Innovation in Entrepreneurial Firms and VC Exits

Transcription

1 Innovation in Entrepreneurial Firms and VC Exits Susanne Espenlaub Arif Khurshed Haitong Li This draft: October, 2016 I study the impact of portfolio companies pre-exit innovation on VC exit choices by comparing the innovation output of portfolio companies through different VC exits (IPOs, M&As and liquidations). Using a sample of VC-backed companies in the US with their patent-based metrics, I find that more innovative companies are associated with higher probability of successful exits than unsuccessful exits. Further, more innovative companies are associated with higher probability of exits through IPOs than through M&As. My identification strategy exploits plausible exogenous sources of variation in state R&D tax credits which affect portfolio companies incentive to innovate and overcome the endogenous problem with two-stage least square instrumental variable models. My results are also robust to controls for a variety of factors, a sample of companies with published patent data (both granted and non-granted), a subsample of companies with VC exits before 2012, and a subsample of companies generating at least one patents. The analysis reveals that the degree of innovation is an important driver in VC exit choices. Key words: innovation, entrepreneurial firms, venture capital, exit Alliance Manchester Business School, The University of Manchester, Booth Street West, M15 6PB, United Kingdom. s: Susanne.espenlaub@manchester.ac.uk, arif.khurshed@manchester.ac.uk, Haitong.li@manchester.ac.uk 1

2 1. Introduction Are innovative portfolio companies more likely to have successful VC exits? This question is important given the critical role of VC exits in the long run growth of both portfolio companies and VC investors and the significant effect of innovation on company competitiveness and capability in product market. Till today, one stream of existing literature on the determinants of the exit route choice studies a broad range of factors, including VC firms and portfolio companies characteristics, IPO and M&A market conditions, financial contracts and so on. And another stream of literature shows the change of innovation output before and after IPO no matter how the company is funded (via VC or not), by testing the impact of company public or private status on its innovation output (e.g. Bernstein, 2015; Wu, 2012) or the impact of VC exits on innovation output (Aggarwal and Hsu, 2013). Even though Schwienbacher (2008) developed a theoretical model linking innovation as a factor of product market competitiveness and VC exits, no empirical work has directly tested the relationship between innovation output and VC exit outcomes. This paper measures innovation based on patent data and studies the impact of portfolio companies pre-exit innovation output on the choices of VC exit. My results are in line with the prediction from Schwienbacher (2008) theoretical model and its main contribution is to show empirically that more innovative companies are associated with higher probability of successful exits than unsuccessful exits, and further, more innovative companies are associated with higher probability of exits through IPOs than through M&As. 1 VC-backed portfolio companies, compared to companies with other funding, provide two significant advantages for this study. Firstly, they are the main innovators in the economy and seek out protected market niche that are too small for larger organizations. In the meantime, previous literature has shown a positive and significant impact of venture capital to technological innovation through their investments in entrepreneurial firms (Kortum and Lerner, 2000). Hence, they have a strong motivation to innovate by the desire to earn profits and further, high innovation level is not only offering profits for the company, but also beneficial for VC firms long-term growth. Accordingly, the innovation productivity could be regarded as another performance measure of the VC-backed entrepreneurial firm and a more innovative entrepreneurial firm is supposed to have greater potential and be more successful. 1 Giot and Schwienbacher (2007) examine factors of exit type and exit time using technological improvement from one round to the other and mention that the degree of innovation is also expected to be an important driver but unfortunately their data do not allow them to proxy the degree of innovation. 2

3 Secondly, when VCs start their investment, they expect an exit after a possible specific period. For a non-venture-backed company with access to alternative financial resources, remaining private is always an option. But for a venture-backed firm, the fact that the VC wants to exit rules out the option of remaining private. Hence, when venture capitalists plan to end the investment and exit from the entrepreneurial firm, they would bring entrepreneurial firms with better performance (e.g. those more innovative) to successful exit choices, including initial public offers (IPOs) and acquisition. I use a sample of all VC-backed portfolio companies from VentureXpert which received their first VC investment from year 1995 to 2010 and their exit year spans from 1996 to The sample covers 9877 observations, including 1231 IPOs, 6391 M&As, and 2255 liquidations. I measure pre-exit innovation of portfolio companies with patent data so that I could compare the degree of innovation in portfolio companies through different exits. Since 75% observations out of the full sample have no patent data, I do a subsample test covering firms with at least one patent. My results show that portfolio companies with successful exits are associated with more pre-exit innovation output and more innovation output increase the probability of exits through IPOs, compared to M&As. Further, I overcome the endogenous problem with a two-stage least square instrumental variable model, using a dummy variable representing the introduction of R&D tax credits across states as an instrumental variable. The results of 2SLS confirms my baseline findings. As far as I know, this is the first paper to use all VC-backed portfolio companies (except those are still under active VC funding) in multiple industries to study the impact of innovation output on the VC exits. This paper makes contributions to the existing literature on VC exit choice, private company decision making on going public or being acquired and the importance of entrepreneurial innovation during the development of the company. The rest of this paper is organized as following. Section 2 includes literature review and hypothesis, followed by data, sample selection, and methodology in Section 3. Section 4 shows the results analysis and section 5 includes several robustness tests. Section 6 concludes. 2 The reason to choose this specific time period is in detail in Section 3. 3

