MANAGERIAL ABILITY AND FIRM INNOVATION

Transcription

1 MANAGERIAL ABILITY AND FIRM INNOVATION Bill B. Francis* Iftekhar Hasan** Gokhan Yilmaz*** September 2014 Abstract Using the managerial ability measure developed by Demerjian et al. (2012) and the NBER patent data through , we find that the managerial ability is positively related to the quantity and quality of future innovative output as measured by the number of granted patents and the average nonself citations received per patent. In addition, innovative efficiency as measured by the fraction of number of patents and total forward citations over the firms R&D capital increases with the managerial ability. Evidence from difference-in-differences estimation using the CEO turnover data show that the relationship between managerial ability and innovation is causal. Furthermore, the patents of firms with more able managers receive significantly more nonself citations while they make significantly less citations to the others and have significantly lower self-citation ratio which indicate that these firms can pursue explorative and novel innovation. Lastly, using the general ability index developed by Custódio, et al. (2013), we find that the patents of firms with more able CEOs cite and are cited by patents that belong to a significantly wider range of fields when the CEOs are generalists rather than specialists. Results are robust to propensity score matching and are obtained after controlling for firm and year fixed effects, past innovative performance, compensation incentives, optimism and a broad vector of relevant control variables that are used in the innovation literature. JEL Classification: G34, O31, O32, O34 Keywords: Managerial Ability, Innovation, Innovative Efficiency, R&D, Patents. * Lally School of Management, Rensselaer Polytechnic Institute. francb@rpi.edu ** Fordham Schools of Business, Fordham University. ihasan@fordham.edu ***Corresponding author. Labovitz School of Business and Economics, University of Minnesota Duluth. gyilmaz@d.umn.edu

2 1. Introduction Innovation is of utmost importance for firms in sustaining their competitive advantage. Increasing market competition and the rapidly changing technological environment urge firms to innovate more to be successful. In this sense, the shareholders wealth is closely linked with the innovative success of their firms. Accordingly, empirical evidence show that innovative efficiency as measured by the number of patents and citations scaled by R&D expenditures can predict future stock returns (Hirshleifer, Hsu and Li (2013)). Among the wide array of factors that determine innovative success, we would expect talent to play a key role. In today s corporate framework where decisions are made by the top management team, the collective managerial ability and ultimately the CEOs ability is particularly important since the CEOs can exert the most influence in decisions made by the top management team. Given this importance, in this paper, we analyze the relationship between the managerial ability and innovation in terms of its quantity, quality and efficiency as measured by the number of patents, the number of forward citations per patent and the number of patents and citation-weighted patents over R&D expenditures, respectively. Existing literature has documented that managers, particularly CEOs matter for their firms in many aspects. We understand this from the heterogeneity that exists in CEO compensation, the non-uniformity of the decisions that they make for their firms and the outcomes of these decisions. For example, CEOs are paid accordingly with their ability, skill set and characteristics which also have very important implications for firm performance. Theoretical literature on CEO compensation has modeled CEOs with various skills in explaining the increasing trend in CEO compensation (e.g. Murphy and Zábojník (2004), Gabaix and Landier (2008)). On the empirical side, evidence show that there is a pay premium for CEOs with transferable skills and reputational, career and educational credentials (e.g. Custódio, Ferreira and Matos (2013) and Fallato, Li and Milbourn (2014)). Our paper relates more closely to the literature on corporate performance which has documented that managerial characteristics and behavior impact firms decisions and performance. For example, Bertrand and Scholar (2003) show that manager fixed effects and management style impact the firms financial policy, investment policy and organizational strategy. Adams, Almeida and Ferreira (2005) find that specific CEO characteristics, in conjunction with organizational 1

3 variables has important implications for firm performance. Kaplan, Klebanov and Sorensen (2012) find that firm performance is related to CEOs general ability and execution skills. Hirshleifer, Low and Teoh (2012) show that overconfident CEOs increase investment in innovation, obtain more patents and patent citations and utilize their R&D expenditures more efficiently in obtaining these outcomes. Given the findings of the existing literature, we would expect managerial characteristics to impact innovation as well which also relates to the investment policy and organizational strategy of firms. There are several other reason why more able managers may promote more innovation and do this more efficiently. Firstly, managers who are known to be more successful would possess more ability. One among many examples is Steve Jobs of Apple Inc. Apple s transition from a troubled company to one of the most innovative global companies is mostly attributed to Job s ability in innovating path breaking products that led sales to skyrocket. After his demise, the common question was if Apple could sustain its success in his absence. Secondly, we would expect more able CEOs to more willingly pursue innovation. Innovation is inherently risky and it may result in the dismissal of the CEO in case it fails. Thus, career concerns of CEOs also play a role in nurturing the innovation. Aghion, Van Reenen and Zingales (2013) show that when career risk associated with risky projects are lesser, institutional ownership leads to more innovation. In a similar respect, we conjecture that CEOs with higher managerial ability would be less concerned about their job prospects compared to CEOs with lesser ability. As a result, they would be more willing or less reluctant in pursuing innovation. Thirdly, it is likely that more able CEOs can obtain more output with less inputs. Their skill in navigating varying economic conditions and a broader understanding of the long-term goals of a firm help them allocate company resources more efficiently. As a result, firms may innovate more and better. Also, more able managers may do a better job in freeing up more inputs to innovate more. Literature, as exemplified above, have successfully quantified and proxied for managerial ability and characteristics. Several measures that have been used are CEOs education, tenure, age at the time of appointment, previous work experience, media coverage and industry-adjusted firm performance (e.g. Rose and Sheapard (1997), Milbourn (2003), Custódio, Ferreira and Matos (2013), Custódio and Metzger (2013)). Other scholars have used behavioral characteristics of 2

