TY - JOUR
AU - Ravix, Jacques-Laurent
AB - Abstract This article investigates the relationship between corporate governance (CG) and innovation for firms of different ages. We blend data drawn from the ISS RiskMetrics and the Bureau van Dijk Orbis databases to perform micro-econometric analyses on a sample of 2200 listed firms. We show that CG may decrease research and development for all firms and that, more importantly, it has a significant and negative effect on the patenting strategy of the younger firms. Our results are consistent with the CG life cycle view, according to which young companies tend to privilege short-termism and value preservation rather than long-term risky innovation strategies. What shown and discussed in this contribution supports the proposition that firm age matters in how CG may alter innovation. 1. Introduction Since the seminal contribution by Jensen and Meckling (1976), a wide body of literature has investigated the principles of corporate governance (CG). While the concept itself applies to any firm involving different kinds of stakeholders, the theoretical and empirical literature has mostly targeted those firms in which ownership and control are separated. Principal-agent theory is the starting point of most discussions on CG as agency problems can affect firm value and performance via expected cash flows for investors and the cost of capital (Shleifer and Vishny, 1997). “Good” CG means that “more of the firm’s profit would come back to (the investors) as interest or dividends as opposed to being expropriated by the entrepreneur who controls the firm” (La Porta et al., 2002: 1147). Most of the extant studies have investigated the effects of good CG on firm performance in terms of profits, revenue growth, share performance, market capitalization or productivity, by stressing international differences due to country-specific institutional settings (La Porta et al., 1997, 1998, 2000b; Gompers et al., 2003; Core et al., 2006; Bebchuk et al., 2008; Aggarwal et al., 2010; Bebchuk and Weisbach, 2010). A somewhat less explored area of investigation concerns the relationship between CG and innovation. The existing studies yield mixed results, due to the different mechanisms through which CG may affect innovation. On the one hand, good governance reduces agency costs, and this should be beneficial to all investments including research and development (R&D) and patent filing, all other things being equal. On the other hand, good governance puts a large emphasis on the interests of the shareholders as a primary goal, and this may be detrimental to innovative activities since shareholders and investors are mostly interested in dividends and short-term returns on investments rather than long-term and uncertain innovation outcomes. Our article contributes to this emerging stream of literature by investigating the impact of CG on innovation, and stresses the importance of firm age in moderating such a relationship. In so doing, we gather theoretical considerations grounded on agency theory that provides expectations on the effects of good governance practices on innovation efforts, with the literature on CG life cycle, according to which firms tend to privilege short-term rather than long-term strategies. The contribution to the extant literature is manifold. First, to our knowledge, there are neither empirical nor theoretical analyses focused on the interplay between CG, age, and innovation. Second, we compare results obtained by using both input and output measures of innovation, i.e. R&D expenditures and patent applications, to get a complete picture of the impact on innovation performance, especially in view of capturing a differential impact at the level of the age of the firm. Third, while previous studies mostly focused on single attribute measures and national data, we use an original data set of listed firms drawn from the ISS RiskMetrics database, one of the largest international database providing a multi-attributes metrics of CG, and merged with the Bureau van Dijk Orbis database. Finally, we contribute to encompass some of the great divides in the literature, and offer a new light on extant and contrasting findings. The results of our study are consistent with the CG life cycle view, according to which firms tend to privilege short-term rather than long-term strategies. We show that good CG is, on average, negatively related to innovation performance and that such a negative relationship is stronger for young listed firms. Moreover, while the net effect is not robust for R&D investments, it is clearly negative for patents, leading younger firms to discard one essential way to develop and signal their innovative capacity. Narrower resource bases, insufficient knowledge, and underdeveloped capacity to manage R&D investments may hinder younger firms. Because of this and of the higher level of risk in developing successful innovative outcomes, investors may be even more motivated to gain rewards quickly, and this source of agency costs may in turn decrease the innovation propensity of young firms. The reminder of the article is organized as follows. In Section 2 we shortly review the literature on CG and innovation, and formulate a set of conjectures about the moderating effect of firm age. Section 3 presents the data and variables, while Section 4 describes the employed methodologies. In Section 5 we show and discuss the empirical findings. Finally, we relate our results to the extant literature and draw some recommendations for policy. 2. Corporate governance, age, and innovation The net effect of “good” CG on firms’ innovative behavior remains theoretically unclear. We have already seen that high standards of governance reduce agency costs but at the same time put a large emphasis on the (short-term) interests of the shareholders as a primary goal. So far, the empirical literature linking CG attributes and innovation performance has failed to provide unambiguous evidence (Belloc, 2012). The largest body of empirical studies focus on the impact of anti-takeover provisions, ascertaining whether the managerial myopia hypothesis (Stein, 1988) or the quiet life hypothesis (Bertrand and Mullainathan, 2003) hold in the context of innovation. According to the first hypothesis, the threat of hostile acquisition can lead managers to avoid undertaking long-term, risky investments because these projects can lead to a wide divergence between market and intrinsic values. Anti-takeover provisions may shield managers from concerns related to short-term performance and permit a more long-term, value-maximizing investment strategy that encourages greater innovation. Alternatively, based on the second assumption, if the presence of takeover protection reduces the effectiveness of the external disciplinary market, then managers may be able to avoid difficult and risky investments, especially if these show that managers are of lower quality. In a recent paper, Atanassov (2013) combines financial data available in S&P’s Compustat database with the National Bureau of Economic Research (NBER) patent file. The data include 13,339 US firms, over the period 1976–2000. The results show a significant decline in the number of patents and citations per patent for firms located in the US states that pass anti-takeover laws compared with the ones that did not. In the meantime, Becker-Blease (2011) uses the Investor Responsibility Research Center Institute (IRRC) and merges the data with financial accounting standards and NBER patent database. The study covers the period 1984–1997 and the sample is composed of 600 US firms. The results show that higher levels of 23 takeover provisions are associated with innovation efforts (R&D expenditures, awarded patents, quality of patents, number of patents awarded per $ of R&D), suggesting that innovation is positively correlated with anti-takeover provisions. The heterogeneity of investors matters with impatient capital having a focus on short-term dividends, while more committed long-run capital takes into consideration the basic characteristics of uncertainty and risk of innovation and recognizes innovation as a major driving force of economic growth. There is a strand of empirical analysis focusing on the quality of the investor. In this context, Fang et al. (2014) show that increased liquidity is associated with a reduction in future innovation. The authors identify as possible determinants increased exposure to hostile takeovers as well as the type of investors involved, especially the ones that do not gather information or do not monitor. In Brown et al. (2013), long-run R&D investments are correlated to a high level of shareholder protections, while Manso (2011) finds that a combination of stock options with long vesting periods, option repricing, golden parachutes, and managerial entrenchment are necessary conditions for innovation. Baranchuk et al. (2014) put a similar conclusion forward, according to which managers are better motivated to pursue innovation when their incentive compensation scheme is over long vesting periods and when anti-takeover protection exists. Finally, Brossard et al. (2013) report a positive relationship of CG on R&D and provide evidence of the negative influence from impatient institutional investors on R&D spending. Other recent empirical investigations on the matter do point to very contrasting results. By way of example, Driver and Guedes (2012) test the possibility of a perverse effect of “good” governance on uncertain, long-term R&D investments. Using UK data, they end up with a long-run negative effect of governance on R&D. O’Connor and Rafferty (2012) employ IRRC data, and obtain a negative but non-robust relationship. Finally, in a study of large French listed business groups, Lhuillery (2011) notes that there is no systematic influence of “good” governance on R&D decisions. The literature on the CG life cycle approach may provide a fertile ground to organize such mixed evidence. It regroups different contributions on CG more largely embedded with innovation theory where heterogeneous firms evolve and change over time (Lazonick and O’Sullivan, 2002; Filatotchev and Wright, 2005; Krafft and Ravix, 2005, 2008; Belloc, 2012; Dietrich and Krafft, 2012). The CG life cycle approach suggests that innovative efforts of firms can hardly be considered equal to other activities and this could lead to potential underestimation of important issues on CG at a firm level. The life cycle characteristics of firms may play a role, as in their young age firms may essentially develop their innovative activity through patents, also signaling by this their ability to innovate to shareholders and investors, even if not all patents represent a commercially exploited innovation. Alternatively, aging firms start putting more emphasis on R&D by developing dedicated departments, inducing an expected positive relation with innovative efforts for shareholders and investors. As far as innovation is taken into account in CG, factors such as appropriability, asymmetric information, and high risk in decision-making induce more careful monitoring, resulting eventually in higher costs of capital. This may involve firms opting for short-term rather than long-term strategies (Holmstrom, 1989). This may also lead to underinvestment in resource creation as motivating the managers to give back free cash flow to shareholders is not necessarily beneficial to innovation when the short-term strategy of dividends yields are systematically preferred to the long-term strategy of re-investment in product and process innovation (Lazonick, 2007). Filatotchev et al. (2006) provide instead a framework for understanding the link which gathers together agency issues with a resource-based view of the firm. In such a context, mature firms are characterized by an extensive resource base, i.e. tacit knowledge that has been accumulated over time as well as production facilities, trade secrets, engineering experience and human capital assets. Mature firms seem to possess all the resources that are needed to manage successful innovative projects. On the contrary, young firms are typically characterized by a narrow resource base and are mostly dependent on external knowledge sources. Based on the arguments discussed so far, we can now spell out our research hypotheses as follows. Good CG practices are likely to affect firms’ innovative outcomes. In particular, it may reduce agency costs, stimulating R&D effort and patent filing. At the same time, good CG, stressing the interests of the shareholders as a primary goal, may be detrimental to innovative activities as a short-term strategy might prevail, while innovation is typically long term. The appreciation of firms’ age structure helps framing this composite evidence, suggesting that the dominance of short-term over long-term strategies and the scope of firms’ resource bases evolve over time and shape the effect of CG on innovation according to firms’ life cycles. In particular, according to the arguments supported by the literature, young age is expected to drive the negative relationship between good CG and innovation. 