4 2. Literature Review and Hypothesis Development 2.1 The Importance of Exit Choices A VC investor provides capital to a start-up with a view to a later exit opportunity, either in the form of an IPO or a sale to another firm (Gompers and Lerner, 1999). It is meaningful to get a better and more comprehensive understanding of the mechanism behind the VC exit route because it is related to the long-term development of both VC firms and entrepreneurial firms. For VC firms, the way the VC exits is a central element in the venture capital cycle (Gompers and Lerner, 2004). They make large investments in entrepreneurial firms and hold an important equity position, in the expectation of a successful exit (an IPO or M&A) from the portfolio companies (Cumming, 2008). Such an illiquid investment could not provide them with dividends during the first few years due to the fact that start-ups initially do not generate profits. Thus, the VC exit route plays a significant role in whether the investment could offer venture capitalists positive returns and further help venture capital firms build their reputations for long-term growth. On the other hand, entrepreneurs receiving VC investment want to have successful exit outcomes not only for financial returns but also for their own business ideas or business dream come true. Hence, these two players have consistent willing to make the portfolio firm successful. The two main successful exit routes include an initial public offering (IPO) and an M&A. Through an IPO, VC-backed firms adjust the firms capital structure and raise more funds to allow the VC investors to divest. Lerner (1994a) finds that VC investors are successful in timing the decision to take the entrepreneurial firm public: they choose to exit their company by IPO when its valuation is at its peak and when the industry valuations are highest. Venture capitalists exits with positive financial returns, and after an IPO entrepreneurs still control their firms. On the other side, through an acquisition, a new acquirer will buy the entrepreneurial firm, so both entrepreneurs themselves and venture capitalists will exit the firm with cash. In this case, entrepreneurs will lose their control of the firm. So considering their private benefits, they generally prefer IPOs to acquisitions (e.g. Zingales, 1995). There are also less successful or even failure exit routes, which are a secondary sale, a buyback, and a write-off. A secondary sale is the sale of the VC investor s shares to a strategic acquirer or another VC company. A buyback is the entrepreneurial firm buying back the VC investor s shares. In some contracts, there is a clause requiring entrepreneurs to buy 4

5 back their shares if an IPO or M&A has not occurred within a specified timeframe (Pearce and Barnes, 2006). Since there is no external capital coming into the company, a buyback is usually less profitable for VCs. A write-off describes that the entrepreneurial firm has failed or when it is unprofitable. In short, the exit routes not only determine the financial return and long-run development of venture capital firms but also affect entrepreneurs own control rights and private benefits. Towards the previous literature on the determinants of the exit choices, I can categorize them into four categories: (1) VC investors characteristics; (2) entrepreneurial firms characteristics; (3) IPO and M&A market conditions, and (4) financial contracts between VCs and portfolio companies. 2.2 Innovation Outcome and VC Exit Choice Main Hypothesis Previous literature towards the relationship between innovation outcome and VC exits is mainly concentrated on the effect of VC exits on the change of innovation output. As mentioned in Section 1, as far as I know, there are one theoretical paper by Schwienbacher (2008) and one empirical paper by Aggarwal and Hsu (2013). Schwienbacher (2008) is a unique theoretical paper on this topic by linking the product market and venture capital exits. The analysis takes the entrepreneur s private benefits from remaining independent in the firm after the exit into consideration,which drives the entrepreneur to attempt to differentiate his product from the competition more than would be optimal from a pure profit maximizing point of view. He could be motivated to carry out more innovative activities to increase the probability of going public. Also, when the new product is sufficiently innovative, an IPO can be more profitable than an M&A. Another important aspect of product market competition is consumer heterogeneity and the model predicts more IPOs in markets with greater consumer heterogeneity. This paper shows how the exit choice depends on product market characteristics, including both innovation and consumer heterogeneity in the industry. The empirical paper by Aggarwal and Hsu (2013) discusses two mechanisms through which the VC exit choice might impact innovation, including project selection and information 5