4 CEOs, such as overconfidence, in explaining corporate phenomena (e.g. Malmendier and Tate (2005, 2008), Malmendier, Tate and Yan (2011), Kaplan et al. (2012), and Hirshleifer et al. (2012)). The managerial ability measures that we mentioned above mostly include the CEOs and their firms in the Execucomp database which starts from On the other hand, Demerjian, Lev and McVay (2012) quantify managerial ability for a broad sample of Compustat firms. They use Data Envelopment Analysis (DEA) and easily obtainable financial data to calculate the managers efficiency in generating revenues given various inputs of the revenue generating process. They find that this measure is strongly related to CEO fixed-effects. The premise behind the advantage of this measure is that it outperforms other ability measures such as historical stock returns and media citations through several validity and robustness tests that authors conduct. In that respect, authors show that the DEA based managerial ability is economically and significantly associated with manager fixed effects, negatively associated with CEO turnover announcement returns and the relative changes in this ability measure is associated with subsequent firm performance. Also, this DEA based ability measure, provided by the authors, starts from 1980 and overlaps more closely with the NBER patents data that spans from 1976 to Therefore, we adopt the DEA based managerial ability measure of Demerjian et al. (2012) in this paper as our key variable of interest in investigating the link between managerial talent and firm innovation. Differently, Custódio et al. (2013) develops a general managerial ability measure that quantifies the generality (transferability) of CEOs skills for the CEOs in the Execucomp universe. They obtain their measure through a factor analysis of the number of different positions the CEOs have held, the number of firms they worked at, the number of industries that they have been exposed to and the number of previous positions that they worked as CEOs. From the combined score that they compute, they define CEOs who score above the median as generalists. Otherwise, the CEOs are classified as specialists. We adopt and use their measure for a subsample test in explaining the link between a firm s innovation and the innovation pursued by other firms. 1 Our empirical findings support our views. Using the DEA based managerial ability measure developed by Demerjian et al. (2012) and the NBER patent database 2 through , we find 1 We thank Cláudio Custódio, Miguel Ferreira and Pedro Matos for making their general ability index publicly available. 2 NBER U.S Patent Data are described in detail in Hall, Jaffe and Trajtenberg (2001). 3

5 that the managerial ability is positively related to the quantity and quality of future innovative output as measured by the number of granted patents, and the average nonself citations received per patent, respectively. In addition, innovative efficiency as measured by the fraction of above innovation measures over the firms R&D capital increases with the managerial ability. To address the possibility that our findings might be driven by the differences in firm characteristics that have access to higher ability CEOs through hiring and the firms that do not have the similar access, we propensity score match these two groups firms based on the control variables that we use in our regressions. This procedure identifies the firms that are similar in all observed aspects but the managerial ability. We denote firms that have above the median DEA based managerial ability as the treated firms. Otherwise, they are denoted as control firms. We compute the average treatment effect, with respect to innovation, on the treated firms (ATT). We find that the firms with above the median managerial ability generate significantly more innovation. We also find that these firms have significantly higher innovative efficiency. In order to provide additional robustness to our estimations and establish causality for the relationship between the managerial ability and innovation, we create a CEO turnover sample using the Execucomp database and employ the difference-in-differences (DiD) methodology. If managerial ability drives innovation, we should observe an improvement in innovation following a CEO change when the new CEO is of higher ability than the replaced CEO. Although the DEA based managerial ability measure is attributable not only to the CEO but to the entire management team, we feel safe in using the CEO turnover as an exogenous event to investigate the causality by relying on the evidence of Demerjian et al. (2012) that the ability measure is highly correlated with the CEO fixed effects (rho=0.835) and that 60.5% of the CEO fixed effects are significant when the DEA based ability measure is regressed on the CEO fixed effects and firm fixed effects with the standard errors clustered by firm and year. Our DiD analysis using this turnover sample show that the relationship between managerial ability and innovation is causal. Management literature has documented that firms have diverse innovation strategies such as exploratory and exploitative innovation (e.g. McGrath (2001), Danneels (2002), Benner and Tushman (2003) and Jansen et al. (2006)). Some firms build and develop their future patents incrementally on their pool of existing patents (i.e. exploitative strategy) and others pursue innovation more radically in areas that are relatively new and unexplored (i.e. explorative strategy). We expect more able managers to be more willing to undertake an explorative strategy 4

