A Comparison of Corporate Innovation Strategies in Public and Private Firms *

Transcription

1 A Comparison of Corporate Innovation Strategies in Public and Private Firms * Huasheng Gao Nanyang Business School Nanyang Technological University S3-B1A-06, 50 Nanyang Avenue, Singapore hsgao@ntu.edu.sg Po-Hsuan Hsu Faculty of Business and Economics University of Hong Kong Pokfulam Road, Hong Kong paulhsu@hku.hk Kai Li Sauder School of Business University of British Columbia 2053 Main Mall, Vancouver, BC V6T 1Z kai.li@sauder.ubc.ca First version: August 2013 This version: August 2013 Abstract We provide a large sample comparison of public and private firm innovations to gain insight into determinants of corporate innovation strategies. We show that public firms have significantly more patents and citations than private peer firms. Consistent with Ferreira, Manso, and Silva (2013), we show that public firms patents are more exploitative with more depth (i.e., exploiting existing knowledge), while private firms patents are more exploratory with more scope (i.e., exploring new ideas). The results based on a sample of IPO firms are different from those based on the cross-section: After going public, a firm s patent and citation counts significantly decline. Our main findings are robust to accounting for endogeneity and truncation biases in measures of innovation. Contrary to popular belief, we conclude that being publicly-listed does not stifle innovation. Keywords: exploitative; exploratory; innovation; public firms; private firms; innovation depth; innovation scope JEL Classification: G31; G32 * We are grateful for helpful comments from Andres Almazan, Heitor Almeida, Geert Bekaert, Chuan Yang Hwang, Alexander Ljunqgvist, and Gustavo Manso, and the Brown Bag seminar participants at the University of Hong Kong. We acknowledge dedicated effort of our team of research assistants Rita Lei Chen, Yi Du, Cherry Guo, Jialing Hu, and Chi-Yang Tsou. Gao and Li acknowledge the financial support from the Social Sciences and Humanities Research Council of Canada. All errors are our own.

2 1. Introduction It is well established that innovation is the engine of economic growth and prosperity (Solow (1957) and Romer (1990)) and a key factor in competitive advantages (Porter (1990)). Several recent studies show that successful innovation also raises firm profitability and market value (see, for example, Hall, Jaffe, and Trajtenberg (2005a), Hirshleifer, Hsu, and Li (2013), and Cohen, Diether, and Malloy (2013)). However, almost everything we know about corporate innovation activities at the micro level is based on evidence from public (i.e., stock-market listed) firms that number only in the thousands. Privately-held (i.e., unlisted) firms, number in the millions, yet there has been a scarcity of data and research on corporate innovation activities within this group. This paper aims to fill a gap in current research by enhancing our understanding of how private firms innovate, and adds fresh evidence to the debate on whether and how being publicly-listed affects innovation incentives and outcomes. Anecdotal evidence seems to suggest that firms that stay privately-held can be more innovative. On February 5, 2013, CEO of Dell Computer, Michael Dell wrote to all Dell employees on his going private decision: Today, we announced a definitive agreement for me and global technology investment firm Silver Lake to acquire Dell and take it private I believe that we are better served with partners who will provide long-term support to help Dell innovate and accelerate the company s transformation strategy. 1 In 2004, when Google went for its initial public offering (IPO), Google adopted a dual class share structure whereby the founders and the CEO held super-voting shares while outside investors hold shares with one-share one-vote. To rationalize this arrangement, in Google s S-1 registration statement, founders Larry Page and Sergey Brin stated: Therefore, we have implemented a corporate structure that is designed to protect Google s ability to innovate and retain its most distinctive characteristics. 2 Recent empirical studies also seem to suggest that firms that stay privately-held are more innovative. Using 472 leveraged buyout (LBO) target firms between 1986 and 2005, Lerner, Sørensen, and Strömberg (2011) show that patents of LBO target firms are more frequently cited. These firms show no deterioration in patent originality and generality. The level of patenting does not appear to consistently 1 See the full press release Dell Enters Into Agreement to Be Acquired By Michael Dell and Silver Lake at: 2 See the full version of An Owner s Manual for Google's shareholders at: 1