6 disclosure under different firm ownership. Innovation is a long process involving experimentation and brings in returns with great risks. From the project selection aspect, managers under public ownership need to report project status on a regular basis, so they may prefer to choose more certain projects yielding steady progress. Ferreira et al. (2012) shows that private ownership is optimal for situations in which managers want to incentivize exploratory innovation and Lerner et al. (2011) also finds that going private from a publicly held status improves innovation outcomes. With regard to acquisitions, Seru (2013) argues that managers in the conglomerate are less willing to fund innovative projects in the first place, because they are not able to assess the true quality of projects. From the information disclosure aspect, the decision to go public likely involves a trade-off between early liquidity and the risks of information disclosure to product market competitors. Both Manso (2011) and Ferreira et al. (2012) show that information disclosure to a broad audience under public ownership can negatively impact innovation quantity and quality by reducing the tolerance for failure. With VC-backed entrepreneurial firms in biotechnology industry, Aggarwal and Hsu (2013) show that that innovation quality is highest under private ownership and lowest under public ownership, with acquisition intermediate between the two. A second stream of related literature is about how IPOs or M&As affect innovation output, no matter whether the portfolio companies are backed by VC investment or not. Some literature shows the impact of public or private company status on its innovation outcome. For instance, Bernstein (2015) compares companies innovation output after IPO with planned but withdrawn IPO companies innovation output in multi-industry and finds that quality of innovation drops after private companies going public and quantity of innovation is unchanged. Wu (2012) collects medical device industry data and finds that after IPO, quantity of innovation rises while quality of innovation drops. Seru (2014) tests the effect of acquisition on innovation outcome. He shows that after acquisition, both quantity and quality of innovation drops, especially when the acquiring conglomerate operates an active internal capital market. In addition, different kinds of VC funding also show different innovative results. For example, Chemmanur et al (2014) argues corporate venture capital firm backed portfolio companies could generate more patents and more citation per patent. Both practitioners and researches want to find out what could bring the firm innovative. But, one thing need to remember that companies strive to improve their innovation level is not the final step in the business line. Why do they pay so much attention on innovation which is so 6

7 risky and have a large probability of failure? That s because innovation could bring them with far more return compared to its risk. Innovation has been proved to be a key driver of company performance (Bloom and Van Reenen, 2002; Nicholas, 2008; and Pastor and Veronesi, 2009). To go deeper, VCs make investment in the portfolio companies and motive them to do innovative activities not only for outstanding company performance but also for successful exits. Only if the payoff from innovation exceeds the costs would VCs make such investment and the payoff will only be realized when VC exits take place. Why does innovation affect VC exit choices? Firstly, innovative companies are crown jewels. They are shining and famous. The successful innovative products are attractive to consumers and well-known by the public. In this sense, it is easy for venture capitalists to advertise their innovative portfolio companies and the public have more information towards the business and development of them, which would make it easier for them to have successful exits, either through an IPO or M&A. Further, IPOs enhance these companies image and publicity (Roell, 1996), which in turn make innovative companies successful in the product market. From another aspect, innovative companies know how to fill market niches, beat the market and get profits from inventions. They are more capable to enter the product market and are more viable in the face of product market competition (Schwienbacher, 2008). Hence, more innovative firms are likely to go public and I predict that M&A is as middling in innovation output, with better outcomes than liquidation and worse outcomes relative to going public. Secondly, innovative companies are eager for external funding, not only to allow VC investors to divest but also to fund further innovation activities. VC firms make investment to entrepreneurial firms based on their appraisals and expectations. On the one hand, they may invest much more in innovative companies because innovative companies have intensive innovation activities and need larger investment than less innovative companies. Considering VCs relative large investment in innovative companies, it is necessary to get sufficient external funding to divest when they exit. 3 On the other hand, venture capitalists never provide more financial support than what they expect entrepreneurial firms need. Since innovative firms hold the belief of having great growth potential, they have stronger motivations towards successful exits, which could ease capital constraints due to heavy 3 Later, the correlation matrix shows a positive relationship between innovation output and total VC investment in the portfolio companies. Thus, more innovative companies are associated with larger VC investment that they received. 7

8 investigation on current innovative activities and bring them with sufficient external funding for further development. Thirdly, VCs are keen to make profits and build their reputations through successful investments. An IPO is typically taken as the most profitable exit route for venture capitalist. Gompers (1995) reports the average annual rate of return for a venture capital in the US is 60% when the exit occurs through an IPO and only 15% for a trade sale. Further, when the new product is sufficiently innovative, an IPO can be more profitable than an M&A (Schwienbacher, 2008), so highly innovative and profitable entrepreneurial firms are more likely to go public than those with more imitative products. Also, IPO exits are viewed as the most successful venture capital investments, while other exit routes tend not. Previous literature use IPO event as a measure of performance of the venture capital investment. Cumming and MacIntosh (2003a) shows that ventures with highest market-to-book ratios more often go public in both Canada and the US. Apart from these, VCs use stage financing to guarantee their successful exits. Innovation activities are risky and time-consuming, so VCs would check whether the company has reached a milestone this round before they make next round investment. And when they provide more funding into the company, they also monitor the overall progress of the company in addition to the innovation achievement. Hence, when VCs hold negative attitude to the future development of the company, they would stop financing. When they expect they will get successful exits at the end of the investment, they will continue investing to the next rounds. So more investment rounds show VCs confidence in the company and it s likely to have more successful exits. 4 Fourthly, imperfect contractibility of innovation could generate agency conflict between entrepreneurs and venture capitalists of innovative companies. Aghion and Tirole (1994) stress the fact that the exact nature of the innovation is ill-defined ex ante and that the two parties cannot contract for delivery of a specific innovation. On the one hand, entrepreneurs with significant discretionary power over the choice of business and R&D strategies to adopt, which in turn potentially affects the riskiness of the project and exit outcomes of the VC. On the other hand, entrepreneurs value private benefits from controlling the company and prefer an IPO so they are motivated to be more innovative to decrease the probability of being 4 My data also show a positive correlation between innovation output and number of rounds. Dimitrova (2015) assumes ventures that have received more investment rounds are potentially more innovative. 8