6 due to the reasons we discussed above. Consistent with our expectation, we find that the patents of firms with more able managers receive significantly more nonself citations while they make significantly less citations to the others. These firms also have significantly lower self-citation ratio which indicate that they can pursue explorative and novel innovation. Next, we examine how patented innovations are linked to other innovations. We find that the patents of firms with more able CEOs cite patents that belong to a significantly wider range of fields (high originality). Also, patents of these firms are cited by other patents that belong to a significantly wider range of fields (high generality). However, these results hold only for a subsample of firms for which the CEOs are generalists rather than specialists. We obtain this subsample using the general ability index developed by Custódio et al. (2013). Higher originality and generality scores indicate that a patent has a widespread impact in that it influences other innovations in diverse fields (Hall et al. (2001)). An important discussion in Hall et al. (2001) is that patents that make more citations to other patents and that receive more citations from other patents inevitably have higher originality and generality measures, respectively. This is because when the number of citations increase, made or received, there is increased tendency that these citations are made to or received from a larger range of fields. We control for such an effect by including lagged values of the number of citations made and number of forward citations received in the regressions for originality and generality, respectively. We obtain our results after controlling for firm and year fixed effects, past innovative performance and a broad vector of control variables that have relevance in exploring the relationship between managerial ability and innovation. In that regard, we construct the same vector of control variables as in Fang, Tian and Tice (2013). Specifically, in our regressions, we control for the logarithm of one plus the market value of equity (LN_MV), R&D expenditures scaled by the total assets (RD_TA), return on assets (ROA), property, plant and equipment scaled by the total assets (PPE_TA), leverage (LEV), capital expenditures scaled by the total assets (CAPEX_TA), Tobin s Q (Q), KZ index (KZ_INDEX), logarithm of one plus the Compustat age of firms to account for firms life cycle (LN_AGE), and industry Herfindahl index and its squared specification (H_INDEX and H_INDEX2). The use of firm and year fixed effects partially mitigates possible endogeneity concerns that are driven by the time invariant observed or unobserved factors that we may be omitting. We also mitigate reverse and simultaneous causality 5

7 concerns by lagging all control variables relative to the dependent variables which we use to measure various aspects of innovation. To see whether our results are driven by the managerial compensation incentives and not necessarily the managerial ability, we control for CEOs delta and vega as presented in Core and Guay (2002) and Coles, Daniel and Naveen (2006). Delta measures the dollar increase in CEOs stock and option compensation for a 1% change in the stock price. Vega measures the dollar sensitivity of stock and option compensation for a 1% change in standard deviation of stock returns. Alternatively, we also control for the scaled wealth performance sensitivity measure that is developed by Edmans, Gabaix and Landier (2009). Scaled WPS is the dollar change in a CEO s wealth for a percentage change in firm value, scaled with the CEO s annual compensation. The main advantage of this measure is that it is independent of firm size. Higher delta, vega and scaled WPS suggest that a CEO is incentivized towards undertaking risky investments such as innovation. We find that our results are robust to the inclusion of these control variables as well. We know from the behavioral literature that overconfident CEOs increase corporate investment (Malmendier and Tate (2005)), particularly in innovation (Hirshleifer et al. (2012)). These CEOs may overestimate the future cash flows from innovation or they may underestimate the proper discount rates. Both cases will result in an overestimation of the true NPV of innovation. This overestimation may encourage CEOs to innovate more. However, if the firms of overconfident CEOs end up investing heavily in R&D and if the actual innovative outcome is less than the expected, the innovative efficiency will decline. Therefore, we control for the CEO overconfidence using the measure developed by Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2011). We still find that the managerial ability measure has significant explanatory power on innovation and innovative efficiency. Overall, we contribute to the literature on managerial talent and its benefits to firm performance. We also contribute to the innovation literature by identifying an important driver of firm innovation. Particularly, we show that the executive talent plays a key role in technological advancement. We organize the rest of this paper as follows. We describe our data and variable construction in Section 2. In Section 3, we test and discuss the relationship between managerial ability and innovation and perform robustness checks. Section 4 concludes. 6

