn’t be a source of concern for my estimation.
ENDOGENEITY AND ROBUSTNESS
As it is very common in the field of finance research, endogeneity poses an issue in this study as well. Concerning the sample data of my first specification between CVC investment and technological innovation as proxied by the yearly patent growth, the former variable might be biased by endogeneity. Additionally, Graph 1 and Graph 2 clearly show that the main measures used as dependent variables have large outliers. Furthermore, due to the fact that technology is an ever positively advancing variable, a positive linear relationship would naturally be expected. Table A presents an OLS regression of my first hypothesis. As expected, there is a positive relationship between the overall Round Number of investments as technological growth. For every 1 percent increase in Patent Growth there is an increase of circa 0.5 in the number of rounds of CVC investment. This relationship is reversed when I use Equity Investment as the main dependent variable. For every percent increase in Patent Growth, there is a decline of USD 2.7 million per deals.
Table B applies an instrumental variable regression for which variables Patentlag3 and Patentlag4 are used as instruments for the endogenous variable Patent growth. Patentlag3 corresponds to the number of patents issued per month lagged by three years and Patentlag4 corresponds to the number of patents issued per month lagged by 4 years. Patent Growth, Equity Investment and Round Number are tested for endogeneity through the Durbin and Wu– Hausman tests. For this test, the null hypothesis is that the variable under consideration is exogenous (Durbin, 1954; Wu, 1974; Hausman, 1978). Here, both of the test statistics9 that
belong to the two stages least squares are highly significant at the 1 percent level (p=0.0000 and p=0.0008 respectively), which leads to the model rejecting the null hypothesis. These results indicate that there is endogeneity in the model and that the variables should be considered as endogenous.
One of the methods to correct for endogeneity and make sure that the model is corrected for omitted variable bias, heteroscedasticity and multicollinearity, is using the difference-in-differences estimation method of regression. Therefore, in order to be able to identify correctly the intensity of the investment growth, I need to first identify a set of exogenous shocks and describe the average reaction of an economy to these shocks. The exogenous shocks that I will use in my empirical examination are namely (1) the Internet Bubble of 2000 and (2) the launch of Amazon Web Services in 2006. Both of these events are isolated from the empirical model that I wanted to test, thus not depended on any of the variables that I examine.
METHODOLOGY
This section will explain the empirical methodology that I used in order to test my two hypotheses. I will first elaborate on the methodology theory. I will then address and justify my empirical approach to solving the endogeneity issue within my model. Lastly, I will present the regression models I used to test each of the hypotheses, as well as the expected and the observed relationship of my main measures.
Empirical Methodology
Difference-in-differences
For my first hypothesis I apply a difference-in-differences estimation method. Difference-in-differences (or “DID” for short) was popularized by Ashenfelter and Card (1985) in order to
study two groups of data for two time periods and observe the outcomes. This regression will estimate the difference between a control group “other firms” and a treatment group “technological centric firms”. In the case where the same units within a group are observed in each time period, the average gain in the second (control) group is subtracted from the average gain in the first (treatment) group. This removes biases in second period comparisons between the treatment and control group that could be the result from permanent differences between those groups, as well as biases from comparisons over time in the treatment group that could be the result of trends (Bertrand, Duflo and Mullainathan, 2004; Cao et al 2005; Lechner, 2011).
This leads to the identification strategy that tackles the endogeneity within my model: the randomized experiment throughout the Internet Bubble and then through the launch of Amazon Web Services. Both of these events had a significant and long lasting effect on the market which was isolated outside my model and completely exogenous as far as it concerns the right hand and left hand variables. This evidence speaks to the binding nature and economic significance of short-sale constraints. However, one of the problems I was faced with when compiling a DID estimation was the fact that there is not a clear beginning date or end date considering my first exogenous shock consideration, the Internet bubble. Due to the vagueness of the time nature of a bubble, I took as reference several other studies which consider March 31st 2000 as the peak dating of the bubble have also considered the Internet bubble crash as the
cut-off date for their estimations (Townsend, 2015; Brunnermeier and Nagel, 2004).