9 acquired (Schwienbacher, 2008). Thus, more innovative companies tend to choose an IPO instead of an M&A. Thus, I have the following hypothesis of my research question. Hypothesis: Portfolio companies with more pre-exit innovation output are likely to have successful exits, and among successful exits, those with more pre-exit innovation output are likely to go public instead of being acquired Measurement of Innovation To measure innovation level or innovation output is a difficult task and different measurements may stress different aspects of innovation. Scholars of technological innovation and economic growth have heavily used two streams of innovation measurement, which are patents and productivity growth (Cumming et al., 2014). The quantity and quality of patent are often taken as a direct measure of innovation. To grant patents is a way to protect new ideas and investments in innovation and creativity, so entrepreneurs apply for patents as soon as they finish developing the new technology. In general, if an entrepreneurial firm is relatively more innovative, it would have more patents. However, patented inventions could vary a lot in quality and many patents are not even introduced to markets. Whether these patents it holds are with high quality or not could not be decided only by the number of patents. A patent is supposed to be of a higher quality if it has more citations after granted. Not all innovations are patented as well. Productivity growth measures include labor productivity growth and total factor productivity growth. Labor productivity growth measure is a partial measure of the productivity growth, because when labor productivity is growing, productivity of other inputs may be going down. On the other hand, total factor productivity theoretically accounts for the productivity of all inputs. Nevertheless, it is difficult to measure all inputs, especially when quality of inputs is changing, due to the rapid development of technology, such as computers and automated machinery. Apart from the above two measures, R&D expenditure is often used as an indicator of innovative activity, while it just represents the input of innovation instead of the output and in fact these two ends of innovation are not equivalent to each other. It is not precise to say that 9

10 the more the firm input, the more it will obtain. In fact, innovation depends both on how much R&D expenditures are spent and on how efficiently they are spent. Hence, R&D expenditure is just an indirect measure of innovation, compared with the direct measures as patent and productivity growth. Practitioners may plan to have a certain amount of innovative output, so they allocate their resources according to their plan. Here in this paper, I investigate the output so I ll not use R&D as a measure. In this paper, I use four variables of patent and citations portfolio companies before VC firms exit to measure a firm s innovation output from both quantity and quality aspects. First variable is number of patents per company. I collect every company s granted patents (if any) whose application dates were between first VC investment and VC exits. Second variable is patents per year, which is number of patents (first variable) divided by time to exit (total years between the first VC investment date and VC exit date). As the time-to-exit period increases, the company probably have more patents applied in total. Therefore, I use the patents per year to measure the innovation efficiency. Third variable is number of citations, which counts total citations of all patents in the portfolio company before latest update of the patent database (before August, 2015). Fourth variable is citations per patent. This is the number of citations each patent has received before latest update of the patent database (before August, 2015) Other Factors Affecting VC Exit Choices Previous literature has covered a large range of factors affecting the exit choices, including VC firms characteristics, portfolio companies characteristics, IPO and M&A market conditions and financial contracts between VCs and portfolio companies. Next, I ll describe these factors briefly and show how I include them as control variables in my model VC Investors Characteristics VC reputation (experience) 5 Using citations as a measure of innovation quality has a drawback, because patents near the end of the data collecting period would possibly have less citations than it should have. Also, some citations are received after the company went public or was acquired, so the exit route will affect the number of citations. For instance, if a company goes public, it will have spill-over effects in the same industry. Other companies may investigate something new and cite its patent as a reference. 10

