The Effect of Institutional Ownership Types on Innovation and Competition

Transcription

1 The Effect of Institutional Ownership Types on Innovation and Competition Paul Borochin School of Business University of Connecticut Jie Yang Board of Governors of the Federal Reserve System Rongrong Zhang College of Business Administration Georgia Southern University Preliminary draft. Please do not quote. This version: December, 2017 Abstract In common ownership, the identity of the common owner matters. We document two countervailing effects of ownership structure of firms by financial institutional owner type: higher within-firm ownership by focused, long-term financial institutions promotes innovation as measured by patent applications and R&D spending. However, more between-firm same-industry connections through ownership by the same institutions leads the connected firms to innovate less and more closely follow industry peers in setting corporate policies. These results contribute to, and potentially help resolve, an ongoing debate about the effects of common institutional ownership on competition. Keywords: Institutional investors, investor type, networks, innovation, patents, product market, competition JEL classification: G30, G32 We thank seminar participants at the University of Connecticut and the University of Toronto for helpful comments and suggestions. The ideas in this paper are solely those of the authors and do not necessarily reflect the view of the Federal Reserve System. All errors are our own. Storrs, CT Phone: (860) Washington, DC Phone: (202) Statesboro, GA Phone: (912)

2 What stands out about the increase in horizontal shareholding resulting from institutional investors is that its potential anticompetitive effects have, until now, gone unnoticed and unaddressed. - Einer Elhauge, 2016, Horizontal Shareholding Harvard Law Review Introduction Is the increase in institutional ownership, and the consequent increase in common ownership of rival firms by the same institution, related to the competitive behavior of the held firms? Institutional ownership of firms has seen a marked rise in the past few decades, with average institutional ownership share of a firm rising from 20% to 30% in the 1980s to over 65% of the total by the 2010s, with residual retail ownership correspondingly falling from 80% to less than 35% of the firm. 1 Over the same time window, the fraction of the average firm held by institutions holding blocks of same-industry rival shares has risen from 4.5% to 28%. 2 Two recent studies find that common institutional ownership affects the competitive behavior of firms in opposite ways in terms of rising profit margins (Azar, Schmalz, and Tecu, 2017) and rising market share (He and Huang, 2017). We contribute to the understanding of the role of institutional ownership on competition in two novel ways. First, we find that portfolio firms differ in their competitive behaviors across different types of institutional owners, consistent with prior findings on the relationship between institutional ownership types and R&D (Bushee, 1998), cite-weighted patents (Aghion, Van Reenen, and Zingales, 2013), and misvaluation and governance (Borochin and Yang, 2017). More importantly, however, we find that common ownership of same-industry peers by dedicated (DED), transient (TRA), and quasi-indexer (QIX) institutional investor 1 Based on data from Thomson Reuters s34 Holdings Database and corroborated by Blume and Kein (2014) and Borochin and Yang (2017). 2 Consistent with this pattern, He and Huang (2017) find that the fraction of firms sharing a common 5% blockholder with a same-industry peer rises from below 10% to about 60% over the same interval. 1

3 types defined by Bushee (1998) have distinct and even more robust effects. Specifically, same-industry common ownership by TRA investors results in less competitive behavior, including innovation and other firm financial policies. Why might financial institutions purposely engage in cross-holding of rival firms, apart from a consequence of broad diversification? The most significant benefit unrelated to firm competition is economies of scale in information gathering for multiple firms in the same industry (O Brien and Bhushan, 1990; Kacperczyk, Sialm, and Zheng, 2005). However, Elhauge (2016) and Azar, Schmalz, and Tecu (2017) propose an alternative benefit stemming from changes in firm competitive behavior: common ownership of sameindustry firms incentivizes both investors and managers to maximize portfolio rather than firm profits. Consistent with Hansen and Lott (1996), common institutional ownership may encourage portfolio firms to avoid wasting resources by competing with each other. Since common institutional ownership is public knowledge, Elhauge (2016) points out that this reprioritization does not even require active communication between commonly-held firms and their institutional owners. In addition to this, Azar, Schmalz, and Tecu (2017) identify the additional channels of institutional shareholder voice and vote as means by which institutions can actively influence the decisions of portfolio firms to maximize their owners portfolio value. These authors suggest that common institutional ownership incentivizes firms to increase prices and profits rather than output or market share. As a counterpoint, He and Huang (2017) argue that common institutional ownership does improve the firm s competitiveness, either overall or at least relative to those rivals outside the common ownership portfolio. This increase in competitiveness may come about due to the role of the common institutional owner in facilitating collaboration between firms. They argue that incomplete contracting may impose frictions, such as risk of expropriation by collaborators, that a common owner can help mitigate. Furthermore, common institutional owners can indirectly or directly facilitate information sharing between same-industry rivals. Due to concerns about expropriation, rivals may conceal private information from each other 2