3 change, and the firms patent portfolios become more focused. They conclude that LBOs do not sacrifice target firms long-term investments as previously thought. It is worth noting that only 64 LBO cases in their sample involve a public-to-private ownership transition. Using 1,478 completed and 323 withdrawn IPO filings over the period , Bernstein (2012) examines the effect of going public on corporate innovation. He finds that IPOs change the nature of projects pursued by the newly public firms as internal innovation becomes less novel and relies on a narrower set of technologies. Further, going public is associated with substantial employee turnover and newly public firms rely more heavily on the acquisition of external technologies. It is worth noting that Bernstein s sample firms are private firms in transition to become publicly-listed. On the other hand, going public may foster more innovation. Since the beginning of the Internet era in the early 1990s, there have been over 600 Internet-related companies gone public starting with Cisco in 1990 to Facebook in Over a more extended period , close to 40% of the IPOs are completed by high-tech firms. 3 The large number of innovative firms in the Internet and high-tech sectors seeking public firm status does not square well with the view that being publicly-listed stifles innovation. Policy makers also take note. In the hope that the IPO market stimulation will jumpstart innovation and job creation, President Obama s Council on Jobs and Competitiveness has urged Congress to amend the Sarbanes-Oxley Act to allow small companies to tap public equity markets (Bernstein (2012)). A number of studies also suggest that going public may foster innovation. Access to public equity markets allows investors to diversify their risk and to share upside potential thus encouraging corporate innovative activities (King and Levine (1993) and Levine (2005)). Brown, Fazzari, and Petersen (2009) show that booms (or busts) in the supply of (external) equity financing to young high-tech firms lead to booms (or busts) in aggregate R&D expenditures. Using international data, Brown, Martinsson, and Petersen (2013) and Hsu, Tian, and Xu (2013) show that public firms in countries with better equity market development tend to invest more in R&D. Using a matched sample of 15,128 private firm-year observations (representing 5,173 unique private firms) to 15,128 public firm-year observations (representing 4,063 unique public firms) from 3 See Professor Jay Ritter s IPO data at: 2

4 , and different measures of corporate innovative activities, we show that public firms have significantly more patents and citations than their private peer firms. The results based on a sample of IPO firms are different from those based on the cross-section: After going public, a firm s patent and citation counts significantly decrease compared to itself prior to the initial public offering. Consistent with Ferreira, Manso, and Silva (2013), we show that public firms patents are more exploitative with more depth (i.e., exploiting existing knowledge), while private firms patents are more exploratory with more scope (i.e., exploring new ideas). Our main findings are robust to accounting for endogeneity and truncation biases in innovation measures. Contrary to popular belief, we conclude that being publicly-listed does not necessarily stifle innovation. Our paper makes contribution to the literature in the following dimensions. First, this is one of the first studies providing large sample evidence on the innovation activities of U.S. privately-held firms. Most studies to-date on private firm innovation employ only a small sample of transitioning firms such as LBO firms or IPO firms. 4 Second, our large sample comparison of innovation activities in public and private firms contributes to the current debate on whether being publicly listed stifles innovation. The findings from our paper have important policy implications with respect to disclosure and IPO regulations. Finally, our measures of corporate innovation activities are multi-faceted, beyond the usual patent and citation counts, but also include exploitative versus exploratory innovation and depth versus scope of innovation. As a result, we gain a better understanding of corporate innovation strategies from different perspectives. The paper is organized as follows. Section 2 reviews the related literature and develops our hypotheses. Section 3 describes the data, our sample, and key variable construction. Section 4 presents the main empirical analysis. Sample selection issues are addressed in Section 5. We conclude in Section Literature Review and Hypothesis Development 2.1. Related Literature 4 The exception is Barrot (2012) who shows that private equity investors horizon plays an important role in innovation activities of private firms based on a sample of 3,241 firms. 3

5 A number of papers have examined the relation between going public/disclosure and corporate innovation. Bhattacharya and Ritter (1983) show that innovative firms may lose their informational advantage if they disclose details of their R&D projects to capital markets in order to raise financing at better terms. Incorporating similar trade-offs, Maksimovic and Pichler (2001) further demonstrate that both technological and competitive risks affect the timing of private and initial public offerings in new industries: Early private financing occurs in industries that are perceived to be risky, with high development costs and low probability of being displaced by technologically superior rivals, while early public financing occurs in industries perceived to be viable, with low development costs and low probability of displacement. Another strand of the literature explores the role of financing frictions in corporate innovation. Brown, Fazzari, and Petersen (2009) show that supply shifts in external equity financing can explain a significant portion of the 1990s R&D boom and decline, especially among young firms. Similarly, Ball, Hoberg, and Maksimovic (2012) show that financial constraints directly affect corporate R&D activities. Further, the main friction that firms face is raising equity capital to fund growth opportunities. In an international study, Hsu, Tian, and Xu (2013) find that more developed equity markets encourage innovation in high-tech industries while more developed debt markets discourage innovation in those industries. Finally, several recent studies focus on comparing differences in innovation strategies and outcome between public and private firms. Ferreira, Manso, and Silva (2012) show that it is optimal to go public to exploit existing ideas and optimal to stay private to explore new ideas because private firms are less transparent than public firms and with an IPO option to timely exit from failures, private firms are more willing to innovate outside their existing knowledge. Lerner, Sørensen, and Strömberg (2011) find that post-lbo, target firms pursue more influential innovations (i.e., patents with more citations) and display no change in patent counts. Bernstein (2012) finds that the quality of internally-generated innovation declines in post-ipo firms relative to firms that remain private. He attributes the decline to both an exodus of skilled inventors and a drop in productivity among remaining inventors Hypothesis Development 4

