R&D Investments, Technology Spillovers, and Stock Returns*

Transcription

1 R&D Investments, Technology Spillovers, and Stock Returns* Jong-Min Oh This Version: March 2015 ABSTRACT I demonstrate that R&D investment is crucial for firms to effectively absorb and benefit from potential technology spillovers. In the presence of large spillovers, I find that firms with higher R&D earn significantly higher subsequent equity returns. Importantly, when exposed to large spillovers, higher R&D firms generate more patents (citations) per available spillover and exhibit higher operating performance. I further demonstrate that the complementary relation between R&D and spillovers is most pronounced among low investor attention stocks. My findings suggest that the market appears to undervalue the complementary role of R&D, and that limited investor attention contributes to this undervaluation. JEL Classification: G11, G12, G14, O31, O32, O33 Keywords: Spillover, Innovation, R&D, Limited attention, Market efficiency * I am indebted to Robert Kieschnick, Michael Rebello, Alessio Saretto, Jun Li, and Arzu Ozoguz. I give special thanks to Nicholas Bloom for his generosity in sharing his resources with me. I am grateful to Harrison Hong, Po- Hsuan Hsu, Wan-Jiun (Paul) Chiou (discussant), Samar Ashour (discussant), Cheolwoo Lee (discussant), Seong Byun, Bernhard Ganglmair, Malcolm Wardlaw, Han Xia, Yexiao Xu, Harold Zhang, and seminar and conference participants at the Financial Management Association (FMA) Annual Meeting 2014 (Nashville, TN), FMA Doctoral Student Consortium Special Job Market Paper Presentation 2014 (Nashville, TN), the Southern Finance Association (SFA) Annual Meeting 2014, and the University of Texas at Dallas for constructive comments and suggestions. I also thank Amanda Besch for editorial assistance. Department of Finance and Managerial Economics, Naveen Jindal School of Management, The University of Texas at Dallas, 800 West Campbell Road SM 31, Richardson, TX jongmin.oh@utdallas.edu

2 R&D Investments, Technology Spillovers, and Stock Returns ABSTRACT I demonstrate that R&D investment is crucial for firms to effectively absorb and benefit from potential technology spillovers. In the presence of large spillovers, I find that firms with higher R&D earn significantly higher subsequent equity returns. Importantly, when exposed to large spillovers, higher R&D firms generate more patents (citations) per available spillover and exhibit higher operating performance. I further demonstrate that the complementary relation between R&D and spillovers is most pronounced among low investor attention stocks. My findings suggest that the market appears to undervalue the complementary role of R&D, and that limited investor attention contributes to this undervaluation. JEL Classification: G11, G12, G14, O31, O32, O33 Keywords: Spillover, Innovation, R&D, Limited attention, Market efficiency

3 I. Introduction Innovative activities drive a firm s technological advancements. They also generate technology spillovers since they are often sources of new ideas and opportunities that add to the existing knowledge pool (e.g., Schumpeter, 1934; Arrow, 1969; Romer, 1986). The literature on innovations has documented that related spillovers ( spillovers hereafter), spillovers that are coming from technologically-related firms, create positive externalities, and provided evidence that spillovers positively affect a related firm s operating performance (e.g., Jaffe, 1986; Bloom, Schankerman, and Van Reenen, 2013). 1 Based on the real effects of spillovers, recent empirical studies have documented that there exists a positive relation between technology spillovers and subsequent stock returns on average. They argue that this positive relation is driven by the market s undervaluation of spillover effects since identifying spillovers is a difficult task for investors (e.g., Hsu, 2011; Jiang, Qian, and Yao, 2013; Chen, Chen, Liang, and Wang, 2013). However, two related questions may arise. First, are the technology spillovers a free lunch or are there additional costs associated with absorbing the available spillovers? 2 Second, what are the underlying mechanisms that drive the relation between technology spillovers and future stock returns? Answering these questions is crucial for investors since identifying these underlying mechanisms will help them properly allocate their resources toward firms that are more likely to generate higher returns on their innovations. It is also important for managers since understanding the relations will allow them to invest in the innovative projects that will ultimately enhance shareholder value. To answer these questions, I examine whether or not cross-sectional differences in a firm s ability to absorb from available technology spillovers matter and affect the relation between spillovers and 1 For example, Xerox developed the first graphical user interface (GUI) for computers in the early 1970s that Microsoft and Apple used to develop new products (Windows and Macintosh), which became very profitable. 2 The evidence in the literature is mixed. The level of exposure to spillovers itself may positively affect the performance of other technologically-related firms by allowing them to incur low cost (i.e., R&D spending) for innovations since spillovers create positive externalities (e.g., Nelson, 1959; Arrow, 1962; Griliches, 1979; Mansfield, 1977 and 1988; Spence, 1984; and Bernstein and Nadiri, 1989). Alternatively, firms may need to build up sufficient absorptive capacity (i.e., high R&D investments) in order to benefit from external information (e.g., Evenson and Kislev, 1976; Cohen and Levinthal, 1989 and 1990; Henderson and Cockburn, 1996). 1