8 2. Data Our sample consists of the updated NBER Patent database, Compustat, Execucomp and the managerial ability data developed by Demerjian et al. (2012). We exclude the firms that operate in the financial and utility industries. Our sample spans from 1980 to 2006 and our baseline model in which we investigate the relationship between the managerial ability and innovation (number of patents and average number of citations per patent) includes firm-year observations. On average, a firm is awarded with 21 patents (NPAT) and each patent received 8 nonself forward citations (CITPPAT) from other patents. Firms, on average, make 143 citations (CMADE) to patents other than theirs. The fraction of citations that a firm makes to its own patents (selfcitations) to the total citations it makes (SELF%) is 5%, on average. Our data includes the commonly used variables in the innovation literature. Specifically, we follow Fang et al. (2013) and control for the logarithm of one plus the market value of equity (LN_MV), R&D expenditures scaled by the total assets (RD_TA), return on assets (ROA), property, plant and equipment scaled by the total assets (PPE_TA), leverage (LEV), capital expenditures scaled by the total assets (CAPEX_TA), Tobin s Q (Q) to account for the growth opportunities, KZ index (KZ_INDEX) to account for the financial constraints as in Kaplan and Zingales (2007), logarithm of one plus the Compustat age of firms (LN_AGE), and industry Herfindahl index and its squared specification (H_INDEX and H_INDEX2) to account for the linear and non-linear relationship between product market competition and innovation (Aghion, Bloom, Blundell, Griffith and Howitt (2005)). In additional tests, we also include the CEO optimism, delta, vega and scaled WPS into our vector controls. Table 1 depicts the descriptive statistics and Table 2 shows the correlations between select variables. In Table 2, the low pairwise correlations of number of patents and citations per patent with the managerial ability is mostly due to the time trend in patenting. Our further investigation show that the associations become positive and significant after inclusion of a time trend into our regressions. After using year fixed effects and our vector of control variables, the time trend variable becomes insignificant in all specifications. For that reason, we drop the time trend variable from our regressions. [INSERT TABLE 1] 7

9 2.1 Construction of Innovation Measures We aggregate the NBER Patent Data from patent level to the firm level using the grant year as our basis in counting the number of patents that is granted to each firm. However, the number of patents alone does not convey information regarding the quality of the patents. Therefore, we count the total number of citations a firm s patents receive from other patents (CREC) as well. When counting the citations, we exclude the citations that a firm makes to its previous patents (selfcitations) as in Fang et al. (2013). Then, we compute the number of nonself citations received per patent (CITPPAT) for each firm by dividing the total number of nonself citations received by the number of patents each firm has. We also count the total number of nonself citations a firm makes to other patents (CMADE). In order to measure the extent a firm cites its previous patents, we create a separate variable (SELF%) by dividing the number of self-citations by the number of total citations a firm makes (nonself citations + self-citations). This measure helps us investigate if a firm pursues exploitative or explorative innovation. The patent database contains only the patents that are eventually granted. This introduces a truncation bias to the dataset in the sense that the patents which are applied for in the later years may not be represented in the dataset if they are not yet granted or if the grant date is close to 2006, the last year the innovation data is available. For example, if a patent is granted in 2004, the patent database will contain only the forward citations that occur until 2006 for this patent. We do not have the knowledge of the forward citations for this patent beyond However, Hall et al. (2001) computes a correction factor (hjtwt) to account for the understatement in the number of forward citations due to the truncation bias. We adjust our forward citation count related measures accordingly by multiplying them with this factor. A second set of innovation variables that we construct (IE1 and IE2) relates to how efficiently the patents and citations are obtained using the R&D expenditures that the firms incur. We follow a similar method as in Hirshleifer et al. (2013) to calculate the IE1, which is based on patent counts. Specifically, we compute IE1 as: IE1=NPATi, t /(R&Di, t-2+0.8r&di, t-3+0.6r&di, t-4+0.4r&di, t-5+0.2r&di, t-6) (1) 8

10 where, NPAT is the number of patents that a firm obtains in a given year. The denominator is the R&D capital which represents the cumulative R&D expenditures in the previous 5 years from fiscal year-end t-2 to t-6, assuming a straight-line depreciation rate of 20% per year (Chan, Lakonishok and Sougiannis (2001), Lev, Sarath and Sougiannis (2005)). The use of the straightline depreciation reflects a reasonable assumption that more recent R&D expenditures contribute more in increasing the innovative output. As we mentioned, the patent counts alone do not convey information regarding the quality of patents. Therefore, similar to Hirshleifer et al. (2013), we compute a second innovative efficiency measure (IE2) that accounts for the forward citations a firm s patents receive. Specifically, we compute IE2 as: IE2=CITTOTi, t/(r&di, t-3+r&di, t-4+r&di, t-5+r&di, t-6+r&di, t-7) (2) where, CITTOT is the truncation adjusted total forward citations a firm s patents receive. The denominator is the R&D capital which represents the cumulative R&D expenditures in the previous 5 years from fiscal year-end t-3 to t-7. The time lag between the numerator and the denominator in specifications (1) and (2) are due the application-grant lag. Application-grant lag refers to the duration, that is, on average 2 years, between the time a patent is applied for and the time it is granted by the USPTO. [INSERT TABLE 2] 2.2 DEA Based Managerial Ability Measure We use the DEA based managerial ability measure of Demerjian et al. (2012). The authors construct this measure, for a broad sample of Compustat firms with easily obtainable financial data, by assessing the managers relative (to their industry peers) efficiency in converting firm resources into revenues. In doing so, they initially rank firms in terms of their efficiency in transforming their inventories, fixed assets, operating leases, general and administrative expenses, past R&D expenditures and intangible assets into sales. After constructing the relative efficiency measure, they extract the part that is attributable to the management team by regressing the firm efficiency on the firm-specific factors (i.e. size, market share, cash availability, life cycle, operational complexity and foreign operations) that may affect it. The residual from this regression 9