6 Being publicly-listed or privately-held leads to a number of important differences. Private firms are subject to more financing frictions than public firms do (Brav (2009) and Gao, Harford, and Li (2013)). More (equity-financed) capital allows public firms to engage in innovative projects. In contrast, private firms mainly rely on debt financing (Brav (2009)), which typically does not encourage innovative behavior. Public firms are mostly owned by institutional investors including mutual funds, pension funds, and alike. Aghion, Van Reenen, and Zingales (2013) show that greater institutional ownership is associated with more innovation. They further demonstrate the underlying mechanism of this causal relationship is that institutional owners increase innovation incentives through reducing public firm managers career risks: CEOs are less likely to be replaced in the face of profit downturns when institutional ownership is higher. Public firms are more likely to employ high-powered managerial incentives. Gao, Lemmon, and Li (2013) show that compared to private firms, public firms tend to use more equity-based compensation and exhibit stronger CEO pay-to-performance sensitivity. Ederer and Manso (2013) provide evidence suggesting that incentive plans that tolerate early failure and reward long-term success lead to more innovation. The above discussions lead to our first null hypothesis regarding levels of innovation in public and private firms: H1 0 : Private firms innovate less than public firms do. On the other hand, due to a lack of stock liquidity and analyst coverage, private firm managers are not under any short-term pressure, thus are willing to engage in long-term investments in R&D and innovation (Lerner, Sørensen, and Strömberg (2011), Fang, Tian, and Tice (2013), and He and Tian (2013)). Further, private firms operate in an environment with little analyst and financial media coverage, which can be beneficial to risky investment like R&D. Finally, firms with greater information asymmetries have stronger incentives to use patenting activities to signal their quality to potential investors (Hsu and Ziedonis (2012) and Cockburn and MacGarvie (2009)). Thus, everything else being 5

7 equal, private firms are more motivated to patent their inventions than public firms. This discussion leads to an alternative hypothesis regarding levels of innovation in public and private firms: H1 A : Private firms innovate more than public firms do. Next, we discuss other differences between public and private firms that have implications on the nature of innovation. Ferreira, Manso, and Silva (2012) point out that private firms are less transparent to outside investors than are public firms, leading to the tolerance-for-failure effect as the key determinant of exploratory innovation in private companies. In contrast, the rapid incorporation of good news into market prices creates incentives for short-termism behavior by public firms. Longer investment horizon of private firm shareholders reinforces the positive effect of low transparency on exploratory innovation. Our second hypothesis regarding the nature of innovation in public and private firms is thus: H2: Private firms innovation is more exploratory with greater scope that public firms does. Our empirical analyses are designed to test these hypotheses. In contrast to prior work that examines private firms in small numbers or firms that are in transition, we contribute to the debate on whether and how public firm status affects corporate innovation by providing some of the first largesample comparisons of innovation in public and private peer firms. In the next section we describe our sample and key variable construction, and present descriptive statistics. 3. Sample Formation, Variable Construction, and Overview Our primary data sources are the Capital IQ database and the USPTO patent database. Capital IQ is an affiliate of Standard & Poor s which produces the Compustat database. 5 We start with a list of unique U.S. firms covered in Capital IQ for the period Our sample starts in 1995 because the data coverage in Capital IQ is poor before We end up with an initial sample of 63,597 public firm- 5 Starting from the mid-1990s, Capital IQ provides data on some private U.S. firms. When available, Capital IQ provides data on firm accounting information with a similar level of detail as provided by Compustat for public U.S. firms. 6

8 year observations and 255,638 private firm-year observations. We obtain the patent and citation data of these firms by manually matching the patent database (by firm names and locations) with the Capital IQ database. We further require that all sample firms have basic financial information. To mitigate the reverse causality concern that innovation drives going public/ private decisions (rather than the public/private status drives innovation), we remove all firm-year observations associated with going-public (1,669 IPOs) or going-private (122) transactions. In other words, our sample private firms always stay private and our sample public firms always stay public. 6 Public firms are those traded on the NYSE, AMEX, or NASDAQ. We end up with 18,201 private firm-year observations (6,520 unique private firms) and 48,461 public firm-year observations (6,823 unique public firms). Prior work has shown that patenting activities tend to vary systematically by industry and larger firms tend to have more patents due to economies of scale (Hall, Jaffe, and Trajtenberg (2005b) and Ciftci and Cready (2011)). 7 We match each private firm with a public firm in the same industry, year, and closest in total assets (with replacement) with an additional requirement that the total assets of matched public firm must be within [75%, 125%] of that of a private firm. The matching (with replacement) procedure helps mitigate the large difference in the size distribution between the two samples and the smaller, but potentially important difference in sample firm distribution across industries. 8 In our final sample, we have 15,128 private firm-year observations representing 5,173 unique private firms and 15,128 matched public firm-year observations representing 4,063 unique public firms for the period Measures of Innovation Patent count 6 It is worth noting that when including those going-public/going-private firms in our analyses, the main findings remain unchanged. 7 Public firms are typically larger firms than private firms. Ciftci and Cready (2011) show that R&D productivity increases with scale as captured by firm size. 8 Based on the sample of public and private firms without matching, our main results still hold and are even stronger because public firms on average are much larger and have more resources in supporting innovative activities. 7