4 in a way that is optimal for the individual firm, but suboptimal for the portfolio of firms or the industry. By helping disseminate firm-specific private information about processes, suppliers, market research, and other inputs into product development (Brown and Eisenhardt, 1995) across all portfolio firms, a common institutional owner may improve each individual firm s product market performance. In addition to these channels, the effect of higher common ownership on executive compensation has been suggested as a channel to promote more aggressive competition due to more use of performance-sensitive compensation (Anton, Ederer, Gine, and Schmalz, 2017), as well as to promote less aggressive competition due to the greater use of relative performance evaluation (Kwon, 2017). Thus, common institutional ownership may have both beneficial and harmful effects on the firm s competitiveness in the product market. We find that the identity and investment objectives of the institutional owner determine which effect dominates. Of the many types of competitive firm behavior, innovation stands out as a particularly important dimension (Porter, 1985; 1990). Innovation is a costly and easily observable way firms can compete with each other, and our study contributes to an ongoing debate in the literature. On one side, Azar, Schmalz and Tecu (2017) find that common ownership by mutual funds results in anti-competitive behavior in airlines. A common owner reduces the incentive for any individual airline to compete on price since the owner holds a claim on the aggregate cash flow from the entire industry, consistent with prior theoretical findings. 3 Conversely, He and Huang (2017) find that common ownership makes firms more competitive by enabling market share growth through expanding the boundaries of the firm, improved pricing power, and higher quality information production. 4 By separating the monolithic block of common institutional ownership into types, we provide a potential resolution to this debate. We also add to a growing body of literature on the relationship between institutional 3 See, e.g., O Brien and Salop (2000), Gilo, Moshe, and Spiegel (2006), Rubin (2006). 4 See also Kacperczyk, Sialm, and Zheng (2005). 3

5 ownership and innovation (see, e.g., Bushee, 1998; 2001; Atanassov, 2013; Aghion, Van Reenen, and Zingales, 2013; and Brav, Jiang, Ma, and Tian, 2016). Indeed, we document two countervailing effects of ownership structure of firms by financial institutional owner type: higher within-firm ownership by long-horizon financial institutions promotes innovation as measured by patent applications. However, more between-firm same-industry connections through ownership by focused and long-horizon institutional owners leads the connected firms to innovate less. Furthermore, same-industry common ownership by both focused and diversified long-horizon instutional investors causes firms to more closely follow industry peers in setting corporate policies. These dampening effects of same-industry common ownership by institutional type on innovation and other firm policies related to competition suggest a more nuanced view of the optimal ownership structure of the firm. Within-firm levels of focused, long-horizon institutional investor ownership is good for promoting innovation and therefore competition. However, between-firm common ownership linkages by the same institutional investor are bad for it. Despite our use of lagged institutional ownership measures at the firm and network levels, these results may be due to self-selection of institutional ownership types into more or less competitive firms. This clustering by institutional investors can lead to more betweenfirm common ownership linkages across the portfolio firms, potentially confounding a causal interpretation between common ownership and competition and innovation. To correct for this, we make use of an exogenous shock to the level of within-firm ownership and the structure of the institutional ownership network. Specifically, we use a difference-in-differences regression around mergers of different types of financial institutions for firms whose same-industry peers are held by the other party in the deal. The control firms in the model are those without same-industry peers in the portfolio held by the other party in the financial institution M&A deal. These mergers of 4

6 different types of financial institutions provide an exogenous shock to the level of ownership by the acquiring type within the firm. They also provide a similarly exogenous shock to the between-firm network structure of institutional ownership of firms in our sample. The financial institutions M&A decisions are likely unrelated to the policies of the firms they own, providing a clean identification of the effect of the resulting changes in institution ownership level by type. We also use the Heckman (1979) approach by instrumenting for network density. We add to the growing literature on the effects of institutional ownertypes, as well as common ownership by type, on competitive behavior within the firm in terms of overall patenting, the impact of the firm s patents, the types of patents the firm pursues, its investment in R&D, and its conformity to its peers in setting operational policies. By separating within-firm institutional ownership and between-firms common ownership by different types of institutions, we contribute to the ongoing debate about the effects of institutional ownership and common ownership on competitive behavior within the firm. By separating the effects of different types of institutional ownership on the firm, we reconcile previously contradictory findings and help define the optimal ownership structure of the firm. 2 Data and Hypothesis Development 2.1 Institutional Ownership and Investor Types We obtain data on institutional investor ownership from the Thomson Reuters Institutional Holdings database. This database lists the positions of institutional investors conducting business in the U.S. with investments over $100 million as reported in 13F forms filed with the Securities and Exchange Commission (SEC) every quarter. For institutional ownership to be meaningful, we only consider institutional ownership positions of 1% or greater relative to firm value. This threshold is set higher than the 0.5% blockholder definition used by Azar, 5