3. Data, measurement, and sample characteristics 3.1 The data set In this article we employ the Corporate Governance Quotient (CGQ) index from RiskMetrics/Institutional Shareholder Services. We focus on overall (aggregate) CG ratings for a large range of international firms. Our sample is constructed using information on 2203 firms in 34 countries (see Table A1 in Appendix for details) and 21 industries. The CGQ is calculated on the basis of a rating system that incorporates eight categories of CG, leading to an improved qualitative measure of 55 governance factors. The study period covered is 2003–2008 which includes the largest number of reporting firms with complete and consistent data. The ISS RiskMetrics has been matched to the Bureau van Dijk Orbis database by using the International Securities Identification Numbers (ISIN) identification code. This passage enabled us to obtain firm-level data from their balance sheets and to assign patent applications to sampled firms to calculate their patent stock for each year. The initial ISS Risk Metrics data set consisted of 3369 firms, spread across 34 countries. After the matching, we were left with 2203 firms spread across the same countries. Table A1 in the Appendix shows the details. At first sight, the cross-country distribution after the matching seems to be similar to that of the original data set. For confirmation, we calculated the Spearman rank correlation coefficient, based on the relative frequency of firms in the different countries, obtaining a coefficient equal to 0.953.1 3.2 Variables 3.2.1 Measuring innovation We use different firm-level variables to proxy the innovative behavior of companies in our sample. Since the seminal contribution by Griliches (1984), scholars have typically adopted two measures of innovation based on the well-known knowledge production function approach. On the one hand, R&D expenditures represent a standard input measure for the innovation process; on the other hand, patent applications are considered as an output. Both variables have of course pros and cons (see, among the many, Pavitt, 1985; Griliches, 1990). The main downsides of patents are related to the fact that they are sector specific and that not all innovations are patentable. Moreover, the propensity to patent is unevenly distributed between small and large firms, and it is not stable over time, due for instance to changing costs of patenting. Nevertheless, previous literature has provided support to the use of patents as a measure of production of new knowledge. These studies show that patents are a reliable proxy for knowledge and innovation, as compared to other measures based upon surveys investigating the dynamics of process and product innovation (Acs et al., 2002). The main limit of R&D statistics is instead that they only refer to inputs of the innovation process, rather than successful outputs. In this respect, R&D efforts are not necessarily conducive to innovations, as some innovative projects are successful and some others are not. Moreover, R&D expenditure data are not comprehensively available, and do not allow the investigation of particular technological fields or relationships with other innovations (Lanjouw et al., 1998). It is worth stressing, however, that empirical analyses have long demonstrated that patents and R&D are dominated by a contemporaneous correlation (Hall et al., 1986). To overcome the intrinsic limitations of R&D and patents, as well as to get an in-depth picture of the impact of CG on the firm’s innovative behavior, we shall carry out empirical investigations by adopting both measures as dependent variables in separate estimations.2 3.2.2 Corporate governance quotient Prior to being acquired by RiskMetrics in 2007, Institutional Shareholder Services operated independently as the world’s largest CG data provider. Institutional Shareholder Services developed its CG rating system to help institutional investors evaluate the impact that a firm’s CG structure and practices might have on performance. The rating is aimed at providing objective and complete information on firm’s governance practices. Importantly, these ratings are not tied to any other service provided by RiskMetrics/Institutional Shareholder Services and firms do not pay to be rated, although they are invited to check the accuracy of the ratings. The only way a firm can improve its rating is to publicly disclose changes to its governance structure and/or practices. The CGQ is the output of a CG scoring system that evaluates the strengths, deficiencies, and overall quality of a company’s CG practices. It is updated daily for over 7500 companies worldwide. Each company’s CGQ rating is generated from detailed analysis of its public disclosure documents (i.e. Proxy Statement, 10 K, 8 K, Guidelines, etc.), press releases, and company Web site. CGQ is calculated by adding 1 point if the firm under scrutiny meets the minimum accepted governance standard. The score for each topic reflects a set of key governance variables. Most variables are evaluated on a standalone basis. Some variables are analyzed in combination based on the premise that CG is improved by the presence of selected combinations of favorable governance provisions. For example, a company whose board includes a majority of independent directors and independent board committees (audit, etc.) receives higher ratings for these attributes in combination than it would have received for each separately. Next, each company’s CGQ is compared with other companies in the same index [here the index is Morgan Stanley Capital International (MSCI) EAFE index].3 For example, Company A scores 24% (or 0.24) for its CGQ index which means that Company A is performing better (outperforming) in relation to CG practices and policies than 24% of the companies in the MSCI EAFE index. Table 1 presents the CG variables. A detailed description of governance standards using the eight categories (board of directors, audit committee, charter/bylaws, anti-takeover provisions, compensation, progressive practices, ownership, and director education) is provided in Krafft et al. (2014). Table 1. Corporate Governance Quotient criteria BOARD  AUDIT  Board Composition  Audit Committee  Nominating Committee  Audit Fees  Compensation Committee  Auditor Rotation  Governance Committee  Auditor Ratification  Board Structure  EXECUTIVE AND DIRECTOR COMPENSATION  Board Size  Cost of Option Plans  Changes in Board Size  Option Re-Pricing  Cumulative Voting  Shareholder Approval of Option Plans  Boards Served On—CEO  Compensation Committee Interlocks  Boards Served On—Other than CEO  Director Compensation  Former CEO's  Pension Plans for Non-Employee Directors  Chairman/CEOs Separation  Option Expensing  Board Guidelines  Option Burn Rate  Response to Shareholder Proposals  Corporate Loans  Boards Attendance  PROGRESSIVE PRACTICES  Board Vacancies  Retirement Age for Directors  Related Party Transactions  Board Performance Reviews  CHARTER/BYLAWS  Meetings of Outside Directors  Features of Poison Pills  CEO Succession Plan  Vote Requirements  Outside Advisors Available to Board  Written Consent  Directors Resign upon Job Change  Special Meetings  OWNERSHIP  Board Amendments  Director Ownership  Capital Structure  Executive Stock Ownership Guidelines  ANTI-TAKEOVER PROVISIONS  Director Stock Ownership Guidelines  Anti-Takeover Provisions Applicable  Officer and Director Stock Ownership  Under Country(local)Laws  DIRECTOR EDUCATION    Director Education  BOARD  AUDIT  Board Composition  Audit Committee  Nominating Committee  Audit Fees  Compensation Committee  Auditor Rotation  Governance Committee  Auditor Ratification  Board Structure  EXECUTIVE AND DIRECTOR COMPENSATION  Board Size  Cost of Option Plans  Changes in Board Size  Option Re-Pricing  Cumulative Voting  Shareholder Approval of Option Plans  Boards Served On—CEO  Compensation Committee Interlocks  Boards Served On—Other than CEO  Director Compensation  Former CEO's  Pension Plans for Non-Employee Directors  Chairman/CEOs Separation  Option Expensing  Board Guidelines  Option Burn Rate  Response to Shareholder Proposals  Corporate Loans  Boards Attendance  PROGRESSIVE PRACTICES  Board Vacancies  Retirement Age for Directors  Related Party Transactions  Board Performance Reviews  CHARTER/BYLAWS  Meetings of Outside Directors  Features of Poison Pills  CEO Succession Plan  Vote Requirements  Outside Advisors Available to Board  Written Consent  Directors Resign upon Job Change  Special Meetings  OWNERSHIP  Board Amendments  Director Ownership  Capital Structure  Executive Stock Ownership Guidelines  ANTI-TAKEOVER PROVISIONS  Director Stock Ownership Guidelines  Anti-Takeover Provisions Applicable  Officer and Director Stock Ownership  Under Country(local)Laws  DIRECTOR EDUCATION    Director Education  Our sample is composed of 2203 non-US firms operating in 34 countries and 21 industries. Table 2 reports information on the composition of our sample according to NACE classification. Almost half of the sample is composed of firms operating in manufacturing, followed by financial and insurance activities. Despite a few exceptions, we have data on firms that are active in almost all sectors. As in the original database, CGQ refers to 55 governance factors spanning the eight categories of CG. The data are thus firm level; all our scores are relative (percentiles), allowing for within-country as well as cross-country differences [the data explicitly consider anti-takeover provisions under national (local) law]. Table 2. Sectoral distribution of sampled firms Industry (NACE Rev. 2)  # Firms  Frequency  A – Agriculture, forestry and fishing  7  0.318  B – Mining and quarrying  62  2.814  C – Manufacturing  811  36.813  D – Electricity, gas, stream and air conditioning supply  56  2.541  E – Water supply; sewerage, waste management and remediation activities  9  0.409  F – Construction  71  3.223  G – Wholesale and retail trade; repair of motor vehicles and motorcycles  165  7.489  H – Transportation and storage  105  4.766  I – Accommodation and food service activities  39  1.777  J – Information and communication  214  9.714  K – Financial and insurance activities  346  15.706  L – Real estate activities  88  3.995  M – Professional, scientific and technical activities  118  5.356  N – Administrative support service activities  48  2.179  O – Public administration and defence; compulsory social security  0  0.000  P – Education  2  0.091  Q – Human health and social work activities  11  0.499  R – Arts, entertainment and recreation  20  0.907  S – Other service activities  16  0.726  T – Activities of households as employers  0  0.000  U – Activities of extraterritorial organisations and bodies  0  0.000  Missing information  15  0.681  Total  2203  100%  Industry (NACE Rev. 2)  # Firms  Frequency  A – Agriculture, forestry and fishing  7  0.318  B – Mining and quarrying  62  2.814  C – Manufacturing  811  36.813  D – Electricity, gas, stream and air conditioning supply  56  2.541  E – Water supply; sewerage, waste management and remediation activities  9  0.409  F – Construction  71  3.223  G – Wholesale and retail trade; repair of motor vehicles and motorcycles  165  7.489  H – Transportation and storage  105  4.766  I – Accommodation and food service activities  39  1.777  J – Information and communication  214  9.714  K – Financial and insurance activities  346  15.706  L – Real estate activities  88  3.995  M – Professional, scientific and technical activities  118  5.356  N – Administrative support service activities  48  2.179  O – Public administration and defence; compulsory social