9 Our first measure is patent count, which is the number of granted patents filed by firm i in year t. This measure has been widely used in the literature to capture the success of a firm s innovation activities. We use the application year (instead of the grant year) to better capture the exact timing of the underlying innovation activities behind a patent. We set patent count to zero for firm-year observations without any patent record. Assessment of the innovation output of a given firm using patent count can only be made relative to some benchmark, for the following reasons. First, technology classes differ in the nature of R&D activities and resources required to produce patentable innovation such that patent counts in two different classes may not be directly comparable. Second, there are technology class-specific time trends in the number of awarded patents that may not fully reflect changes in innovation output. In particular, large increases in the number of awarded patents in some classes over time might reflect the evolution of USPTO practices with respect to what is patentable innovation, and hence patent counts from different years may not be time-consistent measures of innovation output even within the same technology class. We address both issues by summing up patent counts using the scaled number of patents, where we divide the number of granted patents filed in technology class k and year t by the average number of granted patents filed by public (both public and private) firms in the same technology class and year, following Bena and Li (2013). 9 Our adjusted patent count thus accounts for differences across technology classes and differences in USPTO practices over time. The details of three variables we construct, Patent, Patent All, and Patent Pub, are provided in Appendix Citation count Our second measure is citations count, which is the number of citations received by granted patents filed by firm i in year t. This measure is often referred to as citation-weighted patent count in the literature, capturing the impact of corporate innovation and its associated economic value (e.g., Trajtenberg (1990) and Hall, Jaffe, and Trajtenberg (2005b)). 9 This approach is similar to but different from the adjustment used by Bena and Li (2013) because the sample firms we use in this study include both public and private firms. There are over 400 technology classes reported in the USPTO: 8

10 We need to make some adjustment for citation counts because patents continue to receive citations over a long period of time, while we observe at best only up to the last year of the available data (in our case the year of 2010). As a result, patents granted in different years suffer to different extent from this truncation bias in the number of citations received, and their citation counts are not comparable and cannot be aggregated. Further, patents granted in different technological classes are also cited in different ways, which needs to be accounted for before aggregating citation counts. Following Lerner et al (2011), Seru (2011), and Bena and Li (2013), we make the following adjustments to obtain citation count for each sample firm. First, for each patent, we count the number of citations it receives in the grant and subsequent three years. The three-year window is used to mitigate the truncation bias inherent in citation counts. Second, for each patent, we divide its raw citation count in step 1 by the average number of citations received by all approved patents in the same technology class and in the same grant year over the same three-year period by public (both public and private) firms. This adjustment addresses differences in the number of citations received by patents in different technologies and mitigates potential truncation biases. Finally, citation count is to sum up the adjusted citation count for each patent filed in year t over different technological classes. 10 The details of three variables we construct, Citation, Citation All, and Citation Pub, are provided in Appendix Exploitative (depth) versus exploratory (scope) Different from most prior studies on corporate innovation, we also employ novel measures of innovation that capture the nature of innovation exploitative versus exploratory and depth versus scope. Following Benner and Tushman (2002), we use a firm s existing knowledge base the combination of its existing portfolio of patents and citations made by its existing patents over the past five 10 An alternative approach to addressing truncation bias in citations is developed by Hall et al. (2005a), who estimates the shape of the citation-lag distribution for each of the six USPTO technology categories (Chemical, Computers and Communications, Drugs and Medical, Electrical and Electronic, Mechanical, and Others), i.e., the fraction of lifetime citations (defined as the 30 years after the grant date) that are received in each year after the grant year. Assuming that this distribution is stationary and independent of overall citation intensity within each technology category, they estimate the adjusted citations of each patent by dividing the number of citations by the fraction of the population distribution that lies in the time interval for which citations should have been observed. Then, they sum up the adjusted citation counts. It is worth noting that our main findings remain unchanged if we use their adjustment for citation counts. 9

11 years to define the nature of its innovative effort. Exploitative innovation deepens the existing knowledge, while exploratory innovation expands the existing knowledge. A patent is categorized as exploitative if more than 80% of its citations are based on a firm s existing knowledge base (as defined above); while a patent is categorized as exploratory if more than 80% of its citations are based on new knowledge outside a firm s existing expertise (i.e., 80% of the citations are not citing existing patents or cites). At the firm level, our first measure of the nature of innovation, exploitative ratio, is the number of firm i s exploitative patents filed in year t divided by its patent counts filed in the same year. We label this proxy as Exploit. The exploitative ratio ranges between zero and one, and a higher exploitative ratio suggests that innovation effort focuses on deepening a firm s technological expertise by following its current innovation trajectories and developing patents that extend its competitive advantage in existing territories. Our second measure, exploratory ratio, is the number of firm i s exploratory patents filed in year t divided by its patent counts filed in the same year. We label this proxy as Explore. This variable also ranges between zero and one, and a higher exploratory ratio suggests that innovation effort focuses on expanding a firm s technological expertise by deviating from its current innovation trajectories and creating patents that may lead to new competitive advantages in totally different territories. We also employ two related alternative measures to capture the nature of innovation, following Katila and Ahuja (2002). Innovation depth intends to capture the repetitive use of existing knowledge: The more often a firm has used its existing knowledge, the more deeply it understands existing knowledge. A firm s existing knowledge base consists of citations made by its existing patents over the past five years. Our first alternative measure, depth ratio for firm i filing in year t, is the number of repeated citations made by patents filed in year t to firm i s knowledge base divided by the number of total citations made by patents filed in year t. Repeated citations are citations that have been made in the past five years. We label this proxy as Depth. The depth ratio is equal to or greater than zero, 11 and a 11 The depth measure could be above one because some current citations may have been made more than once in the past. For example, if a firm has only one patent that is filed in this year and only cites one prior patent, and if this citation has been made by two patents filed by the firm in the past, then the depth measure is two for this firm in this year. 10

