The Money of Innovation: The Impact of Venture Capital on Innovation in Sweden

The Money of Innovation –

The Impact of Venture Capital on Innovation in Sweden

Master’s Thesis 30 credits

Department of Business Studies Uppsala University

Spring Semester of 2019

Date of Submission: 2019-05-29

Erik Dahlberg Sofia Sörling

Supervisor: David Andersson

Acknowledgements

The authors would like to take the opportunity to thank a few people whose contribution has greatly helped the process of producing this study. A big thank you to our supervisor David Andersson and our fellow students in the thesis group for valuable input. Also, thank you to the department of statistics for help with statistical enquiries. Last but not least, thank you to SVCA for providing data critical for the realization of this study.

May 29, 2019

Erik Dahlberg and Sofia Sörling

Abstract

Innovation leads to economic growth, however, financing innovation comes with major uncertainties and therefore there is a risk of underinvestment in innovation. One type of investor who is prepared to bear these uncertainties is the Venture Capitalist (VC). The question remaining is whether VCs spur further innovation or mainly exploit existing innovation. By counting the patents belonging to 133 Swedish VC financed firms and comparing these to 609 control firms, the difference in the level of innovation is assessed. The result indicate that VC financed firms, on average, become 23% more innovative after receiving financing from a VC.

Thus, it is concluded that VCs spur innovation in the Swedish context.

Keywords: Innovation, Venture Capital, Financing, Patents, Sweden.

Table of Contents

1. Introduction ... 4

1.1 Purpose of Study ... 6

1.2 Research Question ... 6

1.3 Disposition... 6

2. Theoretical framework ... 7

2.1 Financing innovation... 7

2.2 The role of VCs ... 8

2.3 Literature review on VCs impact on innovative behavior ... 9

2.4 Summary and hypothesis ... 12

3. Method ... 13

3.1 Data and variables ... 15

3.1.1 VC data ... 15

3.1.2 Patent data ... 20

3.1.3 Constructing the control group... 22

3.1.4 Pre and post periods ... 26

3.2 Empirical model... 27

3.3 Method criticism ... 29

4. Result ... 30

4.1 Descriptive statistics ... 30

4.2 Result regression model ... 33

4.2.1 Robustness test ... 35

5. Analysis ... 37

6. Conclusion ... 39

7. Suggestions for future research ... 40

References ... 43

Appendix ... I

1. Introduction

In most modern countries economic growth is considered a highly desirable goal since it will provide the opportunity to be better off tomorrow than today (Grossman & Helpman, 1994).

One important cause behind economic growth is innovation, which has the force to renew industrial structures and even make new sectors emerge (Schumpeter, 1939). Further, according to Romer (1989) innovation and technological specialization are key drivers to growth in both developed and developing countries. Bessant and Tidd (2015) as well as Lindholm Dahlstrand and Cetindamar (2000) proclaim that financing is an important part in order to achieve successful innovation.

Financing innovation is however a non-trivial matter, the risks of innovative projects are often relatively high, and the expected return is usually very uncertain (Arrow, 1962; Hall & Lerner, 2010). Furthermore, the most groundbreaking innovative projects are often conducted in smaller firms, which are unable to finance their projects through positive cash flow from other projects (Hall & Lerner, 2010). There is substantial evidence that small firms, compare to larger firms, face larger growth constraints due to difficulties in receiving financing from external sources (Beck & Demirguc-Kunt, 2006). According to the authors this can be one reason why small and medium-sized enterprises (SME) contribute less to growth than they could have, if given more favourable circumstances. Many public policies are structured in a way that doesn’t meet the financial need of smaller firms. For example, governments promote and finance basic research at universities, which is easier for larger firms to commercialize (Heydebreck, Klofsten & Maier, 2000). SMEs are limited and systematically disadvantaged within these policies to fully exploit their potential, thus, they are less able to secure long-term financial solutions compared to larger firms (Heydebreck, Klofsten & Maier, 2000). Therefore, many firms seek for alternative ways to finance their businesses and one actor which finance smaller and often innovative firms are Venture Capitalists (VCs) (Ueda, 2004).

A VC is an equity or equity-linked investor, which usually invest in young privately held firms

(Kortum & Lerner, 2000), usually start-ups and entrepreneurial firms (Cumming & Johan,

2013). The VC is seen as a financial intermediary who is typically very active as either a director,

an advisor, or even as an operating manager of the firm (Kortum & Lerner, 2000). On average,

the typical investment of a VC bears a higher risk than a bank normally accepts in order to grant a loan (Ueda, 2004). However, as a return the VC expect the investment to generate higher return in terms of higher growth and profitability (Ueda, 2004). Furthermore, banks are also reluctant to grant financing to smaller firms due to the information asymmetry. This asymmetry is smaller between a VC and the firm since the VC extensively screen its potential investment targets and are thus more accurate in evaluating the firm’s potential (Ueda, 2004). Venture Capital is sometimes even referred to as the money of invention ¹ . As active owners VCs provide several resources to the entrepreneurial firm on top of the financial, these can be administrative, marketing, and other strategic advices (Cumming & Johan, 2013). Furthermore, it’s not unusual that the entrepreneurial firm get exclusive access to technology, management support and contact with different professionals such as accountants, lawyers and investment bankers through the VC network which increase the value of the entrepreneurial firm significantly (Cumming & Johan, 2013; Block, Colombo, Cumming & Vismara, 2018). Through these characteristics’ VCs are believed to omit some of the hurdles of financing innovation. By being willing to bear a higher risk and invest in smaller firms they can fill a specific role in bridging the financing gaps for smaller innovative, but growing firms (Hall & Lerner, 2010). There is however no academic consensus on whether VCs spur innovation simply by financing innovative firms, or if they also help these firms innovate further (e.g. Kortum & Lerner, 2000;

Engel & Keilbach, 2007).

There is not only an academic interest in the financing of innovation, but also a practical interest from policy makers. Sweden has long been one of the top three countries in the world in stimulating an innovative corporate environment, producing many innovative firms (Global Innovation Index, 2018). In 2018 Sweden was considered the innovation leader in the EU (European Commission, 2018). The importance of creating an innovative corporate climate in Sweden is highly supported by the Swedish government and therefore they have formed the National Innovation Council. This group is aiming to help and develop Sweden’s innovation capacity and competitiveness compared to other countries (Government Offices of Sweden, 2018). This wish to create an environment where innovative firms thrive is also manifested

1 Gompers and Lerner even wrote a book titled “The Money of Invention: How Venture Capital Creates New

Wealth” (2001, Harvard Business School Press)

through the government agency Tillväxtanalys which among other things focus on smaller firms’ ability to secure financing from VCs. Tillväxtanalys report regularly on the development of private equity investments in Sweden (Tillväxtanalys, n.d).