7 Schmalz, and Tecu (2017) but lower than the 5% definition used by He and Huang (2017) as a compromise between capturing meaningful ownership stakes and the diffusal of usable observations across three different types of institutional owners that we consider. We categorize institutional owners into types following Bushee (1998, 2001). Bushee (1998, 2001) categorizes institutional investors into dedicated (DED), quasi-indexer (QIX), and transient (TRA) types based on their portfolio characteristics of turnover and diversification. 5 DED institutions have low portfolio turnover and therefore long investment horizons, as well as focused portfolio holdings and therefore larger stakes in individual firms. This combination of portfolio focus and long horizon aligns their objectives most closely with firm innovation as a firm-specific project with long-horizon payoffs (Bushee, 1998; 2001; Borochin and Yang, 2017). The second category, QIX, is defined by low turnover but diversified portfolio holdings. Like the DEDs, the QIX institutions long horizons make innovation relevant but their diversified position makes firm-specific decisions less so. Their conventional perception as passive investors with little influence on corporate governance has recently been challenged (Boone and White, 2015; Appel, Gormley, and Keim, 2016). The third category, TRA, have high portfolio turnover and high diversification consistent with opportunistic strategies. Whereas both DED and QIX institutional investors have long investment horizons differing in portfolio diversification, the TRA institutional investors have short investment horizons making firm-specific innovation less relevant to their investment decisions. This provides a potentially useful contrast with the preceding two types. Figure 1 Panel A shows the proportions of DED, TRA, and QIX institutional ownership within the average firm through time, computed using cross-sectional averages of firm-level institutional ownership at the 1% or greater blockholding level. We see that the proportional amount of DED ownership decreases over time, while TRA increases and QIX remains largely 5 We are grateful to Brian Bushee for providing this data on his website. 6

8 constant. 6 Panel B of Figure 1 presents cross-sectional averages of same-industry common ownership, defined as 1% or greater blockholding by an institution that holds a block in a same-sic4 industry peer. The figure shows that, consistent with the points raised by Elhauge (2016), same-industry common ownership by institutional blockholders within the average firm is on the rise, even when normalized by the total number of blockholders. Indeed, whereas roughly 10-15% of DED and TRA and 23% of QIX blockholders held significant stakes in same-sic4 peers in 1980, by 2009 this fraction has risen to above 40% for DED and TRA, and above 60% for QIX. Consistent with He and Huang (2017), this provides compelling evidence that common ownership is a significant phenomenon whose implications require investigation. The effects of common ownership by institutional type are uncertain ex ante. DED investors focus on a smaller number of firms that they hold for long-run growth (Bushee, 1998; 2001), and thus have incentives to encourage the firm to optimize for long-run value. QIX investors have a similar long-term view, but may care less about the long-run value of each individual firm due to a diversified portfolio. TRA investors have more short-term views, chasing trends and short-term performance (Borochin and Yang, 2017). We state our hypotheses in null form: Hypothesis 1. Common institutional ownership has no effect on innovation as measured by patents and citations, regardless of type of common institutional owner. Due to the potential for common ownership to serve as an informal coordination mechanism in determining innovation strategies (Jansen, Van Den Bosch, and Volberda, 2006) in addition to other competitive policies (Azar, Schmalz, and Tecu, 2017; He and Huang, 2017), we consider the effects of institutional ownership types: Hypothesis 2. Common institutional ownership does not affect the innovation strategy a firm chooses to engage in, regardless of type of common institutional owner. 6 The three institutional ownership types do not necessarily add up to 100% ownership because some institutions are uncategorized and therefore omitted. 7

9 Finally, we test whether common ownership by institutional types has different effects on operational policy deviations from same-industry peers. investigation comes from the active investment literature: The intuition behind this funds that make significant deviations from the peer group seek to outperform and are therefore pursuing a competitive policy, while those that stick to their peers seek to simply keep up, and are less competitive (Berk and Green, 2004; Fama and French, 2010). Hypothesis 3. Common institutional ownership does not affect the operating policies of the firm relative to its peers, regardless of type of common institutional owner. 2.2 Network Measures We begin by creating a measure of common ownership within the set of SIC4 industries. This enables us to perform tests of whether ownership networks that result in higher common ownership promote competitive policies within the firm (He and Huang, 2017; Kacperczyk, Sialm, and Zheng, 2005) or impede them (Azar, Schmalz, and Tecu, 2017; Elhauge, 2016) depending on institutional owner type. We focus on same-sic4 common ownership following He and Huang (2017) since competition primarily occurs within industries, mapping the institutional ownership connections each firm has with other firms in the same and other industries, defining a linkage between two firms through an institutional owner when that owner reports holding at least a 1% block in both firms in their 13F SEC filing for the quarter. 7 We illustrate a simple network that defines our measure of common institutional ownership following He and Huang (2017) for the DED, TRA, and QIX institution types in Figure 2. To measure the proportion of common DED ownership for Firm 1 in Figure 2, we compute the number of connections Firm 1 has to other firms in the same SIC4 industry 7 In this analysis we use an equal-weighted measure that treats all linkages the same regardless of institutional owner portfolio weight. An alternative specification with strengths of linkages based on portfolio weights produces similar results. 8