4 subsequent stock returns. I posit that when exposed to high spillovers firms with higher R&D investments will likely better absorb and convert the spillovers into value-relevant business improvement. Specifically, the complementary relation between high spillovers and high R&D investments rather than merely high exposure to spillovers will be more likely to affect a firm s innovation-related productivity and profitability. My argument builds on the intuitive idea that a firm should be up-to-speed on current technologies in order to benefit from available technology spillovers. For example, a firm that has already undertaken significant research in related areas would better understand and absorb outside knowledge. Existing studies have argued that R&D investments can enhance a firm s ability to recognize, assimilate, and exploit new external information, namely an absorptive capacity (AC) (e.g., Cohen and Levinthal, 1989 and 1990; Henderson and Cockburn, 1996). Firms with higher R&D investments will therefore likely have a higher AC, allowing them to generate more innovative outcomes (i.e., patents or patent citations) given the same level of spillovers and exhibit higher operating performance. Exploring the complementary relation between technology spillovers and R&D investments ( spillover-ac synergy hereafter) leads to my main hypothesis. I hypothesize that it is spillover-ac synergy rather than the size of potential spillovers alone that positively predicts subsequent stock returns. Although firms with greater spillover-ac synergy exhibit higher future innovation productivity as well as operating performance, investors may have difficulty in processing such favorable information. That is, the investors may have difficulty in recognizing size of available spillovers and, more importantly, to what extent a given firm can absorb the available spillovers. From this perspective limited investor attention likely leads to underreaction to this favorable information. 3 Thus, the positive effects of spillover-ac synergy may not be fully valued by the market, leading to positive subsequent stock returns. In order to test these ideas I follow previous studies and use the ratio of R&D-to-sales (R&D intensity) as a proxy for a firm s absorptive capacity (e.g., Cohen and Levinthal, 1989 and 1990; Henderson and 3 See for example Hirshleifer, Hsu, and Li (2012 and 2013) among others in which the authors demonstrate that limited investor attention contributes to underreaction to the hard-to-process favorable information related to firms various innovative activities. 2

5 Cockburn, 1996). Next, I use a measure for technology spillover pools recently developed in Bloom, Schankerman, and Van Reenen (2013), which captures the extent of a firm s exposure to other firms R&D efforts in the similar technological field. Specifically, the technology spillover pool for each firm constitutes the sum of other firms R&D spending weighted by technology similarities. The technology similarity between each pair of firms is estimated by calculating the distance between each firm s technology positions identified as the composition of each firm s patent portfolio. 4 Using these two measures, I find that spillover-ac synergy rather than the size of spillover pool per se positively predicts cross-section of future stock returns. Forming portfolios by double-sorting firms into R&D intensity and spillover pool terciles, I first confirm the existing studies on spillovers that the level of spillover exposure is positively related to future stock returns (e.g., Hsu, 2011; Jiang, Qian, and Yao, 2013; and Chen, Chen, Liang, and Wang, 2013). For example, the return spread between firms with high and low spillovers within high R&D group is 0.58% per month (t = 2.85). Importantly, however, this return predictability does not appear to be significant among firms with low R&D investments. A portfolio of firms with both large spillover pools and high R&D intensity ( PlusSpill portfolio) outperforms a portfolio of firms with the same size spillover pools but low R&D intensity ( OtherSpill portfolio). For example, the value-weighted PlusSpill portfolio earns a 0.78% (t = 3.97) monthly sizeadjusted four-factor alpha, while the OtherSpill portfolio earns 0.12% (t = 0.86). The hedge portfolio that is long in PlusSpill firms and short in OtherSpill firms earns a monthly alpha of 0.66% (t = 2.76) or a yearly alpha of around 8%. The results suggest that the market s undervaluation, if any, is concentrated on firms with both high spillovers and high R&D rather than on firms with just high spillover exposure. I confirm these portfolio analysis results using Fama-MacBeth (1973) cross-sectional regressions of stock returns on lagged R&D intensity, spillovers, the interaction between these two variables, and 4 Using technology similarities to consider related firms rather than only considering firms within the same industry has the advantage of capturing the technology spillovers generated by firms outside the same product market industry (e.g., Bloom et al., 2013). For example, Xerox Corp. and Apple Inc. compete in different product market industries, but share patents in conductors and insulators (USPTO Class 174), circuit makers and breakers (200), radiant energy (250), and many other categories reflected in the measure of technology spillovers used here. 3

6 additional sets of control variables. These results are also robust to potential risk-based explanations such as increased product market competition (e.g., Bloom et al., 2013), financial constraints (e.g., Li, 2011), and sensitivity to down-market periods. Furthermore, the effect of Spillover-AC synergy is robust to inclusion of the other innovation-related factors that have shown to affect future stock returns such as patenting activities (e.g., Deng, Lev, Narin, 1999) and innovative efficiency (e.g., Hirshleifer, Hsu, and Li, 2013). These findings suggest that the spillover-ac synergy rather than high exposure to spillovers alone is likely to be undervalued by the market and that I have uncovered new innovation-related feature that can positively affect the future stock returns. I next explore why the positive relation between spillovers and subsequent stock returns is concentrated in firms with high R&D investments. Specifically, I examine whether or not spillover-ac synergy has significantly stronger real effects on a firm s future innovation productivity and firm profitability, but is less likely to be fully recognized by the market. First, I show that when firms are exposed to large spillovers, firms with higher R&D investments indeed better absorb and convert available spillovers into more successful innovative outcomes (i.e., patent generations or patent citations), allowing them to have higher innovation related productivity given the same level of spillovers. Moreover, firms with spillover-ac synergy not only have higher innovation productivity but also exhibit superior operating performance. For example, firms with higher R&D intensity and higher spillover exposure exhibit significantly higher returns on assets (ROA) and profit margins (PM) over the next year. These results imply that the high future abnormal returns of the PlusSpill portfolio can be at least in part driven by the undervaluation of the increased innovation-related productivity and strong firm fundamentals associated with spillover-ac synergy. Second, I investigate the source of the market s inability to fully recognize the positive effects of spillover-ac synergy. Specifically, I test whether or not the limited investor attention contributes to the positive alpha of the PlusSpill portfolio. If limited attention contributes to the positive alpha of the PlusSpill portfolio, I would expect that the positive relation between spillover-ac synergy and future 4