One of the groups is exposed to a treatment which here is technology- centric CVC in the second period but not in the first period. The second group is not exposed to the treatment thus including all the rest of the companies, during either period. A standard differences-in-differences regression model looks like this:
𝑦 = 𝛽0+ 𝛽1𝑑𝐵 + 𝛿0𝑑2 + 𝛿1𝑑2 ∗ 𝑑𝐵 + 𝑢
Which, when applied in accordance to my study leads to the DID estimation of:
𝛿̂1 = (𝑦̅𝐵,2− 𝑦̅𝐵,1) − (𝑦̅𝐴,2− 𝑦̅𝐴,1)
Where y is the CVC investment (outcome of interest), d2 is a dummy for either the Internet bubble or the launch of Amazon Web Services (the dummy variable for the second time period)
that captures all the factors that may have caused changes in CVC investment (y) in the absence of the time treatment. In the end however, 𝛿̂1 is the ultimate coefficient of interest for us. This is the estimation that multiplies the d2*dB, the interaction term which tells us if the expected mean change in the outcome from the before-time-period to the after-time-period was indeed different.
Tobit
The second hypothesis is formulated to estimate the relationship between the probability of success of a CVC investment and technological innovation, when controlled for tech-centric CVC and observed before and after the respective time dummies. This will be estimated through a Tobit regression. The deal success is based on the probability of the incumbent firm receiving a follow up investment thus making the boundaries of the variable either 0 or 1. With these estimation requirements a model is needed that allows the dependent variable to fall within specific boundaries. The standard Tobit model is a regression that applied to these criteria. A standard Tobit system of regressions looks like this:
𝑦𝑖 = { 𝑦 𝑖𝑓 𝑦0 𝑖𝑓 𝑦𝑖∗ > 0
𝑖∗ ≤ 0
Where 𝑦𝑖∗ is the latent variable 𝑦
𝑖∗= 𝛽 + 𝑢𝑖 𝑢𝑖~𝑁(0, 𝜎2) Estimation Model
For robustness checks, the first hypothesis is tested initially through a simple OLS and TSLS. This will be addressed in the ENDOGENEITY AND ROBUSTNESS section. Because I expected a different relationship than that which I observe I further examine my hypothesis using a DID and a combination of a DID-Tobit for the first and second hypothesis respectively.
The first regression that is used to test for first hypothesis in this study takes the following DID form: 𝐶𝑉𝐶 𝐼𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡 = 𝛼 + 𝛽1 𝑇𝑒𝑐ℎ𝑛𝑜𝑙𝑜𝑔𝑖𝑐𝑎𝑙 𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝜀 or 𝛾𝑖,𝑡 = 𝛽1+ 𝛽2𝑇𝑒𝑐ℎ 𝐶𝑉𝐶 + 𝛽3𝐷𝑜𝑡𝐶𝑜𝑚 + 𝛿1𝑇𝑒𝑐ℎ𝐶𝑉𝐶 ∗ 𝐷𝑜𝑡𝐶𝑜𝑚 + 𝑢 and 𝛾𝑖,𝑡 = 𝛽1+ 𝛽2𝑇𝑒𝑐ℎ 𝐶𝑉𝐶 + 𝛽3𝐴𝑊𝑆 + 𝛿1𝑇𝑒𝑐ℎ𝐶𝑉𝐶 ∗ 𝐴𝑊𝑆 + 𝑢
From the beginning of this study I expected to see a positive relationship with technological innovation and the CVC investment. Given the Gompers and Lerner (2000) research, where they noticed an increase in CVC regardless of strategic fit, the main assumption I will be testing is to see whether technological innovation is related to this phenomenon. The regression model used to test for the second hypothesis takes the form:
H2: Technological innovation leads to more successful CVC investments
𝐶𝑉𝐶 𝑠𝑢𝑐𝑐𝑒𝑠𝑠 = 𝛼 + 𝛽1 𝑇𝑒𝑐ℎ𝑛𝑜𝑙𝑜𝑔𝑖𝑐𝑎𝑙 𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝜀
To summarize my hypothesis, Table A. lays out an overview of the hypothesis, along with each of the respective dependent variables that I will be using the in the continuity of my study, according to the existing literature that. In this table I have noted my expectations on the findings according to the previous literature on CVCs and finally on the last column I have added the observed relationship that the research of this study shows.