10 through institutional owners of that type. 8 Of its same-industry peers Firm 1 has two connections to Firm 2 (through DED2 and DED3), one connection to Firm 3 (through DED2), and one with Firm 4 (through DED1), with the total number of connections to same-sic4 firms thus calculated to be four. We follow the same process to calculate TRA and QIX same-industry common ownership counts. 2.3 Innovation Data Our primary measure of firm competitive behavior is innovation. We measure innovation using log-transformed patent and citation counts from the USPTO data obtained through the Harvard Dataverse from 1980 to Only those patents that are eventually granted are included in this database. We follow the innovation literature in using the application date of the patent, rather than the date it is granted, as the effective date of the patent (Griliches, Pakes, and Hall, 1998; Cohen, Diether, and Malloy, 2013; Fang, Tian, and Tice, 2014). The effective date for citations is when they are recorded in another patent application. The typical lead time for a patent from application to grant date is two years (Hall, Jaffe, and Trajtenberg, 2001). We match the patent application year and quarter date with Thomson 13 F quarterly filing dates to establish the actual time of innovation activity (Griliches, Pakes, and Hall, 1988). Because a simple count of patents does not differentiate between more and less impactful patents, we create a second measure of innovation output by counting the total number of non-self-citations for each firm s patent portfolio by quarter. To more precisely define the types of innovation we follow prior innovation literature (Almeida, Hsu, and Li, 2013; Phelps, 2010; Benner and Tushman, 2002; Katila and Ahuja, 2002; Sorensen and Stuart, 2000) in identifying patent subtypes to measure the degree of risk-taking in the innovation that takes place. We classify a patent as a safer exploitative type if at least 60% of the patents it cites constitute the firm s existing knowledge base, coming from its own previously filed patents over the past five years and other companies 8 This is analogous to the NumCross measure in He and Huang (2017). 9

11 patents cited by the firms patents filed over the past five years. Conversely, we classify a patent to be a riskier exploratory type if at least 60% of the patents it cites are outside of the firm s existing knowlege base. Some patents are thus neither exploratory nor exploitative. The two primary dimensions that distinguish exploratory versus exploitative innovation are either a proximity to existing technologies, products, and services or a proximity to existing customer or market segments (Jensen, Van Den Bosch, and Volberda, 2006). Of these, exploratory innovations are fundamentally novel approaches aimed to meet the needs of new markets, whereas exploitative innovations are more incremental and aimed at existing ones. Figure 3 Panel A presents the average number of total patents applied for, as well as exploratory and exploitative patent types, by firm throughout our sample. 9 The post peak and subsequent dropoff in patent applications is due to the requirement that the application be subsequently granted, with the latest grant date occurring in 2009 in our sample and a typical two year period between grant date and application date (Hall, Jaffe, and Trajtenberg, 2001). Figure 3 Panel B shows the average number of patents applied for by firms with above-median DED, TRA, and QIX ownership through time. While firms with abovemedian dedicated or transient ownership have more patent applications in the early part of the sample, they are subsequently overtaken by firms with above-median quasi-indexer ownership. Figure 3 Panel C shows that on average, firms with above-median same-sic4 industry connectedness through dedicated institutional ownership hold more patents than those with above-median connectedness by quasi-indexer ownership, followed by transient ownership. This contemporaneous difference suggests that common ownership has an effect on firm patenting, but could be the result of self-selection by dedicated owners into more innovative 9 This average is computed conditional on patenting and therefore does not include firms that have never received a patent. 10

12 firms, or could be explained by other firm characteristics related to more common ownership such as size or age. This difference in patenting motivates our more detailed investigation into the effects of common ownership on innovation in the rest of the paper. 2.4 Financial Statement Data We collect information on firm characteristics as control variables for the relationship between institutional investor ownership levels and networks with innovation. We define firm characteristics based on financial statement data obtained from Standard and Poor s Compustat North American quarterly database from 1980 to All value amounts are measured in 2000 dollars using CPI to adjust for inflation. We remove any firms with negative book asset value, market equity, book equity, capital stock, sales, dividends, debt, and inventory. These firms are prone to data errors, are likely to be distressed or severely unprofitable, or are otherwise likely to be outliersl. We also delete observations in which book assets or sales growth over the quarter is greater than 1 or less than -1 and remove firms worth less than $5 million in 2000 dollars in book value or market value to remove observations that have abnormally large changes due to acquisitions or small asset bases. Next, we remove outliers defined as firm-quarter observations that are in the first and 99th percentile tails for all relevant variables used in our analysis. Following standard practice in the literature, we remove all firms in the financial and insurance, utilities, and public administration industries as they tend to be heavily regulated. Merging institutional investor data to corporate financial data based on a firm s CUSIP and year-quarter gives us a sample of 207,634 firm-quarter observations spanning 8,359 firms. Table I Panel A provides the summary statistics. The average (median) firm in our sample has 40.2% (36.8%) institutional ownership, with 5.6% (3.4%) of the firm owned by dedicated institutional investors, 24.0% (23.6%) by transient, and 68.1% (68.8%) by quasi-indexer institutional investors. The average (median) number of same-sic4 common ownership connections is.6 (0.0) through dedicated institutions, 1.1 (0.0) through transient, and