1.1 Purpose of Study

The purpose of the study is to measure the effect Swedish VCs have on the degree of innovation in the Swedish firms they invest in.

1.2 Research Question

The purpose of this study is aimed to be fulfilled by answering the following research question:

How will the degree of innovation in Swedish firms be affected from a VC investment?

1.3 Disposition

Previous section presented the setting for the importance of financing innovation and VCs’

potential role as an investor to mitigate the financial constraints for smaller innovative firms. To

our knowledge the relationship between VCs and level of innovation has never been studied

within the Swedish context before. In chapter 2 the challenges of financing innovation on a

theoretical level is further explained, so are other important concepts. The chapter also contain

the results from previous studies which have found different answers to the question whether

VCs spur innovation or not. In Chapter 3 the method used is presented. For this study, a

quantitative method referred to as Difference in differences-method was applied to isolate and

measure the effect on innovation from receiving an investment from a VC. Chapter 4 present

the statistical results from the tests. All test results are statistically significant on 1% level and

show a positive association between VCs and innovation. These results are analysed in chapter

5 and the main analysis is that the results are congruent with several of the previous studies

within this field. The main conclusions are presented in chapter 6, it is concluded that VCs are

expected to have a positive impact on firms in which they invest in. The expected increase in

innovation given financing from a VC is 23%. In chapter 7 suggestions for further research are

presented.

2. Theoretical framework

2.1 Financing innovation

Before describing the difficulty of financing innovation, a short definition of innovation is useful. Schumpeter (1939) define innovation using a production function. If a certain function of means of production yield a specific output, then an innovation is to create the same output by changing the form of the function while keeping quantities constant (Schumpeter, 1939).

Innovation, as described by a production function combining capital and labor, can also be defined as making new things given the same amount of input (Schumpeter, 1939). Therefore, the definition of innovation ultimately lands in “setting up a new production function”

(Schumpeter, 1939, p. 84), which according to Sweezy (1943) can be interpreted as “doing things differently in the realms of economic life” (Sweezy, 1943, p. 93).

According to Kerr and Nanda (2014) there are some challenges regarding financing innovation

and assert that under frictionless circumstances all net present value (NPV) projects are financed

and conducted, however, the world is not friction free. According to Arrow (1962) there are

three things which impede private actors from investing in innovation: uncertainty, indivisibility

and inappropriability. Uncertainty impedes investing in innovation since the outcome is

unknown. Since there is no way to insure against this uncertainty, it simply has to be accepted

as a hurdle. Kerr and Nanda (2014) refer to this as the investors inability to evaluate the

innovation before an investment, where the only way of learning about the potential of the

innovation is to actually invest in it. With indivisibility, Arrow (1962) refer to the fact that the

investment cost in an R&D project is often large and fixed in the sense that the output is binary,

either it will fail or succeed. According to Arrow (1962) there is no range between fully succeed

or fully fail, so, a decrease in spending will not return a proportionally smaller output but rather

make the project fail in total. Inappropriability is another challenge, even if there is a patent

system in place to help protect intellectual property, there is never a full guarantee that someone

else will not be able to exploit some of the value of the innovation as it becomes publicly known

(Arrow, 1962). Further, the marginal cost of spreading the innovation is little or close to nothing

when viewing information about the innovation as a commodity, and this discourage actors from

investing in innovation (Arrow 1962). Private actors simply have little incentive to invest and

develop innovative projects. As a consequence, Arrow (1962) assert that this results in systematic underinvestment in innovation and a subsequent welfare-loss. Arrow (1962) therefore advocates for a centralized, governmental or non-profit organization to administer investments in innovation in order to achieve a more optimal allocation of resources. Kerr and Nanda (2014), on the other hand, assert that, as the modern patent system is structured and working today, there is a consensus around the financial markets ability to select and finance the innovations with the highest potential. However, the authors agree on four recurring hurdles that are similar to the ones mentioned by Arrow (1962) and which prevent investing in innovation. These four are: (1) the potential outcomes of the projects are uncertain, (2) the return of the investments are skewed and resembles a pareto distribution ² , (3) even though the innovator usually has a better idea of the innovation’s potential compared to an investor, it is largely unknown to both. Lastly, (4) innovative activities are often associated with intangible assets and human capital, which pose the risk of losing key employees before their contribution to a project is completed. In addition, Kerr and Nanda (2014) also assert that smaller firms compared to their larger peers tend to focus more on groundbreaking innovation, which in turn is riskier and further impede these firms’ possibilities to secure financing. However, it is within these more riskier investments where VCs recognizes opportunities according to Hall and Lerner (2010).

2.2 The role of VCs

The relationship between an entrepreneur and a VC can be studied through the lens of the Agent- Principal theory, where the entrepreneur is the agent and the VC who invest in the company is the principal (Arthurs & Busenitz, 2003). The VC can mitigate the risks of conflicts by structuring financial contracts between herself and the entrepreneur to create incentives for the entrepreneur to behave in a way that is in line with the interest of the VC (Kaplan & Strömberg, 2001). However, the authors also claim that there is no agency problem during the time when the goals of the VC and the entrepreneur are aligned.

2 Meaning that most projects will yield no or little return, while a few ones are likely to have very large returns

(Upton and Cook, 2014)

VCs are usually seen as active owners, who think and act like owners with a responsibility that covers far more than just being a financial source and are in some cases even regarded as an extension of the entrepreneur (Arthurs and Busenitz, 2003). The difference between being an active owner compared to a passive owner, according to Kim, Nofsinger and Mohr (2010), is that an active owner focuses more on the business performance and not only on the associated risk and return of the investment. Therefore, the agency conflict might be mitigated between active owners and upper managers because they have similar ideas in how to operate. Kaplan and Strömberg (2001) conducted an extensive study of VCs characteristics as a principal. The authors advocate that VCs mitigate the principal-agent dynamic through three means of control:

contracting, screening and monitoring (Kaplan & Strömberg, 2001). VCs extensively screen the firms before investing and thus gain a deeper understanding of the firms in which they invest in (Kaplan & Strömberg, 2001). According to Hall and Lerner (2010) this reduces the information asymmetry, which usually increase the cost of capital for innovative firms with proportionally higher R&D spending. The VCs also form the contractual basis for the investment in a way which provides them with a strong influence over the control of the company (Kaplan &

Strömberg, 2001). After the investment VCs usually keep monitoring and influencing the firm in order to keep the operations in line with their interests and goals (Kaplan & Strömberg, 2001).