11 VC reputation is a key factor in VC exit choice and the latter in turn affects the building of VC reputation and VC s future activities.vc investors will be able to raise greater follow-on funds only if the performance of their prior funds has been successful. When experienced venture capitalists join in the syndication in the later rounds, the firm is usually doing well (Lerner, 1994b). Sorensen (2007) finds companies funded by more experienced VCs are more likely to go public primarily due to their ability to source better investments. Nahata (2008) proves that companies backed by more reputable VCs are more likely to exit successfully, access public markets faster, and have higher asset productivity at IPOs. Usually, researchers use VC firm age as VC reputation measures for simplicity. Here I have a little argument towards the use of VC firm age as a measure of VC reputation. On the one hand, Gompers (1996) verifies that younger VC firms are under great pressure to build their reputation so they take their portfolio companies public sooner to raise new funds for following investment. On the other hand, Giot and Schwienbacher (2007) shows that more experienced VC firms are associated with more IPO exits. Apart from firm age, there are still other VC reputation measures, the number of IPO firms a VC has previously backed (Lee and Wahal, 2004), capital under management (Gompers and Lerner, 2000; Krishna et al., 2011), 6 and cumulative dollar capitalization share of IPOs (Nahata, 2008). 7 Also, Hochberg, Ljungqvist, and Lu (2007) show that better-networked VCs tend to have better fund performance and the portfolio companies of better-networked VCs are more likely to have successful exit. I measure VC reputation by VC firm age and firm size. VC firm age is the difference between the founding date of the VC firm and the first investment date by this VC firm in the portfolio company. VC firm size is the total amount of capital committed by a VC firm to all the companies in its portfolio as an alternate measure of VC. VC Syndication Usually venture capital deals are associated with more than one VC investors. Syndication allows portfolio companies to have access to adequate financial funding and VCs to diversify investment risk. Syndicated VCs can help improve the performance of portfolio firms and 6 Gompers and Lerner (2000) states that the capital committed by a VC in the past could indicate its future fund-raising ability, and larger firm size increases the probability of successful exits. 7 Nahata (2008) measures VC reputation by cumulative dollar capitalization share of IPOs which consistently predicts portfolio companies performance. 11

12 facilitate successful exits through IPO or acquisition by greater complementarities of skills, connections and quality verification (Lockett and Wright, 2001; Giot and Schwienbacher, 2007). Hence, I expect a larger VC syndicate size increases the probability of exit as an IPO or M&A. Time to exit VC funds have on average limited lifespan of 10 years with possible extensions of 1 to 3 years. Hence, as a VC fund approaches the end of its life, the need to exit from its investments becomes ever more urgent. I measure time to exit as the difference between first VC investment date and exit date and it could have either positive or negative effects on the exit choices. For one thing, the longer the time to exit, the more urgent the firm has an exit through an IPO or M&A. For the other, VC firms may not have the opportunity to choose a successful exit route near the end of the VC fund and the company goes to liquidation Portfolio Company Characteristics Age of the portfolio company VCs typically invest in relatively young and unexperienced companies. As the age of the portfolio company increases, the company accumulates experiences to survive to later stage. Older portfolio companies are supposed to be more capable to face the pressure of product market and have less information asymmetry problems. This is likely to help VC exits through IPOs or M&As. Therefore, I expect the age of the portfolio company has a positive correlation with the probability of exits through an IPO or M&A. Size of the VC investment VCs invest large amounts of illiquid capital in portfolio companies in the expectation of successful exits, so VC s investment amount could fully reflect VC s confidence towards the portfolio company. In this sense, the amount of the VC investment in a portfolio company is expected to influence the decision to exit the portfolio company. Also, it is possible that portfolio companies with more innovative activities are in greater financial needs, so venture capitalists invest a great amount of capital in it. VCs with large investment have a strong motivation towards successful exits. Therefore, I would expect the probability of successful exits to increase with investment size. 12

13 2.3.3 IPO and M&A Market Conditions Since Lerner (1994a), several studies have shown that VCs time their exits, particularly the IPOs, with better market conditions. Both the IPO and M&A market conditions could affect the exit outcome as an IPO or M&A. The higher the number of IPO (or M&A) transactions, the more conducive the environment is for VCs trying to exit successfully during that period. Nahata (2008) measures two variables to proxy the state of IPO and M&A market conditions respectively. First measurement is the quarterly number of IPOs, which counts the number of IPOs over a period of 3 months prior to the portfolio company s exit. This measure is lagged by a quarter so as to allow a typical company and its investors up to 3 months to prepare for an impending exit by way of an IPO or an acquisition. Second measurement is the quarterly number of M&A deals prior to the portfolio company s exit. I follow Nahata (2008) measurements of the IPO and M&A market conditions and control for them in the regression. IPO data are from SDC Global New Issue Database and M&A data are from SDC Merge and Acquisition Database. Apart from the market condition, portfolio companies in the same industry may show similar performance. Firms in industries with high market-to-book ratios have greater growth options and could be more likely to go public (Gompers and Lerner, 1999). Firms in industries with high concentration and high intensity of product market competition could be more likely to be acquired (Bayar and Chemmanur, 2012). Also, Schwienbacher (2008) theoretically shows that consumer heterogeneity in one industry affects the product market competition and further the VC exit choices. At this stage, I don t quantitative these measures of industry s different characteristic, and instead, include industry dummies in the regression to control for the industry-specific characteristics Financial Contracts between VCs and Portfolio Companies VCs negotiate with entrepreneurs for certain control rights according to their appraisal of entrepreneurial firms quality, so the contracts between these two parts determine future VC involvement in the development and exit choice (e.g., Hellmann, 2006; Cumming, 2008). Particularly, Cumming (2008) uses survey data of 223 European VC investments to study the impact of VCs control rights on exit choice. The results show that the probability of an acquisition is approximately 30% more when VCs have effective contractual control or veto rights, such as drag-along rights, board control, and the ability to replace the founding 13