13 (2.0) through quasi-indexer ownership. Patenting activity is significantly skewed with the average (median) firm applying 4.6 (0.0) patents per quarter, with a standard deviation of 30.6 patents and a right tail of 90 quarterly patent applications at the 99th percentile. Panel B of Table I presents the correlation matrix for the key variables in our study. The within-firm level of DED ownership is negatively correlated with all innovation outcomes except cites per total patent count, while that of TRA ownership is positively correlated to both exploratory patents and citations per patent stock. QIX ownership is positively correlated to all innovation outcomes except citations per patent stock. Meanwhile, the number of same-industry connections through DED and QIX ownership are both positively correlated with all innovation measures other than citations normalized by patent counts, with the opposite result for TRA connections. Notably, the common ownership network measures have greater correlation magnitudes with innovation than the levels of ownership, motivating the importance of considering between-firm ownership linkages. We further motivate our study by comparing firm characteristics in two-sample t-tests across levels and networks of institutional ownership in Table II. Panel A compares firms with above-median levels of institutional ownership across the three types in the left set of columns of the table. We find that the three types of institutional investors hold firms with significantly different characteristics. Specifically, we find that isqix firms have more patents than ist RA firms, which in turn have more than isded firms, with average counts of 7.9 to 4.0 to 2.0 patents granted per quarter respectively, all significant at the 1% level. This greater level of innovation by isqix firms extends to the number of exploratory and exploitative patents and citation counts with the same level of significance. Furthermore, levels of within-firm ownership by institution type are related to other firm policy decisions as seen in Panel A of Table II. In a comparison of within-firm ownership, firms with above-median levels of same-industry connections through different institutional owner types have different levels of R&D expense, leverage, cash holding, size, market to book ratio, ROA, and PP&E to total assets in all cases. The CAPEX ratio is not significantly 12

14 different between firms with DED and QIX linkages, but is in the other two comparisons. Notably, the univariate relationship between same-industry connections and innovation differs from that observed previously between the levels of ownership and innovation. Table II Panel B reports the t-test results across samples with above-median same-industry connections through DED, TRA, and QIX insitutions. We see that firms with above-median numbers of same-industry connection through DED owners have more patenting than those with above-median TRA connectedness significant at the 5% level, and more than those with above-median QIX connectedness at the 10% level. There is no significant difference in patent counts between firms with above-median TRA and QIX connectedness. These results are similar, but slightly more significant for exploratory patents. Firms linked by DED investors have more exploitative patents than those linked by either TRA or QIX, both significant at the 1% level. However, firms linked by TRA investors have more citations per total patent count than either DED or QIX, and those linked by DED investors have more citations per patent count than QIX, all significant at the 1% level. These results suggest that the measures of same-industry common ownership by institutional types provide information that is significantly different from that of the levels of institutional ownership by type within each firm. These differences between firms with above-median industry linkages through different institutional owner types again extend to other corporate policy decisions in terms of R&D expense, leverage, cash holding, size, market to book ratio, ROA, and PP&E to total assets for all comparisons, with an insignificantly different CAPEX ratio between firms with DED and QIX linkages, but significantly different in the other two comparisons. These differences motivate a subsequent investigation of how corporate decisions related to competition are affected by networks of different institutional ownership types. The univariate findings in Table II provide preliminary evidence that there is a relationship between same-industry common ownership by institution type and patenting activity as a form of competitive behavior. We next examine these relationships in a panel 13

15 context with control variables to account for potential confounding effects. 3 Baseline OLS Results In this section we consider the effects of common ownership by institutional investor types on competitive behavior within the firm as measured by patenting activity, its impact as measured by citations, and the types of innovation strategies pursued in terms of exploratory versus exploitative patenting. We estimate the effect of contemporaneous levels of within-firm ownership %T Y P E, as well as same-industry common ownership Cross-Holding T Y P E as described in Section 2.2, by institutional type on variables of interest Y i,t : the one-quarter forwards of the natural log of patent applications filed and of the number of citations received normalized by the stock of patent applications filed by the firm to date. Innovation as proxied by patenting is a costly strategy that can help the firm more effectively compete by raising barriers to entry and providing superior product. The number of citations received relative to total patent count is a proxy of the significance of this innovation. We decompose the level of ownership by type, as well as the level of between-firms same-industry common ownership by institutional type, into the lag of the level and the contemporaneous change: X i,t X i,t 1 + X i,t 1,t This decomposition gives us two distinct advantages. First, it allows us to study the impact of levels as well as changes of firm-specific and common ownership by institutional type on innovation. Second, while the lagged level controls for likely persistence of ownership, the contemporaneous change in ownership to act as a shock. It thus allows us a first attempt to address the potential selection bias between institutional ownership and competition by 14