7 stock returns will be most pronounced among firms with higher degrees of limited investor attention. Running Fama-MacBeth regressions within subsamples of high and low investor attention groups, 5 I find that firms within the low (high) investor attention group have large (small) and significant (insignificant) slope coefficients on the interaction between R&D intensity and spillovers. Furthermore, the difference in the magnitudes between the low and high attention groups is substantial and statistically significant. These findings support the notion that the market does not fully recognize and process information on the positive real effects of spillover-ac synergy, leading to positive future abnormal returns. The rest of the paper is organized as follows. Section II discusses the relevant literature. Section III discusses the data and key variables. Section IV presents the main results on the positive relation between spillover-ac synergy and future stock returns. Section V examines the real effects of spillover- AC synergy. Section VI provides evidence on misvaluation of the spillover-ac synergy. Section VII provides robustness results. Section VIII concludes. II. Related Literature and Contributions This paper is related to three strands of literature. First, This paper contributes to a large and growing body of work showing that the market does not fully incorporate information regarding a firm s innovative activities (i.e., R&D investments) as well as innovation efforts of peer firms (i.e., spillovers) into stock prices (e.g., Lev and Sougiannis, 1996; Deng, Lev, and Narin, 1999; Chan, Lakonishok, and Sougiannis, 2001; Eberhart, Maxwell, and Siddique, 2004; Hsu, 2011; Hirshleifer, Hsu, and Li, 2012 and 2013; Cohen, Diether, and Malloy, 2013; Jiang, Qian, and Yao, 2013; and Chen, Chen, Liang, and Wang, 2013). Recent studies by Chen, Chen, Liang, and Wang (2013) and Jiang, Qian, and Yao (2013) document an average positive effect of peer firm s R&D activities on a given firm s subsequent stock returns. Unlike Chen et al. (2013) and Jiang et al. (2013), my findings provide evidence that exposure to large potential spillovers is a necessary but not sufficient condition for firms to have economically 5 I use analyst coverage and firm size as proxies for investor attention. Firms with low analyst coverage or small size are classified as low attention group (e.g., Hong, Lim, and Stein, 2000; and Hirshleifer and Teoh, 2003). 5

8 meaningful subsequent positive abnormal returns. Firms need to invest in their own R&D to benefit from other firms R&D efforts. Likewise, I provide evidence that information on a firm s high level of past R&D investments alone is not sufficient to generate significant positive alpha, consistent with previous studies (e.g., Chan, Lakonishok, and Sougiannis, 2001; Li, 2011; Cohen, Diether, and Malloy, 2013). However, I provide further evidence that in the presence of large technology spillovers, a firm s high R&D investments contain information not only on high innovative input levels but also on high absorptive capacity such that firms with high R&D investments earn significantly higher alphas when they are exposed to larger spillovers. Therefore, what seems to be misvalued by investors is not the information on the size of technology spillovers or the level of R&D investments alone, but instead on the strong complementary relation between the spillovers and R&D investments (Spillover-AC synergy). My findings are also related to the literature on how other firms innovative activities have an impact on a given firm s managerial decisions. For example, Qiu and Wan (2015) have provided evidence that spillovers have positive relation with a given firm s cash holding since the firm wants to meet the potential needs of future innovations influenced by the technological opportunities. Bena and Li (2014) and Sevilir and Tian (2012) have documented a firm s incentive to acquire an innovation-intense targets to create synergy on innovations. By providing evidence that firms need to maintain high innovative capacity to better benefit from external technological opportunities, findings in my paper suggest that a firm s incentive to hold more cash or incentive to acquire the innovation-intense targets may be a rational managerial response to benefit from its own as well as peers innovative activities. Finally, this paper adds to the limited attention literature that has shown that the market fails to fully incorporate information into stock prices when the information lacks saliency (e.g., Klibanoff, Lamont, and Wizman, 1998; Huberman and Regev, 2001; Hirshleifer and Teoh, 2003; Lev, 2004; Peng and Xiong, 2006; DellaVigna and Pollet, 2009; Hirshleifer, Lim, and Teoh, 2009 and 2011; Hou, Peng, and Xiong, 2009). I provide evidence in line with these studies by finding that the complementary relation between a firm s R&D and spillovers is most pronounced among low investor attention stocks. 6

