Managerial Risk-Taking Incentive and Firm Innovation: Evidence from FAS 123R *

Transcription

1 Managerial Risk-Taking Incentive and Firm Innovation: Evidence from FAS 123R * Connie Mao Temple University Chi Zhang Temple University This version: December, 2015 * Connie X. Mao, Department of Finance, Fox School of Business and Management, Temple University, Philadelphia, PA Tel: (215) ; Fax: (215) ; cmao@temple.edu. Chi Zhang, Department of Finance, Fox School of Business and Management, Temple University, Philadelphia, PA 19122; chi.zhang@temple.edu. We would like to thank seminar participants at Temple University for their helpful comments and discussions. All errors are solely ours.

2 Managerial Risk-Taking Incentive and Firm Innovation: Evidence from FAS 123R Abstract We investigate how incentives derived from CEO compensation affect firm innovation. Our baseline results show that CEOs wealth-risk sensitivity (vega) is positively related to both innovation input and output quantity and quality. To establish causality, we exploit compensation changes instigated by the FAS 123R accounting regulation that mandate stock option expensing, where the FAS 123R creates a shock in managerial vega that is exogenous to firms investment opportunities. Our identification tests indicate a positive and causal effect of CEOs wealth-risk sensitivity (vega) on innovation activities. Furthermore, we find that dampened managerial risk-taking incentive after the implementation of FAS 123R is associated with significant reduction in the fraction of innovation projects related to firms core business. It implies that managers decrease innovation focus (or increase diversification) in their innovation portfolios to curtail business risk when their risk-taking incentive is reduced. JEL Classification: G30, G32, G34, D8, O31 Keywords: Executive compensation, risk-taking incentive, innovation, wealth-risk sensitivity, FAS 123R 1

3 1. Introduction Innovation is a key driver of firms future competitiveness and productivity. The innovation process is, however, risky, long, and idiosyncratic (Holmstrom, 1989). Unlike routine tasks, such as mass production and marketing, innovative projects involve the exploration of new and untested methods that have a high probability of failure. As a result, innovation entails commitment of a firm s resources and managerial effort (Aghion and Tirole, 1994; Manso, 2011). Managers are in general more risk averse than shareholders due to their undiversified wealth and human capital, concern for their reputations, and private benefits of control; hence, they will take less risky projects and sometimes even forgo risky projects with positive NPV (Amihud and Lev, 1981; Hirshleifer and Thakor, 1992; John, Litov, and Yeung, 2008). Compensation contracting is one of the mechanisms to align managers interests with those of shareholders. Smith and Stulz (1985) argue that, to incentivize risk-averse managers to take on risky yet positive NPV projects, firms financial decisions and managerial compensation contracts should ensure managerial wealth to be a convex function of firm value. As such, stock options have been widely used in corporate employee compensation contracts (Hall and Murphy, 2003). Stock options add convexity to managers payoffs and hence increase the sensitivity of managerial wealth to firm risk (vega). Consistent with this proposition, prior literature has documented that high vega leads to riskier investment and financing policies; e.g., greater leverage ratio, more R&D investments, less diversifications, and greater underfunding and risky asset allocation in pension policy (Guay, 1999; Rajgopal and Shevlin, 2002; Coles, Daniel, and Naveen, 2006; Chava and Purnanandam, 2010; Anantharaman and Lee, 2014). Nevertheless, Hayes, Lemmon, and Qiu (2012) find that the decline in the usage of stock option following the adoption of FAS 123R does not result in less risky investment and financial policies. 2