16 controlling for the existing relationship between the two. We address potential endogeneity issues more rigorously in Section 4. We include the level and network measures of ownership by type for DED, TRA and QIX, as well as year, quarter, and industry fixed effects. We follow Peterson (2009) in clustering standard errors by firm and quarter. Our baseline OLS model for each institutional owner type is described below: Y i,t = α + β 1 %T Y P E i,t 1 + β 2 %T Y P E i,t 1,t + β 3 Cross-Holding T Y P E i,t 1 + β 4 Cross-Holding T Y P E i,t 1 + fe t + fe j + ε i,t (1) Table III Panel A presents the results from Eq. 1 for patents and citations for the DED institutional type in columns (1) and (2), TRA in columns (3) and (4), and QIX in columns (5) and (6). We find that the level and change in within-firm DED ownership is related to a reduction in the log of next-quarter patent applications, but is positively related to next-quarter citations normalized by patent count. The relationship of same-industry DED common ownership is the opposite, with a positive relationship with next-quarter patent applications but a negative one with next-quarter citations. All effects are significant at the 1% level. Both levels and changes of within-firm TRA ownership is negatively related to future patents and citations, while TRA common ownership levels are positively related to patent counts and negatively related to citations with significance at the 1% level. The change in TRA common ownership has no significant relationship with patenting, and a weakly positive one with citations significant at the 10% level. Finally, levels and changes of withinfirm QIX ownership have a positive relationship with future patenting at the 1% level, and a negative one with citations at the 1% and 10% levels respectively. The levels and changes of QIX common ownership also have a positive relationship with patenting and a negative relationship with citations, all at the 1% level. The differences across both within-firm ownership, and more importantly same-industry common ownership, by institutional owner type provide preliminary evidence that the type 15

17 of institutional owner matters. Despite controlling for year, quarter, and industry effects on firm-specific innovation, the constant term is significant at the 1% level in all specifications of Table III thus far suggesting the model does not fully explain future patenting and citation activity. Furthermore, the observed relationships between ownership levels and networks could be caused by self-selection of institutional owners into firms with certain characteristics that correlate with innovation. Therefore, we introduce control variables related to innovation into our baseline model to obtain our full model at quarterly frequency: Y i,t = α + β 1 %T Y P E i,t 1 + β 2 %T Y P E i,t 1,t + β 3 Cross-Holding T Y P E i,t 1 + β 4 Cross-Holding T Y P E i,t 1 + β 5 %INST i,t 1 + β 6 lnt A i,t 1 + β 7 MtB i,t 1 + β 8 RD i,t 1 + β 9 ROA i,t 1 + β 10 P P E i,t 1 + β 11 Lev i,t 1 + β 12 CAP EX i,t 1 + β 13 HHI i,t 1 + β 14 HHI 2 i,t 1 + β 15 LT CR i,t 1 + β 16 lnage i,t 1 + fe t + fe j + ε i,t (2) where %IN ST is the overall fraction of the firm s market capitalization reported owned by institutional investors in 13F filings, lnt A is log-transformed total assets of the firm to control for firm size, MtB is the market to book value of equity ratio to control for growth options, RD is the R&D expenditure to control for the current investment in innovation, ROA is the return on assets as a measure of performance, P P E is the plant, property, and equipment net of depreciation as a measure of the physical capital in the firm, Lev is the ratio of total debt to total assets, CAP EX is capital expenditure, HHI is the Herfindahl- Hirschman index for the firm s SIC4 industry as a measure of its competitiveness, HHI 2 is the square of HHI to account for nonlinearities in industry concentration, LT CR is an indicator variable for whether the firm has a credit rating to control for access to financing of innovation, and lnage is the log of firm age. The control variables are defined in Appendix A. As before, we include year, quarter, and industry fixed effects and double-cluster standard errors by firm and year-quarter. 16

18 We tabulate the results from the model in Eq. 2 in columns (1) through (6) of Table III Panel B. Introducing controls does not affect the relationship of levels and changes of common DED ownership with patenting, but eliminates its significance for changes in DED common ownership with citations in columns (1) and (2). The relationship between TRA common ownership and patenting in columns (3) and (4) becomes negative and positive respectively, while that for QIX common ownership becomes insignificant except for a negative relationship between the level of QIX ownership and future citations. 4 Robustness and Identification Tests 4.1 Institutional Owner Merger as an Exogenous Shock The control variables related to innovation and competitive behavior which we included in the prior section may not fully account for a self-selection issue in our sample: institutions of different types may select their portfolio of firms based on some other characteristics related to future competitive behavior, leading to an omitted variable bias. They may also select their portfolio weights based on their expectation of the owned firms competitive behavior itself, resulting in reverse causality between competitive behavior and ownership by specific institutional types based on their investment preferences for innovation. We address this concern using exogenous changes in the level of common ownership by institutional types introduced through the mergers of the three Bushee (1998) types of financial institutions. As suggested by He and Huang (2017), a merger of two financial institutions, each of which owns a block in one of two same-industry rivals, forms a new common ownership linkage between the two rivals after the combination of their two institutional owners into one. Furthermore, a merger of two financial institutions of different types will result in a new level of within-firm institutional ownership for any firms in which either the target or acquirer was a blockholder pre-merger. For example, if a DED acquirer were to absorb a QIX target s portfolio, any same-industry rivals held by either the 17