11 is the managerial ability measure. It is important to note that this ability measure is attributable not only to the CEO but to the entire management team of a firm. Nevertheless, as we mentioned, the authors present strong evidence that the DEA based ability measure is largely attributable to the CEOs. 3. Empirical Findings 3.1 Patent Quantity, Quality and Innovative Efficiency We hypothesize that more able managers are willing to pursue more innovation in terms of patents that are of higher quality in terms of more forward citations. We also hypothesize that these managers are more efficient (more patents and citations per dollar of R&D) in terms of innovation than managers of lesser ability. First, we investigate whether higher managerial ability is associated with higher number of patents and forward citations per patent. Specifically we run the following models; (Ln_NPATi,t+n)=αi+Xi,t β+µi,t+ ƞi,t+ɛi,t (3) (Ln_CITPPATi,t+n)=αi+Xi,t β+µi,t+ƞi,t+ɛi,t (4) where n={2,3,4}, Ln_NPAT is the natural logarithm of one plus the number of patents a firm obtains in year t+n, Xi,t is the vector of control variables, and µi,t and ƞi,t are the firm and year fixed effects, respectively. The use of the firm and year fixed effects help us partially mitigate endogeneity concerns that are related to time invariant unobserved and omitted factors. We also mitigate reverse and simultaneous causality concerns by lagging all control variables relative to the dependent variables. The results are tabulated in Table 3. We find a positive and significant relationship between managerial ability and the number of subsequent patents (columns 1, 2, and 3) and the citation counts per patent (columns 4, 5, and 6) a firm obtains. The relationship is also economically significant. One standard deviation increase in the managerial ability is associated with [(e ) x 0.14] = 1.4% (at T+3), to 1.8% (at T+4) increase in number of patents. Number of citations received per patent increase by [(e ) x 0.14] = 1.3% (at T+3) and 1.9% at (T+4) for a standard deviation increase in the managerial ability. The coefficients of the control variables have the 10

12 expected magnitudes and signs. Size, R&D investment and firm age are significantly and positively related to the number of patents that firms obtain. Similarly, capital expenditures, growth opportunities and firm age are significantly and positively associated with the citations per patent. [INSERT TABLE 3] Next, we examine if managers of higher ability pursue innovation in a more efficient way. Specifically, we regress the innovative efficiency variables (IE1 and IE2) that we construct on our vector of control variables. We tabulate the results in Table 4. In columns 1, 2 and 3, our dependent variable Ln_IE1t+n reflects how efficient a firm is in obtaining patents with its R&D capital. We find that there is a positive and significant relationship between this efficiency measure and managerial ability. In columns 4, 5, and 6, our dependent variable Ln_IE2t+n measures the innovative efficiency in terms of the total citations a firm receives to its patents. Therefore, larger citation-based innovative efficiency indicates better quality patents per dollar spent on R&D. We find positive and statistically significant relationship between the citation-based innovative efficiency and the managerial ability. We control for firm and year fixed effects in both the patentcount based and citation-count based efficiency regressions and cluster the robust standard errors at the firm level. The signs and magnitudes of the coefficients of the control variables are as expected. For example, we document a negative relationship between size (LN_MV) and both measures of innovative efficiency which is consistent with the literature that smaller firms are more innovative given their scarce resources. The negative relationship with the R&D investment (RD_TA) suggests that the efficiency declines as more R&D is employed as a fraction of total assets (diminishing returns to scale). Leverage (LEV) also has a negative relationship with the efficiency. On the other hand, the growth opportunities (Q) that firms possess have positive and significant relationship with the efficiency. Similarly, product market competition (H_INDEX) is positively and significantly associated with efficiency. We mentioned earlier that Demerjian et al. (2012) uses R&D expenditures as one of the inputs in calculating firms efficiency in generating revenues. After obtaining an efficiency score for each firm, this efficiency score is regressed against a number of firm specific characteristics and managerial ability is the residual from this regression. Our specifications in which we use innovative efficiency (innovative output scaled by the R&D expenditures) as our dependent 11