9 III. Data, Innovation and Spillover Measures, and Descriptive Statistics A. Data I build my sample primarily from three sources of data: the National Bureau of Economic Research (NBER) U.S. Patent Data, the Center for Research in Security Prices (CRSP) database, and the COMPUSTAT. The NBER U.S. Patent Data contains detailed information on all utility patents granted by the U.S. Patents and Trademark Office (USPTO) from 1976 to The use of patent data restricts my sample to either firms that have received patent grants during my sample period or firms that have been confirmed to have zero patent grants. The NBER patent data provides additional matching data linking the patent assignee classification to COMPUSTAT firm-level identifiers which I use to merge patent data with the COMPUSTAT and CRSP data (e.g., Hall, Jaffee, and Trajtenberg, 2001). I use the common stocks trading on the NYSE, AMEX, and NASDAQ exchanges from the CRSP monthly stock file. I add delisting returns to each firm s monthly returns if it is delisted during my sample period. I filter firm-year observations from the COMPUSTAT database following the standard procedures implemented in previous studies by dropping financial firms (SIC codes ). I mitigate backfilling bias following previous literature: firms must be listed on COMPUSTAT at least for two years before they are included in my sample. I require firms to have nonnegative and nonmissing total book value of assets or equity, and also have positive R&D expenditures. Following Fama and French (2006), and Ciftci and Cready (2011), I exclude firms with either extremely small size or sales volume in order to mitigate the influence of these small firms. 6 I remove firm-year observations with abnormally large jumps in either sales or number of employees since these are likely to reflect significant restructuring activities such as mergers and acquisitions even if the company keeps the same firm identifier (e.g., Bloom, Schankerman, and Van Reenen, 2013). Specifically, I follow Bloom et al. (2013) and drop a firm-level observation if either the sales growth or employee growth exceeds 200% or 6 I exclude firms with total book value of assets or sales less than US $5 million or book value of equity below US $2.5 million. However, I find similar results even if I include these firms in my sample. 7

10 falls by more than 66% (approximately the top and bottom 0.5 percentiles). I finally merge the resulting data with the NBER patent data. I use sample period ranges from 1976 to 2006 for constructing variables. However, for all of the analyses I use sample period from July of 1982 to June of 2007 since some of the measures are constructed using multiple years of past data. B. Measure of Technology Spillover Pool My hypothesis builds on the idea that spillovers create positive externalities (knowledge pool) that other firms and entities can absorb and benefit from. I accordingly estimate a technology spillover pool measure based on the similarity of technologies among firms (e.g., Jaffe, 1986; and Bloom, Schankerman, and Van Reenen, 2013). Following Bloom et al. (2013) I first construct a vector of a firm s patent shares within each technology class assigned by the USPTO in order to identify the firm s position within the technology space. I then measure the technology similarity by calculating the Mahalanobis distance between each pair of firm positions within the technology space. However, I depart from Bloom et al. (2013) and use only information on a firm s patent shares up to time t in constructing spillover pool measure at time t in order to prevent any potential look-ahead bias. 7 Specifically, I first define the matrix (T, N) at time t as X t = [, ], where X i,t is a vector of a firm s patent shares within each T technology class, X i,t = (x i1,t, x i2,t x it,t ), and x iτ,t is firm i s (i = 1, 2,, N) proportion of patents in technology classification τ over the period up to time t. I also use rolling method as a robustness check instead of using whole information up to time t since a firm s innovative focus on certain technology areas may change over time. I accordingly calculate a firm s patent shares at time t within each T technology class using number of patents granted in technology class τ over the previous three years (t to t-2). 8 7 I also use the original measure incorporating information over the whole sample periods and find similar results, confirming that the original measure does not suffer from a serious look-ahead bias (unreported). 8 I also test five-year rolling window (unreported) and find similar results. 8

11 Each row of X t contains a firm s patent shares within the T technological classes and each column stands for firm i s composition of patent shares across all T patent classes. I then normalize X t with a firm s patent share dot product obtaining the (T, N) matrix such that each element of is the uncentered correlation between firms. 9 For example, element (i, j) of is the correlation between firm i and j based on their patent portfolio similarity. 10 I next define a (T, T) matrix at time t as measure between patent classes. Accordingly, in which each element is now the correlation captures technology spillovers across patent classes within a firm. For example, if patent classes i and j coincide frequently within the same firm then will be close to 1 (with ). I then calculate the Mahalanobis distance technology closeness measure as: where the (i, j) element in matrix ( ) measures the correlation weights of the overlap in patent shares between firm i and j at time t by how close their different patents shares are to each other. Finally, a measure of technology spillover pool for each firm i in year t is: where is firm j s R&D expenditure in year t. The then represents dollar value of potential technology spillover pool for firm i in year t. Table 1 reports the average Spilltech along with standard innovative activity measures (i.e., number of patents and citations) for selected Fama-French (1997) 48 industries. I select the top and bottom 10% 9 For example, can be expressed as. 10 Jaffe (1986) constructs pool of technology spillover measure as the weighted sum of other firms R&D with weights calculated as. More specifically, Jaffe (1986) measures the (i, j) element in P t as the uncentered correlation between firm i and j s patent portfolios and is calculated as:. However, this measure ignores the possibility of technology spillovers across technology classes. 9