security  0  0.000  P – Education  2  0.091  Q – Human health and social work activities  11  0.499  R – Arts, entertainment and recreation  20  0.907  S – Other service activities  16  0.726  T – Activities of households as employers  0  0.000  U – Activities of extraterritorial organisations and bodies  0  0.000  Missing information  15  0.681  Total  2203  100%  3.3 Descriptive statistics Micro-level accounting data come from the Bureau van Dijk Orbis data set. Beside demographic characteristics such as age and size, we introduce in our analysis a measure of cashflow with the aim at capturing firm operating performance which is likely to affect a firm’s innovative behavior (Brown et al., 2009). The full list of variables used in this study is finally reported in Table 3. Table 3. Definition of variables Variables  Definition  CGQ  Corporate Governance Quotient from RiskMetrics/Institutional Shareholder Services  ΔCGQ  Variation in CGQ index  ln(Age)  Logarithm of firm’s age  ln(SZ)  Logarithm of firm’s total turnover  ln(RDI)  Logarithm of R&D expenditure over total turnover (R&D-to-sales intensity)  Δln(RDI)  Growth rate of R&D-to-sales intensity  Patents  Number of patent applications  CF  Firm’s cashflow (in millions of $)  Variables  Definition  CGQ  Corporate Governance Quotient from RiskMetrics/Institutional Shareholder Services  ΔCGQ  Variation in CGQ index  ln(Age)  Logarithm of firm’s age  ln(SZ)  Logarithm of firm’s total turnover  ln(RDI)  Logarithm of R&D expenditure over total turnover (R&D-to-sales intensity)  Δln(RDI)  Growth rate of R&D-to-sales intensity  Patents  Number of patent applications  CF  Firm’s cashflow (in millions of $)  Table 4 shows basic descriptive statistics, while in Figure 1 we plot the kernel densities of the main variables under investigation, i.e. CG, R&D-to-sales intensity (RDI), patent applications, and age. Table 4. Descriptive statistics Variables  Mean (standard)  Minimum  1st quartile  Median  3rd quartile  Maximum  CGQ  0.47 (0.26)  0  0.26  0.45  0.69  1  ΔCGQ  0.17 (2.57)  −1  −0.16  −0.02  0.12  3.41  Age  54.38 (47.90)  0  17  44  81  536  ln(SZ)  14.76 (1.86)  1.39  13.27  14.49  15.56  19.94  ln(RDI)  0.98 (1.12)  0  0.05  0.65  1.53  12.39  Δln(RDI)  −0.01 (0.68)  −5.98  −0.06  0  0.05  6.77  Patents  2.29 (4.98)  0  0  0  2  47  CF  0.80 (2.39)  −7.30  0.04  0.16  0.55  44.51  Variables  Mean (standard)  Minimum  1st quartile  Median  3rd quartile  Maximum  CGQ  0.47 (0.26)  0  0.26  0.45  0.69  1  ΔCGQ  0.17 (2.57)  −1  −0.16  −0.02  0.12  3.41  Age  54.38 (47.90)  0  17  44  81  536  ln(SZ)  14.76 (1.86)  1.39  13.27  14.49  15.56  19.94  ln(RDI)  0.98 (1.12)  0  0.05  0.65  1.53  12.39  Δln(RDI)  −0.01 (0.68)  −5.98  −0.06  0  0.05  6.77  Patents  2.29 (4.98)  0  0  0  2  47  CF  0.80 (2.39)  −7.30  0.04  0.16  0.55  44.51  Figure 1. View largeDownload slide Kernel density for main variables. Note: Kernel densities are computed by pooling all the observations and estimated with an Epanechnikov kernel. Figure 1. View largeDownload slide Kernel density for main variables. Note: Kernel densities are computed by pooling all the observations and estimated with an Epanechnikov kernel. First, we notice that a considerable proportion of firms in our sample (almost 10%) do not perform R&D activities,4 while only a few companies invest more than their turnover (Figure 1). These firms are primarily young and operating in high-tech industries. We also account for variation in R&D intensity by computing the log-difference for each subsequent year. On the output side of the innovation process, we observe that the average number of patent applications per year is 2.29. The statistical distribution displays a positive skewness so that the mass of the density is clearly concentrated on the left tail. It should be noted that almost two-thirds of the total number of observations has value equal to zero (no patent filing). We will explicitly take into account this fact by adopting econometric tools designed for the presence of many zeros in the response variable. CGQ ranges from 0 to 1, with mean and median equal to 0.47 and 0.45, respectively. Its kernel density (Figure 1) displays wide support, hence confirming the strong heterogeneity in CG practices. This evidence motivates us to explicitly account for idiosyncratic firm fixed effects in our econometric setting. We decompose the standard deviation of CGQ into between (σ-between = 0.2355) and within (σ-within = 0.1313) components to verify whether firms are prone to change their governance practices over time. The non-negligible value of the within component suggests that this is the case; thus, besides the CGQ in level we also account for changes in CG by calculating its growth rates. Firms in our sample are also quite heterogeneous in terms of age. The latter ranges in fact from 0 (newborn companies) to 536 (old established enterprises) years old, with a mean and median of 54.38 and 44, respectively. To compress the scale, we apply a log transformation. Basic statistics suggest that, though we do have information on new nascent firms, our sample is primarily composed of incumbent established units. This evidence will drive us, when selecting a cut-off point to distinguish young/medium-aged vs. mature firms, to look for an age threshold which is a reasonably good compromise between sample size and coherence. To appreciate a first screenshot of the contemporaneous relationship between the entire set of variables, Table 5 reports the pair-wise correlation matrix (significance at 5% level are indicated by asterisks). There is a negative association between corporate governance index GGQ and age, as with the size of the firm. Beyond some expected relationships (for instance, the positive correlation between age and size), it should be noted that CGQ and patent applications are negatively and significantly correlated. RDI and CGQ appear, on the contrary, characterized by a positive association. However, when we look at the correlation between age and innovation variables (R&D intensity and patent applications), we detect negative relations. Overall, we can conclude that the relationships which are in place appear to be very complex, thus the need of a multivariate approach to properly capture the underlying dynamics. Table 5. Correlation matrix Variables  CGQ  ΔCGQ  ln(Age)  ln(SZ)  ln(RDI)  Δln(RDI)  Patents  CF  CGQ  1                ΔCGQ  −0.0042  1              ln(Age)  −0.1910*  0.0068  1            ln(SZ)  −0.0509*  0.0065  0.2334*  1          ln(RDI)  0.0635*  −0.0159  −0.0845*  −0.3843*  1        Δln(RDI)  −0.0122  −0.0214  0.0229  −0.0743*  0.2703*  1      Patents  −0.1519*  −0.0206  0.1640*  0.1550*  0.1471*  0.0214  1    CF  0.0964*  0.0118  0.0085  0.4387*  −0.0309*  0.0018  0.0798*  1  Variables  CGQ  ΔCGQ  ln(Age)  ln(SZ)  ln(RDI)  Δln(RDI)  Patents  CF  CGQ  1                ΔCGQ  −0.0042  1              ln(Age)  −0.1910*  0.0068  1            ln(SZ)  −0.0509*  0.0065  0.2334*  1          ln(RDI)  0.0635*  −0.0159  −0.0845*  −0.3843*  1        Δln(RDI)  −0.0122  −0.0214  0.0229  −0.0743*  0.2703*  1      Patents  −0.1519*  −0.0206  0.1640*  0.1550*  0.1471*  0.0214  1    CF  0.0964*  0.0118  0.0085  0.4387*  −0.0309*  0.0018  0.0798*  1  Note: * P-value < 0.05. 4. Methodology RDI and patent applications are used as response variables in a standard parametric setting to ascertain the average relationship between CG, age, and innovation, and establish some comparisons with the extant literature. We set different specifications. We first model (Fixed Effects—within transformation) the variation in RDI as a function of CG, age, and a set of key controls. The baseline-specified model is the following:   ΔlnRDIi,t=α+β1lnRDIi,t-1+β2CGQi,t-1+β3ΔCGQi,t+β4lnAgei,t+ (1)  +β×Xi,t-1+ui+εi,t while the fully specified model is as follows:   ΔlnRDIi,t=α+β1ln⁡RDIi,t-1+β3ΔCGQi,t+β4lnAgei,t+ (2)  +ΣkγkCGQi,t-1×dAgek+β×Xi,t-1+ui+εi,t for each firm i at time t. X is a vector of control variables such as size and cashflow.5 All the non-time-varying determinants (e.g. technological opportunities) which are likely to influence R&D activities are subsumed in the fixed-effect term ui. The lagged variables partially reduce the potential endogeneity between the set of covariates and the innovation proxy, but at this stage we refrain from giving any causal interpretation. In Equation (2) the variable CGQi,t−1 is interacted with dummy variables identifying age groups according to the distribution of firm age (Figure 1). In particular, we choose the quartiles of the age distribution as cutoffs, that is, age_1: age ≤ 25° percentile; age_2: 25° percentile < age ≤ 50° percentile; age_3: 50° percentile < age ≤ 75° percentile; age_4: age > 75° percentile. In this way we cover the full age distribution of sampled firms which leads us to drop CGQi,t−1 from the equation. For the sake of clarity, we label the four age classes as follows: (i) young firms, (ii) medium-aged firms, (iii) old firms, (iv) very old firms. Appendix reports descriptive statistics of the main variables broken down by these four groups. Some differences across groups are worthy to remark. First, young firms are characterized by a higher CGQ score. Second, mature firms tend to be more innovative than the younger counterparts (when we look at the medium values of RDI and patent applications). Third, both size and cashflow increase with the age the firm. The advantage of selecting the cutoff points by splitting the distribution of age rather than choosing specific arbitrary values is that our criterion is completely data driven. However, a series of robustness checks with different thresholds have been undertaken (see Section 5.1). We start by regressing the CGQ index on the variation in RDI. Afterward we augment the model with several explanatory variables to verify whether our estimations are robust across different configurations. We control for legal and institutional differences across countries with a set of country dummies and, though the time window we span is rather short, we include time dummies to account for potential macro-economic changes. Subsequently, we model the innovative effort in level by implementing the Arellano and Bond (1991) two-step robust generalized method of moments (GMM) estimators. The implementation of the dynamic model is derived from Equation (2), by considering that ΔlnRDIi,t=lnRDIi,t-lnRDIi,t-1. Equation (2) can be rewritten as follows:   lnRDIi,t-lnRDIi,t-1=α+β1ln⁡RDIi,t-1+β3ΔCGQi,t+β4lnAgei,t+  +ΣkγkCGQi,t-1×dAgek+β×Xi,t-1+ui+εi,t (2a) This leads us to the following dynamic specification:   lnRDIi,t=α+γ1ln⁡RDIi,t-1+β3ΔCGQi,t+β4lnAgei,t+ (3)  +ΣkγkCGQi,t-1×dAgek+β×Xi,t-1+ui+εi,t where γ1=β1+1. Equation (3) suffers from endogeneity due to the presence of the lagged dependent variable, and the estimates of the other covariates might be biased due to reverse causality between R&D investments and firms’ characteristics. The GMM estimator mitigates these problems by exploiting lags of the regressors as internal instruments; it involves however a certain degree of arbitrariness in the specification choices which we do believe deserve a few comments. First, in all our models we treat age and year dummies as exogenous variables, while the index of corporate governance CGQ, the size of the firm, and its cashflow are always regarded as endogenous to RDI, thus instrumented. Second, different lags of the endogenous covariates are used as instruments, based on the Arellano-Bond tests for serial correlation and on Sargan and Hansen tests for overidentifying restrictions. The outcome of these tests is always maximized when we take a lag structure that ranges from t-2 to t-5. Third, we do not collapse the set of internal instruments as their number in all specifications is rather small and the risk of instruments proliferation is virtually absent (Roodman, 2009). We finally apply small-sample correction for the asymptotic variance of the two-step GMM as proposed in Windmeijer (2005) to correct for the typical downward bias in the computation of standard errors. Turning to the innovation outcome, as highlighted in Section 3.3, the patent applications variable presents a very skewed distribution with the presence of many zeros. Moreover, the conditional variance exceeds the conditional mean to a large extent. Thus, to analyze the effect of CG on patent applications, it seems appropriate to abandon the Ordinary Least Squares (OLS) setting and adopt a Zero-Inflated Negative Binomial model (henceforth, ZINB), explicitly designed for the nature of our response variable. Zero-inflated models estimate two equations simultaneously, one to describe the relationship between the response variable and the set of covariates and one to model the excess of zeros.6 The equation to be estimated through the ZINB is the following:   Patentsi,t=α+β1ln⁡RDIi,t-1+β3ΔCGQi,t+β4lnAgei,t+ (4)  +ΣkγkCGQi,t-1×dAgek+β×Xi,t-1+εi,t We substantially re-estimate the model in Equations (1) and (2), substituting patent applications as the response variable.7 As for the zero-inflation, we use R&D intensity as an inflator since we expect firms will lower R&D investment to exhibit a lower propensity to patent. 