Hall and Lerner (2010) support these claims and add that VCs also monitor by claiming board seats to a larger extent compared to other owners and also usually prohibit the CEO from also being chairman. VCs also handle the financial risks by setting up the investment in stages or rounds so that a failing project can be cut off from further financing at a later stage (Hall &

Lerner, 2010). VCs also often syndicate the investment so that more than one VC invest in the same company at the same time which reduce the risk for each investor (Hall & Lerner, 2010).

Even though VCs cannot be seen as an optimal financing resource for all small firms (Hall &

Lerner, 2010), they can still fulfil a function for those firms that actually receive an active ownership from a VC.

2.3 Literature review on VCs impact on innovative behavior

There is no academic consensus on whether VC investments inherently make a company more

innovative or if the contribution is basically cash which secure financing to current innovations,

different studies of different markets at different times have reached different results. Kortum

and Lerner (2000) pioneered the field of studying VCs and their impact on firm’s innovative behavior through their extensively cited study from 2000. The authors found the relation between VCs and the level of innovation were in particularly interesting to study after a policy shift in 1979 which allowed pension funds to invest in VCs. This policy shift had a tremendous impact on VCs, it resulted in a substantial increase in the supply of capital for VCs and, subsequently, in more capital to be invested by the VCs (Kortum & Lerner, 2000). Kortum and Lerner (2000) research cover US firms during the period between 1965 and 1992 and they use count of patent registrations in order to measure the level of innovation. The study is analysed on a macro level, and the findings show that innovation have a positive correlation with VC investments. A similar study was conducted by Popov and Roosenboom (2012), which cover European firms over the period 1991 to 2005 and use count of granted patents as a proxy for innovation. Their result also strengthens the notion that VCs have a positive effect on innovation, however the result is weak in its statistical significance and therefore they are critical to the interpretation that VCs actually spur innovation (Popov & Roosenboom, 2012). One argument for the slightly different outcomes and why Kortum and Lerners (2000) result not necessarily will repeat for European firms is claimed to be that American VCs differ substantially in the way they operate compared to their European equivalents (Popov &

Roosenboom, 2012). On the other hand, the findings of a different study conducted by Da Rin and Penas (2007) is in line with those of Kortum and Lerner (2000). The authors study Dutch firms over the years 1998 to 2004 (Da Rin and Penas, 2007). According to the authors, VCs have a strong impact on their portfolio firm’s innovation strategy. They argue that VC financed firms tend to “make” rather than “buy” their innovations. Further, they also “make” and “buy”

innovation to a higher degree than their non-VC financed peers and should therefore be considered more innovative (Da Rin & Penas, 2007) ³ .

Even though several studies argue for a positive effect on innovative behavior given financing from a VC there are several studies, which are critical to these findings (Engel & Keilbach, 2007; Lahr & Mina 2016; Hellman & Puri, 2000). Engel and Keilbach (2007) study the relationship between VC and innovation on new startup firms in Germany and similar to Kortum

3 Authors note, “make” refer to developing innovations in-house while “buy” refer to acquiring firms to access their

intellectual property.

and Lerner (2000) they use patents as their proxy for innovation. However, Engel and Keilbach (2007) study the relationship on a micro level instead of industry aggregate like Kortum and Lerner (2000) did. Initially they also find a significant positive result of VCs effect on innovation, however, after analysing the data sample of VC financed firms and all the other firms they notice significant differences in firms’ characteristics between the groups (Engel &

Keilbach, 2007). One of the differences was the patenting behaviour before the entry of VCs (Engel & Keilbach, 2007). The authors claim that there is a sort of causality problem as to whether VCs increase innovation in their portfolio firms or if innovative firms attract VCs. This is something Popov and Roosenboom (2012) also touch upon in their discussion, whether VCs actually have a positive impact on innovation or if they only follow the innovation. This is also in line with Hellman and Puri (2000), which assert that innovative firms are more likely than imitator firms to receive financing from VCs.

In contrast to Kortum and Lerner (2000) who state that the result was indifferent to industry

sector, Engel and Keilbach (2007) state that industry is one important factor to consider when

studying the effect given a VC investment. Therefore, Engel and Keilbach (2007) changed their

method to only compare VC financed firms to non-VC financed firms which shared multiple

similarities with the VC financed firms. Even after controlling for these similarities the positive

correlation remains, however, the result is no longer statistically significant on a 5% level. The

authors conclude that VC financed firms do show a significantly larger number of patent

registrations compared to all other firms, but since they do so even before the investment the

authors interpret it as VCs favoring more innovative firms already in their screening process

(Engel & Keilbach, 2007). Another study by Lahr and Mina (2016) assert that VCs might even

have a negative impact on firms’ innovative behavior. Lahr and Mina (2016) measuring VCs

effect on firms’ patenting behavior by using survey data from 2004 to 2005 on 940 UK and US

firms. Lahr and Mina (2016) use probit models in order to estimate the probability of VC

financed firms to apply for or being granted patents compared to their peers. Lahr and Mina

(2016) assert that VCs do not increase the probability for firms to apply for or being granted

more patents, rather the opposite, the patenting activities decreases given a VC investment. They

summarize the result as a support for the idea that VCs excel in finding innovative firms and

focusing on the commercialization of their innovations, rather than spurring future innovation.

2.4 Summary and hypothesis

To summarize the presented literature, it is reasonable to say that financing innovation is a struggle for smaller, especially start-up firms. VCs can be viewed as an alternative financing source for these firms. VCs can also be viewed as active owners which exercise a strong influence over the firms in which they choose to invest. Hence, compared to other investors, the development of the VC financed firm is therefore assumed to be in line with the interest of the investor. Previous literature has also claimed a relationship between venture capital and innovation, and most research presented have found the relationship to be positive (Kortum &

Lerner, 2000, Da Rin and Penas, 2007, Popov & Roosenboom, 2012), only the study by Lahr and Mina (2016) is claiming the opposite relationship. It is therefore reasonable to expect that there is a positive relationship between VCs and innovation in the Swedish context. To test the effect in the Swedish context and thereby answering the research question the following hypothesis is constructed.

H0: Swedish firms do not innovate more after receiving financing from a Venture Capitalist.

H1: Swedish firms innovate more after receiving financing from a Venture Capitalist.

3. Method

The aim of this study is to measure the effect in level of innovation for firms who receive financing from VCs. The simplest form of measuring this is to measure the average innovation level in firms before an investment from a VC and compare it to the average innovation level after the investment. However, there are two main problems with this approach. Firstly, the effect is not isolated from other external factors such as e.g. changes in public policies or shocks to the financial systems which could have an impact on firm’s innovative behavior. Secondly, as mentioned earlier, several studies claim that VCs tend to invest in already innovative firms.