4 In this paper, we are interested in exploring how the risk-taking incentive that is derived from the compensation of top executives affects firm innovation. Given the particular risky, long-term, and idiosyncratic nature of firm innovation, prior literature has developed models of optimal executive compensation contracts that motivate managers to invest in the exploration of new ideas that are highly risky. Manso (2011) shows that motivating innovation demands tolerance for early failure and reward for long-term success. Theoretical work by Laux (2015) demonstrates that the optimal pay package consists of stock options (to encourage the discovery of innovative ideas) and either restricted stock (to combat excessive risk taking) or severance pay (to combat excessive conservatism). Consistent with Manso (2011), Ederer and Manso (2013) document that, in a controlled laboratory setting, the combination of tolerance for early failure and reward for long-term success is effective in motivating innovation. They also find evidence that the threat of termination can weaken incentives for innovation, while golden parachutes can mitigate the adverse effects. Lerner and Wulf (2007) find that more long-term incentives (such as stock options and restricted stock) for executives who are responsible for firms R&D operation are associated with larger patent counts and more citations per patent. In addition, Francis, Hasan, and Sharma (2011) find a significant positive relationship between CEO s wealth sensitivity to volatility (vega) and innovation. In contrast, there is no relationship between pay-for-performance sensitivity and firm innovation. Prior empirical work has been focused on the association between CEO compensation and innovation rather than establishing a causal relationship. The challenge in establishing causality is because CEO compensation is endogenous. Unobservable firm heterogeneity, correlated with both CEO compensation and firm innovation, could bias our results (i.e., the omitted variable 3

5 concern). In this paper, we take advantage of the accounting regulation FAS 123R in 2005 to identify any causal effect of managerial risk-taking incentive in compensation on firm innovation. Prior to the implementation of FAS 123R, firms were allowed to expense stock options at their intrinsic value. The implementation of FAS 123R in 2005 eliminated firms ability to expense options at intrinsic value and required them to expense all stock-based compensation at its fair value. Such a change removes the accounting advantages associated with option grants relative to restricted stock grants; therefore, firms respond to it by significantly cutting the stock options they grant to employees (Hayes, Lemmon, and Qiu, 2012). 1 However such a change in accounting treatment of options will hardly affect firms investment opportunities and hence corporate investment policy. As a result, the implementation of FAS 123R imposes an exogenous shock to managerial risk-taking incentive via its effect on stock option grants, hence providing an ideal setting that allows us to analyze the causal effect of CEO compensation on firm innovation. In addition to the use of R&D expenditures as a measure of innovation activities, we obtain information from the National Bureau of Economic Research (NBER) Patent Citation database and use the number of patents granted to a firm and the number of future citations received by each patent to measure innovation output. Specifically, the former captures the quantity of innovation and the latter proxies the quality of innovation. Our use of patenting to capture firms innovation output has become standard in the innovation literature (e.g., Acharya et al., 2013; Aghion et al., 2013; Nanda and Rhodes-Kropf, 2013). Using a sample of 15,741 firm-year observations during , we first confirm from our baseline ordinary least squares (OLS) results that there is a significant positive relationship between CEOs risk-taking incentive (vega) and innovation output number of patents and 1 Hayes, Lemmon, and Qiu (2012) document that the fraction of stock options out of total compensation decreased by about 17% on average after the FAS 123R became effective in

6 citations per patent, as well as innovation input R&D investments. Second, as with Hayes, Lemmon, and Qiu (2012), we find a significant drop in new option grants and managerial risk-taking incentive (vega) after the adoption of FAS 123R. Furthermore, reduction in risk-taking incentive (vega) around the FAS 123R implementation is associated with significant declines in the number of patents and citations per patent. Third, we employ a difference-in-differences (DiD) approach that compares the change in innovation output of firms that are more affected by FAS 123R (i.e., treatment firms) to those of firms that are less affected by the accounting regulation (i.e., control firms). Following Hayes, Lemmon, and Qiu (2012), we sort our sample into two groups based on the perceived accounting costs of option expensing. We proxy the perceived accounting costs of option expensing as the average value of the pro forma option expenses (deflated by fully diluted shares used to calculate earnings per share) the company reported during the pre-fas 123R period of This variable measures the amount by which earnings per share would be reduced if the firm had to recognize compensation expenses based on the fair value of its options. We define firms with pro forma option expenses above the sample median in the pre-fas 123R period as having high accounting impact from FAS 123R, while the remainder have low accounting impact from FAS 123R. We conjecture that if the change in CEOs risk-taking incentive (vega), as a result of the FAS 123R, affects firm innovation, then treatment firms with high accounting impact should respond more to the accounting regulation (i.e., should have a larger reduction in innovation) than control firms with low accounting impact. We document a significantly greater reduction in patent quantity and quality after the adoption of FAS 123R in the treatment group than in the control group. The results are robust with an alternative measure of treatment versus control 5