5. Econometric results and discussion The results of the estimations using RDI as response variable are shown in Table 6 and Table 7. In particular, Table 6 reports the results obtained by estimating a static model through fixed-effect panel techniques. Column (1) shows the baseline model. The growth rate of R&D intensity is regressed just against the CG index (level and growth rate). Only the CG growth rate seems to yield a statistically significant coefficient, the sign of which is negative though significant only at 10% level. This supports the conjecture that the improvement of governance mechanisms leads to a decrease in R&D intensity, due to the shareholders value maximization target, which leads managers to prefer value preservation and a short-term horizon instead of value creation and long-term development. Table 6. CGQ effect on Δln(RDI)—FE Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.1121  −0.0671          (0.0749)  (0.0634)          ΔCGQt  −0.0071*  −0.0077***  −0.0080***  −0.0080***  −0.0082***  −0.0080***  (0.0040)  (0.0025)  (0.0026)  (0.0026)  (0.0028)  (0.0028)  ln(Age)t    0.0172  −0.0179  0.0279  −0.1074  −0.0829    (0.1937)  (0.1940)  (0.2029)  (0.2413)  (0.2573)  ln(RDI)t−1    −1.0686***  −1.0691***  −1.0700***  −1.0876***  −1.0907***    (0.0311)  (0.0307)  (0.0309)  (0.0438)  (0.0446)  ln(SZ)t    −0.5491***  −0.5493***  −0.5704***        (0.0717)  (0.0712)  (0.0731)      ln(SZ)t−1          0.0379  0.0405          (0.0418)  (0.0440)  CFt−1        −0.0009    −0.0100        (0.0106)    (0.0101)  CGQ*Young      −0.3491**  −0.3591**  −0.3625*  −0.3566*      (0.1504)  (0.1512)  (0.2118)  (0.2156)  CGQ*Medium-aged      −0.1576  −0.1565  −0.2434*  −0.2377      (0.1144)  (0.1158)  (0.1423)  (0.1451)  CGQ*Old      −0.0267  −0.0244  −0.0999  −0.0993      (0.0765)  (0.0771)  (0.0862)  (0.0875)  CGQ*Very old      0.0255  0.0230  −0.0024  −0.0024      (0.0821)  (0.0827)  (0.0826)  (0.0833)  Time dummies  yes  Yes  Yes  yes  yes  yes  N  3754  3712  3712  3668  3712  3668  R2  0.0023  0.6997  0.7006  0.7054  0.5882  0.5891  Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.1121  −0.0671          (0.0749)  (0.0634)          ΔCGQt  −0.0071*  −0.0077***  −0.0080***  −0.0080***  −0.0082***  −0.0080***  (0.0040)  (0.0025)  (0.0026)  (0.0026)  (0.0028)  (0.0028)  ln(Age)t    0.0172  −0.0179  0.0279  −0.1074  −0.0829    (0.1937)  (0.1940)  (0.2029)  (0.2413)  (0.2573)  ln(RDI)t−1    −1.0686***  −1.0691***  −1.0700***  −1.0876***  −1.0907***    (0.0311)  (0.0307)  (0.0309)  (0.0438)  (0.0446)  ln(SZ)t    −0.5491***  −0.5493***  −0.5704***        (0.0717)  (0.0712)  (0.0731)      ln(SZ)t−1          0.0379  0.0405          (0.0418)  (0.0440)  CFt−1        −0.0009    −0.0100        (0.0106)    (0.0101)  CGQ*Young      −0.3491**  −0.3591**  −0.3625*  −0.3566*      (0.1504)  (0.1512)  (0.2118)  (0.2156)  CGQ*Medium-aged      −0.1576  −0.1565  −0.2434*  −0.2377      (0.1144)  (0.1158)  (0.1423)  (0.1451)  CGQ*Old      −0.0267  −0.0244  −0.0999  −0.0993      (0.0765)  (0.0771)  (0.0862)  (0.0875)  CGQ*Very old      0.0255  0.0230  −0.0024  −0.0024      (0.0821)  (0.0827)  (0.0826)  (0.0833)  Time dummies  yes  Yes  Yes  yes  yes  yes  N  3754  3712  3712  3668  3712  3668  R2  0.0023  0.6997  0.7006  0.7054  0.5882  0.5891  Note: This table reports coefficients of Fixed-Effects (FE) estimations of Equation (2) with firm-level fixed effects. The response variable is Δln(RDI) and all other explanatory variables are defined in Table 3. To identify how the relationship between CG and innovation is moderated by age, we interact the CGQ index with four age classes. Robust standard errors clustered at firm level in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. Table 7. CGQ effect on ln(RDI)—GMM Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.3122**  −0.3852**          (0.1590)  (0.1640)          ΔCGQt  −0.0224  −0.0186  −0.0115  −0.0078  −0.0124  −0.0106  (0.0167)  (0.0182)  (0.0197)  (0.0179)  (0.0193)  (0.0242)  ln(Age)t    0.0641  −0.0774  0.0218  0.0071  −0.0166    (0.1724)  (0.2053)  (0.1996)  (0.1875)  (0.1792)  ln(RDI)t−1    −0.0536  −0.1434  −0.1569  −0.1028  −0.0666    (0.1264)  (0.1502)  (0.1236)  (0.2130)  (0.1768)  ln(SZ)t    −0.5488***  −0.4667***  −0.3593**        (0.1920)  (0.1777)  (0.1591)      ln(SZ)t−1          0.0661  0.0599          (0.1534)  (0.1408)  CFt−1        0.0028    0.0022        (0.0084)    (0.0080)  CGQ*Young      −1.1318*  −1.0511*  −1.1338*  −0.9500      (0.6051)  (0.5738)  (0.6408)  (0.5874)  CGQ*Medium-aged      −0.2014  −0.2459  −0.3377  −0.2693      (0.2225)  (0.2101)  (0.2191)  (0.2229)  CGQ*Old      −0.2359  −0.1669  −0.2479  −0.1737      (0.1838)  (0.1712)  (0.1899)  (0.1847)  CGQ*Very old      −0.1042  −0.0660  −0.0843  0.0100      (0.2051)  (0.1906)  (0.2083)  (0.2015)  Time dummies  yes  yes  yes  yes  yes  yes  N  3007  2743  2743  2705  2743  2705  AR(1)  0.000  0.007  0.060  0.035  0.109  0.031  AR(2)  0.452  0.515  0.293  0.172  0.740  0.828  Sargan  0.347  0.030  0.352  0.560  0.248  0.425  Hansen  0.408  0.532  0.805  0.563  0.640  0.672  Number of instruments  16  37  47  57  46  56  Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.3122**  −0.3852**          (0.1590)  (0.1640)          ΔCGQt  −0.0224  −0.0186  −0.0115  −0.0078  −0.0124  −0.0106  (0.0167)  (0.0182)  (0.0197)  (0.0179)  (0.0193)  (0.0242)  ln(Age)t    0.0641  −0.0774  0.0218  0.0071  −0.0166    (0.1724)  (0.2053)  (0.1996)  (0.1875)  (0.1792)  ln(RDI)t−1    −0.0536  −0.1434  −0.1569  −0.1028  −0.0666    (0.1264)  (0.1502)  (0.1236)  (0.2130)  (0.1768)  ln(SZ)t    −0.5488***  −0.4667***  −0.3593**        (0.1920)  (0.1777)  (0.1591)      ln(SZ)t−1          0.0661  0.0599          (0.1534)  (0.1408)  CFt−1        0.0028    0.0022        (0.0084)    (0.0080)  CGQ*Young      −1.1318*  −1.0511*  −1.1338*  −0.9500      (0.6051)  (0.5738)  (0.6408)  (0.5874)  CGQ*Medium-aged      −0.2014  −0.2459  −0.3377  −0.2693      (0.2225)  (0.2101)  (0.2191)  (0.2229)  CGQ*Old      −0.2359  −0.1669  −0.2479  −0.1737      (0.1838)  (0.1712)  (0.1899)  (0.1847)  CGQ*Very old      −0.1042  −0.0660  −0.0843  0.0100      (0.2051)  (0.1906)  (0.2083)  (0.2015)  Time dummies  yes  yes  yes  yes  yes  yes  N  3007  2743  2743  2705  2743  2705  AR(1)  0.000  0.007  0.060  0.035  0.109  0.031  AR(2)  0.452  0.515  0.293  0.172  0.740  0.828  Sargan  0.347  0.030  0.352  0.560  0.248  0.425  Hansen  0.408  0.532  0.805  0.563  0.640  0.672  Number of instruments  16  37  47  57  46  56  Note: This table reports coefficients of the two-step robust GMM estimations of Equation (3). The response variable is ln(RDI) and all other explanatory variables are defined in Table 3. AR(1) and AR(2) are the P-values for the Arellano-Bond tests for the first- and second-order autocorrelation. Sargan and Hansen are the P-values for the tests of overidentifying restrictions. To identify how the relationship between CG and innovation is moderated by age, we interact the CGQ index with four age classes. Windmeijerrobust standard errors in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. The second column shows the estimation including some control variables, such as age, past levels of R&D, and size. The results on CG are persistent as far as both the level and the growth rate are concerned. Moreover, the statistical significance of the coefficient of ΔCGQ is largely improved. The negative sign of the lagged level of R&D is expected, the R&D growth rate being the dependent variable. The lagged level of R&D expenditures can be in fact interpreted as a measure of size of the innovative efforts. Our results are consistent with longitudinal and cross-sectional investigations that have for long found negative correlation between size and growth of R&D efforts (Mansfield, 1968; Mitchell et al., 1995). As for the size of the company, we find a negative and significant effect on the growth of R&D intensity. This evidence contradicts the conjecture that large firms should have greater economies of scale and scope at their disposal, together with an easier access to capital. Our result is more in line with the literature which hypothesizes a negative relation between size and innovation propensity, based on the potential loss of managerial control in research allocation, typical of large companies (Cohen, 2010).8 In Columns (3) and (4), the differential effects of the CGQ variable across different age classes is appreciated by taking the quartiles of the distribution of the variable age and then calculating the interactions with CGQ. The results show that the coefficient is negative and significant only as far as firms belong to the first quartile of age distribution. This means that the effect of CG on innovation is moderated by the age of the firm, with young firms being the only category in which such effect is negative and statistically significant. As stressed in Section 2, this result is consistent with the resource-based view approach to corporate life cycle, according to which younger firms have on average a narrower knowledge base and weaker absorptive capacity. It is fair to note that so far our econometric specification includes the simultaneous value of size.9 However, regressing RDI against the simultaneous level of sales engenders severe reverse causality problems so that estimation results are likely to be biased. For this reason, in Columns (5) and (6) we report the estimations obtained by including the lagged value of sales as a proxy for size. The results do not seem to be much affected by this change since the coefficient is still negative and significant and the magnitude is largely stable. We notice, however, that the standard error has increased so that the coefficient is now significant only at 10% level, while it was at 5% in the previous specifications. Reworking Equation (1), the static panel model can be turned into a dynamic one, as shown in Equation (3). The results of the two-step difference GMM estimation are reported in Table 7. We follow the same format of Table 6. Column (1) shows the results for the baseline model. Here the situation is reversed, with respect to Table 5, as only the lagged level of CGQ has a significant (and negative) coefficient, while the growth rate of GCQ is not significant. The sign of the coefficient still suggests that improved CG is associated with lower levels of R&D intensity, supporting the idea that good governance can have a perverse effect on uncertain, long-term R&D investments. The situation is not altered by the introduction of control variables in Column (2). These latter but size have non-significant coefficients. Size is instead characterized by a negative and significant coefficient, which is consistent with our previous results. In Columns (3) and (4) we have introduced the four dummies identifying the quartiles of the distribution of sampled firms. Also, in these estimations, the only group showing a negative and significant coefficient is the one including the youngest firms, though the effect we find is always significant only at 10% level. This result is also robust to the introduction of the cashflow variable which should minimize the confounding effect of liquidity constraints, above all, as far as young firms are concerned, or that of over-investments also in risky projects as far as older firms are concerned. Columns (5) and (6) include the lagged value of size, instead of contemporaneous one, so as to minimize issues due to reverse causality. The coefficient of CG for younger firms keeps being negative and significant only in the first case, as in Column (6), when we add cash flow as a control