This means that measuring only the effect of before and after an investment will not necessarily measure the effect given VCs involvement since the innovative behavior could be expected to increase anyways (Engel & Keilbach, 2007; Lahr & Mina, 2007). To overcome these problems the method used will be an extension of the “before-after” design which instead of just following one group before and after a treatment, a control group is assigned in order to validate the effect of the treatment affecting the treatment group, this design is called difference in differences design (DiD) (Clair & Cook, 2015. Lee, 2016). Thus, when using this method, the construction of a comparable control group is highly important (Ishise, Kitamura, Kudamatsu, Matsubayashi

& Murooka, 2019)

The DiD design is a so called quasi-experimental design which can be considered a useful design

for making causal claims (Clair & Cook, 2015). In a true experiment the researcher selects the

participants randomly. By doing this the researcher can avoid selection bias and also control for

causal effects, such as determine the timing and size of treatment, while observing the effect

(Bärnighausen, Røttingen and Rockers, 2017). In a quasi-experiment the researcher selects the

participants based on certain criteria and observe their behaviour with less or no control of the

causal circumstances (Bärnighausen, Røttingen and Rockers, 2017). Therefore, to be able to

assess the impact of VC financing, the innovative behavior of the VC financed firms during the

pre and post period to the treatment must be quantified. Second, the hypothetical state of

innovative behavior, if the treatment had not taken place, also need to be quantified in order to

verify the treatment effect. The hypothetical part is of course not observable and therefore it

needs to be estimated in some way (Heckman, Ichimura & Todd, 1998). See equation 1 for

illustration. Let “(1)” denote VC financing (treatment) and “(0)” denote non-VC financing. 𝑽𝑪 =

1 indicates VC financed firms. The expected outcome in innovative behavior of the treated group will be denoted 𝑌 ^(') and the outcome of the non-treated group will be denoted 𝑌 ⁽⁾⁾ . The latter expression in equation 1, marked “c”, is therefore the counterfactual part. In other words, 𝑉𝐶 = 1 represent all firms financed by a VC, this is divided in 𝑌 ^(') and 𝑌 ⁽⁾⁾ . 𝑌 ⁽⁾⁾ represents the unobservable, hypothetical, state of these firms if they would not have received VC financing.

𝜃 ^(') = 𝐸.𝑌/ ^(') − 𝑌/ ⁽⁾⁾ |𝑉𝐶 = 1 3 = 𝐸.𝑌/ ^(') |𝑉𝐶 = 1 3 − 𝐸.𝑌/ ⁽⁾⁾ |𝑉𝐶 = 1 3 (1) c

The sample of VC financed firms in this study is not randomly collected from the population, which would have been the case in a true experiment. In contrast, these firms are selected because of their special characteristic of being a VC financed firm. Thus, the control group cannot be randomly selected since this would result in selection bias, the control group has to be constructed (Lechner, 1998). Engel and Keilbach (2007) also argue that VC financed firms distinguish substantially from all other firms in two ways: firstly, VC tend to invest only in firms that have already undergone an extensive pre-investment screening process and secondly, firms will not even seek for VC financing if they don’t think they will fulfill the performance criteria set by the VCs. In their study they highlighted that VCs are attracted to firms which are already innovative and therefore criticizing the method of Kortum and Lerner (2000) who doesn’t take this into account. According to Engel and Keilbach (2007) this leads to a statistical bias through self-selection. However, the statistical bias can be corrected by explicitly modelling the selection process and constructing a control group that is as close to the counterfactual part as possible (Heckman et al., 1998; Keilbach 2005).

The importance of finding a suitable control group is crucial in order to validate the treatment

effect (Heckman et al., 1998; Keilbach 2005). The control group should include firms that could

be considered to be twin pairs to the firms in the VC financed sample. One approach of

identifying these twin firms is through the Conditional Independence Assumption (CIA) stated

by Rubin (1977). The CIA assert that different firms i can have identical realizations of variable

𝑋 ₅ (denoted 𝑥 ₅ ) differ significantly in their target variable only because of their treatment. In this

study the treatment is the VC financing and the target variable is patent registrations. If applying

this assumption on this study the treatment is the VC financing and the target variable is innovation. See equation 2 below, where 𝑌 ⁽⁾⁾ represents innovation without treatment, 𝑉𝐶 = 1 represent the group of firms that will receive VC financing, and 𝑉𝐶 = 0 represent the control group.

𝐸.𝑌 ⁽⁾⁾ |𝑉𝐶 = 1, 𝑋 = 𝑥 3 = 𝐸.𝑌 ⁽⁾⁾ |𝑉𝐶 = 0, 𝑋 = 𝑥 3 (2)

This denotes the assumption that the expected value of a VC financed firm innovation level is equal to the control group firm for this company given identical values of different x variables.

The only difference in innovation level between these firms is when the VC financed firm receive capital from the VC. Given that equation 2 is true the estimated difference in innovation level given a VC financing can be estimated as in equation 3. Where 𝑌 ^(') represents the estimated level of innovation in a firm which have received investment from a VC, 𝑌 ⁽⁾⁾ if not.

The methodological choices included in the process of constructing a comparable control group for this study is explained under the Data chapter, section 3.1.3 Constructing the control group.

𝜃> ^(') = 𝐸.𝑌? ^(') |𝑉𝐶 = 1, 𝑋 = 𝑥3 − 𝐸.𝑌? ⁽⁾⁾ |𝑉𝐶 = 0, 𝑋 = 𝑥3 (3)

3.1 Data and variables

3.1.1 VC data

This study will answer the research question by analyzing Swedish firms. Since there is no general database covering all investments by VCs in Sweden the starting point of collecting data have been through the Swedish Private Equity & Venture Capital Association, SVCA ⁴ . For this study SVCA were kind enough to provide the necessary data of the VCs investments over the period 1993 to 2018. The dataset contains company name and organization number of the VC financed firm, company name of the VC or VCs who made the investment and in what year the investment was made, for some investments there is also information on which year the VC

4 The Swedish Private Equity & Venture Capital Association is the industry body and public policy advocate for the

private equity and venture capital industry in Sweden (SVCA, 2019, https://www.svca.se/om-svca/)

divested. All of the data was collected by SVCA by identifying VC investments in Sweden with the use of public sources ⁵ . See table 1 for a summary of the dataset’s descriptive statistics and figure 1 for an illustration of the frequency of the VC investments during the period 1993-2018.

Figure 1 Frequency of VC investments during the period 1993-2018

According to SVCA the collected data is less representative in the earlier years due to the fact that this data was not systematically collected by that time, later attempts to collect historical data proved harder the more time that had passed, and thus earlier years data is not complete.