7 firms. These results lend strong support to the causal effect of managerial risk-taking incentive in compensation contract on firm innovation. The effect of the FAS 123R regulation on innovation output patents and citations could result from two distinct channels. One is the managerial risk-taking incentive channel, by which investments in risky innovative projects are dampened due to reductions in option-based compensation as a result of the accounting regulation. Consequently, this leads to reductions in innovation output. An alternative channel is the capital budgeting channel, by which managers reduced R&D investments after FAS 123R simply because of capital budgeting constraints resulted from the higher costs of option expensing under FAS 123R. Different from our findings with innovation output, we find at best a statistically weak association between the change in R&D investments and the change in risk-taking incentive (vega) around the adoption of FAS 123R. In the DiD analysis, we observe little significant difference in post-fas 123R changes in R&D investments between the treatment and control firms. These results are inconsistent with the capital budgeting channel. Instead, our findings demonstrate that the FAS 123R regulation has a materially adverse effect on managerial risk-taking incentive, through which firm innovation is curtailed. Though managers did not cut R&D investments after the regulation, they might invest in projects that are less risky and also less innovative, resulting in lower level of innovation output. Patenting activity captures a firm s innovation better than R&D because patenting is an innovation output variable, which encompasses the successful usage of all (both observable and unobservable) innovation inputs. In contrast, R&D expenditures only capture one particular observable quantitative input (Aghion, Van Reenen, and Zingales, 2013) and are sensitive to 6

8 accounting norms, such as whether they should be capitalized or expensed (Acharya and Subramanian, 2009). Finally, we are able to observe the scope of a firm s innovation, which allows us to explore the change in the focus of the innovation projects after FAS 123R implementation. The NBER database provides detailed classifications of each patent s technology class, which can be mapped to standard industry classifications. Hence, we can construct proxies that capture a firm s innovation scope and compare it against its core business. We find a significantly greater reduction in the percentage of patents related to firms core business after the adoption of FAS 123R in the treatment group than in the control group, which implies that dampened risk-taking incentive in managerial compensation after FAS 123R implementation leads firms to invest in more diversified (or less focused) innovation projects. While reducing the focus of the innovation projects by working on projects unrelated to firms core business could curtail investment risk due to the diversification effect, it substantially curbs overall innovation output since unrelated innovation projects could be out of the scope of managers and firms innovation expertise. Our paper contributes to the literature in two ways. First, while previous studies mainly investigate the association between CEO compensation and corporate innovation, we take advantage of the accounting regulation FAS 123R to explore the causal effect of managerial risk-taking incentive derived from their compensation packages on firm innovation. Second, our study also contributes to the literature on the causal effect of stock option usage on firms investment and financing policies. Existing literature has focused on firms risk policies through the lens of R&D investments, capital expenditures, leverage ratio, cash holding etc (Coles, Daniel, and Naveen, 2006; Chava and Purnanandam, 2010; Hayes, Lemmon, and Qiu, 2012). In 7