variable, it is no longer significant.10 In sum, when RDI is used as a measure of firms’ innovation efforts, the analysis of the effects of innovation does not provide a stable picture. On the one hand, when adopting a fixed-effect panel data estimator, only the growth rate of CGQ is significant (and not the level), while CGQ for younger firms yields a negative and significant effect on R&D intensity across the different specifications. On the other hand, in the estimation through dynamic panel techniques, only the lagged value of CGQ shows a negative and significant (10% level) coefficient. Moreover, CGQ for younger firms yields statistically significant effects across all different specifications but that including the lagged value of size and cashflow. We have seen that while R&D intensity is a standard measure of input of the innovation process, patents can be considered as an output indicator. Both measures have their pros and cons. For the purposes of our analysis, patents seem to be better suited to grasping the effects of corporate life cycle related to the accumulation of technological competences and absorptive capacity. R&D efforts are not necessarily conducive to patented innovations, as some innovative projects are successful and some others are not. The share of unsuccessful innovative projects is likely to decrease as firms get more mature as an effect of learning dynamics (Dosi, Faillo, and Marengo, 2008). Besides this, the use of R&D expenditures as a variable is not fully reliable due to different regulatory settings concerning the reporting of these expenditures in different countries. Thus, we estimate the effects of CGQ on firms’ patenting activity, and the results are reported in Table 8. Since the number of patents is a count variable and a large share of zeros is observed in the data set, we implement a ZINB estimation, as explained in Section 4. Table 8. CGQ effect on patent applications—ZINB Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.2557**  −0.2607**          (0.1261)  (0.1319)          ΔCGQt  −0.0080  −0.0063  −0.0069  −0.0065  −0.0068  −0.0064  (0.0105)  (0.0107)  (0.0105)  (0.0106)  (0.0105)  (0.0106)  ln(Age)t    0.0632**  −0.0023  0.0212  −0.0025  0.0204    (0.0315)  (0.0587)  (0.0591)  (0.0585)  (0.0588)  ln(RDI)t−1    0.0345  0.0396  0.0236  0.0502*  0.0308    (0.0252)  (0.0254)  (0.0254)  (0.0265)  (0.0274)  ln(SZ)t    0.0476***  0.0475***  0.0207        (0.0159)  (0.0158)  (0.0183)      ln(SZ)t−1          0.0535***  0.0274          (0.0160)  (0.0191)  CFt−1        0.0207**    0.0192**        (0.0091)    (0.0094)  CGQ*Young      −0.6276**  −0.5777**  −0.6492**  −0.5876**      (0.2615)  (0.2634)  (0.2588)  (0.2617)  CGQ*Medium-aged      −0.2052  −0.1473  −0.2125  −0.1524      (0.1909)  (0.1925)  (0.1911)  (0.1927)  CGQ*Old      −0.2153  −0.2153  −0.2170  −0.2158      (0.1509)  (0.1526)  (0.1512)  (0.1527)  CGQ*Very old      −0.1972  −0.2142  −0.2053  −0.2194      (0.1560)  (0.1550)  (0.1575)  (0.1561)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  Yes  Yes  Yes  Yes  Country dummies  Yes  Yes  Yes  Yes  Yes  Yes  Inflation:              ln(RDI)t  −7.9037***  −8.7567***  −8.6102***  −8.3749***  −8.7365***  8.4436***  (1.6645)  (1.7682)  (1.7627)  (1.7972)  (1.8144)  (1.8291)  N  4023  3712  3712  3668  3712  3668  Vuong  11.39***  10.29***  10.24***  9.94***  10.14***  9.86***  Log likelihood  −9572.68  −9043.59  −9040.27  −8980.36  −9039.40  −8979.96  Variables  (1)  (2)  (3)  (4)  (5)  (6)  CGQt−1  −0.2557**  −0.2607**          (0.1261)  (0.1319)          ΔCGQt  −0.0080  −0.0063  −0.0069  −0.0065  −0.0068  −0.0064  (0.0105)  (0.0107)  (0.0105)  (0.0106)  (0.0105)  (0.0106)  ln(Age)t    0.0632**  −0.0023  0.0212  −0.0025  0.0204    (0.0315)  (0.0587)  (0.0591)  (0.0585)  (0.0588)  ln(RDI)t−1    0.0345  0.0396  0.0236  0.0502*  0.0308    (0.0252)  (0.0254)  (0.0254)  (0.0265)  (0.0274)  ln(SZ)t    0.0476***  0.0475***  0.0207        (0.0159)  (0.0158)  (0.0183)      ln(SZ)t−1          0.0535***  0.0274          (0.0160)  (0.0191)  CFt−1        0.0207**    0.0192**        (0.0091)    (0.0094)  CGQ*Young      −0.6276**  −0.5777**  −0.6492**  −0.5876**      (0.2615)  (0.2634)  (0.2588)  (0.2617)  CGQ*Medium-aged      −0.2052  −0.1473  −0.2125  −0.1524      (0.1909)  (0.1925)  (0.1911)  (0.1927)  CGQ*Old      −0.2153  −0.2153  −0.2170  −0.2158      (0.1509)  (0.1526)  (0.1512)  (0.1527)  CGQ*Very old      −0.1972  −0.2142  −0.2053  −0.2194      (0.1560)  (0.1550)  (0.1575)  (0.1561)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  Yes  Yes  Yes  Yes  Country dummies  Yes  Yes  Yes  Yes  Yes  Yes  Inflation:              ln(RDI)t  −7.9037***  −8.7567***  −8.6102***  −8.3749***  −8.7365***  8.4436***  (1.6645)  (1.7682)  (1.7627)  (1.7972)  (1.8144)  (1.8291)  N  4023  3712  3712  3668  3712  3668  Vuong  11.39***  10.29***  10.24***  9.94***  10.14***  9.86***  Log likelihood  −9572.68  −9043.59  −9040.27  −8980.36  −9039.40  −8979.96  Note: This table reports coefficients of the Zero Inflated Negative Binomial estimations of Equation (4). The response variable is patent applications and all other explanatory variables are defined in Table 3. Vuong is the statistic for the test of ZINB versus negative binomial model. To identify how the relationship between CG and innovation is moderated by age, we interact the CGQ index with four age classes. Robust standard errors clustered at firm level in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. Table 9. Sensitivity analysis (I) Variables  (Age < 15)  (Age < 20)  (Age < 15)  (Age < 15)  (Age < 20)  (Age < 20)  Patents  Patents  Δln(RDI)  ln(RDI)  Δln(RDI)  ln(RDI)  CGQ*Young  −0.5006**  −0.5623***  −0.4970  −0.7400  −0.2130  −0.4388  (0.2195)  (0.2062)  (0.3731)  (0.6870)  (0.1826)  (0.4377)  CGQ*Old  −0.2379*  −0.2175  −0.0821  −0.1593  −0.1075  −0.1664  (0.1345)  (0.1345)  (0.0735)  (0.1578)  (0.0716)  (0.1645)  ΔCGQt  −0.0066  −0.0068  −0.0079***  −0.0097  −0.0080***  −0.0071  (0.0106)  (0.0105)  (0.0027)  (0.0254)  (0.0027)  (0.0290)  ln(Age)t  0.0281  0.0057  −0.1392  −0.0030  −0.0909  0.0776  (0.0392)  (0.0422)  (0.2425)  (0.1912)  (0.2423)  (0.1911)  ln(RDI)t−1  0.0477*  0.0481*  −1.0854***  −0.1686  −1.0865***  −0.1911  (0.0263)  (0.0262)  (0.0430)  (0.2371)  (0.0441)  (0.2296)  ln(SZ)t−1  0.0535***  0.0550***  0.0426  −0.0793  0.0397  −0.0641  (0.0161)  (0.0162)  (0.0416)  (0.1270)  (0.0415)  (0.1286)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −8.8104***  −8.7107***          (1.8087)  (1.8009)          N  3712  3712  3712  2743  3712  2743  R2      0.5892    0.5875    Vuong  10.18***  10.17***          AR(1)        0.069    0.082  AR(2)        0.836    0.687  Sargan        0.016    0.012  Hansen        0.444    0.550  # of instruments        43    43  Variables  (Age < 15)  (Age < 20)  (Age < 15)  (Age < 15)  (Age < 20)  (Age < 20)  Patents  Patents  Δln(RDI)  ln(RDI)  Δln(RDI)  ln(RDI)  CGQ*Young  −0.5006**  −0.5623***  −0.4970  −0.7400  −0.2130  −0.4388  (0.2195)  (0.2062)  (0.3731)  (0.6870)  (0.1826)  (0.4377)  CGQ*Old  −0.2379*  −0.2175  −0.0821  −0.1593  −0.1075  −0.1664  (0.1345)  (0.1345)  (0.0735)  (0.1578)  (0.0716)  (0.1645)  ΔCGQt  −0.0066  −0.0068  −0.0079***  −0.0097  −0.0080***  −0.0071  (0.0106)  (0.0105)  (0.0027)  (0.0254)  (0.0027)  (0.0290)  ln(Age)t  0.0281  0.0057  −0.1392  −0.0030  −0.0909  0.0776  (0.0392)  (0.0422)  (0.2425)  (0.1912)  (0.2423)  (0.1911)  ln(RDI)t−1  0.0477*  0.0481*  −1.0854***  −0.1686  −1.0865***  −0.1911  (0.0263)  (0.0262)  (0.0430)  (0.2371)  (0.0441)  (0.2296)  ln(SZ)t−1  0.0535***  0.0550***  0.0426  −0.0793  0.0397  −0.0641  (0.0161)  (0.0162)  (0.0416)  (0.1270)  (0.0415)  (0.1286)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −8.8104***  −8.7107***          (1.8087)  (1.8009)          N  3712  3712  3712  2743  3712  2743  R2      0.5892    0.5875    Vuong  10.18***  10.17***          AR(1)        0.069    0.082  AR(2)        0.836    0.687  Sargan        0.016    0.012  Hansen        0.444    0.550  # of instruments        43    43  Note: This table reports coefficients of the robustness checks: FE, GMM and ZINB estimations. Two age classes have been defined according to two arbitrary thresholds of firm’s age, namely, 15 and 20 years. All variables are defined in Table 3. AR(1) and AR(2) are the P-values for the Arellano-Bond tests for the first- and second-order autocorrelation. Sargan and Hansen are the P-values for the tests of overidentifying restrictions. Vuong is the statistic for the test of ZINB versus negative binomial model. To identify how the relationship between CG and innovation is moderated by age, we interact the CGQ index with four age classes. Robust standard errors in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. Consistently with previous estimations, we begin by reporting the baseline model in Column (1). In line with GMM estimations in Table 7, only the lagged level of CGQ is negative and significant, while the growth rate is not. On average, this suggests that the better the CG score, the lower the innovative output for firms. The persistence of this result provides further robustness to the hypothesis concerning the adverse effects of good governance on risky investments involving innovation. The inclusion of control variables in Column (2) does not alter the picture. It should be noted that age yields a positive and significant coefficient which is largely in line with a resource-based view of the firm. Also (contemporaneous) size is characterized by a positive and significant coefficient. In Columns (3) and (4), the effect of CGQ is interacted with the four dummies identifying the quartiles of the firms’ age distribution. Once again, the results are in line with previous estimation and show that only for firms in the first quartile, i.e. younger firms, the effects of CGQ are negative and significant, and this holds also when cashflow is included in the estimation. We can notice, however, that the size coefficient loses its significance when cashflow is added to the model. As we discussed in Section 2, liquidity constraints (i.e. low cashflow) should hinder innovative outcomes and are typically stronger in the case of small and young companies. As a consequence, these relations are such that omitting cashflow from the estimation causes a positive bias on the size coefficient. Finally, in Columns (5) and (6), we replicate the previous two estimations by including the lagged level of size instead of the contemporaneous one. The results are consistent with the evidence discussed so far, as CG is negative and significant only as far as younger firms are concerned. For firms belonging to the second, third, and fourth age group, we do not obtain any significant coefficient on CGQ. 5.1 Sensitivity analysis A reasonable objection could be that the results we have shown so far may be to some extent driven by the choice of the cutoffs for the age classes, even though these are based on the distribution of the variable. For the sake of a robustness check, we report in Table 9 the estimations of the effects of CG on innovation across differently aged firms by choosing arbitrary thresholds. Columns (1) and (2) report the results of the ZINB estimation by using 15 years and 20 years, respectively, as critical values to discriminate between young and mature firms. In the first case (the least inclusive one), CG is characterized by a negative and significant coefficient on both old and young firms, but the magnitude of the latter is definitely larger (twice as large) and statistically more significant. If we extend the group of young firms so as to include companies aged up to 20 years, then only CG for young firms is negative and significant, and the statistical significance is dramatically improved compared with the previous estimation. Columns (3) to (6) report the estimation results of the model in which R&D intensity is the dependent variable. In these estimations, we notice that the coefficient is never significant, neither for young nor for old firms. The result is robust to different techniques [Fixed Effects (FE) and GMM] and to different thresholds for the age classes. The conclusion we can derive is that results on RDI are quite sensitive to different specifications, whereas the evidence on patent applications is very robust. A further empirical issue is related to the fact that missing data are a frequent complication of any statistical