14 entrepreneur as CEO. On the contrary, weak VC control rights increases the probability of IPO exits and write-off exits. So, there is a statistically and economically significant positive correlation between acquisitions and the use of VC veto and control rights. Yet, since the contracts data are confidential and not available from any database, I could not include the control rights as control variables in my regression. I describe innovation output and control variables in my empirical analysis in Table 1, including four main categories of variables: innovation output, VC firm characteristics, portfolio company characteristics and market conditions. [Insert Table 1 here] 3 Data and Sample Selection 3.1 Baseline Sample The full sample covers portfolio companies receiving their first VC investment from 1/1/1995 to 31/12/2010 in VentureXpert. The sample starts from year 1995 for the purpose to collect portfolio companies data from Capital IQ which covers poor information on private firms before The sample ends by year 2010 because it takes on average three to five years for VCs to exit from the portfolio companies. Totally, I have VC-backed portfolio companies, including those exiting through an IPO, M&A or liquidation and those still under active VC investment. Table 2 shows the distribution of portfolio companies according to the year when they received their first VC investment (from year 1995 to 2010) in four categories: IPOs, M&As, liquidations, and active. More than half of the VC investment exit from portfolio companies before year 2004 and approximately 40% exit through M&As. Compared with IPOs and liquidations, more and more VC investment exit through M&As towards the end of my sample. [Insert Table 2 here] Among the VC-backed portfolio companies, companies have exited through one of the possible exit routes (IPO, M&A and liquidation) by 31/12/ I focus my attention towards these companies which go public, be acquired or be liquidated, and I drop 8 The date when I collected the portfolio companies information from VentureXpert. 14

15 those companies under active VC investment. For these portfolio companies, I searched Capital IQ, Factiva, Yahoo and Google to fill missing data of company founding date and exit date. For those portfolio companies which finally went public, I matched my sample with Jay Ritter s founding dates of IPO firms for correction. After deleting those without founding date or without size of VC investment, I get portfolio companies. I further remove all records where time to exit are small than 14 days or larger than 20 years, and get companies. Then, I define a lead VC in a syndication. I follow Nahata (2008) procedure to define a lead VC as the VC firm that participated in the first round and made the largest total investment in the company across all rounds of funding. Usually, the lead VC originates the deal and is among the first venture investors in the start-up. If some VC investments don t disclose their identities and there is no such VC firm starting its investment in the first round, I apply the same criteria to determine the lead VC based on identities of the second round investors. The lead VC is often the most active investor and extensively helps in monitoring, professionalizing the company, and locating follow-on VCs for syndicated investments. For very a few portfolio companies, VentureXpert doesn t provide VC firms or funds information. 9 So, after deleting firms without exit date or without VC firms information, I have 9877 portfolio companies in my sample, including 1231 IPOs, 6391 M&As, and 2255 liquidations. Table 3 shows the distribution of portfolio companies according to the year that VC exits (from year 1995 to 2010) in three exit categories. [Insert Table 3 here] 3.2 Innovation Data For my empirical analysis, I get patent and citation data from the EPO Worldwide Patent Statistical Database (PATSTAT, Autumn 2015). A new edition of the PATSTAT data is released twice a year, in Spring and Autumn. Typically, the date of data extraction from the source databases is end of January for the PATSTAT Spring Edition and end of July for the PATSTAT Autumn Edition. It is the most comprehensive patent and citation database for statistical analysis. To be specific, for US patent data, PATSTAT covers USPTO s Published Applications and Published Grants. Moreover, it includes not only patent of invention, but 9 VentureXpert uses a term undisclosed firm to show that VC firms don t provide their firm name or other information in the investment. 15

16 also design rights. 10 I could get the patent data for each company by matching this database through company names and geographical locations. I collect every company s patents whose application dates were between its first VC investment date and the exit date. I also collect those patents citations from the latest version of the patent database (Autumn 2015 version). 3.3 Methodology I estimate an unordered multinomial logit model of the probabilities of alternative VC exit routes, specifically the probabilities of VCs exiting through IPOs, M&As, and liquidation (using the liquidation outcome as the base case). The multinomial logit regression allows us to examine the likelihood of a VC choosing a specific route relative to alternative routes. The probability of a VC exiting through a specific route is given in Equation (1): P ij = P(Exit = j X it ) = Exp(X it β j ) J 1+ Exp(X it β j ) j=1, (1) where j = 1 is an IPO exit, j = 2 is an M&A exit; X it is a vector of variables that influence the probability of an exit through any route. I estimate the following two simultaneous models to examine the likelihoods of VC exits: P IPO ln ( ) = X P it β j=1 + ε IPO, Liquidation P M&A ln ( ) = X P it β j=2 + ε M&A, Liquidation I use lead VC firm age (alternatively, lead VC firm size), size of syndication, time to exit, age of portfolio company, size of VC investment in the portfolio company, quarter lagged IPO activity and M&A activity, industry dummies and year dummies as control variables. A positive β j means that the probability of choosing the exit route j increases relative to the probability of choosing the liquidation exit route. I also measure relative risk ratio (similar to odds ratio in a simple logit model) which shows the probability of a VC exiting through a 10 Design rights protect the external, visible features of the appearance of a product. The main forms of intellectual property rights established and protected by law are patent, trademarks, designs, and copyright. The PATSTAT does not cover trademarks and copyrights, so I could not use them. 16