9 contrast, our paper emphasizes the quantity, quality, and scope of firm innovation using patent data. Examining patent information as a form of investment policy in innovative projects is particularly appealing since it allows us to explore the effect of managerial risk-taking incentive on not only the quantity but also the quality of innovation. This unique feature makes firm innovation an outcome variable that is superior to R&D investments, capital expenditures, and leverage, because one cannot easily judge the change in the quality of these outcome variables. Furthermore, we are able to observe the scope of a firm s innovation based on the technology classification code of each patent, hence capable of exploring the effect of managerial risk incentive on the focus of innovation projects. Such detailed measures of the quality of firm innovation help us gain insights on the impact of managerial risk-taking incentive on firms investment policy in technology innovation. The rest of the paper is organized as following: Section 2 describes data and summary statistics, Section 3 presents the empirical results, and Section 4 concludes the paper. 2. Sample Selection, Variable Construction, and Summary Statistics In this section, we describe the sample selection process and how each variable is constructed. We also summarize our sample statistics. 2.1 Sample Selection We use several databases to construct our sample. We start with the ExecuComp database and focus on non-financial and non-utility firms by excluding firms with SIC codes between 6000 and 6999 and between 4900 and We collect data on salary, bonus, grants of stock options, grants of restricted stock, and long-term incentive awards for the CEOs of each firm. We also obtain the same data on top executives from these firms who are responsible for research & 8

10 development or innovations whenever the data are available. We obtain managerial incentive data (delta and vega) from Professor Naveen s website ( Following Gu, Mao, and Tian (2014), we obtain the patent and citation data from the latest version of the NBER Patent Citation database for We next supplement the information for patents granted from 2007 to 2010 that is provided by Kogan et al. (2012), available at Patent citations over from 2007 to 2010 are constructed using the Harvard Business School (HBS) patent and inventor database available at To construct control variables, we obtain firm accounting information from the Compustat database and institutional holdings information from the Thomson Reuters Institutional Holdings (form 13F) database. We end up with a sample of 15,741 firm-year observations during the sample period of 1992 to 2008 for our baseline analysis. We end our sample period in 2008 because, though we have patent and citation data through 2010, there is an average of two to three years lag between patent application year and grant year; thus, we can only quantify the total number of patents applied for (and eventually granted) through For the analyses that examine the impact of the adoption of FAS 123R in 2005, we restrict the sample period to 2002 through 2008, which evenly encompasses the pre- and post-fas 123R eras. Since FAS 123R became effective for large public firms for the first reporting period beginning June 15, 2005, we exclude fiscal year 2005 to avoid cloudy information in this transitory year. Applying these rules leads to a sample of 6,552 firm-year observations during the sample period of 2002 to Variable Construction 9

11 2.2.1 CEO Compensation and Managerial Incentives Since there are dramatic changes in the reporting requirements for executive compensation after December 15, 2006, we follow Hayes, Lemmon and Qiu (2012) closely in the calculation of bonus, restricted stocks, stock options, and long-term incentive awards. Prior to 2006, bonus, current and previously granted restricted stocks, and stock options are obtained directly from the Annual Compensation table in ExecuComp. Under the new requirements, firms continue to present annual components of compensation, such as salary and bonus, in a summary compensation table, but the details of option and equity awards now appear in two additional tables, the Plan-Based Award and Outstanding Equity Award tables. Also, some bonuses have been reclassified as non-equity incentive compensation. There are two types of managerial incentives: delta and vega. According to Coles, Daniel, and Naveen (2013) and the prior literature (e.g., Guay, 1999), delta is defined as the dollar change in an executive s wealth for a 1% change in stock price, and vega is defined as the dollar change in executive wealth for a 1% change in volatility. Therefore, vega captures CEOs risk-taking incentive in their compensation contract Innovation Output We measure patenting activities as the proxies for firm innovation. Our first measure is the total number of patent applications filed in a given year (and eventually granted), which is a measure of innovation quantity. As Hall, Jaffe, and Trajtenberg (2001) point out, there are two or three years lag between patent application year and grant year. We thus choose the application year instead of grant year because the actual timing of the patented innovation is closer to the application year. Our second measure is the number of non-self-citations each patent receives in 10