12 industries based on their average Spilltech rankings. As shown in Table 1 both Spilltech and the innovation measures exhibit huge variation across industries, suggesting that it is important to account for simple industry effects. I accordingly construct my main measure for a firm s technology spillover pool by adjusting for industry effects ( ). Specifically, I scale Spilltech it by corresponding industry average Spilltech in the same year based on Fama-French 48 industry classifications. As a robustness check I also estimate the firm-by-firm correlation matrix TECH t using the three-year rolling method and construct a rolling-based technology spillover pool measure,. C. R&D Intensity and Innovation-Related Productivity Measures Since my hypothesis focuses on a firm s absorptive capacity (AC) I need a proxy for a firm s AC that enables a given firm to effectively benefit from spillovers. Following Cohen and Levinthal (1989 and 1990), and Kumar and Li (2013), I use a firm s R&D intensity as a proxy for AC. I use R&D expenditure scaled by sales volume (RD/sales) as an R&D intensity measure. RD/sales is one of the most widely used measures for a firm s R&D intensity (e.g., Cohen and Levinthal, 1989 and 1990; Lev and Sougiannis, 1996; Chan, Lakonishok, and Sougiannis, 2001; Li, 2011; Kumar and Li, 2013; Cohen, Diether, and Malloy, 2013; and Almeida, Hsu, and Li, 2013). As a robustness check I also use R&D expenditure scaled by the total book value of assets (RD/AT) as an alternative measure for R&D intensity. I next construct a firm s innovation-related productivity measures in order to test whether or not firms with spillover-ac synergy exhibit higher future productivity in generating successful innovative outcomes per unit of available spillover. I build two measures for the innovation-related productivity: 1) Citations/Spill defined as the total number of adjusted patent citations received from the grant year to 2006 scaled by industry-adjusted spillovers in year t,, and 2) Patents/Spill defined as the total number of adjusted patents applied in year t scaled by industry-adjusted spillovers in year t,. 10

13 Since previous literature on innovation has documented that a firm s measures for innovative activities (i.e., number of patents being granted or citations received) show large variations across different technology classes and application years, I follow the literature and adjust for these issues (e.g., Seru, 2010; Bena and Garlappi, 2012; Almeida, Hsu, and Li, 2013; and Hirshleifer, Hsu, and Li, 2013). Specifically, I calculate adjusted patents as the number of patents within each technology class divided by the cross-sectional average number of patents applied in the same year and assigned to the same technology class by the USPTO. I also calculate adjusted citations by scaling the number of citations received from each patent by the average number of citations received by patents applied in the same year and assigned to the same technology class. My innovation-related productivity measure Citations/Spill (or Patents/Spill) differs from existing innovation-related measures used in previous literature. For example, Gu (2005) and Deng, Lev, and Narin (1999) focus on innovative output (i.e., number of patents or citations) by scaling innovative output by a firm s total assets, whereas Citations/Spill (or Patents/Spill) measures how productive is a firm in generating innovative outputs per unit of available technology spillover. Citations/Spill (or Patents/Spill) is related but different from the innovative efficiency measures used in Hirshleifer, Hsu, and Li (2013), and Almeida, Hsu, and Li (2013) in that innovative efficiency measures focus more on the productivity of a firm s own R&D investments whereas Citations/Spill (or Patents/Spill) focuses on a firm s productivity in generating innovative outputs per unit of available technology spillover. My innovation-related productivity measures therefore allow me to directly test whether or not a firm s high R&D intensity enhances a firm s ability to effectively absorb outside knowledge and convert it into a given firm s own innovative outputs when combined with a large technology spillover pools. D. Descriptive Statistics Panel A of Table 2 presents the average characteristics for the different portfolios formed by doublesorting independently on a firm s R&D intensity (RD/sales) and technology spillovers ( ). 11

14 Specifically, I form portfolios every year at the end of June for year t by independently double-sorting based on the 30 th and 70 th percentiles of a firm s R&D intensity and technology spillover levels from the fiscal year ending in calendar year t-1. I then hold these portfolios over the next 12 months (i.e., July of year t to June of year t+1). The average number of firms in the RD low /Spill high and RD high /Spill high portfolios, which are the main focus of this paper, is 82 and 89 respectively. The average (median) size as measured by market capitalization at the end of June of year t for these two portfolios is $6, ($1,071.6) million and $7, ($769.5) million respectively. Moreover, untabulated results show that firms in the RD low /Spill high and RD high /Spill high portfolios account on average for 7.11% and 8.36% of the total U.S. market capitalization respectively. Firms in these two portfolios (about 15.5% of the U.S. total equity market) therefore constitute an economically meaningful portion of the U.S. stock market. In addition, RD high /Spill high portfolio firms tend to be growth stocks compared to firms in the RD low /Spill high portfolio. Within each R&D intensity group, there seem to be significant variations in the industry-adjusted spillover measure. The Spill adj for Spill low and Spill high group are 0.46 and 1.94 respectively within the RD low group, and 0.55 and 1.97 respectively within the RD high group. Additionally, the average innovation-related productivity is higher for firms in RD high group than firms in RD low group for each spillover group. For example, firms in the RD high /Spill high portfolio show greater innovation-related productivity than firms in same Spill high group that have low R&D intensity (RD low group). Moreover, the innovation-related productivity increases with the potential spillover pools within each RD group. However, one thing worth noting is that firms Citations/Spill within the RD high /Spill low portfolio is higher than Citations/Spill in the RD low /Spill high portfolio firms, suggesting that larger spillover pools do not necessarily guarantee higher innovation-related productivity. Instead, R&D intensity seems to play an important role in determining a firm s innovation productivity. I will explore this possibility in Section V.A. 12