12 higher ratio indicates that a firm revisits its existing knowledge base and thus may have deep understanding of its current technology territories. Innovation scope intends to capture the frequency a firm explores new knowledge outside of its existing expertise: The more often a firm has used new knowledge, the broader knowledge it acquires. Our second alternative measure, scope ratio for firm i filing in year t, is the number of new citations made by patents filed in year t outside firm i s knowledge base divided by the number of total citations made by patents filed in year t. We label this proxy as Scope. New citations are citations that have never been made in the past five years. The scope ratio ranges between zero and one, and a higher ratio indicates that a firm acquires more new knowledge in building its patent portfolio and thus may have a better chance to achieve new competitive advantages. The details of three variables we construct, Exploit, Explore, Depth, and Scope are provided in Appendix Summary Statistics Table 1 provides time-series distribution of our sample that consists of 15,128 public firm-year observations and 15,128 private firm-year observations. We show that other than the first year of the sample period, there is an increasing trend in Capital IQ s coverage of private firms. After 2000, the coverage experiences a slow decline. Nonetheless, over the entire sample period, our sample has over 700 private firms and their matching public firms each year. The median value of total assets of public and private firms in our sample is 376 million and 374 million, respectively. Private firms slowly increase their size over time. Importantly, about 13% of public firms have awarded patents, while only about 6% of private firms have awarded patents. The univariate statistics are suggestive of public firms being more innovative than private firms. We further note that starting from 2008, there is a decline in the fraction of public (private) firms with awarded patents, probably because of the time lag for a filed patent to be officially approved, which is on average two to three years (Hall et al. (2005b)). Table 2 provides summary statistics for our sample firms. Definitions of all variables are provided in Appendix 1. All dollar values are in 2010 dollars. All continuous variables are winsorized at the 1 st and 99 th percentiles. 11

13 Panel A compares innovation measures between public and private firms. We first show that public firms are more innovative in terms of both quantity and quality of innovation output. On average, public firms produce 0.82 patents per year, while private firms produce 0.28 patents per year. Similarly, the mean patent count (adjusted by public firm averages) in public firms is 0.25, while the mean in private firms is Using adjustment based on all firm averages, gives similar results. The two-sample tests of differences reject the null that public and private firms have the same level of innovation quantity at the 1% level. The mean HJT-adjusted citation counts show that on average, public firms patents receive 8.79 citations, while private firms patents receive 2.96 citations. Similarly, the mean citation count (adjusted by public firm averages) in public firms is 0.80, while the mean in private firms is Using adjustment based on all firm averages, gives similar results. The difference is statistically significant at the 1% level in both t-tests and Wilcoxon tests, rejecting the null that public and private firms have the same level of innovation quality. When examining the nature of innovation in public and private firms, we show that on average, 20% of public firms patents are exploitative, while 16% of private firms patents are exploitative. Such a difference is statistically significant at better than the 5% level, indicating that public firms innovation is more exploitative than private firms. In contrast, on average, 57% of public firms patents are exploratory, while 63% of private firms patents are exploratory. Again such a difference is statistically significant at the 1% level, suggesting that private firms focus more on exploring new technological territories than public firms do. Using alternative measures to capture the nature of innovation, we show that the average depth ratio in public firms is 102%, while the average depth ratio in private firms is 76%. The difference is statistically significant at the 1% level. In contrast, the average scope ratio in public firms is 66%, while the average scope ratio in private firms is 72%. The difference is statistically significant at the 1% level. These two measures suggest that private firms patents are more novel while public firms patents are more in depth. Overall, the four measures of the nature of innovation show that public and private firms have distinct innovation strategies: Private firms explore new knowledge, while public firms exploit existing knowledge. 12