17 given route relative the probability of a liquidation exit. The overall fit of the model is determined by a likelihood ratio test and pseudo R-square. Industry dummies include BIOTECH (biotechnology), COMMEDIA (communication and media), COMPUTER (computer related), MEDICAL (medical/health/life science), and SEMIC (semiconductor/other elect). According to VentureXpert industry classification, if the given firm belongs to one of the industries, the specified industry dummy is equal to 1. For companies not in any of the five industries, these five industry dummies are all zero. After multinomial logit regression with the three exit outcomes, I estimate a simple logit model with a subsample of IPO and M&A exits to test the probability of VCs exiting through IPO or M&A conditional on the innovation measurements and control variables. 4. Results and Analysis 4.1 Summary Statistics Table 4 describes summary statistics (columns 1 to 20) and differences tests (columns 21 to 26), including my interested innovation output variables and control variables. The table summarizes the data for all observations (columns 1 to 5), as well as the subsamples of IPOs (columns 6 to 10), M&As (columns 11 to 15) and liquidations (columns 16 to 20). Two types of difference tests, Wilcoxon tests with median and t-tests with mean, provide a number of insights into the difference between the subsamples. [Insert Table 4 here] On the firm level, the average number of patents for all the observations is 2.24 and the average number of patents per year is The average number of citations per firm is 0.32 and the average of number of citations per patent is In my sample, around 75% of the VC-backed companies have no granted patent before exits, so the average number could not represent the average innovation level of all the observations. The firm with the maximum number of patents (369) exited through IPO. The maximum patents per firm in M&As is 294 and maximum patents per firm in liquidations is 113. According to univariate tests, portfolio companies through successful exits (IPOs and M&As) have significantly more patents, more patents per year, more citations and more citations per patent than unsuccessful exits 17

18 (liquidations). Also, portfolio companies exiting through IPOs have significantly more innovation output than M&As. For the control variables, portfolio companies through successful exits are backed by more experienced VC firms (older and larger VC firms) compared to those through unsuccessful exits. The former have larger syndications and longer time to exit and receive more investment from VC firms and more rounds of investment. Also, portfolio companies through successful exits experience more IPO activity and more M&A activity one quarter before exits. Compared with M&As, portfolio companies through IPOs are backed by more experienced VC firms (older and larger VC firms). They have larger syndications and less time to exit, receive more investment from VC firms and more rounds of investment, and experience more IPO activity and more M&A activity one quarter before exits. Table 5 is the correlation matrix across exit outcomes, innovation output, VC investors and portfolio companies characteristics, and market conditions. Significant correlations at 5% level appear in boldface. Table 5 shows similar relationship between exits routes and other variables as what have shown in the univariate tests. One implication of Tables 5 is that portfolio companies with more pre-exit patents have more other innovation output and receive more investment from more experienced VCs and larger syndications. Also, patents per year has a positive correlation with time to exit, which show that companies with more patents are not due to longer time before exits. [Insert Table 5 here] 4.2 Multinomial Logit Regression In this section, I describe my regression models specifications and results. Table 6 reports four multinomial logit regression models of the impact of VC firms and portfolio companies characteristics and market conditions on exit outcomes (base case is the liquidation). In addition to the control variables, I also control for the industry and exit year with dummies. I present both the estimated coefficients (with p-value in the parenthesis below) and relative risk ratios from the multinomial logit estimates to illustrate explicitly economic significance as well as statistical significance. I do not present the marginal effects 18