13 variable and an ability measure in which R&D expenditures are used as an input as key dependent variable may spark the question whether having R&D involved in both sides of the regression cause a misspecification. We believe that this concern is not warranted because managerial ability is simply a residual and one cannot conjecture on the variability of this residual with the level of R&D expenditures. Also, the correlation between managerial ability and R&D expenditures scaled by total assets is quite low which undermine such concern. 3.2 Exploratory and Exploitative Innovation Different firms may pursue innovation differently. For example, firms may strategically choose to build on their cumulative knowledge by developing patents that extend their previous innovations incrementally, in which case we name it exploitative innovation. Firms that pursue exploitative innovation may also exploit the innovations developed by other firms without violating the protection rights granted by the USPTO. On the other hand, some firms choose to innovate in an explorative manner. Such firms aim to develop new products and technologies that are not closely related to their previous innovations or the innovations developed by other firms. We hypothesize that the type of innovation that firms pursue (exploitative or explorative) is also related to the ability of their managers. This is a legit argument because managers who possess higher ability can better grasp the latest technology and identify products that will create higher demand. These managers are also able to undertake more risk by pursuing novel (explorative) innovation since their career concerns are less severe than the managers who have lesser ability. We use two measures to determine the extent the firms rely on their previous patents or the patents that are developed by the other firms. [INSERT TABLE 4] Our first measure is the fraction of the number of total nonself forward citations received by a firm to the number of total nonself citations it makes. CR/CM = Total Nonself Citations Received/Total Nonself Citations Made (5) 12

14 Larger values of this ratio indicate that the firms pursue explorative innovation because their patents cite a small number of others while they are cited by many others. Conversely, lower values of the ratio indicate that the firms pursue exploitative innovation. Our second measure is the self-citation ratio. We calculate it as the fraction of the self-citations a firm makes to its patents to the total citations the firm makes including both the self and nonself citations. Since most firms pursue a combination of exploitative and explorative innovation, larger values of this ratio suggest that a firm pursues exploitative innovation to a larger extent (by citing its own previous patents to a larger extent) compared to a firm that has a lower self-citation ratio. As we mentioned, we hypothesize that firms with more able managers would pursue more explorative innovation. We test this hypothesis and tabulate the results in Table 5. In all regressions we use firm and year fixed effects and cluster the standard errors at the firm level. We find that, managerial ability has a positive and significant association with the CR/CM ratio which indicates that firms with more able managers pursue more explorative innovation (columns 1, 2, and 3). In columns 4, 5, and 6, we use generalized estimating equations with a logit link because selfcitation ratio is a continuous variable that take values between 0 and 1. This estimation uses semirobust standard errors. We find that managerial ability is negatively and significantly related to self-citation ratio which suggest that when managers are of higher ability their firms pursue more explorative innovation. We also find that firm size is negatively associated with the CR/CM ratio and positively associated with the self-citation ratio, which indicate that larger firms pursue more exploitative innovation. In addition, we find that the older firms and firms with larger capital expenditures pursue exploitative innovation. 3.3 Originality and Generality of Patents Previously, we have shown that the managerial ability has a positive association with the quantity and the quality of innovation and the innovative efficiency. We also have shown that the firms with more able managers pursue more novel innovation than firms with managers of lesser ability. In this section, we investigate the relationship between the managerial ability and the breadth (broadness) of the innovation that firms pursue, namely the generality and originality of firms patents. These measures quantify how the patented innovations are linked to other innovations. 13

15 Hall et al. (2001) define generality as the one minus the Herfindahl concentration index of the citations a patent receives from other patents that belong to a diverse range of fields. If a patent is cited by other patents that belong to wide range of fields the generality ratio will be high. If the citing patents belong to a narrow range of fields the generality ratio will be low. A high generality measure indicates that the impact of a patent is outspread to diverse areas, in contrast to a low generality measure that suggests that the impact of a patent has been limited. Originality is quantified the same way but instead it uses the citations made by a patent. So, if a patent cites other patents that belong to a diverse range of other fields, the originality measure will be high. Originality and generality measures are constructed at the patent level in the NBER Patent Dataset. In order to compute the originality and the generality measures at the firm level, we take the mean of the originality and generality measures of individual patents that are granted in a year to a firm. [INSERT TABLE 5] We hypothesize that more able managers possess a broader understanding of technology, trends and the innovations in diverse areas. We expect these managers to reflect this broader understanding in the innovation they pursue. As a result, the patents that are developed under a management with higher ability would cite and get cited by patents that belong to a diverse range of fields. Therefore, we expect these patents to have larger originality and generality. On the other hand, we cannot ignore the heterogeneity in the skill sets of managers in the sense that two managers may be identical in terms of a quantified ability measure such as DEA based ability but one may be a generalist and the other may be a specialist. This heterogeneity would reflect differently on the innovation that these two otherwise identical managers pursue. Custódio et al. (2013) classify CEOs as generalist and specialists based on a factor analysis of the number of different positions the CEOs held, the number of firms they worked at, the number of industries that they have been exposed to and the number of previous positions that they worked as CEOs. From the combined score that they compute, they define CEOs who score above the median as generalists. Otherwise, the CEOs are denoted as specialists. We acknowledge that the association between managerial ability, originality and generality of a firm s patents also depend on this heterogeneity in managers skill sets. Therefore, we investigate this relationship separately for the 14