15 Panel B of Table 2 reports pair-wise correlations among R&D intensity, spillover, and other firm characteristics. Surprisingly, a firm s R&D intensity (RD/sales) and spillovers (Spill adj ) have very low correlation (1%), indicating that there exists substantial variations in R&D intensity (RD/sales) unrelated to a firm s technology spillover exposures (Spill adj ). One possible explanation would be that how much is spent on R&D (R&D intensity) and how a firm should allocate its R&D spending across different technology areas (exposure to spillovers) are two different dimensions of R&D spending. I exploit this variation in the R&D intensity across firms in order to examine an effect of a firm s increased absorptive capacity on subsequent stock returns. Lastly, one thing worth noting is that and Patents are both increasing with a firm s market value, suggesting that it is important to control for firm size. IV. R&D Investments, Technology Spillovers, and Equity Returns A. Empirical Methodology In this section I briefly explain the key empirical methodologies for testing whether or not spillover- AC synergy positively predicts future stock returns. I use a calendar-time portfolio regression approach (e.g., Hirshleifer, Hsu, and Li, 2012 and 2013; and Cohen, Diether, and Malloy, 2013) and Fama- MacBeth (1973) cross-sectional regression approach. Since Panel B of Table 2 suggests that spillovers show strong positive correlation with firm size, it is important to adjust for the size effects when analyzing portfolio returns. I calculate the monthly size-adjusted portfolio returns following Hirshleifer et al. (2012 and 2013). At the end of June of year t, I sort firms independently into two size groups (S small or S big ) based on the median NYSE market capitalization breakpoints at the end of June of year t, R&D intensity (RD/sales) terciles (i.e., RD low and RD high below the 30 th and above the 70 th percentiles of RD intensity), and spillover (Spill adj ) terciles (i.e., Spill low and Spill high below the 30 th and above the 70 th percentiles of Spill adj ). R&D intensity and spillovers are measured in fiscal year ending in the calendar year t-1. I consequently form portfolios from the intersection of the firm s size, R&D, and spillover sorts. 13

16 I then hold these portfolios from July of year t until June of year t+1 and calculate the value-weighted monthly returns for each portfolio. Finally, I calculate six size-adjusted monthly portfolio returns by averaging the value-weighted portfolio returns across the different size groups. 11 I mitigate the effects of market microstructure by requiring firms to have stock prices higher than US $5 at the time of portfolio formation. I then employ the Carhart (1997) four-factor model in order to adjust for the well-known risks. For the Fama-MacBeth regression approach I run a cross-sectional regression of monthly stock returns minus the risk-free rate on lagged R&D intensity, technology spillovers, and the interaction between these two variables along with firm characteristics. The independent variables are from fiscal year ending in the calendar year t-1 except size and momentum. B. Portfolio Returns I provide portfolio analysis results in this section. Panel A of Table 3 presents the monthly sizeadjusted portfolio returns in excess of one-month Treasury bill rates for each portfolio using RD/sales and spillover measure Spill adj. First, the monthly portfolio return patterns within RD high group are consistent with the previous spillover literature that has shown that the level of exposure to spillovers positively predicts subsequent stock returns (e.g., Hsu, 2011; Chen, Chen, Liang, and Wang, 2013; and Jiang, Qian, and Yao, 2013). For example, the last column of Table 3 shows that the spread between Spill high and Spill low portfolio within the RD high group is 0.58% per month (t = 2.85). However, the level of spillovers seems to have positive stock return predictability only for firms in the RD high group. A portfolio of firms exposed to large technology spillover pools that also have high R&D investments ( PlusSpill portfolio hereafter) outperforms a portfolio of firms with high amount of spillover pool but low R&D investments ( OtherSpill portfolio hereafter). The value-weighted monthly size-adjusted four-factor alpha for the PlusSpill portfolio (0.78% with t = 3.97) is approximately seven times greater 11 For example, monthly portfolio return of firms with both high R&D intensity and spillovers can be calculated as. Results without the size-adjustments are in the Appendix Table B2. 14

17 the magnitudes compared to the monthly alpha of the OtherSpill portfolio (0.12% with t = 0.86). Moreover, the monthly portfolio return of the OtherSpill portfolio not only demonstrates lower abnormal returns but is also statistically insignificant at the 10% significance level. This suggests that firms that do not invest enough in their own innovations do not appear to be able to effectively absorb and benefit from spillover pools even though the pool is large. This is consistent with previous studies showing the role of a firm s own R&D investments in absorbing external knowledge (e.g., Evenson and Kislev, 1976; Cohen and Levinthal, 1989 and 1990; and Kumar and Li, 2013). Column 7 of Table 3 presents the monthly alphas of a spread portfolio that is long in PlusSpill firms and short in OtherSpill firms. This zero-investment portfolio earns a 0.66% monthly alpha with t = However, despite the size adjustments, higher loading on the size factor,, for the R&D high firms compared to R&D low firms may suggest that higher alphas of the PlusSpill firms may simply reflect the size differentials of the R&D high versus R&D low within the Spill high group. I will address this issue further in the next section. One thing worth noting is that the stock return spread between RD high and RD low firms for the Spill high group is greater than that of the Spill low group, implying that high R&D contains information not only on a firm s innovative input level but also on its high absorptive capacity. Panel B of Table 3 reports the robustness of the portfolio results by using the alternative spillover pool measure discussed in Section III.B. Specifically, I use in order to allow for the timevarying components of a firm s technology proximities to other firms. Using alternative technology spillover pool measures exhibits similar results. The value-weighted monthly size-adjusted four-factor alpha for the PlusSpill portfolio (0.71% with t = 3.45) is significantly greater than the alpha of the OtherSpill portfolio (0.11% with t = 0.61) with a spread portfolio earning a 0.60% (t = 2.08) monthly alpha. Additionally, the monthly return for the OtherSpill portfolio is statistically insignificant. These results confirm the results in Panel A. As an additional robustness check I use RD/AT as a proxy for a firm s R&D intensity when sorting firms into R&D intensity terciles; these results are provided in the Appendix. As shown in Table B1, the 15