14 Panel B compares firm characteristics between public and private firms. 12 We show that our matching scheme worked well that public and private firms in our sample are similar in size. Further, we show that on average, public firms have significantly lower leverage than private firms, consistent with the common belief that public firms have better access to equity financing and thus rely less on debt financing. Finally, public firms have significantly higher fixed assets as measured by property, plant and equipment (PPE), and are significantly older than private firms. 4. Main Results 4.1. Innovation Output To compare innovation strategies in public and private firms and to test our hypotheses, we run the following panel data regression (see, for example, Aghion, Van Reenen, Zingales (2013) and Fang, Tian, and Tice (2013)): Ln(1 + Innovation Output) = α + β 1 Public + β 2 Ln(Total assets) + β 3 Leverage + β 4 CF + β 5 Capex + β 6 R&D + β 7 Acquisition + β 8 PPE + β 9 Ln(Firm age) + Industry Year FEs + ε, (1) where Ln(Innovation level) denotes the logarithmic value of one plus patent count or citation count (defined in Section 3.2 and Appendix 1). We take the logarithmic transformation because patent counts and citation counts are often zero and highly skewed. Public is an indicator variable that takes the value of one if the firm is a public firm in that year, and zero otherwise. We include industry times year fixed effects to control for unobserved industry- and year-specific heterogeneity such as competitive pressure and industry-specific business cycles. The coefficient on the Public indicator variable thus measures the difference in innovation output between public and private firms that cannot be accounted for by differences in firm characteristics and industry and year effects. Table 3 presents the regression results when the dependent variable is measures of innovation quantity. In column (1), when we use the simple patent count as the dependent variable (Ln (1+ Patent)), 12 Note that none of the correlations among firm characteristics is high enough to warrant multi-collinearity concerns for our multivariate analyses. 13

15 the coefficient on the Public indicator variable is and significant at the 1% level, suggesting that public firms produce 9.5% (e ) more patents than private firms do. In column (2) ((3)), when the dependent variable is adjusted patent counts using different benchmark firms (public firms only versus full sample of both public and private firms), the coefficient on the Public indicator variable remains positive and significant at the 1% level. In terms of the economic significance, we show that compared to private firms, public firms produce 4.9% (e ) to 5.1% (e ) more patents. In brief, public firms produce about 5% to 9% more patents than similar private firms, supporting our hypothesis H1 0. Table 3 also identifies other factors that are related to corporate innovation quantity. We show that firm size is positively related to innovation quantity, as is operating cash flows. Not surprisingly, firms with greater R&D expenditures produce more patents, and older firms produce more patents. On the other hand, leverage, acquisition expenditures, and PPE are all negatively related to innovation quantity. All these findings are consistent with prior studies; see, for example, Ciftci and Cready (2011), Aghion, Van Reenen, Zingales (2013), and Fang, Tian, and Tice (2013)). Table 4 presents the regression results when the dependent variable is measures of innovation quantity. In column (1), when we use the HJT-adjusted citation count as the dependent variable (Ln (1+ Citation)), the coefficient on the Public indicator variable is and significant at the 1% level, suggesting that public firms patents receive 16.3% (e ) more citations than private firms do. In column (2) ((3)), when the dependent variable is adjusted citation counts using different benchmark firms (public firms only versus full sample of both public and private firms), the coefficient on the Public indicator variable remains positive and significant at the 1% level. In terms of the economic significance, we show that compared to private firms, public firms patents receive 7.5% (e ) more citations. In brief, public firms produce more impactful patents with about 8% to 16% more citations than similar private firms, again supporting our hypothesis H1 0. Table 4 also identifies other factors that are related to corporate innovation quality. We show that large old firms are more likely to produce high impact innovation. Further, firms with greater expenditure in capex and R&D are also associated with higher impact innovation, while firms with high leverage, acquisition spending, and tangible assets are associated with lower impact innovation. All these findings 14

16 are consistent with prior studies; see, for example, Aghion, Van Reenen, Zingales (2013) and He and Tian (2013)). In summary, the evidence in Tables 3 and 4 collectively provide support to our hypothesis H1 0 that private firms innovate less than public firms do, possibly due to different degrees of financial constraint and resulting financing mix, different ownership composition, and different equity incentives to top management in these two groups of firms. Next, we further examine whether and how public firm status impacts corporate innovation strategies as predicted by our hypothesis H Innovation Strategies So far, we have shown that public firms on average are more innovative than private firms in terms of both innovation quantity and quality, it is premature to rave about public ownership given that there are many more private firms than public firms contributing significantly to the US economy (Farre- Mensa (2012)). We need to take a careful look at these two groups of firms innovation strategies as captured by the exploitative versus exploratory nature and innovation depth versus innovation scope. In an influential paper, Ferreira, Manso, and Silva (2013) highlight the importance differences in public and private firms incentives to innovate. Private firms are less transparent to outside investors than are public firms. Further, private firms can time the market by choosing an early exit strategy if their innovative effort fails. The combination of these two effects makes private firms more tolerant of failures and thus more inclined to invest in innovative projects than public firms do. To empirically test their model and our hypothesis H2, we run the following panel data regression: Ln(1 + Innovation Strategy) = α + β 1 Public + β 2 Ln(Total assets) + β 3 Leverage + β 4 CF + β 5 Capex + β 6 R&D + β 7 Acquisition + β 8 PPE + β 9 Ln(Firm age) + Industry Year FEs + ε, (2) where Ln(Innovation Strategy) denotes the logarithmic value of one plus one of the four measures for innovation strategies the exploitative ratio, the exploratory ratio, innovation scope, and innovation depth (defined in Section 3.2 and Appendix 1). We take the logarithmic transformation because patent counts 15