16 firms with generalist and specialist CEOs. In doing this, we again rely on the evidence of Demerjian et al. (2012) that DEA based managerial ability is largely attributable to CEOs. Originality and generality measures take continuous values between 0 and 1. We regress them on the DEA based managerial ability measure and our vector of control variables using tobit models as in Chemmanur, Loutskina and Tian (2013) and Aggarwal and Hsu (2013). Hall et al. (2001) mention that there is a natural tendency for the patents that receive many citations to have greater generality measures because it is more likely that these patents are cited by other patents that belong to broader range of fields. This is also the case for the originality measure. When a patent makes many citations to other patents, it is more likely that the cited patents belong to broader range of fields. Hence, the citing patent will have a higher originality measure. To address this tendency, we control for the truncation adjusted forward citations received and the citations made in our models for generality and originality, respectively. We use year and industry fixed effects and the robust standard errors are clustered at the firm level. Industry fixed effects are based on the Fama-French 12 industry classification. 3 We tabulate the results in Table 6 and Table 7 for generalist CEOs and specialist CEOs, respectively. In Table 6, we find that the associations between the managerial ability and originality and the managerial ability and generality are positive and significant when firms are managed by generalists CEOs. For the firms with specialist CEOs, the relationship between the managerial ability and originality is positive and marginally significant. We do not find a statistically meaningful association with the generality measure. In that sense, findings are in line with what we would expect. The level of the managerial ability matters in terms of originality and generality that firms with managers who have broad knowledge of different fields can produce innovation that have broader impact. We should mention that the negative pseudo-r 2 s in the originality regressions in Table 6 and Table 7 are not reflective of the variation in originality that is explained by the independent variables. For many models, including tobit, the pseudo R 2 is not informative. 3 Details for the Fama-French industry definitions can be found on Kenneth R. French s academic web page. 15

17 3.4 Innovation, CEO Characteristics and Compensation So far, we have used firm level controls and find that the positive and significant relationship between the managerial ability and innovation are robust to the inclusion of these control. However, existing studies show that managerial optimism (i.e. Campbell et al. (2011), Hirshleifer et al. (2012)) and compensation (i.e. Coles, Daniel and Naveen (2006), Edmans et al. (2009)) are important determinants of managerial risk taking which, in turn, impact firms investment decisions such as R&D and innovation. Since we mentioned that the DEA based ability measure is largely explained by the CEO fixed effects, it may be the case that CEO compensation and optimism are correlated with the managerial ability measure to the extent they are time invariant. Thus, compensation incentives and CEO optimism may be driving our results rather than the managerial ability measure itself. In order to address this issue, we control for the CEO optimism as constructed by Campbell et al. (2011), CEOs delta and vega as presented in Core and Guay (2002) and Coles et al. (2006) and the scaled wealth-performance sensitivity (WPS) that is constructed by Edmans et al. (2009). 4 The optimism measure developed by Campbell et al. (2011) is based on the CEOs decisions regarding the stock option exercise or hold. They identify the highly optimistic CEOs as the CEOs who hold onto their stock options that are more than 100% in the money for at least during two years. These CEOs are considered highly optimistic from the beginning of the first year that they exhibit this behavior. [INSERT TABLE 6 & 7] CEOs compensation delta measures the dollar increase in CEOs stock and option compensation for a 1% change in the stock price. Vega measures the dollar sensitivity of stock and option compensation for a 1% change in standard deviation of stock returns and the scaled WPS is the dollar change in a CEO s wealth for a percentage change in firm value, scaled by the CEO s annual compensation. The main advantage of WPS is that it is independent of the firm size so that the compensation incentives are comparable between small and large firms. Overall, higher delta, vega and scaled WPS suggest that a CEO is incentivized towards undertaking risky investments such as innovation. 4 We thank Alex Edmans for sharing the scaled wealth - performance sensitivity measure publicly in his academic web page. 16

18 In Table 8 Panel A, we present our findings on the managerial ability and innovative efficiency measures after controlling for our baseline vector of control variables, CEO optimism and CEOs compensation delta and vega. In Panel B, we control for optimism and WPS, instead of delta and vega. In all regressions, we use firm and year fixed effects and the standard errors are clustered at the firm level. For brevity, we do not report the results when dependent variables are at T+2 and T+4 and the results when the dependent variables are number of patents. However, findings are consistent with the results that we present in Table 8. In columns 1 and 2 of Panel A, our dependent variable is the patent quantity based innovative efficiency (IE1). We find that the relationship between innovative efficiency and the managerial ability stays positive and significant even after controlling for optimism, delta and vega. In columns 3 and 4, we find similar results when the dependent variable is the citation based innovative efficiency (IE2). In Panel B, similar to Panel A, dependent variables are the patent quantity based innovative efficiency (IE1) and the citation based innovative efficiency in columns 5-6 and columns 7-8, respectively. The WPS dummy (WPS_DUM) takes a value of 1 if a CEO s scaled wealth performance sensitivity is above the median of the wealth performance sensitivity of all CEOs in a given fiscal year. It takes a value of zero, otherwise. Similar to what we find in Panel A, the relationship between the managerial ability and the patent quantity and citations based innovative efficiencies stay positive and significant. 3.5 Robustness [INSERT TABLE 8] Propensity Score Matching Firms that are capable of hiring more able CEOs and firms that do not have the similar capacity may be systematically different and it is possible that our results are driven by these differences in firm characteristics and not the managerial ability. Therefore, we propensity score match the firms based on our baseline vector of control variables to show that our results are robust to endogenous observable factors. The propensity score matching procedure warrants that the distribution of observed covariates will be similar between the treated and control firms, conditional on the propensity score. In that respect, we identify the firms that are similar in all aspects but the 17