18 results are consistent with the main results in Table 3 that the PlusSpill portfolio outperforms the OtherSpill portfolio. Overall, the results in Table 3 demonstrate that there exists a strong complementary relation between a firm s high R&D investments and high exposure to spillovers ( spillover-ac synergy ); this spillover- AC synergy rather than the size of the spillover pool alone positively predicts subsequent stock returns. This implies that spillover-ac synergy is most likely to be undervalued relative to the Carhart fourfactor model benchmark. That is, the market does not seem to distinguish between spillover-ac synergy and merely high exposure to spillovers. My findings therefore suggest that the positive future abnormal stock returns found in the previous spillover literature (e.g., Hsu, 2011; Chen, Chen, Liang, and Wang, 2013; and Jiang, Qian, and Yao, 2013) are mainly driven by firms enjoying spillover-ac synergy. C. Fama-MacBeth Cross-Sectional Regression Results In this section I confirm the main results that spillover-ac synergy rather than the size of spillovers alone have stock return predictability. I employ a monthly Fama-MacBeth (1973) cross-sectional regression of individual excess stock returns (individual stock returns over the risk-free rate) on R&D intensity, spillovers, and firm control variables. The key independent variables are ln(1+rd/sales), ln(1+spill adj ), and the interaction between these two variables ln(1+rd/sales)*ln(1+spill adj ). Coefficient of interest is the coefficient on the interaction term. I then use the familiar cross-sectional controls shown to predict future stock returns in order to mitigate concerns that my main results in the previous section are not driven by some cross-sectional firm characteristic variation that I cannot fully control for in portfolio analyses. The control variables I use are firm size, book-to-market, and momentum. Firm size ln(me) is the natural log of the market capitalization at the previous month, book-to-market ln(b/m) is the natural log of the book value of equity in the fiscal year ending in calendar year t-1 divided by the market value of equity at the end of calendar year t-1, and momentum r -12,2 is the prior 12-month returns with a one-month gap between the 16

19 holding period and current month. For one of these models I additionally include industry fixed-effects based on the Fama and French 48 industry classifications. Table 4 presents the results for the monthly Fama-MacBeth regressions. Column 1 shows the results without interaction terms in order to compare the results to previous studies using a spillover measure alone (e.g., Jiang, Qian, and Yao, 2013; and Chen, Chen, Liang, and Wang, 2013). By examining the coefficients on the ln(1+spill adj ) the level of spillover pools alone seem to positively predict future stock returns. However, when I include interaction terms as in Columns 2 and 3 the results show that a firm s level of R&D investments play an important role in determining a technology spillover s future stock return predictability. That is, the marginal value of potential technology spillovers is increasing in a firm s R&D intensity. For example, Column 2 shows that the coefficient on the ln(1+rd/sales)*ln(1+spill adj ) term is positive (4.5%) and statistically significant, while the coefficient on the ln(1+spill adj ) term becomes economically and statistically insignificant. Since the average ln(1+spill adj ) for firms in the Spill high portfolio is 1.1 and the average ln(1+rd/sales) for firms in the RD high (RD low ) portfolio is 0.2 (0.01), high R&D firms with large spillover pools will earn future stock returns of 0.99% on average whereas low R&D firms will earn only 0.05% returns on average even though the spillover pools are large. Additionally, Column 3 reports the results running the same model as in Column 2 but including industry dummies in order to control for any industry-level characteristics that may be driving my results. Adding industry-fixed effects does not change either the economic or statistical significance of the results. I also use dummy variables for the R&D intensity to better compare the return differences between the high and low R&D firms with given level of exposure to the spillovers. As in the portfolio analysis, I sort firms by their R&D intensity (RD/sales) of the year t-1 and define firms as R&D low (R&D high ) being below the 30th percentiles (above the 70th percentiles) of RD/sales. The results are consistent with the results in Table 4 and are provided in the Appendix Table B3. Finally, as a robustness check, I test whether the higher alpha of the R&D high /Spill high shown in Table 3 may simply be driven by effect of the size differentials between the R&D high and R&D low groups within 17