17 and citation counts are often zero and highly skewed. The coefficient on the Public indicator variable thus measures the difference in innovation strategies between public and private firms that cannot be accounted for by differences in firm characteristics and industry and year effects. Table 5 presents the regression results. Columns (1)-(2) compare public and private firms in exploitative versus exploratory nature of their innovation. We show that the coefficient on the Public indicator variable is positive and significant at the 5% level when the dependent variable is the exploitative ratio. This suggests that public firms are more likely to utilize existing knowledge in their innovation effort. In contrast, we show that the coefficient on the Public indicator variable is negative and significantly at the 5% level when the dependent variable is the exploratory ratio. This suggests that public firms innovation effort is less likely to be exploring new knowledge. In brief, these findings suggest that public firms focus more on exploiting their existing knowledge than private firms, which may help public firms to maintain current competitive advantages and produce innovations more relevant to their current knowledge domain. Columns (3)-(4) compare public and private firms in innovation depth versus scope. We show that the coefficient on the Public indicator variable is positive and significant at the 5% level when the dependent variable is innovation depth. This suggests that public firms are more likely to deepen existing knowledge in their innovation effort. In contrast, we show that the coefficient on the Public indicator variable is negative and significantly at the 5% level when the dependent variable is innovation scope. This suggests that public firms innovation effort is less likely to beyond exiting knowledge. In brief, these findings suggest that private firms are more likely to expand their expertise and to create new competitive advantages compared to public firms. All results are in strong support of our hypothesis H2 that private firms innovation are more exploratory with greater scope than public firms. In summary, our empirical investigation provides a richer picture of the influence of public firm status on corporate innovation strategies. On the one hand, public ownership encourages innovation, on the other hand, public ownership is associated with less novel innovation. Our findings are consistent with Almeida, Hsu,and Li (2013) who show that explorative innovation activities tend to result in lower efficiency in intangible investment, which leads to lower innovation output. Private firms in our sample are more likely to pursue novel innovation leading to lower patenting output. Our findings also provide 16

18 direct support to the predictions in Ferreira, Manso, and Silva (2013) that due to different tradeoffs, public firms innovation are less exploratory with less scope than private firms. 5. Additional Investigation 5.1. Reconciling with Prior Work Given that prior work has shown that going public is associated with reduced patenting output (Bernstein (2012)), it is important for us to figure out whether our sample and results better capture the public versus private firm dichotomy and its impact on innovative incentives and outcome. For this purpose, we examine a set of firms that change their ownership status from privately-held to be publiclisted over the sample period There are 328 IPOs and 2,296 firm-year observations (i.e., sample firms switch from private to public) with available information over the seven-year period from year -3 to year +3 around the IPO so that all the IPO firms will have the same seven years of data available to compare their innovation output before versus after IPOs. Following Bernstein (2012), we run the following panel data regression: Ln(1 + Innovation Measure) = α + β 1 IPO( 3) + β 2 IPO( 2) + β 3 IPO( 1) + β 4 IPO(+1) + β 5 IPO(+2) + + β 6 IPO(+3) + β 7 Ln(Total assets) + β 8 Leverage + β 9 CF + β 10 Capex + β 11 R&D + β 12 Acquisition + β 13 PPE + β 14 Ln(Firm age) + Firm FEs + ε, (3) where Innovation Measure denotes one of the many different innovation measures we have used before. IPO( 3), IPO( 2), and IPO( 1) are the indicator variables for the event years that are three, two, and one year before IPO, respectively. IPO(+1), IPO(+2), and IPO(+3) are the indicator variables for the events that are in one, two, and three year after IPO, respectively. Following Bernstein (2012), we skip the IPO year, and include firm fixed effects in regressions so that we could examine within firm variation under the public/private status. For brevity, only the coefficients on the six IPO event year indicator variables are reported in Table 6. Panel A shows that when the dependent variable is patent count, there is significant increase in patenting output in the year right before the IPO, and there is generally a declining trend in patenting 17