19 managerial ability. We denote firms that have above the median DEA based managerial ability as the treated firms. Otherwise, they are denoted as the control firms. Finally, we compute the average treatment effect, with respect to innovation, on the treated firms (ATT). First, we run a fixed effects probit model to compute the probability (propensity score) of a firm to have an above the median managerial ability conditional on our baseline vector of characteristics. We use firm and year fixed effects and the robust standard errors are clustered at the firm level. Coefficients to this regression are tabulated in Table 9, Panel A. Later, we match the treated and the control firms, which are in the region of common support, based on their propensity scores, without replacement. Thus, a control firm is matched to a treatment firm only once. We make sure that the difference in means of all baseline characteristics between the treated and the control firms are statistically insignificant before computing the average treatment effect on the treated firms. In Table 9, Panel B, we show the average treatment effect of higher managerial ability on the number of patents, number of citations per patent, and the quantity and citations based innovative efficiency (at T+2 and T+3 for brevity). We find strong evidence that our findings of positive and significant relationship between the managerial ability and innovation are not endogenously driven by other observable firm characteristics. [INSERT TABLE 9] Difference-in-Differences Using CEO Turnover Sample In order to provide additional robustness to our estimations and establish causality for the relationship between managerial ability and innovation, we create a CEO turnover sample using the Execucomp database between 1992 and 2006 and employ the difference-in-differences (DiD) methodology. The idea underlying our DiD analysis is that we can eliminate observed and unobserved differences between two groups of firms (with high low managerial ability) by using an exogenous event such as the CEO turnover. Demerjian et al. (2012) document that their ability measure is highly correlated with the CEO fixed effects (rho=0.835). They also find that 60.5% of the CEO fixed effects are significant when the DEA based ability measure is regressed on the CEO fixed effects and firm fixed effects with the standard errors clustered by firm and year. Thus, we feel safe in using CEO turnover as an exogenous event to investigate causality although the DEA 18

20 based managerial ability measure is attributable not only to the CEOs but to the entire management team. In this sense, if managerial ability drives innovation in an exogenous manner, we should observe an improvement in innovation following a CEO change when the new CEO is of higher ability than the replaced CEO. We start preparing our CEO turnover sample by computing the years of uninterrupted employment of each CEO at each firm. For the CEO firm pairs which we have complete information regarding the date the CEO became the CEO, we calculate years of employment as the elapsed time between the date a CEO became the CEO and the ending date of the most recent fiscal year. When the date a CEO became the CEO is not available, we use the number of fiscal years a CEO appears as the CEO of a particular company. We require each CEO and his predecessor to serve continuously for at least 3 years at the company, excluding the fiscal years that they became the CEO. Firms that operate in the financial and utility industries are excluded from the sample. We are able to identify 1285 CEO turnovers which satisfy our criteria. Next, we identify if managerial ability increased after the turnover event. We partition firms that experienced an increase in the managerial ability into quartiles, and denote the firms in the top quartile as the treated firms. Firms in lower quartiles, firms that did not experience any increase in managerial ability and firms that experienced a decline in managerial ability are denoted as control firms. Then we propensity score match the treated and control firms prior to their turnover events based on the probability (propensity score) of being a treated firm. We look for the matches on the common support. We allow a control firm to match multiple treated firms. This procedure gives us 47 unique treated firms (240 firm-year observations) and 83 unique control firms (434 firm year observations). Table 10 Panel A presents the pre-turnover differences in terms of characteristics between the treated and control firms. As can be seen, after the matching treated and control firms are not substantially different in the statistical sense. Following the matching we run a fixed effects model as in the Table 10 Panel B to investigate whether our innovation variables are significantly explained in the post-turnover period by the firms that experienced a substantial increase in managerial ability. Specifically, we are interested in the coefficient of the variable TREATED*AFTER in Panel B which is our DiD estimate. Regressions include less than 674 firm year observation based on the availability of the dependent variables. We find evidence that, after a CEO turnover that results in a substantial increase in managerial ability, innovation increases significantly in terms quantity, quality and efficiency. 19