As illustrated in figure 1, the number of VC investments have a substantial increase around year 2005. To our knowledge, there is no reason to believe that there have been any special market conditions or policy changes affecting operations of VCs on the Swedish market around that year. Thus, it is assumed that the lower number of investments before 2005 is explained by the previously mentioned lower data availability and therefore the investments in the dataset representing investments in year 2004 or earlier will be excluded from the sample. The dataset

5 SVCA collected the data using their own judgment of what they considered to be VC investments, generally categorized as a financial investor with a limited time horizon for the investment, investing in the early stages of a firm’s life.

0 20 40 60 80 100 120 140

199 3 199 6

199 7 199 8

199 9 200 0

200 1 200 2

200 3 200 4

200 5 200 6

200 7 200 8

200 9 201 0

201 1 201 2

201 3 201 4

201 5 201 6

201 7 201 8

Total

covers both single investments, made by one VC, and syndicated investments, made by two or more VC firms. Thus, there are 270 unique investment combinations in the dataset.

Table 1 Data information Panel A: Data information

Period investment years No. of investments

No. of VC companies

No. of unique combinations*

1993-2018 1,209 145 270

Panel B: Descriptive statistics investment duration No. of investments with exit date

(% of total) Mean Median Minimum Maximum

402 (33%) 5.5 5.0 0 23

Note: Panel A include general information of the dataset. *Is the number of unique combinations of VCs in the dataset. It should be interpreted as of the 1209 investments there are 270 different combinations of venture capital investments. Panel B include descriptive statistics on investment duration based on all observations in the dataset that had an exit date.

As described in table 1, 33% of the VC financed firms in the dataset have an exit date registered.

Based on all these firms with exit date, the maximum and minimum investment duration for a

VC is 23 respectively 0 years and the average duration is 5.5 years and median are 5 years. See

additional illustration in figure 2. This average is assumed to hold also for the investments

missing data on exit date. This information indicate that the effect of VCs should be studied

around five years after an investment.

Figure 2 Frequency of venture capital investment duration from the dataset.

Registration year and industry sector for all VC financed firms are extracted from SERRANO (Swedish House of Finance, n.d). The SERRANO database is a database containing financial as well as other types of information about Swedish firms (Weidenman, 2016). This database covers most legal forms in the Swedish business community and it include data from 1990 to 2016 (Weidenman, 2016). In the sample of observed VC financed firms the entry of venture capital seems to occur in the early years or even in the same year as the firm is registered, see figure 3. This is interpreted as VCs attracts to start-ups rather than mature firms.

0 10 20 30 40 50 60 70

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

FR EQ U EN CY

YEARS

Figure 3 Percentage of VC financed firms' age at the time of investment. There are a few observations older than 30 years and a few observations that are not even registered at the time of investment, these are not included above.

Finally, VC financed firms seems to represent different industries, however, industries as IT and Electronics, Health and Education and Corporate services are more common industries compared to others, but no industry seems to dominate the sample. See figure 4 below.

Figure 4 The allocation of industries among VC financed firms.

0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

PERCEN TA GE

YEARS

Corporate Services 20%

Telecom Media 6%

IT & Electronics 25%

Finance & Real Estate 4%

Health & Education 21%

Convinience goods 2%

Shopping goods 8%

Construction industry

1%

Idustrial goods 9%

Materials 1%

Energy &

Environment 1%

Other

2%

3.1.2 Patent data

To be able to quantify a firm’s innovative behavior and using it in statistical models, different measurements have been suggested as proxies for innovation. One of these is the investments in, or employment cost for, Research and Development (R&D), but this is far from a perfect measurement (Crosby, 2000). R&D spending or employment cost for R&D employees can be a flawed proxy since it only captures the input value for the innovative efforts with no guarantees for resulting in an innovative output. Acs, Anselin and Attila (2002) presents a similar view where the R&D only provide a measure of budgeted resources for innovative efforts. Another commonly used proxy for representing a firm’s innovative behavior is the number of patents belonging to the firm (Engel & Keilbach, 2007; Kortum & Lerner, 2000). Patents provide a better proxy than R&D costs since it measures an innovative output and the data is available regardless of whether the owner of the patent is public or not (Acs, Anselin & Attila, 2002).

Patents provide a way for inventors to protect intellectual property. When a patent is granted the inventor is given legal, exclusive, right to make, use or sell the invention for a limited time ⁶ . In order for a patent application to be granted the invention will be scrutinized by the patent office.

The invention needs to be deemed novel, inventive and have a wide industrial applicability ⁷ . This means that a granted patent should signify an innovation, and it should be desirable for firms to patent their innovations when possible since this would grant them a time limited monopoly on whatever economic value the innovation might be able to exploit. The usefulness of patents as a proxy for innovation can be verified by previous studies which have shown that there is a significant positive correlation between the input value of R&D expenditures and patent applications and patent grants (Pakes & Griliches, 1980; Trajtenberg, 1990a). Acs, Anselin and Attila (2002) examines patents correlation to a completely different measure of innovation, a literature-based measure of innovations. This is portrayed as a very reliable measure where information about innovations is retrieved through the new products section of trade and technical journals. Acs, Anselin and Attila (2002) asserts that patents provide a fairly reliable measure of innovative activity.

6 Definition of patent, European Patent Office Glossary https://www.epo.org/service-support/glossary.html#p

7 Conditions for a patent, Patent och Registreringsverket https://www.prv.se/en/patents/applying-for-a-

patent/before-the-application/conditions-for-a-patent/

What is important to keep in mind is that all innovations are not possible to protect through patent, nor are all innovations desired to be protected by patent (Crosby, 2000). Some of the other limitations of just counting patents as a proxy for innovation lies in that patents differ in their innovative importance and their economic impact (Trajtenberg, 1990a: Hall, Jaffe &

Trajtenberg, 2005: Bernstein, 2015). In order to avoid some of the drawbacks of just counting patent registrations Trajtenberg (1990b), Hall et al. (2005) and Bernstein (2015) argue for the use of patent citations as an alternative measurement, where a higher number of patent citations should reflect a more innovative or valuable patent ⁸ . Citations has also been used to determine the economic value of patents (Griliches, 1990). Therefore, the literature promoting patent citations as a better proxy, is more preoccupied with the innovative impact of the innovation itself or the economic value of the patent. These aspects are outside the aim of this study. So, even though patent registrations do not cover the whole spectrum of innovative behavior it’s still an available, quantitative and systematic measure of innovation worth considering (Pakes

& Griliches, 1990). Therefore, patent registrations will be used as a proxy for innovation and as the dependent variable in this study.