inference, particularly when it comes to deal with firm-level data. BvD Orbis, though widely exploited by researchers in past few years, suffers from missing information as far as R&D expenditure is concerned. This might be due to imputation errors or simply to the fact that firms do not disclose such an information. In the analyses shown so far, our conservative strategy was to run analysis on the set of observations with no missing values. A first drawback of this choice, however, is related to the loss of a large fraction of the observations. Another problem is that if the missing completely at random assumption is not fulfilled, estimates might be biased (Little and Rubin, 2014). In our framework, this assumption implies that missingness should be independent of the main explanatory variable, that is, the CG index. We have tested this assumption and unfortunately we could not find its support in our data. Thus, following Horton and Kleinman (2007), we have undertaken further estimates to check whether our main findings are influenced by the treatment of missing values in the R&D proxy. The first strategy consists of an “ad-hoc” adjustment, based on recoding missing values to some common value, creation of an indicator of missingness as a new variable, and including both these variables along with their interaction in the regression model. Quite intuitively our choice is to turn R&D expenditure missing values into 0. Results are reported in Table 10. Table 10. Sensitivity Analysis (II). CGQ effect on innovation—R&D missing values turned to zeros Variables  (1)  (2)  (3)  (4)  (5)  (6)  Patents  Patents  Δln(RDI)  Δln(RDI)  ln(RDI)  ln(RDI)  CGQ*Young  −0.7657***  −0.7907***  −0.2106*  −0.2135*  −1.1985*  −0.9247*  (0.2440)  (0.2509)  (0.1219)  (0.1232)  (0.6761)  (0.5612)  CGQ*Medium-aged  −0.2277  −0.2747  −0.0692  −0.0704  −0.2180  −0.2060  (0.1765)  (0.1789)  (0.0770)  (0.0782)  (0.1673)  (0.1559)  CGQ*Old  −0.2147  −0.2334  −0.0409  −0.0406  −0.1670  −0.1614  (0.1433)  (0.1459)  (0.0624)  (0.0634)  (0.1627)  (0.1441)  CGQ*Very old  −0.2139  −0.2430  0.0674  0.0704  0.1387  0.2334  (0.1554)  (0.1570)  (0.0629)  (0.0634)  (0.3285)  (0.2898)  ΔCGQt  0.0040  0.0016  −0.0039**  −0.0039**  0.0503  0.0039  (0.0103)  (0.0100)  (0.0016)  (0.0016)  (0.0657)  (0.0624)  ln(Age)t  0.0028  0.0158  −0.0015  0.0254  −0.3681  −0.3962  (0.0561)  (0.0570)  (0.1147)  (0.1217)  (0.2962)  (0.3023)  ln(RDI)t−1  0.0110  0.0183  −1.0166***  −1.0239***  −0.2124  −0.1999  (0.0261)  (0.0267)  (0.0349)  (0.0359)  (0.1868)  (0.1824)  ln(SZ)t−1  0.0341**  0.0229  −0.2842***  −0.2902***  0.3383  0.2567  (0.0165)  (0.0207)  (0.0497)  (0.0516)  (0.2218)  (0.2127)  CFt−1    0.0235**    −0.0011    0.0225    (0.0093)    (0.0078)    (0.0177)  R&D missing dummy and interaction  yes  yes  yes  yes  yes  yes  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −13.8024***  −13.8997***          (2.6027)  (2.8545)          N  5658  5397  5661  5591  4312  4249  Vuong  13.11***  12.64***          R2      0.628  0.629      AR(1)          0.0507  0.0114  AR(2)          0.5352  0.3214  Sargan          0.0286  0.0234  Hansen          0.8449  0.8396  # of instruments          45  55  Variables  (1)  (2)  (3)  (4)  (5)  (6)  Patents  Patents  Δln(RDI)  Δln(RDI)  ln(RDI)  ln(RDI)  CGQ*Young  −0.7657***  −0.7907***  −0.2106*  −0.2135*  −1.1985*  −0.9247*  (0.2440)  (0.2509)  (0.1219)  (0.1232)  (0.6761)  (0.5612)  CGQ*Medium-aged  −0.2277  −0.2747  −0.0692  −0.0704  −0.2180  −0.2060  (0.1765)  (0.1789)  (0.0770)  (0.0782)  (0.1673)  (0.1559)  CGQ*Old  −0.2147  −0.2334  −0.0409  −0.0406  −0.1670  −0.1614  (0.1433)  (0.1459)  (0.0624)  (0.0634)  (0.1627)  (0.1441)  CGQ*Very old  −0.2139  −0.2430  0.0674  0.0704  0.1387  0.2334  (0.1554)  (0.1570)  (0.0629)  (0.0634)  (0.3285)  (0.2898)  ΔCGQt  0.0040  0.0016  −0.0039**  −0.0039**  0.0503  0.0039  (0.0103)  (0.0100)  (0.0016)  (0.0016)  (0.0657)  (0.0624)  ln(Age)t  0.0028  0.0158  −0.0015  0.0254  −0.3681  −0.3962  (0.0561)  (0.0570)  (0.1147)  (0.1217)  (0.2962)  (0.3023)  ln(RDI)t−1  0.0110  0.0183  −1.0166***  −1.0239***  −0.2124  −0.1999  (0.0261)  (0.0267)  (0.0349)  (0.0359)  (0.1868)  (0.1824)  ln(SZ)t−1  0.0341**  0.0229  −0.2842***  −0.2902***  0.3383  0.2567  (0.0165)  (0.0207)  (0.0497)  (0.0516)  (0.2218)  (0.2127)  CFt−1    0.0235**    −0.0011    0.0225    (0.0093)    (0.0078)    (0.0177)  R&D missing dummy and interaction  yes  yes  yes  yes  yes  yes  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −13.8024***  −13.8997***          (2.6027)  (2.8545)          N  5658  5397  5661  5591  4312  4249  Vuong  13.11***  12.64***          R2      0.628  0.629      AR(1)          0.0507  0.0114  AR(2)          0.5352  0.3214  Sargan          0.0286  0.0234  Hansen          0.8449  0.8396  # of instruments          45  55  Note: Robust standard errors in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. The evidence confirms that the main pattern of significance is unaffected. It is worth noting that the coefficient of ln(SZt−1) appears to be negative and significant in Models (3) and (4). This result can be due to effect of the adjustments on firms with erratic missing values of R&D investments. In this case, it is likely to observe simulated large values for the dependent variable Δln(RDI). Since smallest firms are not required to declare their R&D expenses (Stoneman and Toivanen, 2001), R&D missing values are more likely to be found in this size class, suggesting that those artificial high levels of Δln(RDI) are more likely to feature small firms than large ones. As a result, the negative relationship between size and Δln(RDI) might be a specific artificial outcome of this correction procedure. Further checks carried out dropping outliers in the dependent variables, and the results reported in Table 11 would provide support to this interpretation.11 Table 11. Sensitivity Analysis (III). CGQ effect on innovation, R&D data generated with multiple imputations Variables  (1)  (2)  (3)  (4)  (5)  (6)  Patents  Patents  Δln(RDI)  Δln(RDI)  ln(RDI)  ln(RDI)  CGQ*Young  −0.7050**  −0.6487**  −0.3677*  −0.3778*  −0.5259*  −0.5642  (0.2769)  (0.2813)  (0.2208)  (0.2228)  (0.3274)  (0.9280)  CGQ*Medium-aged  −0.1303  −0.0718  −0.2886**  −0.2850**  −0.4001  −0.4052  (0.2042)  (0.2045)  (0.1388)  (0.1406)  (0.4297)  (0.4151)  CGQ*Old  0.0684  0.0550  −0.1543  −0.1487  0.3068  0.3011  (0.1667)  (0.1680)  (0.1110)  (0.1119)  (0.3496)  (0.3224)  CGQ*Very old  0.0214  −0.0101  −0.0763  −0.0741  0.0831  0.1682  (0.1897)  (0.1866)  (0.1178)  (0.1191)  (0.3897)  (0.3747)  ΔCGQt  0.0073  0.0066  −0.0151***  −0.0152***  0.0503  0.0039  (0.0097)  (0.0095)  (0.0049)  (0.0049)  (0.0657)  (0.0624)  ln(Age)t  −0.0327  −0.0050  −0.2726  −0.3100  −0.3681  −0.3962  (0.0647)  (0.0652)  (0.2435)  (0.2474)  (0.2962)  (0.3023)  ln(RDI)t−1  0.0395  0.0133  −1.1832***  −1.1873***  −0.2124  −0.1999  (0.0318)  (0.0321)  (0.0223)  (0.0224)  (0.1868)  (0.1824)  ln(SZ)t−1  0.1114***  0.0737***  −0.0057  −0.0110  0.3383  0.2567  (0.0195)  (0.0230)  (0.0327)  (0.0335)  (0.2218)  (0.2127)  CFt−1    0.0354***    0.0065    0.0225    (0.0098)    (0.0134)    (0.0177)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −11.3447***  −11.0904***          (2.5772)  (2.5883)          N  5454  5397  5647  5591  4300  4249  Vuong  8.12***  8.01***          R2      0.626  0.628      AR(1)          0.026  0.078  AR(2)          0.232  0.655  Sargan          0.670  0.521  Hansen          0.722  0.579  # of instruments          51  55  Variables  (1)  (2)  (3)  (4)  (5)  (6)  Patents  Patents  Δln(RDI)  Δln(RDI)  ln(RDI)  ln(RDI)  CGQ*Young  −0.7050**  −0.6487**  −0.3677*  −0.3778*  −0.5259*  −0.5642  (0.2769)  (0.2813)  (0.2208)  (0.2228)  (0.3274)  (0.9280)  CGQ*Medium-aged  −0.1303  −0.0718  −0.2886**  −0.2850**  −0.4001  −0.4052  (0.2042)  (0.2045)  (0.1388)  (0.1406)  (0.4297)  (0.4151)  CGQ*Old  0.0684  0.0550  −0.1543  −0.1487  0.3068  0.3011  (0.1667)  (0.1680)  (0.1110)  (0.1119)  (0.3496)  (0.3224)  CGQ*Very old  0.0214  −0.0101  −0.0763  −0.0741  0.0831  0.1682  (0.1897)  (0.1866)  (0.1178)  (0.1191)  (0.3897)  (0.3747)  ΔCGQt  0.0073  0.0066  −0.0151***  −0.0152***  0.0503  0.0039  (0.0097)  (0.0095)  (0.0049)  (0.0049)  (0.0657)  (0.0624)  ln(Age)t  −0.0327  −0.0050  −0.2726  −0.3100  −0.3681  −0.3962  (0.0647)  (0.0652)  (0.2435)  (0.2474)  (0.2962)  (0.3023)  ln(RDI)t−1  0.0395  0.0133  −1.1832***  −1.1873***  −0.2124  −0.1999  (0.0318)  (0.0321)  (0.0223)  (0.0224)  (0.1868)  (0.1824)  ln(SZ)t−1  0.1114***  0.0737***  −0.0057  −0.0110  0.3383  0.2567  (0.0195)  (0.0230)  (0.0327)  (0.0335)  (0.2218)  (0.2127)  CFt−1    0.0354***    0.0065    0.0225    (0.0098)    (0.0134)    (0.0177)  Time dummies  Yes  Yes  Yes  Yes  Yes  Yes  Industry dummies  Yes  Yes  –  –  –  –  Country dummies  Yes  Yes  –  –  –  –  Inflate:              ln(RDI)t  −11.3447***  −11.0904***          (2.5772)  (2.5883)          N  5454  5397  5647  5591  4300  4249  Vuong  8.12***  8.01***          R2      0.626  0.628      AR(1)          0.026  0.078  AR(2)          0.232  0.655  Sargan          0.670  0.521  Hansen          0.722  0.579  # of instruments          51  55  Note: Robust standard errors in parentheses: * P < 0.10, ** P < 0.05, *** P < 0.01. The second strategy is more sophisticated and consists of multiple imputation techniques. By mean of Predictive Mean Matching method (Rubin, 1976) we create plausible R&D values for missing observations, and use these values to “fill-in” the missing values. This process is repeated (five times was adequate to obtain convergence), resulting in the creation of a number of datasets in which R&D expenditure is not missing. Each of these data sets is analyzed and results are next combined to increase the efficiency of the estimated coefficients. Results are reported in Table 11 and again we can conclude that our findings are in line with those discussed in the previous section. 5.2 Relationship with the literature In a nutshell, the results of our empirical investigations show that CG yields negative effects on innovation and that this negative relationship mostly applies to younger firms with high standards of governance practice. While the negative effect is not so robust for R&D, it is particularly strong and significant for patent applications. These results thus broadly contribute to the increasing strand of empirical literature that investigates the impact of good CG on innovation in line with the CG life cycle approach. We generalize the prediction by Holmstrom (1989) and Lazonick, (2007) saying that CG negatively impacts innovation. Based on a multi-attribute measure of CGQ, the negative relationship is largely confirmed in our results, beyond pure moral hazard issues. The effect is even stronger for firms that perform well in scores, suggesting that short-termism and value preservation prevail even more in their decision-making. The inconclusive and mixed results stressed in the literature can be clarified by considering different firm ages. Contradictory findings opposing long-run R&D investments correlated to a high level of shareholder protection (Brown et al., 2013) versus stock options with long vesting periods, golden parachutes, and managerial entrenchment as necessary conditions for innovation (Manso, 2011), may be related to the non-linearities linked to firm age. In the same vein, arguments concerning liquidity and impatient capital can be rationalized by age. Ownership structure, as well as investor characteristics, may differ radically between young firms just gone public and more mature listed firms. Innovation will be stronger, as managers are encouraged to opt for value creation and re-investment in product and process innovation rather than value preservation and distribution of dividends, while investors are more committed to long-run perspectives than short-term ones. However, this should even be more so as firms are young and face a lot of uncertainty. Our analysis opens up a new perspective by taking into account the interacting effects of a firm's age while drawing light on contrasting results in the literature based on a closer identification of what actually drives the observed shift in regressions. Some further theoretical efforts are expected to provide systemic account on the changes of corporate practices across the corporate life cycle, in the same vein as O’Connor and Byrne (2015) and Saravia (2014) on the quality of