19 output post the IPO, with significant decline in year two and/or year three depending on the measures of patenting output. Panel B shows that when the dependent variable is citation count, there is some increase in citations in the years before the IPO, and there is a significant declining trend in citations post the IPO. Results from both panels are consistent with Bernstein (2010) who show that going public is associated with unchanged innovation quantity but declining innovation quality, supporting our hypothesis H1 A. Panel C shows that when the dependent variable is one of four innovation strategy measures, there is an increasing trend in the exploitative nature of innovation after going public, a declining trend in the exploratory nature of innovation after going public. Similarly, after going public, innovations are increasingly to rely on existing expertise, and less likely to explore new knowledge, consistent with crosssectional evidence in Table 5, supporting our hypothesis H2. In summary, we show that IPO firms exhibit different patterns of innovation output compared to cross-section comparison of public and private firms. On the other hand, the evidence on innovation strategies in IPO firms is consistent with cross-sectional evidence. We have the following explanations for our findings. IPO firms may not accurately depict innovation output of privately-held firms for the following reasons. First, a pre-ipo firm may exaggerate its patent filings prior to IPO in order to make itself more appealing to public equity investors, leading to a declining trend in innovation output around IPO. Supporting this argument, Teoh, Welch, and Wong (1998) show that extensive earnings management in pre-ipo firms may explain post-ipo long-run market underperformance. Second, given fluctuations in innovation output over time, a firm may time the IPO market by going public right at the peak of its innovation effort. Lastly, Lowry (2003) show that firms choose to go public when demands for capital and aggregate growth as well as the level of investor optimism are high. This selection issue may further result in observing greater innovation output in pre-ipo firms. On the other hand, IPO firms may not be able to drastically change their innovation strategies, leading to similar pattern using IPO firms or cross-sectional evidence Sample Selection 18

20 Going public is not an exogenous event. Most firms stay public (or private) for reasons that correlate with their investment decisions (see for example, Asker et al. (2012) and Maksimovic et al. (2012)). To account for the possible selection issue, we employ an instrumental variable (IV) approach to estimate a selection equation. Under the IV approach, the innovation output and the public status of a firm can be modeled as follows: 13 Ln(Innovation Measure) = Xβ + β 1 Public + ε, Public = Zγ 2 + ω, Public = 1, if Public > 0; Public = 0, if otherwise (2) The dependent variable is the logarithm of one plus various innovation measures. X is a list of firm-level control variables used in earlier regressions. The coefficient of key interest is β 1, on the public firm indicator variable, Public. The variable Public* indicates the latent propensity of a firm staying public. For the purpose of identification, we need instrumental variables that affect a firm s propensity of staying public, but do not affect its innovation activities directly other than through the effect of being public. That is, the vector of Z in Equation (2) must contain variables in addition to a full overlap with the vector of X. The public firm indicator variable is allowed to be endogenous in the sense that corr(ε, ω) 0. A positive (negative) association indicates that innovation outputs and strategies of public firms are higher (lower) based on unobservable heterogeneity. Thus, an estimate for β 1 is upward (downward) biased if the endogeneity is not properly accounted for. To allow for time-varying unobserved heterogeneity across firms, we resort to the treatment regression using the maximum likelihood estimator given in Maddala (1983, Chapter 5), where the public firm indicator variable is treated as endogenous. Following Gao, Lemmon, and Li (2013), our first instrumental variable is the industry-level underwriter concentration (constructed as # of IPOs by the top 5 lead underwriters in a given industry)/total # of IPOs in that industry) to proxy for the high cost of going public IPO underpricing. This variable passes the exclusion restriction because it is hard to expect this measure to be related to 13 See Li and Prabhala (2007) for an overview of dealing with selection issues versus treatment effects in corporate finance. 19

21 corporate innovation policies. We compute the underwriter concentration measure using the entire time period so that our instrument is only related to the cross-sectional decision to be public but is exogenous to a firm s innovation policy, mitigating the IPO timing concern. Moreover, Officer (2007) shows that due to poor stock liquidity, private firm targets are acquired at a discount relative to comparable public firm targets. Our second IV is this private firm illiquidity discount, constructed as follows. Using completed acquisition deals from , we first obtain the price-to-ebit ratio for each target firm. We then compute a private target s illiquidity discount as the difference between its own price-to-ebit ratio to the median price-to-ebit ratio of all public targets in the same industry. The private firm illiquidity discount is obtained as the median discount across all private targets in the same industry. Again, it is hard to expect that this measure could be related to an individual firm s innovation activities. The results from estimating the treatment regression in Equation (2) are reported in Table 7. The identification relies on both the instrumental variable and the non-linearity of the propensity of going public. As shown in Column (1) of Panel A, both the industry-level underwriter concentration and private firm illiquidity discount have negative and significant coefficients in the first stage probit regression. Both instrumental variables are significant at the 1% level. In columns (2)-(4), the dependent variables are various measures of patent counts used in Table 3. The coefficient on the Public indicator variable captures the effect of being a public firm on innovation output level, taking into account the possible selection of staying public decisions. The effects are, again, positive, as the coefficients are around with statistical significance at the 1% level, which is much larger than that of the coefficients in Table 3 Panel A without controlling for the selection issue. Moreover, the exogeneity test rejects the null hypothesis that ρ = 0 at the 5% level. These results indicate that there is a selection issue associated with being a public firm. In our sample, firms that choose to stay public tend to have lower patent counts conditional on observable firm characteristics. That is, these firms would otherwise be associated with lower patent output (after controlling for observable characteristics) had they not been public. Lerner et al. (2011) show that firms invested by private equity tend to have lower patent output. Our selection effect is consistent with their findings. To the extent that the treatment regression framework is valid, such a selection effect makes the interpretation of a causal effect stronger as it renders the effect of being public under-estimated using an 20