Table A: Summary of hypotheses
The table reports the respective hypotheses that I addressed in this study in the first column. The second column shows the dependent variables that I will be using in each regression model. The last two columns show the expected relationship versus the observed the relationship from the results. Regression model Dependent variable Time cut-off Expected relationship Observed relationship
H1 DID CVC investment AWS + -
DotCom - -
H2 Tobit Success of CVC AWS + -
DotCom - -
RESULTS AND ANALYSIS
This section addresses the two specifications that were previously explained in the METHODOLOGY section, by initially testing the first hypothesis, and thus studying the relationship between technological innovation and CVC investment and then addressing the second hypothesis, therefore shedding light on the impact that technological innovation has on the strategic fit of CVC investments and their portfolio companies.
Results
The first hypothesis aims to test the impact that technological innovation has on the CVC investment. As mentioned previously, for robustness checks I will take two variables into count
that proxy CVC investment in order to see which one provides the best fit for the ongoing study. The first panel presents the Round Number as the main dependent variable. Round number is specified as the number of investment rounds that the incumbent firm has received from a specific CVC. The second panel presents Equity Investment as the main dependent variable. Equity investment is specified as the total amount of equity invested by the CVC on the incumbent firm.
Table 3 presents the results of the first difference-in-differences regression for the first hypothesis. This estimation uses the Amazon Web Services as the time cut-off which is a dummy that takes the value of 1 starting from year 2006, and also uses a technological dummy as treatment (Tech) which takes the value of 1 if the industry focus of the CVC is in technology and 0 if otherwise. The main independent variable is Patent growth which is calculated as the yearly percentage growth of the monthly patents. No of Investors is the total number of investors (CVC, IVC, PE, etc) that have invested in the incumbent firm. Return S&P500 and 5Y Bond return is the S&P 500 market return and the 5 year return on the US bond market respectively, through the yearly observations as presented in the data sample.
Table 4 presents the results of the second test with differences-in-differences regression concerning the first hypothesis. Unlike the first estimation, this difference-in-differences regression is applied by using the DotCom bubble crash as the time cut-off. The time dummy takes the value of 1 for all times that are dated after March 31st 2000. Here I also use the same technological dummy as a treatment as in Table 3. Similarly to Table 3 the first panel shows the regression output where the Round Number as the main dependent variable while the second panel shows the regression output where the Equity Investment as the main dependent variable. The main independent variable is Patent growth which is calculated as the yearly percentage growth of the monthly patents. No of Investors is the total number of investors (CVC, IVC, PE, etc) that have invested in the incumbent firm. Return S&P500 and 5Y Bond return is the S&P 500 market return and the 5-year return on the US bond market respectively, through the sample observation years. Robust standard errors are shown in parentheses10.
Table 5 presents the results of the Tobit regression that test the second hypothesis. This regression explains the probability of success of a CVC during periods of high technological growth. CVC success is the dependent variable which is a dummy equal to 1 if the incumbent firm has received a follow up in investment and 0 otherwise. The main independent variable
remains the same, as the Patent growth. The first panel of the table shows a Tobit DID in which I have used the launch of Amazon Web Services as a time treatment with a cut-off on 2006. The second panel is a Tobit DID with DotCom bubble crash as a time treatment, with a cut off on March 31st 2006. Tech AWS and Tech DotCom are treatment dummies which take the value of 1 if the CVC is a primary investor in the tech sector and 0 otherwise. Robust standard errors as usual are shown in parentheses4. For all regressions I have adopted industry fixed effects,
time fixed effects and firm fixed effects where it was possible.