The patent data used for the study is retrieved from PAtLink, which include all patents belonging to Swedish firms during the years 1990 to 2014 (Swedish House of Finance, n.d). PAtLink is constructed from a unique patent identifier extracted from PATSTAT and merged together with a unique organization identifier extracted from SERRANO. By having access to the SERRANO database and the PAtLink database it is possible to construct a control group that is comparable to the sample of observations received from SVCA. However, since the PAtLink data only contains data until 2014 and the assuming investment duration is entry year plus five years the study will limit its observed VC investments to entry year no later than 2009. This is further explained under the 3.1.4 Pre and Post section below. The number of investments that will be removed from the sample due to this assumption is 828 investments, which represents a considerable part of the total. However, the number of investments remaining after this will be 283 investments and can still be considered as a big enough sample to represent the group of VC financed firms in Sweden. Further, the importance of estimating the right number of years

8 Patent citations acts as references to previous patents or as the number of other patents that have cited and used

this current patent as a reference (Trajtenberg, 1990b).

as a post period is assumed to outweight the loss of these observations. A few of the VC firms were not matched with data in SERRANO. The reason for this is unknown and the loss is considered as minor and will not have any significant impact on the study. The final sample of VC financed firms before constructing the control group is therefore 265 firms. See table 7 in appendix for a summary of all removed and reduced numbers of VC investments based on all assumption mentioned above.

3.1.3 Constructing the control group

In order to find which variables to base the criteria to match the treatment firm with its control firms this study takes inspiration from Engel and Keilbach (2007) who consider industry, firm size and previous patenting to be important characteristics in the matching process. A method similar to that of Boucly, Sraer and Thesmar (2011) is applied to construct the control group.

Boucly, Sraer and Thesmar (2011) match firms given three criteria and these are business sector ⁹ , firm profitability and firm size, all three in the year before treatment year. Return on assets (ROA) is used as a proxy for profitability and number of employees is used as a proxy for size. Given that firms belong to the same two-digit sector as the target firm, Boucly, Sraer and Thesmar (2011) allow a maximum of ±50% deviance for ROA and ±50% in number of employees between the control firm and the treatment firm in order for a company to qualify as a control firm. This study will use the same criteria and proxies as Boucly, Sraer and Thesmar (2011) together with an additional criterion, the level of innovation the year before treatment year and allowing ±50% deviance. The proxy for this is the number of patent registrations the year before treatment year. It is motivated to add this criterion since Engel & Keilbach (2007) have emphasized the causality problem between innovation level and VC financed firms. By considering the level of innovation at the time when VCs probably screen the company before an investment, the estimated control firm for each VC financed firm will have a comparable level of innovation. Thus, the critique that VC financed firms would be innovative regardless of financing from a VC or not, will not hold.

9 Business sector is determined by the 2 first digits of the firms SNI07 number. SNI stands for “Standard för svensk näringslivsindelning” (standard for Swedish business segmentation).

https://www.scb.se/dokumentation/klassifikationer-och-standarder/standard-for-svensk-naringsgrensindelning-sni/

Boucly, Sraer and Thesmar (2011) emphasize the tradeoffs built into this technique, the ±50%

deviation could for example be reduced in order to find better fitting control firms. However, this would lead to a higher number of treatment firms would not match any control firm at all and the risk of response loss will increase. Since this study adds one more criterion to match the firm the approach of using the same ±50% deviance as Boucly, Sraer and Thesmar (2011) for all variables is assumed to be restrictive enough.

For each and every treatment firm, minimum one and maximum five control firms have been selected, if there is no control firm for a particular treatment firm the treatment firm have been removed from the sample. If there are more than five firms suggested as control firms for one treatment firm, the five firms closest to the treatment firm have been selected, in accordance with the technique of Boucly, Sraer and Thesmar (2011). The authors employ a distance calculation to see which control firms are the most similar to the treatment firm. The authors define distance as the sum of the squares of the differences between the treatment and the control firm’s ROA and the treatment and the control firm’s number of employees. A similar calculation has been employed in this study to distinguish the closest five firms to the treatment firm. See equation 4.

𝐷 _5,A = ^(BCD

^FBCD

⁾

BCD

+ ^(JKL

^FJKL

⁾

JKL

+ ^(MDN

^FMDN

⁾

MDN

(4)

𝐷 _5,A = The difference in ROA, number of employees (Emp) and patent registrations (PAT) between treatment firm i and control firm j the year before treatment year.

Where 𝐷 _5,A represents the difference in ROA (denoted ROA), number of employees (denoted

Emp) and patent of registrations (denoted PAT) between the treatment firm i and the control firm

j at one year before treatment year. The equation returns a relative range and is interpreted as

the distance between the firms where a 𝐷 _5,A closer to zero indicate more similar firms and

subsequently the five observations with the lowest 𝐷 _5,A values are chosen. 𝐷 _5,A can only assume

positive values.

When conducting this matching process there were initially 265 VC investments where 133 of these could be matched with at least one control firm. This represents 50% of all investments in the sample. As summarized in table 7 in the appendix the total number of control firms in the control group are 609 firms which result in a total of 742 observed firms. The number of VC financed firms and respectively control firms in the sample seem to be relatively the same for each investment year and no investment year is dominating the sample. See figure 5.

Figure 5 No. of VC financed firms and control firms for each investment year, 2005-2009

The response share of 50% is interpreted as relatively low but expected. The most common reason for the removed investments was that the year of investment was the same year as the company was registered and therefore no year before investment year existed. It might be tempting to change matching year to investment year because of this number of removed investments. However, this is not an option since it would either interfere with the measured treatment effect at a certain time, i.e. patent registrations during investment year, or the study will not consider the innovation level in the VC financed firms before an investment. Only a few numbers of investments were removed due to not matching any control firms, given the four criteria for the VC financed firms. When using four criteria instead of three as Boucly, Sraer

0 20 40 60 80 100 120 140 160 180

2005 2006 2007 2008 2009

VC financed

firms

Control firms

and Thesmar (2011) the number of responses is expected to decrease. However, the decision to include patent registrations is still motivated because it does not only increase the quality of the matching firm it will also mitigate the causality problem which Engel and Keilbach (2007) pointed out as a criticism towards Kortum and Lerner (2000). See part 2, table 7 in appendix for a summary of all removed investments after the matching process.

To be able to assess whether the selected control group is a comparable group to the VC financed group the variables mean, median, minimum, maximum and variance have been calculated and the result is presented in table 2. The result indicates that the mean and median value of all variables between the two groups do not differ noticeably from each other. To be certain of this indicative result, a paired two sample t-Test for the two groups difference in mean value have been conducted for all variables. Since the two groups differ in number of observations and a t- test like this assumes the same number of observations a random sample of 133 firms have been selected from the control group. This procedure has been repeated five times to be certain that the results of the t-tests are rigid. The test result shows that none of the differences in mean value for all variables differ from zero on a 5% significance level. The interpretation of the result is that the selected control group for this study is a good comparable group for the VC financing firms that are remaining in the sample. The output of the t-tests is presented in table 8 in the appendix.