19 target or the acquirer would experience an exogenous increase in DED common ownership. Furthermore, the firms in the QIX target s portfolio would experience an exogenous increase in within-firm DED ownership. More generally, a merger between two different institutional owner types, one of whose portfolios includes at least one firm with one or more same-sic4 rivals in the other s portfolio, will result in exogenous shocks to both the common ownership of all the rival firms as well as an exogenous shock to within-firm ownership by the acquirer s type for the firms in the target s portfolio. On the other hand, a merger between two institutions of the same type will result in only the exogenous shock to common ownership of any same-industry rivals. Here we assume that the combined institutional investor retains the characteristics of the acquirer after the consummation of the merger, which is consistent with the permanent classification of financial institutions developed by Bushee (1998) that we use in this study. This post-merger change in within-firm level and between-firm common ownership by institutional type is likely to be exogenous to the decisions made by the firms owned by the merging financial institutions. Indeed, the primary reasons for financial institution mergers identified in the literature are empire building (Gorton and Rosen, 1995), cost savings (Houston, James, and Ryngaert, 2001), and an expansion of market share (Jyaraman, Khorana, and Nelling, 2002). Furthermore, while institutional acquirers of different types may not fully retain blocks acquired as part of their targets portfolios, liquidity and trading costs will likely prevent them from exiting these positions completely over a short time horizon (Keim and Madhavan, 1996; Madhavan and Cheng, 1997). Thus, institutional mergers are quite likely to result in exogenous changes to the owned firms within- and between-firm levels of institutional ownership by type Quasi-Natural Experiment Setup We begin by identifying financial institution acquirers and targets (those with SIC codes between 6000 and 6999) from the SDC M&A database. We then match them by name to the 18

20 Thomson s34 institutional ownership dataset to identify the firms in which these institutions hold a block of 1% or more of total value. 10 This results in 40 financial institution merger events between 1982 and 2008 that affect 1,283 firms. Of these, five involve DED acquirers, six involve TRA acquirers, and 29 involve QIX acquirers using the permanent Bushee (1998) classification. 11 We draw our sample of treated and control firms from those with at least a 1% blockholding by one of the merging financial institutions one quarter prior to the M&A deal consummation. Of these, the treated firms are those with a same-sic4 rival in the other M&A party s portfolio of blockholdings, while the control firms are those without one. Following He and Huang (2017), this definition attempts to hold constant all institutional owner-specific details about control firms relative to treatment firms, with the only difference that treatment firms become cross-held with an SIC4 rival after the deal whereas the control firms do not. We look at one-quarter-ahead results over the two years prior and two years after the M&A event (discarding the quarter of the event itself). We interact two indicator variables, post treat, to isolate the effect of a treatment firm after the institutional M&A event. Here post is the indicator variable that takes on the value of 1 after the financial institution merger, with a value of 0 prior, while treat takes on the value of 1 if the firm meets our treatment definition having same-industry rivals in the portfolio of the other financial institution participating in the M&A deal that are not included in its institutional owner s own portfolio. We include firm-deal dummies and cluster standard errors by firm. 4.2 Patents and Citations We first apply this exogenous shock in common ownership to verify our OLS results for innovation outcomes in Table III. Our resulting difference-in-differences model for the next- 10 We set the blockholder threshold at 1% to find a middle ground between meaningful ownership and a significant number of deals. This lower threshold than the 5% used by He and Huang (2017) is necessary because we further separate financial institution mergers by type. 11 We use the permanent classification to avoid changes in institutional type due to M&A or other events. 19

21 quarter variable of interest Y i,t+1, the logs of granted patent applications and citations normalized by patent stock, is specified below: Y i,t+1 = α + β 1 post t + β 2 post t treat i + fe i,j + ε i,t (3) Table IV summarizes the results. Panel A of Table IV presents the difference-in-difference coefficients without controls, showing that exogenous increases in DED common ownership of rival firms due to their consolidation into a combined DED acquirer s portfolio have more granted patent applications but fewer citations per stock of patents in columns (1) through (2). The pattern is reversed for TRA acquirers, whose portfolio firms have fewer granted patents but more citations normalized by patent stock due to a post-acquisition increase in TRA common ownership in columns (3) through (4). Increases in common ownership by QIX institutions due to the owning financial institution s acquisition by a QIX acquirer produces results similar to the DED case with more granted patents as well as citations normalized by patent stock in columns (5) through (6). We next introduce control variables into our model to account for potential confounding effects: Y i,t+1 = α + β 1 post t + β 2 post t treat i + β 3 %DED i,t + β 4 %T RA i,t + β 5 %QIX i,t + β 6 %INST i,t + β 7 lnt A i,t + β 8 MtB i,t + β 9 RD i,t + β 10 ROA i,t + β 11 P P E i,t + β 12 Lev i,t + β 13 CAP EX i,t + β 14 HHI i,t (4) + β 15 HHI 2 i,t + β 16 LT CR i,t + β 17 lnage i,t + fe i,j + ε i,t Since the merger of DED, TRA, or QIX financial institutions with those of a different Bushee (1998) type will cause not only the common ownership, but also the level of within-firm institutional ownership to change, we control for pded, pt RA, and pqix to isolate the change in common ownership due to the merger. As before, we also control for %INST, the overall fraction of the firm s market capitalization owned by institutional investors in 13F 20