Analysis
First Hypothesis
On Table 3, we focus on the indicator of a differences-in-differences which is the interaction term or the DID measurement (𝛽3). As such the interaction coefficient represents the difference
between those changes. Here this interaction of coefficients is -0.143. In this case, if investment increases in the control group by 1 unit between the pre- and post- periods, this estimate tells us that tech-centric CVCs see a decline in investment after the launch of Amazon Web Services. Meaning that a 1 percentage point increase in patent growth negatively affects the round numbers of investment by 0.14 rounds. Since the minimum amount of rounds it 1 this estimate is economically insignificant. The first panel on the Table 3 shows no statistical significance concerning its p values and standard errors. In contrast, on the second panel, CVC investment is now proxied through the Equity Investment variable as the main dependent variable. For all columns of the second panel these results however, become highly significant statistically and economically. Here, as the coefficient of interest, DID corresponds to the interaction term between the AWS launch dummy and the treatment for Tech-centric CVCs. This variable measures whether Tech-centric CVC experience a higher amount of equity investment after the launch of Amazon Web Services. Here, we again observe a negative relationship with a DID coefficient of -10.92. Meaning that a 1 percentage point increase in patent growth (technological innovation) leads to a decrease in equity investment for tech- centric CVCs by USD 10.9 million. This estimate holds a highly significant economic value. Furthermore, this coefficient is significant at a 5% level which also makes it highly statistically significant too. Moreover, we see the same pattern emerge when using the crash of the DotCom bubble as the time dummy on Table 4. Here we also focus on the indicator of a difference-in-differences which is the interaction term or the DID measurement (𝛽3). For the first panel, Round Number is the dependent variable. In this case if in the control group investment increases by 1 unit between the pre-and post-periods, the interaction coefficient is positive by both economically
and statistically insignificant at 0.279. This estimate tells us that tech-centric CVCs see an increase in investment rounds after the burst of DotCom bubble (March 31st 2000). None of
the estimations are significant at any level. This estimation is not economically significant either.
These results indicate that while technological growth increases, the investment of tech-centric CVCs decreases, in which these firms see a high decline in the equity amount but a smaller decline in number of rounds. This can mean that while these investments for tech-centric CVCs are characterized by a large fall in equity investment in the presence of technological growth, although theres a decline in the number of rounds as well, this decline is very small, meaning that they still continue to invest on several rounds. This could mean that although these CVCs are more cautious to invest in more rounds for an incumbent, they still are willing to do so. However, they seem to be much more reluctant when it comes to the amount of equity invested. Thus, leading me to believe that due to low trust of the environment they might invest multiple times in a company, yet in fewer amounts of equity. The reason that this happens for tech-centric CVCs after the launch of AWS could probably be due to the highly competitive environment that AWS made possible. Another possibility is that if Amazon Web Services makes it easier for a much larger number of small start-ups to be available to the market, thus lowering the marginal amount of equity invested per company, while keeping a somewhat stable number of rounds as is usual for CVCs. Similarly, to Table 3, the second panel of Table 4 gives contrasting estimates when Equity Investments is used as the main dependent variable instead. In contrast, on the second panel, when the main dependent variable is now proxied through the Equity Investment variable, we yet again see a negative relationship, which is highly significant economically and statistically. Here, the coefficient of interest DID, which corresponds to the interaction term between the DotCom bubble dummy and the treatment for Tech-centric CVCs is -11.11. This indicates that for every percentage increase in yearly patent growth, equity investment sees a decline for tech-centric CVCs by USD 11 million. Since the minimum amount of equity invested by CVCs is USD 0.3 million, this makes this coefficient economically significant. Furthermore, this coefficient is significant at a 5% level which also makes it statistically significant too.
These results lead me to believe that tech-centric CVC firms appear to be more prone to higher rounds of investment in start-ups yet, not surprisingly, far more hesitant to invest large amounts of equity in start-ups after the DotCom bubble crash. This could mean that they are more hesitant to invest more equity due to the overall decline in confidence for tech markets after
the bubble burst. The reason for higher rounds might be due to the fact that these CVCs would need to still maintain a relationship with their incumbents, thus overall offering more rounds of investment.