Table 2 Descriptive statistics of VC financed firms and control firms

Variable Mean Median Minimum Maximum Variance

VC financed firms:

ROA -0.3199 -0,1000 -3.7524 0.5823 0.3814

Employees 6.2180 2.0 0 59 118.3385

Patent registrations 0.4211 0.0 0.0 9 1.9577

Control firms:

ROA -0.2749 -0.087 -4.1333 0.6551 0.3270

Employees 4.4466 2.0 0 71 74.4647

Patent registrations 0.1724 0.0 0 12 0.8370

3.1.4 Pre and post periods

In order to make the DiD design work it is important to define pre and post period for the treatment group and apply the same timing on the control group. For the pre period the year before the investment year serve as the only observed year. It could be argued that the screening process for a VC is longer than only one year before the investment year, however this year represent the year which the treatment firms and control firms are matched and therefore should be very similar to each other. If including more years earlier than only this year there is a risk that other factors during these earlier years can have a negative impact on the measured effect because the assumption of CIA (Rubin, 1977) will be violated. For the post period, the investment year and additional five years after the treatment is chosen as the post period. As previously mentioned, the reason for this is that the VCs in our data hold their investments for about five years on average and therefore should VCs impact on innovation be expected during this period. One convenience with the nature of the data is that since the treatment effect occurs in different years for different observations there is a low risk that a single time anomaly will greatly affect the result of the study. See figure 6 for illustration of the pre and post period for all 742 firms in our sample.

Figure 6 Illustration of the time-window pre and post the treatment

In figure 6 the treatment, i.e. when a VC financed firm receive an investment, is referred to as 𝑡 ₎ . For all VC financed firms there are at least one control firm that have the same hypothetical investment year as one VC financed firm. The pre and post period for all 742 firms in the sample is defined as the year before investment year, 𝑡 _F' , and investment year plus additionally five years, 𝑡 _),…, 𝑡 _Q . This means that for each firm in the total sample of 742 firms there are 7 observations and as a result the number of observations should end up in total 5,194 observations. However, when collecting the data for all firms covering the whole time-window 361 observations are lost. The reason for this can be explained by that the firms have been deregistered from SERRANO sometime during these 7 years and therefore no financial data exist on these firms for these years. The final data sample in this study therefore ends up with 4,833 firm years.

3.2 Empirical model

As figure 6 illustrates there could exist a difference in patent registrations between VC financed firms (treatment group) and control firms (non-treatment group) in the post period. There can also be a difference in patent registrations between pre and post period observations within the treatment group and as well between pre and post observations within only the non-treatment group. Therefore, all these potential effects must be considered in the regression model. The DiD model used in this study is a linear regression model including three dummy variables (model 1), defined in equation 5. The model aims to describe the variance of the dependent variable, patent registrations (denoted Y). However, to get the underlying data more normally distributed the variable log 𝑌 which is the logarithm of patent registrations, will be used instead of Y. Using the logarithm of patent registrations do not only improve the assumption of normal distribution, it also improves the assumption of homoscedasticity since the log transformation stabilizes the variance for the underlying series (Lütkepohl & Xu, 2012). Transforming to logarithm data is also a common approach when using patent data (e.g. Kortum & Lerner, 2000;

Bernstein, 2015). Further, to reduce potential serial correlation between the error terms the error term will be clustered on firm level as recommended in Bertrand, Duflo and Mullainathan (2004).

log 𝑌 = 𝛼 + 𝛽 _' 𝑃𝑂𝑆𝑇 + 𝛽 _Z 𝑉𝐶 + 𝛽 _[ (𝑃𝑂𝑆𝑇 ∗ 𝑉𝐶) + 𝜀 (5)

The dummy variable VC takes the value 1 given observations of VC financed firms (treatment group) otherwise 0. POST takes the value 1 given observations from the post period otherwise 0, and the interaction variable between POST and VC, VC*POST, takes the value 1 given observations of VC financed firms (treatment group) from the post period otherwise 0. Hence, this dummy’s coefficient is therefore the effect evaluated as the expected impact on patent registrations given financing from a VC. 𝜀 represents the error term. The model will also be modified in three different ways to test fixed effects for years and industry. First modification will test for fixed effects for years (Model 2), second, the clustering of error terms will be on industry level instead of firm level (Model 3). Model 4 will test for both fixed effects for years and industry and using robust standard error to reduce potential serial correlation between the error terms. The two control variables, years and industry, are expressed in equation 6 as YEAR and IND respectively, all other dummies have the same interpretation as in figure 5.

log 𝑌 = 𝛼 + 𝛽 _' 𝑃𝑂𝑆𝑇 + 𝛽 _Z 𝑉𝐶 + 𝛽 _[ (𝑃𝑂𝑆𝑇 ∗ 𝑉𝐶) + 𝛾𝑌𝐸𝐴𝑅 + 𝜃𝐼𝑁𝐷 + 𝜀 (6)

YEAR is a control variable testing for year fixed effects, since the sample cover years from 2004 to 2014 each coefficient is the average effect on innovation level given an observation in the particular year relative to the year 2004. IND is a control variable, which testing for industry effects. Since the sample cover 14 different industries each coefficient is the average effect on innovation level given an observation in the particular industry relative to the Energy and Environment industry.

Lastly, it is necessary in a standard linear regression model to control for no multicollinearity

between the independent variables. However, because of how the model is structured, i.e. every

observation in the sample represent more than one, but never more than two independent

variables in the model, there is expected to exist some level of linear association between two

independent variables. Since this is how the model is structured the linear association is regarded

as acceptable.

3.3 Method criticism

One of the most crucial aspects of the method of this study is the construction of the control group. There are alternative methods available to do this, Engel and Keilbach (2007) used propensity score matching in order to find which variables to use for their control group. Since this method was deemed too complicated and time consuming, this study had to use a simpler method used by Boucly, Sraer and Thesmar (2011) and applied the same proxies for firm size and profitability. Here it can be argued that other proxies or ratios could be used. How to define the size of a firm can be debated in philosophical terms but number of employees is a measure that has been used as a proxy in research focused on firm size (Kumar, Rajan & Zingales., 1999) and is easy to use since the data is available. The same goes for return on assets (ROA), which has been used extensively in previous research as a measure of profitability and was also readily available in SERRANO and was therefore also deemed as reasonable. To include patenting as a selection criterion is a decision which has less support in previous literature. However, Engel and Keilbach (2007) found that previous patenting behavior had an impact on later patenting regardless of VC financing. Therefore, it is reasonable to include patenting as a criterion in order to find firms that will likely follow the same trend in patenting. By doing this there is a risk that the result and the losses of observations will be depending on what methodological choices have been made, but it is questionable if any other method would have been superior.