12 subsequent years, which is a measure of innovation quality. To mitigate the truncation problems associated with patent counts and citation counts, we follow Hall, Jaffe, and Trajtenberg (2001, 2005) and Fang, Tian, and Tice (2014) and compute adjusted patent counts by dividing raw patent counts with the sum of application-grant lag distribution. We also compute adjusted citation counts by dividing raw citation counts with the fraction of predicted lifetime citations. We merge the patent data with Compustat. Following the innovation literature, we set the patent counts to zero for firm-year observations that are not matched to the patent database, because our patent sample covers the entire universe of firms that have filed patents with the USPTO. The distribution of patent grants in our final sample is right skewed, with its median at zero. We winsorize these variables at the 99 th percentile and then use a natural logarithm of one plus patent counts (Ln(1+Pat)) and a natural logarithm of one plus number of citations per patent (Ln(1+Cite)) as the main measures of innovation output in our analysis. In addition, we use RD/Assets (R&D expenses scaled by total assets) as a measure of innovation input Control Variables In the regressions explaining executive compensation, we follow Hayes, Lemmon, and Qiu (2012) and control for firm size (a natural logarithm of total assets), CEO tenure, cash compensation (the sum of salary and bonus), and firm-fixed effects. In the models explaining innovation activities, we follow Hirshleifer, Low, and Teoh (2012) and control for a set of firm characteristics that might affect a firm s future innovation. In the baseline regressions, the control variables include Ln(Sales), a natural logarithm of total sales; Ln(PPE/EMP), a natural logarithm of net plant, property, and equipment divided by number of employees; Tobin s Q, which is the ratio of market value of assets (book value of assets minus book value of equity plus 11

13 market value of equity) to book value of total assets; Ln(1+Tenure), a natural logarithm of one plus CEO tenure; Institutional Holding, which is calculated as the arithmetic mean of the four quarterly institutional holdings reported in 13F; the Herfindahl Index (HHI) and its squared term. HHI is computed based on annual sales in each four-digit SIC industry code. We describe variable definitions in details in Appendix A. 2.3 Summary Statistics Table 1 presents the summary statistics of CEO compensation and firm characteristics. Our final sample consists of 6,552 firm-year observations during (excluding 2005). In Panel A of Table 1, we find that that the average annual CEO s compensation is about $6.7 million. Among different types of compensation, the mean values of CEOs stock options are the second largest in both dollar terms and in percentage terms. Mean values of patent counts and citations per patent are 24.1 and 2.8, respectively; however, median values are both zero. This is because about 52.8% of observations have zero patents. To adjust for the right-skewed distributions of patents and citations, we utilize the natural logarithm of one plus patent counts and one plus citations per patent in our regression analyses. In Panel B of Table 1, we compare firm characteristics and CEO compensations for the pre- and post-fas 123R period. The adoption of FAS 123R has a noticeable impact on CEO compensation. While total CEO compensation increases greatly during the post-fas 123R period, we observe a significant drop in stock option grants. The average option grants in CEO compensation decrease from $2.2 million to $1.4 million, and the percentage of options out of total compensation decrease from 36.6% to 25.8%. We also find significant changes in managerial incentives and innovation output after the adoption of FAS 123R. Average CEOs 12

14 delta decreases from 1,002 to762 and average CEO vega decreases from 175 to 128. In terms of innovation output, both patent counts and citations per patent decrease largely in the post-fas 123R period. For instance, the average number of patent counts decreased from 34.1 to 15.1 after FAS 123R was adopted. 3. Empirical Results The summary statistics above show large changes in CEO compensation, managerial incentives, and innovation input and output after the adoption of FAS 123R. In this section, we will first examine the relationship between managerial incentive and innovation activities in an Ordinary Least Square (OLS) regression framework, then investigate the effect of FAS 123R on managerial incentives and firm innovation in a multiple regression framework. 3.1 Baseline Ordinary Least Square (OLS) Regressions We start with baseline ordinary least squares (OLS) regressions using a sample of 15,741 firm-year observations from the ExecuComp database during 1992 to Summary statistics of the sample is reported in Panel C of Table 1. In particular, we estimate the following regression model, and the results are reported in Panel A of Table 2: Ln(1+Pat) or Ln(1+Cite) = α+ β*ln(1+vega) + Controls + Year Fixed Effects+ Industry Fixed Effects + ϵ. (1) Because innovative projects take time to develop, we examine the relationship between CEOs risk-taking incentive (vega) derived from their compensation contracts in year t and firm innovation in the next three years. Specifically, the dependent variables Ln(1+Pat) are a natural logarithm of one plus the total number of patents filed (and eventually granted) in years t+1, t+2, and t+3, and results are reported in columns (1) (3), respectively. The dependent variables 13