Second Hypothesis
Table 5 presents the results of a Tobit regression while using a DID estimator. The model here aims to explain the probability of success of a CVC investment (in incumbent firms) during periods of high technological growth. The dependent variable is CVC Success which is defined as a dummy that takes the value of 1 if the incumbent firm has received a follow up in investment and 0 if otherwise. The first panel is a Tobit DID that uses Amazon Web services as a time treatment, with a cut of on January 2006. The second panel is a Tobit DID that uses DotCom bubble crash as a time treatment, with a cut off on March 31st 2000. Tech (AWS and DotCom) are treatment dummies which take the value of 1 if the CVC is a primary investor in the tech sector and 0 otherwise. The model is counted for industry, year and firm fixed effects.
As previously considered, the variable of interest remains to be the main interaction term, DID. However, in all cases, this variable is highly significant at 1% level. In the first panel, regressions (1) through (3) indicate a similar negative relationship between tech-centric investments in CVC incumbent’s success and technological innovation, when time is treated for the launch of Amazon Web Services. Here, the DID coefficient is the interactive variable of the tech-centric CVCs and the time dummy controlled for the AWS launch. This coefficient takes the value of -0.117. This means that for every percentage increase in yearly patent growth the chances that an incumbent firm will receive a follow up funding, controlled for the tech-centric CVCs are 11.7 percent lower. Economically, this variable is highly significant as the chances of success start from 0 to one thus 0 to 100 percent in percentage terms. The second panel of Table 5 which uses the DotCom crash as a time dummy, reports opposing results compared to the first one. Following the same pattern, the second panel indicates a positive relationship. The interaction term takes the value of 0.0195, showing that for every percentage increase in the yearly patent growth the chances that an incumbent firm pertaining to a tech-centric CVC, will receive a follow up funding are 1.95 percent higher. This coefficient however is neither statistically significant at any confidence level nor economically significant as on a scale from 0 to 100 percent, a 2 percent increase is not necessarily a high value.
These are puzzling findings when compared to my initial expectations for the hypothesis and the theoretical background that I explored. While Table 3 indicates that indeed equity
investment tends to be substantially (USD 10.9 million) lower for tech-centric CVCs after the launch of Amazon Web Services, Table 4 shows that this type of retreatment of investor behaviour for tech-centric corporate ventures prevails even during other types of shocks (with an indicator of equity investment decline of about USD 11 million as well). Lastly, Table 5 also reports conflicting results when it comes to the probability of success of a CVC investment. While after the launch of Amazon Web Services, I expected more successful ventures sponsored by tech-centric CVCs, the first panel showed that this is not the case. There is actually a significantly 11.7 percent lower probability that an incumbent will receive a follow up investment, which could be explained by a lack of confidence on the investment side.
This, further prompts me to apply the same study and compare CVC success with new investments that they make. So far, CVC success was identified as the follow up investment from CVC to their incumbents. Variable New will be used as a dependent variable to control for cases that CVCs decide to instead pursue new ventures instead of continuing to invest on their incumbents. Table 6 and Table 7 presents the results of the Tobit-DID regression. Subsequently, this table elaborates on the Table 5 for a Tobit regression by taking into consideration the New (Investment) as a dependent variable as well. The regression explains the probability of success of a CVC during periods of high technological growth through two different measures. Here, New is a dependent variable which is a dummy equal to 1 if the incumbent firm has just received an investment and 0 otherwise. Patent growth remains as the main independent variable. The first panel is a Tobit-DID that uses the Follow up measurement as a dependent variable. The second panel is a Tobit-DID that uses the variable New as a dependent variable. The results of both these tables do not indicate a change in the relationship of DID interaction variable, nor in the significance of the coefficients from which a different meaning could be inferred from.
These results indicate that when there is an increase in technological growth, after the launch of AWS, there is a decline in the probability of success for the tech-centric CVCs investment, in other words, these firms that are incumbent to the tech-centric CVC portfolios are less likely to receive a follow up in their investment. Similar conclusions could be inferred from the second panel, after the burst of the DotCom bubble. This could happen due to a number of reasons. First, due to the widely available tools and the high competition, after the launch of the AWS, CVCs are less willing to follow up in an investment and might prefer to jump ship onto a new start-up that might signal better operational performances. Concerning the aftermath of the DotCom bubble, this could have happened due to the low confidence in the markets at