22 filings, lnt A, the log-transformed total assets of the firm, MtB, the market to book value of equity, RD, the R&D expenditure, ROA, the return on assets, P P E, the plant, property, and equipment net of depreciation, Lev, the leverage ratio, CAP EX, capital expenditure, HHI, the Herfindahl-Hirschman index for the firm s SIC4 industry, HHI 2, the square of HHI, LT CR, an indicator variable for whether the firm has a credit rating, and lnage, the log of firm age. The control variables are defined in Appendix A. We again include deal-firm fixed effects in all specifications and cluster by firm. These results are summarized in Panel B of Table IV. Our results are similar to those in to the difference-in-differences results without controls in Panel A of of Table IV as well as the OLS results with controls in Panel B of Table III. This evidence supports a causal interpretation of our prior findings in Table III. Exogenous increases in DED common ownership result in more granted patents but fewer citations per stock of patents in columns (1) and (2), while increases in TRA common ownership result in fewer granted patents but more citations normalized by patent stock in columns (3) and (4). An exogenous increase in QIX common ownership still results in more granted patents but fewer citations normalized by patent stock in columns (5) and (6). Overall these results support a causal interpretation of the relationship of common ownership with firm innovation, and show that the effects of common ownership depend on the type of institution. Indeed, the negative effect increases in TRA commonownership have on future patenting is consistent with the anticompetitive interpretation of common institutional ownership by Azar, Schmalz and Tecu (2017). Notably, this effect is reversed for DED and QIX institutions consistent with the beneficial effects of common ownership on product market competition found by He and Huang (2017). This difference across the Bushee (1998) types provides a potential explanation for the apparently conflicting findings in the two prior papers. 21

23 4.3 Innovation Types We next consider the effects of common ownership by institutional types on competitive behavior in more detail by separating patent types into exploratory and exploitative following Almeida, Hsu, and Li (2013). Exploratory patent types are shown to be related to more risky, groundbreaking innovation in new product markets while exploitative patent types are safer and more iterative, focusing on existing markets. 12 Based on the observations of Porter (1985) and Porter (1990), we expect exploratory (exploitative) innovation to obtain more (less) competitive advantage for the patenting firm. The type of competitive strategy pursued by the firm in terms of exploratory versus exploitative innovation may also help explain the observed differences in the impact of innovation observed in Tables III and IV. We take onequarter forwards of log-transformed counts of these two patent types as dependent variables Y i,t+1 in our difference-in-differences model without controls in Eq. 3. Table V Panel A presents the results for the two types of patenting, which futher confirm the differential effects of common ownership by institutional type documented in our prior results. An exogenous increase in DED common ownership results in both higher exploratory and exploitative patenting with significance at the 5% and 1% levels respectively. However, an increase in TRA common ownership results in lower exploratory patenting at the 1% significance level, and an insignificant effect on exploitative patenting. Finally, an exogenous increase in QIX common ownership results in an increase in exploitative patenting but an insignificant effect on exploratory. To control for potential confounding effects we introduce controls as specified in Eq. 4, including deal-firm fixed effects and clustering by firm. We present the results in Table V Panel B, which are largely consistent with the results in Panel A. In the presence of controls, exogenous increases in DED common ownership continue to result in more exploitative patenting, though the effect on exploratory patenting becomes insignificant. Increases in 12 See, e.g., Phelps, 2010; Jansen, Van Den Bosch, and Volberda, 2006; Benner and Tushman, 2002; Katila and Ahuja, 2002; Sorensen and Stuart,

24 TRA common ownership continue to reduce exploratory patenting, without a significant effect on the exploitative type. Finally, increases in QIX common ownership increase exploitative patenting but do not affect exploratory patent counts. It is useful to note that the firms experiencing increases in DED and QIX common ownership see decreases in citations per patent count in Table IV as well as increases in exploitative patenting in Table V. We leave the mechanism by which exploitative patenting appears to be correlated with lower citation counts to future work, but its characterization in Jansen, Van Den Bosch, and Volberda (2006) suggests that exploitative patenting might be less impactful, and may attract fewer citations, due to its more safe and conventional nature. 5 Explanatory Channels The results thus far support the rejection of Hypothesis 1 and 2: the quasi-natural experiment of mergers by different institutional types supports the baseline OLS findings that common ownership of same-industry rivals by the three Bushee (1998) types has different effects on firm patenting. DED and QIX common ownership both result in higher patenting, whereas TRA common ownership causes lower patenting. Furthermore, DED and QIX common ownership causes firms to seek more exploitative innovation and receive fewer citations relative to current patent count while that by TRA institutions causes them to do less exploratory patenting and receive more citations per patent count. We now turn to an investigation of the potential mechanisms by which common ownership affects innovation. 5.1 Industry Competitiveness Common institutional ownership may promote anti-competitive results found by Azar, Schmalz and Tecu (2017), which we observe to be confined to TRA common ownership in the case of patenting, through the channel of industry competition. High concentration, 23