20 the Spill high group as evidenced by the higher loading on the size factor,, for the R&D high group. In addition to firm size control, I accordingly include control variable, the interaction between size and the spillovers ln(me)*ln(1+spill adj ), to combat with this potential concern. Column 4 of Table 4 shows that including the ln(me)*ln(1+spill adj ) term does not affect the coefficient on ln(1+rd/sales)*ln(1+spill adj ). In sum, the results in Column 3 are consistent with the portfolio analysis results that firms exposed to large spillover pools outperform in the future when these large spillover pools are coupled with high past R&D spending (high AC). That is, firms without sufficient AC (low R&D intensity) do not seem to enjoy the full benefits of technology spillovers, and stock return predictability therefore seems to be concentrated among firms with high spillover-ac synergy. V. Real Effects of Spillover-AC Synergy In this section I examine whether or not spillover-ac synergy has real effects. I accordingly test whether spillover-ac synergy is positively related to a firm s innovation productivity and future operating performance. A. Effect of Spillover-AC Synergy on Innovation-Related Productivity My findings so far show that when firms are exposed to large spillovers, firms with high R&D outperform firms with low R&D investments. In this section I explore one of the possible channels through which firms with high spillover-ac synergy earn higher future stock returns. If there exists some heterogeneity across firms in absorbing outside knowledge then firms with higher absorptive capacity (AC) will likely to better absorb and convert the available spillovers into valuerelevant firm performance (e.g., Evenson and Kislev, 1976; and Cohen and Levinthal, 1989 and 1990; and Kumar and Li, 2013). I therefore hypothesize that a firm s level of AC proxied by R&D intensity is positively related to a firm s innovation-related productivity defined as the number of future patents granted (or patent citations received) per unit of available technology spillover. Furthermore, this gain in 18

21 the productivity would be greater for firms that have larger spillover pools. I accordingly run the following annual Fama-MacBeth regressions: where is the innovation-related productivity measure. I measure a firm s innovation productivity at year t in two ways as described in Section III.C: the number of adjusted patent citations received from the grant year until 2006 by a firm s patents applied in year t per unit of a firm s spillover pool at year t (Citations/Spill), and the total number of adjusted patents applied in year t per unit of a firm s spillover pool in year t (Patents/Spill). (RD/sales) at year t-1, is the natural logarithm of one plus R&D intensity is the natural logarithm of one plus Spill adj at year t-1, and the X i,t-1 is a vector of control variables. I use size as measured by the natural log of market capitalization at the end of year t-1, ln(me i,t-1 ), book-to-market as measured by the log of book value to the market value of equity as ln(b/m i,t-1 ), firm leverage at the end of year t-1 as ln(1+lev i,t-1 ), and finally firm age at the end of year t-1 as measured by the years since a firm s first appearance on COMPUSTAT ln(age i,t-1 ). I also include industry fixed-effects based on Fama and French 48 industry classifications. To compare the impacts of each independent variable, I standardize all variables used in this model. Table 5 provides the results. Column 1 in both Panel A and B illustrates the relations between a firm s R&D intensity and future innovation-related productivity controlling for the basic firm characteristics (i.e., firm size and book-to-market). Both Citations/Spill and Patents/Spill show significantly positive relations with a firm s past R&D intensity. As shown in Column 2 for both panels, including the spillover measure does not seem to change the strong relation between a firm s R&D intensity and future innovation productivity. These results are therefore consistent with previous studies documenting R&D investments as a key source for a firm s ability to absorb new external information (e.g., Evenson and Kislev, 1976; Cohen and Levinthal, 1989 and 1990; and Kumar and Li, 2013). 19

22 I next include the interaction term between R&D intensity and spillovers in Column 3 for both Panel A and B in order to test whether or not the impact of a firm s R&D investments on future innovationrelated productivity varies with the size of potential spillover pools. In Column 4 for both panels, I also include additional control variables (i.e., firm leverage and firm age). These results show that the positive impact of a firm s R&D intensity on future innovation productivity seems to be strengthened when the size of the spillover pool is also high. Specifically, the coefficients on the interaction terms are positive and statistically significant for both Citations/Spill and Patents/Spill. For example, for Citations/Spill the non-standardized (standardized) marginal gross R&D effect for the average firms in the high spillover group is 2.7 (0.24), whereas it is 1.4 (0.13) for the average firms in the low spillover group. The results in Table 5 therefore imply that firms with high R&D and high spillover exposures will likely be the most productive in generating future innovative outcomes per unit of available spillover. Importantly, the overall effect of R&D intensity (sum of coefficients on the R&D intensity plus the interaction term) exhibits one of the strongest determinants of a firm s subsequent innovation-related productivity (i.e., the largest in magnitudes except for the firm size) when combined with the interaction effects. For example, in the presence of high potential spillovers the Citations/Spill of the average firms in the RD high group will have approximately 20 times greater Citations/Spill than the average firms in the RD low group. This implies that it is crucial for firms to invest highly in their own R&D and remain up-tospeed on current technologies in order to effectively benefit from spillovers. As a robustness check I also use unadjusted patents or citations in calculating innovation-related productivity. Specifically, I calculate the two alternative innovation-related productivity measures as: 1) (Citations/Spill) Alt defined as the total number of patent citations received from the grant year t to 2006 scaled by stock of spillovers in year t-2, and 2) (Patents/Spill) Alt, defined as the total number of patents granted in year t scaled by stock of spillovers in year t These results are provided in the Appendix. 12 Following the previous literature in constructing R&D stock, I calculate the stock of spillovers at year t as: (e.g., Chan, Lakonishok, and Sougiannis, 2001; and Hirshleifer, Hsu, and Li, 2013). 20