15 Ln(1+Cite) are a natural logarithm of one plus the total number of citations received per patent in years t+1, t+2, and t+3, and the results are reported in columns (4) (6), respectively. Control variables are as defined in Appendix A. We include year fixed effects and industry fixed effects based on the two-digit SIC code. Our primary interest is the relationship between CEOs risk-taking incentive and innovation activities. We find that the coefficient estimates on managerial risk-taking incentive (Ln(1+Vega)) are positive and statistically significant at the 1% level in all columns, suggesting that larger vega is associated with greater patent counts and citations per patent in future years. The results appear economically significant as well. For instance, as it moves from the 25 th to the 75 th percentile, Ln(1+Vega) increases by 2.110, which is associated with a 12.0% (2.110*0.057) increase in patent counts for the subsequent year. The signs of coefficient estimates on the control variables are consistent with conventional wisdom. As with Hirshleifer et al. (2012), we find that the coefficient estimates on Ln(Sales) and Ln(PPE/EMP) are positive, while the coefficient estimates on HHI are negative. In addition to innovation output patents and citations, we investigate the relationship between the risk-taking incentive and innovation input R&D investments. We estimate the following regression model: RD/Assets =α+β*ln(1+vega)+ Controls+ Year Fixed Effects +Industry Fixed Effects+ϵ. (2) Regression results are reported in Panel B of Table 2. Following the existing literature (e.g., Brown et al., 2013), we set missing values in R&D to zeros. The dependent variables are RD/Assets in years t+1, t+2, and t+3 in columns (1) (3) respectively, where RD/Assets is R&D expenses divided by total assets. Similar to the results with Ln(1+Pat) and Ln(1+Cite) as dependent variables, the coefficient estimates on Ln(1+Vega) are all positive and statistically 14

16 significant at the 1% level. Overall, we find that managerial risk-taking incentive is positively related to innovation input, innovation quantity, and quality. Though it is a common practice in the prior literature to set missing values in R&D to zeros, a missing value in R&D expenditures recorded in the Compustat database does not necessarily indicate a lack of innovation activities. Koh and Reeb (2015) have documented that 10.5% of firms with missing values in R&D expenditures generate innovation output patents, and the frequency is 14 times greater than firms that report zero R&D expenses. 2 Their findings imply that some firms that report missing values in R&D have non-negligible innovation activities. To mitigate potential reporting bias in R&D expenses, we follow Koh and Reeb (2015) and estimate a Heckman selection model. In the selection model, the dependent variable is a binary variable that equals one if a firm reports non-missing values in R&D expenses in a sample year, and zero otherwise. We include Former AA Clients, Pseudo-Blank R&D Firms, PPE divided by Number of Employees, Lagged Tobin s Q, Firm Age, Leverage, and Proportion of Firms Reporting R&D in the Industry to estimate the probability that a firm reports a non-missing value in R&D expenses. Detailed definitions of these variables are in Appendix A. The inverse Mills ratio from the selection model is included in the OLS regressions explaining RD/Assets in years t+1, t+2, and t+3 in columns (4) (6) of Panel B of Table 2 respectively, in a sample of observations with non-missing R&D. Similar to the results of OLS in columns (1) (3), the coefficient estimates on Ln(1+Vega) are all positive and statistically significant at the 1% level in years t+1 and t+2. The results confirm that managerial risk-taking incentive is positively related to innovation input. 2 Moreover, compared to firms with positive level of R&D expenses, the pseudo-blank R&D firms (firms that have missing R&Ds but non-missing patent activities) tend to obtain patents with broader contributions, greater citation breadth, and lengthier competitor discovery periods despite having fewer patents. 15