CG in mature firms, and Filatotchev et al. (2006) on the quality of resource base across different firms age, or the growing literature on antitakeover provisions (Atanassov, 2013). In our data set, the governance standard is such that incorporations in a state without antitakeover provisions or opting out of such protections is viewed favorably, but the weight of the different attributes of CG is indeed likely to change across the stages of firm evolution. O’Connor and Byrne (2015) suggest on that point that individual governance provisions such as independence, accountability, and transparency can have differential importance at different moments. As a matter of fact, CG and its impact on innovation changes across the different percentiles and we get to know more about what drives these changes from stage to stage throughout the life cycle. In these efforts, we need to compare the results obtained by using both R&D and patents as a measure of innovation, due to the basic limitations of R&D statistics. The quality of corporate financial reporting on R&D activity and intangibles in general is often inadequate for economic analysis purposes. Therefore, R&D investments can be a source of greater information asymmetries between ownership and management and may not be properly valued by the market. In addition, national accounting laws often do not require corporation to disclose the amount of their annual R&D expenditures. Patent statistics mitigate the bias caused by these problems although they are concerned by other issues (Griliches, 1990; Pavitt, 1985) which however do not dramatically affect their explanatory power. The fact that we find a differential effect on R&D and patents suggests that these two indicators still have to be used jointly to assess the impact on CG in the economic system. Moreover, considering R&D and patents also contribute to understand how the differential effect operates in a heterogeneous population of firms composed of older and younger firms. 6. Conclusions Empirical analyses of the relationships between CG and firm performances have mostly focused on the impact on financial performances and market value. Only recently some contributions have begun to investigate the impact of CG on innovation, by showing in most cases that good governance practices are associated with low levels of innovation. No attention has been devoted in this framework to the differential impact of CG on innovation across the different stages of a firm's life cycle. This article aims to fill this gap by investigating whether firm's age moderates the relationship between CG and innovation and, if so, in which direction. We carried out empirical analyses on a sample of listed firms extracted by the ISS RiskMetrics database, observed in the time period 2003–2008. The results presented along this article provide support to the idea that high CG scores are associated with low (but not robust) levels of R&D and low and significant patenting activity, suggesting that good managers are likely to maximize shareholders’ utility by privileging value preservation rather than value creation. In this framework, the effect of age is such that young new listed firms are characterized by an even stronger negative relationship between CG and innovation. The impact of good governance practices is augmented by a lack of necessary competences in younger firms, which ensure effective management of successful innovation projects, and by a progressive disincentive in patenting given the short-termism orientation of investors and shareholders. Given the huge development of good CG in developed economies now struggling to innovate and generate economic growth, it is certainly important to consider the significant evidence that CG may not always have the expected positive results. Previous contributions have shown that negative effects were obtained on the inputs of innovation lying in R&D expenses and for large populations of firms. The current article reinforces the idea by arguing that the negative effect occurs on the output of innovation as well, namely patents, and especially for younger firms. The future research agenda and related policy recommendations will have to address the basic fact that innovation gives room to agency costs because of its intrinsic and uncertain nature; and that systematic efforts to eradicate these agency costs may simply eradicate innovation at the same time. In that perspective, one can try to extend the current work by getting a clearer picture of the ongoing process through the identification of the key attributes of CG that are the most likely to block the generation of innovation measured by R&D and patents, and also the impact on the nature of innovation, namely, radical versus incremental types of innovation. These further steps will of course structure our own future research. Footnotes 1 Unfortunately, the cross-sector distribution is available only after the matching, so we cannot compare the ex ante distribution with that of ex post. 2 To be as much inclusive as possible, we have included in our analysis both patent applications filed to local patent offices and international applications under the Patent Cooperation Treaty. 3 This is a stock market index of foreign stocks from the perspective of North American investors. The index is market capitalization weighted (meaning that the weight of securities is determined based on their respective market capitalizations). The index aims to cover 85% of the market capitalization of the equity markets of all countries that are a part of the index. It is maintained by Morgan Stanley Capital International. EAFE is Europe, Australia, Asia and the Far East. 4 BvD Orbis is known to suffer the presence of many missing data in innovation-related variables. In our main analysis, missing values on R&D expenditure have not been replaced with zeros, however. It might be the case that our choice matters in some instances, but as will be shown in Section 5.1 our results are unaffected by different treatments of missing values. We thank an anonymous referee for having pointed out this issue. 5 We anticipate that some of the specifications presented in Section 5 will contain contemporaneous variables. This of course entails severe problems of reverse causality but allows us to compare our findings with those coming from cross-section studies and/or others adopting an empirical setting with contemporaneous variables. 6 We implemented the Vuong test to verify whether ZINB model is actually more appropriate for our data than a standard negative binomial (NB) model. The Vuong test is a classical likelihood-ratio test for model selection in which the null hypothesis is that two models (ZINB and NB in this case) are equally close to the true data generating process. We obtain, as will be shown in Section 5, large and positive Vuong statistics in all specifications that provide evidence of the superiority of ZINB over NB model. 7 The computation burden (i.e. the convergence of the likelihood maximization problem is not achieved) of the ZINB model did not allow us to introduce firm-level fixed effects. As robustness check, we have re-estimated a conditional Poisson (with no zero-inflation but with fixed effects) to account for the unobserved heterogeneity. Results are consistent with the ones we present throughout the article and are available upon request. 8 Acs and Audretsch (1987) show that size may also be interpreted as a proxy for market concentration and product market competition, thus leading to different effects depending on the sector in which firms operate. We cannot exclude a priori such a hypothesis in our data but its testing goes largely beyond the purpose of the article. 9 Though we are aware of the endogeneity bias that such a choice may introduce, we report these estimates mainly with the aim at comparing our findings with those of other studies adopting an empirical setting with contemporaneous variables. 10 It is fair to note that in some specifications (3 and 5), the p-value of AR(1) is larger than 0.05, thus not fully satisfactory, although the best we could obtain. The poor quality of the tests implies that results might be sensitive to the specification choices of the GMM estimation. Overall, this implies that the negative association between CG and R&D must be regarded as not very robust. 11 Also in this case we can notice the positive bias on the size coefficient when cashflow is not included into the model (see Columns 1 and 2). References Acs Z. J., Anselin L., Varga A. ( 2002), ‘ Patents and innovation counts as measures of regional production of new knowledge,’ Research Policy , 31, 1069– 1085. Google Scholar CrossRef Search ADS   Acs Z. J., Audretsch D. B. ( 1987), ‘ Innovation, market structure, and firm size,’ The Review of Economics and Statistics , 69, 567– 574. Google Scholar CrossRef Search ADS   Aggarwal R., Erel I., Stulz R., Williamson R. ( 2010), ‘ Differences in governance practices between US and foreign firms: measurement, causes and consequences,’ Review of Financial Studies , 23( 3), 3131– 3154. Google Scholar CrossRef Search ADS   Arellano M., Bond S. ( 1991), ‘ Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations,’ The Review of Economic Studies , 58( 2), 277– 297. Google Scholar CrossRef Search ADS   Atanassov J. ( 2013), ‘ Do hostile takeovers stifle innovation? Evidence from antitakeover legislation and corporate patenting,’ The Journal of Finance , 68, 1097– 1131. Google Scholar CrossRef Search ADS   Baranchuk N., Kieschnick R., Moussawi R. ( 2014), ‘ Motivating innovation in newly public firms,’ Journal of Financial Economics , 111( 3), 578– 588. Google Scholar CrossRef Search ADS   Bebchuk L., Cohen A., Ferrell A. ( 2008), ‘ What matters in corporate governance?,’ Review of Financial Studies , 22, 783– 827. Google Scholar CrossRef Search ADS   Bebchuk L., Weisbach M. ( 2010), ‘ The state of corporate governance research,’ Review of Financial Studies , 23( 3), 939– 961. Google Scholar CrossRef Search ADS   Becker-Blease J. R. ( 2011), ‘ Governance and innovation,’ Journal of Corporate Finance , 17( 4), 947– 958. Google Scholar CrossRef Search ADS   Belloc F. ( 2012), ‘ Corporate governance and innovation: a survey,’ Journal of Economic Surveys , 26( 5), 835– 864. Google Scholar CrossRef Search ADS   Bertrand M., Mullainathan S. ( 2003), ‘ Enjoying the quiet life? Corporate governance and managerial preferences,’ Journal of Political Economy , 111( 5), 1043– 1075. Google Scholar CrossRef Search ADS   Brossard O., Lavigne S., Sakinc M. ( 2013), ‘ Ownership structures and R&D in Europe: the good institutional investors, the bad and ugly impatient shareholders,’ Industrial and Corporate Change , 22( 4), 1031– 1068. Google Scholar CrossRef Search ADS   Brown J. R., Fazzari S. M., Petersen B. C. ( 2009), ‘ Financing innovation and growth: cash flow, external equity, and the 1990s R&D boom,’ The Journal of Finance , 64( 1), 151– 185. Google Scholar CrossRef Search ADS   Brown J. R., Martinsson G., Petersen B. C. ( 2013), ‘ Law, stock markets, and innovation,’ The Journal of Finance , 68( 4), 1517– 1549. Google Scholar CrossRef Search ADS   Core J., Guay W., Rusticus T. ( 2006), ‘ Does weak governance cause weak stock returns? An examination of firm operating performance and investors’ expectations,’ The Journal of Finance , 61, 655– 687. Google Scholar CrossRef Search ADS   Cohen W. M. ( 2010), ‘ Fifty years of empirical studies of innovative activity and performance,’ Handbook of the Economics of Innovation , 1, 129– 213. Google Scholar CrossRef Search ADS   Dietrich M., Krafft J. (eds) ( 2012), Handbook on the Economics and Theory of the Firm . Edward Elgar: Cheltenham. Google Scholar CrossRef Search ADS   Dosi G., Faillo M., Marengo L. ( 2008), ‘ Organizational capabilities, patterns of knowledge accumulation and governance structures in business firms: an introduction,’ Organization Studies , 29( 8–9), 1165– 1185. Google Scholar CrossRef Search ADS   Driver C., Guedes M. J. C. ( 2012), ‘ Research and development, cash flow, agency and governance: UK large companies,’ Research Policy , 41( 9), 1565– 1577. Google Scholar CrossRef Search ADS   Fang V. W., Tian X., Tice S. ( 2014), ‘ Does stock liquidity enhance or impede firm innovation?,’ The Journal of Finance , 69( 5), 2085– 2125. Google Scholar CrossRef Search ADS   Filatotchev I., Wright M. ( 