A consequence due to the choices for the time window of the study was that a substantial part of the VC investments was excluded from the sample. Even though the sample was still considered large enough to conduct the statistical tests, the loss of data raises the question whether any systematic effects have been lost due to the loss of observations. In this case the defined post period can be discussed, whether five years is a too long time after the investment to use and only cover for example three years, for the benefits of keeping almost one third of the lost VC investments. There was however no other circumstance indicating that shorter or longer post period was a natural choice. A longer pre period to reinforce the common trend assumption would have further reduced the sample. This due to the fact that good matches would be harder to find, but also since the VCs invest in young firms, many times the same year as the company is formed.

4. Result

4.1 Descriptive statistics

The final sample size of 4,833 observations covers the period 2004 to 2014. As presented in table 3 the number of observations is less in the earlier and later part of the studied period. This is expected since the investment period covers the years in the middle of this period, 2005-2009, and subsequently these years are expected to have more observations. The descriptive statistics of the final sample show a mean value in patent registrations of 0.22 for all observations, 0.91 for VC financed firms and 0.07 for the control group. These calculations are based on observations for each group included both pre and post period. The maximum number of patent registrations for one observation is 33 and minimum is 0. The variance in patent registrations between VC financed and control group firms differ, where VC financed firms have higher sample variance among them. The whole sample is highly skewed, with a value of 10.99, however the skewness value for the logarithm of patent registrations which is applied in the regression model show a much lower value, 4.99. The kurtosis value is also high for all groups, but especially for the control group which have a kurtosis of 468.50. This is mainly because there were more control firms for the VC financed firms which had zero or low number of patents. The kurtosis for the logarithm of patents registrations, which will be used in the regression model, is also high, but much lower, 29.59 ¹⁰ . See table 3 below and frequency table 9 in appendix.

10 The number of observations is 4,833, therefore, is according to the Central Limit Theorem (CLM) a big enough

sample to assume that the error terms will be approximately normally distributed (Israel, 1992).

Table 3 Descriptive statistics

Panel A: Descriptive statistics

Mean Median Standard Error Sample Variance Min Max Kurtosis Skewness Count

PAT 0.224 0 0.0192 1.7758 0 33 176.12

Log:29.59 10.99 Log: 4.99

4833

PAT_VC 0.912 0 0.0901 7.2008 0 33 43.53 5.46 888

PAT_Control 0.070 0 0.0104 0.4257 0 23 468.50 17.79 3945

Panel B: Number of observations per year

Year Frequency

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

83 176 377 542 726 708 684 593 488 303 153

Note: Descriptive statistics of patent registrations for all observations, VC observations and control group observations over the whole time-window.

To see a potential effect given VC financed firms the effect is illustrated in figure 7. The figure

illustrates the mean number of patent registrations in pre period and post period, allocated on

VC financed firms and control firms. For the year before the investment the difference in mean

value is rather small but there is a clear difference in trends onwards. The VC financed firms

has an upward trend and hold a rather stable and higher level of mean patents, though there are

some dips in year 𝑡 _Z and 𝑡 _d . At the same time the graph shows a downward trend for the non-

VC financed firms which is then sustained with little fluctuation in year 𝑡 _' to 𝑡 _Q . This

illustration indicates that VC financed firms have a positive effect on patent registrations or at

minimum, VC financed firms maintain the level of patent registrations after an investment. The

control group on the other hand seems to have a negative trend in patent registrations after the

pre-period. This indicates that when these firms have been selected as control firms, they have

showed a number of patent registrations necessary to be required as a matching firm but have not continued to register new patent in the same pace as VC financed firms in the following years.

Figure 7 Showing VC financed firms and control group firms for the entire time window of the study. This graph is showing the mean value of patents registered in relation to year relative to investment. Line at year 𝒕

indicate investment year.

When calculating the mean value of logarithm patenting registrations as in line with the difference in differences logic the average difference between before and after for each group is 0.16 for treatment group and -0.05 for non-treatment group. When considering the difference in difference between pre and post period between these groups the difference is calculated as 0.21.

In table 4 the mean value for each group and each period is presented as well as the difference between these. It is the difference in differences which will be statistically tested in section 4.2 Regression model.

Table 4 Difference in Differences output between the two groups

Before After Difference

Treatment 0.1734 0.3293 0.1559

Non-treatment 0.0695 0.0200 -0.0495

Difference 0.1039 0.3093 0.2055

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

t-1 t0 t1 t2 t3 t4 t5

Non VC VC

4.2 Result regression model

The models presented in the previous section was set up in the software Stata and returned the following results, see table 5.

Table 5 The result of regression models

Note: The dependent variable for all models is the logarithm of patent registrations. For model 1 and 2 the error term is clustered on firm level. For model 3 the error term is clustered on industry level and for model 4 the robust standard error is applied. Model 2 and 4 have control variables for year fixed effect and model 4 also have fixed effect for industry. See table 10 and 11 in Appendix for full output for model 2 and 4. *Means statistically significant at the 10% level, ** means statistically significant at the 5% level and *** means statistically significant at the 1% level.

The value of the model’s coefficients is interpreted as follows: a positive (negative) sign before a coefficient is interpreted as the related regressor have a positive (negative) effect on the dependent variable. Subsequently, since the model is a semi-log model the value itself is approximately the relative effect in number of patent registrations given 1 for any of the explanatory variables. However, since all variables are dummy variables and not continuous variables the relative effect is instead carried out according to Halvorsen and Palmquist (1980) as in equation 7 and the percentage effect is then equal to equation 8.

𝑔 = {exp(𝛽 _[ ) − 1} (7) 100 ∗ 𝑔 = 100 ∗ {exp(𝛽 _[ ) − 1} (8)

As visible in table 5, for model 1, the coefficient for the interaction term (POST*VC) describing the relative change in patenting, calculated as in equation 6, given that a company has received VC financing is 0.2281 and significant at the 1% level. The interpretation of this result is that given an investment from a VC the number of patent registrations for a Swedish firm will on

Model 1 2 3 4

Log (PAT) Log (PAT) Log (PAT) Log (PAT)

POST*VC 0.2055*** 0.2053*** 0.2055** 0.2047***

POST -0.0495*** -0.0500*** -0.0495** -0.0615***

VC 0.1039** 0.1042** 0.1039*** 0.1009**

Constant 0.0695*** -0.0176** 0.0695* -0.1363***

Observations 4,833 4,833 4,833 4,833

F (3, 741) 24.76*** 6.92*** 15.71*** 11.15***

Adj. 𝑅

0.1009 0.1037 0.1009 0.1374

Year effect No Yes No Yes

Industry effect No No No Yes