19 here for the reason that marginal effects on mean or median could not represent any observations in the sample. For example, I have five industry dummies, and the means of these five dummies are all greater than zero and smaller than one. Again, the observation at the median of one variable is not the observation at the median of other variables. Instead, I report the relative risk ratios here to show the probability of a VC exiting through a given route relative to the probability of a liquidation exit. [Insert Table 6 here] The results show that portfolio companies with successful exits are associated with more preexit innovation output. All of the four interested variables are significantly positive for both IPOs and M&A (except for citation per patent for M&A). Take panel (1) as an example. The estimated coefficient on Log(Patent) is significantly positive at 1% level and the RRR is greater than one. 11 To be specific, one unit increase in Log(Patent) (equivalent to % increase in number of patents) will results in 38.6% higher probability of exiting through an IPO than the probability of an exit through liquidation and 9.9% higher probability of exiting through an M&A than the probability of an exit through liquidation. The control variables also show their importance in the model. More syndicated investments are associated with successful exits. If the portfolio companies are older when they receive their first VC investment, the probability to have successful exits is higher. Larger VC investment in the portfolio company also increases the probability of successful exits. IPOs have shorter period between first VC investment and exit and M&As have longer period between first VC investment and exit. Although active IPO market one quarter before exit have positive impact on successful exits, active M&A market condition have no significant impact. The models are significant at 1% level and have pseudo 2 R s around Table 7 reports four logit regression models of the impact of VC firms and portfolio companies characteristics and market conditions on comparisons between IPOs and M&As. The results show that more innovation output increase the probability of exits through IPOs. Based on odds ratio from logit regression estimates, one unit increase in Log(Patent) (equivalent to % increase in number of patents) will results in 25.6% higher probability of exiting through an IPO than the probability of an exit through an M&A. Larger 11 To take the log of patent data, ages and sizes aims to make them more normally distributed so as to do the regression, like Aggarwal and Hsu (2013). 19

20 VC syndications, less time to exit, older portfolio companies, larger VC investment in the portfolio companies and active IPO market are all associated with higher probability of exits through IPOs. The models are significant at 1% level and have pseudo 2 R s greater than [Insert Table 7 here] 5. Robustness Tests In this section, I report the results of several robustness tests that further verify the positive effect of innovation on VC exits. I first report the results of the tests that help mitigate endogeneity concerns about innovation output of portfolio companies, and next discuss the results of the additional tests (1) with inclusion of product market competition, (2) with inclusion of portfolio companies investment stages when they received their first VC investment, (3) with samples including both granted and non-granted patent publications, (4) with subsamples including portfolio companies that VC exited before the end of 2012, and (5) with subsamples including observations with at least one patent. 5.1 Endogeneity Concerns about Portfolio Companies Innovation Output One major concern that could undermine the importance of the findings reported so far is the potential endogeneity of portfolio companies innovation output, that is, successful companies with better entrepreneurs are at the same time more likely to generate more innovation output and have successful VC exits. The innovation activity could be endogenously chosen and could be related to unobservable factors that also determine VC exit choices, which leads to correlation between innovation and VC exits. For instance, portfolio firms with successful business line and performance have more incentive to carry out innovation activity to enlarge their business and simultaneously their VC investors are more likely to lead them to IPO or sell them to an acquirer. From another side, VCs have preplanned exits when they make their investment in the portfolio companies, and the preplanned exit choices would affect the portfolio companies attitude towards innovation activities and this causes a reverse causality. I overcome the empirical identification challenge of finding plausibly exogenous sources of variation in innovation following Falato and Sim (2014) which provides detailed changes of R&D tax credits across states in the US. From year 1981 to year 2011, nearly every U.S. state has introduced some type of tax incentive for R&D and the introduction of R&D tax credit 20

21 could be an exogenous time-series factor increasing firms incentives to innovate. 12 Hall and Van Reenen (2000) did a survey to report the evidence on the effectiveness of tax incentives for R&D and Falato and Sim (2014) also show that R&D expenses respond strongly to changes in state R&D tax credits. Hence, portfolio companies in one state where there is R&D tax credit should generate more patents than those companies in that specific state before an introduction of state R&D tax credit, or than those companies in states where there is no R&D tax credit. I take the implementation of state R&D tax credits as a dummy variable and use it as an instrumental variable of innovation output. And further, the state R&D tax credit should serve as a valid instrument for innovation output because the state-level R&D policy should not related to VC exit choice other than through its effect on innovation activities. [Insert Table 8 here] The results for the 2-stage least square regressions are presented in Table 8. I use the instrument to predict four innovation output variables (number of patent, patent per year, citation, and citation per patent, from panel (1) to (4)) in the first stage. The results from the first stage regression show that the instrument variable has significant positive impacts on innovation output of portfolio firms as our expectation. The second stage regression uses predicted innovation outputs in multinomial logit model as our baseline regression, including state dummies to control for the state fixed effect. The results from the second stage regressions confirm the finding that more innovative firms are likely to go public or be acquired instead of be liquidated. To conclude, after controlling for endogeneity, portfolio firms innovation outputs are still significantly positive correlated to successful VC exits. 5.2 Additional Robustness Tests As discussed in Section 2.3, the intensity of product market competition would affect the exit choices (Bayar and Chemmanur, 2012) and VC-backed private firms are less likely to go public in an industry with high intensity competition. Hence, it is necessary to control for industry-specific product market characteristics in the regressions. Following Bayar and 12 Falato and Sim (2014) provides a list of changes in state R&D tax credit. For my sample period from year 1995 to 2010 when firms received their first VC investment, I only take the introduction of state R&D tax credit during this period into consideration and regard those companies in states which implemented the policy before year 1995 as treated observations. 21