2005), The Life Cycle of Corporate Governance . Edward Elgar: Cheltenham. Filatotchev I., Toms S., Wright M. ( 2006), ‘ The firm's strategic dynamics and corporate governance life-cycle,’ International Journal of Managerial Finance , 2( 4), 256– 279. Google Scholar CrossRef Search ADS   Gompers P., Ishii J., Metrick A. ( 2003), ‘ Corporate governance and equity prices,’ The Quarterly Journal of Economics , 118, 107– 155. Google Scholar CrossRef Search ADS   Griliches Z. (ed.) ( 1984), R&D, Patents and Productivity . University of Chicago Press: Chicago. Google Scholar CrossRef Search ADS   Griliches Z. ( 1990), ‘ Patent statistics as economic indicators: a survey,’ Journal of Economic Literature , 28, 1661– 1707. Hall B. H., Griliches Z., Hausman J. A. ( 1986), ‘ Patents and R&D: is there a lag?,’ International Economic Review , 27, 265– 283. Google Scholar CrossRef Search ADS   Holmstrom B. ( 1989), ‘ Agency costs and innovation,’ Journal of Economic Behavior and Organization , 12, 305– 327. Google Scholar CrossRef Search ADS   Horton N. J., Kleinman K. P. ( 2007), ‘ Much ado about nothing: A comparison of missing data methods and software to fit incomplete data regression models.,’ The American Statistician , 61( 1), 79– 90. Google Scholar CrossRef Search ADS   Jensen M., Meckling W. ( 1976), ‘ Theory of the firm: managerial behaviour, agency costs and ownership structure,’ Journal of Financial Economics , 3, 305– 360. Google Scholar CrossRef Search ADS   Krafft J., Ravix J. L. ( 2005), ‘ The governance of innovative firms: an evolutionary perspective,’ Economics of Innovation and New Technology , 14( 3–4), 1– 25. Krafft J., Ravix J. L. ( 2008), ‘ Corporate governance and the governance of knowledge: Rethinking the relationship in terms of corporate coherence,’ Economics of Innovation and New Technology , 17( 1–2), 79– 96. Google Scholar CrossRef Search ADS   Krafft J., Qu Y., Quatraro F., Ravix J. L. ( 2014), ‘ Corporate governance, value and performance of firms: new empirical results on convergence from a large international database,’ Industrial and Corporate Change , 23( 2), 361– 397. Google Scholar CrossRef Search ADS   Lanjouw J. O., Pakes A., Putnam J. ( 1998), ‘ How to count patents and value intellectual property: the uses of patent renewal and application data,’ Journal of Industrial Economics , 46, 405– 432. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A. ( 2008), ‘ The economic consequences of legal origin,’ Journal of Economic Literature , 46, 285– 332. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A., Vishny R. ( 1997), ‘ Legal determinants of external finance,’ The Journal of Finance , 52, 1131– 1150. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A., Vishny R. ( 1998), ‘ Law and finance,’ Journal of Political Economy , 107, 111ss3– 1155. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A., Vishny R. ( 2000a), ‘ Agency problem and dividend policies around the world,’ The Journal of Finance , 55, 1– 33. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A., Vishny R. ( 2000b), ‘ Investor protection and corporate governance,’ Journal of Financial Economics , 58, 3– 27. Google Scholar CrossRef Search ADS   La Porta R., Lopez-de-Silanes F., Shleifer A., Vishny R. ( 2002), ‘ Investor protection and corporate valuation,’ The Journal of Finance , 57, 1147– 1170. Google Scholar CrossRef Search ADS   Lazonick W. ( 2007), ‘ The US stock market and the governance of innovative enterprise,’ Industrial and Corporate Change , 16( 6), 983– 1035. Google Scholar CrossRef Search ADS   Lazonick W., O’Sullivan M. (eds) ( 2002), Corporate Governance and Sustainable Prosperity , Palgrave: New York, NY. Google Scholar CrossRef Search ADS   Lhuillery S. ( 2011), ‘ The impact of corporate governance practices on R&D efforts: a look at shareholders’ rights, cross-listing, and control pyramid,’ Industrial and Corporate Change , 20( 5), 1475– 1513. Google Scholar CrossRef Search ADS   Little R. J., Rubin D. B. (eds) ( 2014), Statistical Analysis with Missing Data . John Wiley & Sons: Hoboken, New Jersey. Mansfield E. (ed.) ( 1968), Industrial Research and Technological Innovation . W.W. Norton: New York, NY. Mitchell W., Roehl T., Slattery R. J. ( 1995), ‘ Influences on R&D growth among Japanese pharmaceutical firms, 1975–1990,’ The Journal of High Technology Management Research , 6, 17– 31. Google Scholar CrossRef Search ADS   Manso G. ( 2011), ‘ Motivating innovation,’ The Journal of Finance , 66( 5), 1823– 1860. Google Scholar CrossRef Search ADS   O’Connor M., Byrne J. ( 2015), ‘ Governance and the corporate lifecycle,’ International Journal of Managerial Finance , 11( 1), 23– 43. Google Scholar CrossRef Search ADS   O’Connor M., Rafferty M. ( 2012), ‘ Corporate governance and innovation,’ Journal of Financial and Quantitative Analysis , 47( 02), 397– 413. Google Scholar CrossRef Search ADS   Pavitt K. ( 1985), ‘ Patent statistics as indicators of innovative activities: possibilities and problems,’ Scientometrics , 7, 77– 99. Google Scholar CrossRef Search ADS   Roodman D. ( 2009), A Note on the Theme of Too Many Instruments. Oxfort Bulletin of Economics and Statistics , 71, 135– 158. Google Scholar CrossRef Search ADS   Rubin D. B. ( 1976), ‘ Inference and missing data,’ Biometrika , 63( 3), 581– 592. Google Scholar CrossRef Search ADS   Saravia J. ( 2014), ‘ The lifecycle of the firm, corporate governance and investment performance,’ Corporate Ownership and Control , 11( 2), 212– 226. Google Scholar CrossRef Search ADS   Shleifer A., Vishny R. ( 1997), ‘ A survey of corporate governance,’ Journal of Finance , 52, 737– 783. Google Scholar CrossRef Search ADS   Stein J. C. ( 1988), ‘ Takeover threats and managerial myopia,’ The Journal of Political Economy , 96, 61– 80. Google Scholar CrossRef Search ADS   Stoneman P., Toivanen O. ( 2001), ‘ The impact of revised recommended accounting practices on R&D reporting by UK firms,’ International Journal of the Economics of Business , 8, 123– 136. Google Scholar CrossRef Search ADS   Windmeijer F. ( 2005), ‘ A Finite Sample Correction for the Variance of Linear Efficient Two-Step GMM Estimators,,’ Journal of Econometrics , 126, 25– 51. Google Scholar CrossRef Search ADS   Appendix Table A1. Firms’ country distribution before and after the matching Original ISS RiskMetrics data set   Data set after the matching with BvD Orbis   Country  Frequency  %  Country  Frequency  %  Australia  158  4.69  Australia  135  6.13  Austria  38  1.13  Austria  24  1.09  Belgium  37  1.1  Belgium  30  1.36  Bermuda  35  1.04  Bermuda  50  2.27  Canada  296  8.79  Canada  48  2.18  Cayman Islands  13  0.39  Cayman Islands  17  0.77  Denmark  43  1.28  Denmark  26  1.18  Finland  60  1.78  Finland  31  1.41  France  144  4.27  France  86  3.90  Gabon  1  0.03  Gabon  1  0.05  Germany  155  4.6  Germany  98  4.45  Greece  84  2.49  Greece  46  2.09  Guernsey  1  0.03  Guernsey  0  0.00  Hong Kong  123  3.65  Hong Kong  67  3.04  Ireland  30  0.89  Ireland  12  0.54  Israel  2  0.06  Israel  2  0.09  Italy  122  3.62  Italy  79  3.59  Japan  754  22.38  Japan  630  28.60  Jersey  1  0.03  Jersey  0  0.00  Liberia  2  0.06  Liberia  1  0.05  Luxembourg  5  0.15  Luxembourg  3  0.14  Marshall Island  2  0.06  Marshall Island  1  0.05  Netherlands  99  2.94  Netherlands  48  2.18  Netherlands Antilles  3  0.09  Netherlands Antilles  1  0.05  New Zealand  24  0.71  New Zealand  21  0.95  Norway  50  1.48  Norway  20  0.91  Panama  1  0.03  Panama  1  0.05  Portugal  22  0.65  Portugal  14  0.64  Singapore  90  2.67  Singapore  56  2.54  South Korea  14  0.42  South Korea  14  0.64  Spain  86  2.55  Spain  55  2.50  Sweden  94  2.79  Sweden  40  1.82  Switzerland  116  3.44  Switzerland  62  2.81  United Kingdom  664  19.71  United Kingdom  484  21.97  Total  3369  100  Total  2203  100  Original ISS RiskMetrics data set   Data set after the matching with BvD Orbis   Country  Frequency  %  Country  Frequency  %  Australia  158  4.69  Australia  135  6.13  Austria  38  1.13  Austria  24  1.09  Belgium  37  1.1  Belgium  30  1.36  Bermuda  35  1.04  Bermuda  50  2.27  Canada  296  8.79  Canada  48  2.18  Cayman Islands  13  0.39  Cayman Islands  17  0.77  Denmark  43  1.28  Denmark  26  1.18  Finland  60  1.78  Finland  31  1.41  France  144  4.27  France  86  3.90  Gabon  1  0.03  Gabon  1  0.05  Germany  155  4.6  Germany  98  4.45  Greece  84  2.49  Greece  46  2.09  Guernsey  1  0.03  Guernsey  0  0.00  Hong Kong  123  3.65  Hong Kong  67  3.04  Ireland  30  0.89  Ireland  12  0.54  Israel  2  0.06  Israel  2  0.09  Italy  122  3.62  Italy  79  3.59  Japan  754  22.38  Japan  630  28.60  Jersey  1  0.03  Jersey  0  0.00  Liberia  2  0.06  Liberia  1  0.05  Luxembourg  5  0.15  Luxembourg  3  0.14  Marshall Island  2  0.06  Marshall Island  1  0.05  Netherlands  99  2.94  Netherlands  48  2.18  Netherlands Antilles  3  0.09  Netherlands Antilles  1  0.05  New Zealand  24  0.71  New Zealand  21  0.95  Norway  50  1.48  Norway  20  0.91  Panama  1  0.03  Panama  1  0.05  Portugal  22  0.65  Portugal  14  0.64  Singapore  90  2.67  Singapore  56  2.54  South Korea  14  0.42  South Korea  14  0.64  Spain  86  2.55  Spain  55  2.50  Sweden  94  2.79  Sweden  40  1.82  Switzerland  116  3.44  Switzerland  62  2.81  United Kingdom  664  19.71  United Kingdom  484  21.97  Total  3369  100  Total  2203  100  View Large Table A2. Descriptive statistics by age categories Age category/Variable  Mean (standard)  Minimum  Median  Maximum  Young:           CGQ  0.57 (0.26)  0  0.58  0.99   ΔCGQ  0.13 (1.02)  −1  0  3.14   Age  9.23 (4.59)  0  9  17   ln(SZ)  13.69 (4.59)  1.39  13.87  19.13   ln(RDI)  1.26 (2.33)  0  0.51  12.39   Δln(RDI)  −0.04 (0.95)  −5.98  0  6.77   Patents  1.12 (3.59)  0  0  43   CF  0.91 (2.56)  −5.23  0.10  26.97  Medium aged:           CGQ  0.50 (0.27)  0  0.51  1   ΔCGQ  0.31 (4.19)  −0.98  −0.01  3.01   Age  29.52 (8.20)  18  29  44   ln(SZ)  13.77 (1.97)  4.22  13.90  19.39   ln(RDI)  0.90 (1.07)  0  0.49  7.27   Δln(RDI)  0.01 (0.76)  −4.81  0  4.45   Patents  1.74 (4.51)  0  0  44   CF  0.57 (1.70)  −4.06  0.11  32.06  Old:           CGQ  0.41 (0.25)  0  0.36  1   ΔCGQ  0.09 (1.34)  −0.98  −0.04  2.16   Age  61.15 (10.33)  45  59  81   ln(SZ)  14.60 (1.70)  2.77  14.71  18.83   ln(RDI)  0.89 (0.88)  0  0.64  4.96   Δln(RDI)  0 (0.54)  −3.97  0  4.09   Patents  3.28 (5.70)  0  0  45   CF  0.73 (2.14)  −7.30  0.16  27.99  Very old:           CGQ  0.44 (0.25)  0  0.40  0.99   ΔCGQ  0.15 (2.50)  −1  −0.03  1.78   Age  119.29 (45.63)  82  108  536   ln(SZ)  14.96 (1.66)  5.16  14.97  19.94   ln(RDI)  0.95 (0.84)  0  0.85  4.97   Δln(RDI)  0 (0.53)  −3.71  0  3.78   Patents  2.96 (5.47)  0  0  47  Age category/Variable  Mean (standard)  Minimum  Median  Maximum  Young:           CGQ  0.57 (0.26)  0  0.58  0.99   ΔCGQ  0.13 (1.02)  −1  0  3.14   Age  9.23 (4.59)  0  9  17   ln(SZ)  13.69 (4.59)  1.39  13.87  19.13   ln(RDI)  1.26 (2.33)  0  0.51  12.39   Δln(RDI)  −0.04 (0.95)  −5.98  0  6.77   Patents  1.12 (3.59)  0  0  43   CF  0.91 (2.56)  −5.23  0.10  26.97  Medium aged:           CGQ  0.50 (0.27)  0  0.51  1   ΔCGQ  0.31 (4.19)  −0.98  −0.01  3.01   Age  29.52 (8.20)  18  29  44   ln(SZ)  13.77 (1.97)  4.22  13.90  19.39   ln(RDI)  0.90 (1.07)  0  0.49  7.27   Δln(RDI)  0.01 (0.76)  −4.81  0  4.45   Patents  1.74 (4.51)  0  0  44   CF  0.57 (1.70)  −4.06  0.11  32.06  Old:           CGQ  0.41 (0.25)  0  0.36  1   ΔCGQ  0.09 (1.34)  −0.98  −0.04  2.16   Age  61.15 (10.33)  45  59  81   ln(SZ)  14.60 (1.70)  2.77  14.71  18.83   ln(RDI)  0.89 (0.88)  0  0.64  4.96   Δln(RDI)  0 (0.54)  −3.97  0  4.09   Patents  3.28 (5.70)  0  0  45   CF  0.73 (2.14)  −7.30  0.16  27.99  Very old:           CGQ  0.44 (0.25)  0  0.40  0.99   ΔCGQ  0.15 (2.50)  −1  −0.03  1.78   Age  119.29 (45.63)  82  108  536   ln(SZ)  14.96 (1.66)  5.16  14.97  19.94   ln(RDI)  0.95 (0.84)  0  0.85  4.97   Δln(RDI)  0 (0.53)  −3.71  0  3.78   Patents  2.96 (5.47)  0  0  47  View Large © The Author 2017. Published by Oxford University Press on behalf of Associazione ICC. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
TI - Corporate governance and innovation: does firm age matter?
JF - Industrial and Corporate Change
DO - 10.1093/icc/dtx031
DA - 2018-04-01
UR - https://www.deepdyve.com/lp/oxford-university-press/corporate-governance-and-innovation-does-firm-age-matter-dXneYv1R51
SP - 349
EP - 370
VL - 27
IS - 2
DP - DeepDyve
ER -