17 3.2 The Effect of FAS 123R on CEO Compensation and Managerial Incentives The association we find in Table 2 does not imply any causality. Both the risk-taking incentive and innovation activities are endogenously determined by firm characteristics, so the positive association is likely driven by unobservable firm characteristics. We thus employ the adoption of the accounting regulation FAS 123R to address the issue of causality. Before 2005, firms can choose expensing their option usage at fair value or at intrinsic value. The accounting regulation FAS 123R was adopted in 2005, which requires firms to expense stock options at fair values. Thus, FAS 123R eliminates the accounting advantage associated with stock options, which in turn reduces firms desires to grant stock options in CEO compensation contracts. As a result, FAS 123R created a shock to managerial incentives that is exogenous to firms investment opportunities and, therefore, innovation outcomes. Therefore, FAS 123R offers an ideal context to test how managerial risk-taking incentive affects firm innovation. For this purpose, we first want to confirm that FAS 123R is a valid shock to CEO compensation and managerial incentives. In Table 3, we examine the changes in each individual compensation component as well as CEOs managerial incentives (delta and vega) before and after FAS 123R was adopted. Using a sample of ExecuComp firms during (excluding 2005), we estimate the following regression model: Executive Compensation = α+ β*post-123r + Controls + Firm Fixed Effects + ϵ, (3) where dependent variables are Total Compensation in column (1), the percentage of each component of total compensation (P_Salary, P_Bonus, P_Option, P_Stock, and P_Long-term Incentive) in columns (2) (6) respectively, and CEOs delta and vega in columns (7) and (8) respectively. Post-123R is a dummy variable that equals one for observations in the post-fas 123R period ( ) and that equals zero for observations in the pre-fas 123R period 16

18 ( ). Following Hayes, Lemmon, and Qiu (2012), we include several control variables: Size is a natural logarithm of total assets; Ln(1+Tenure) is a natural logarithm of one plus the number of days between a given fiscal year-end and the day an executive became CEO; and Cash Compensation is the sum of salary and bonus. We also include firm fixed effects in the regressions. To a large extent, our results in columns (1) (6) confirm the findings in Hayes, Lemmon, and Qiu (2012). The adoption of FAS 123R is associated with a significant reduction of options usage in CEOs compensation by 13.6%. While both percentages of salary and bonus decrease after FAS 123R was implemented, there is a significant increase in grants of restricted stocks and long-term incentive awards, which in turn increases the amount of total compensation. This indicates that firms substitute away from options toward other types of pay-for-performance based compensation; e.g., restricted stocks and long-term incentive awards. In columns (7) and (8), we investigate how the adoption of FAS 123R alters CEOs wealth-performance and wealth-risk sensitivity (delta and vega). The coefficient estimates on the Post-123R dummy are negative and statistically significant in both regressions, suggesting a significant drop in CEO delta and vega due to FAS 123R. Such a decrease in vega by is equivalent to a 9.3% (16.26/175) decrease from the mean value of vega in the pre-fas 123R period. Similarly, a decrease in delta by is equivalent to a 14.2% (142.4/1,002) decrease from the mean value of delta in the pre-fas 123R period. In brief, the adoption of FAS 123R leads to a significant decrease in options grants to CEOs, which in turn diminishes CEOs wealth-risk sensitivity (vega) and dampens their risk-taking incentive. 17

19 3.3 Change in Managerial Risk-Taking Incentive and Change in Firm Innovation around FAS 123R In this section, we examine how reduction in vega around FAS 123R affects firm innovation. Unlike the simultaneous equations framework or an instrumental variable approach to address endogeneity issues, our identification strategy allows us to exploit the cross-sectional relation between change in innovation and change in risk-taking incentive, while controlling for other potential determinants of firm policies that might have changed over time. According to Himmelberg, Hubbard, and Palia (1999), when the unobservable characteristics of the contracting environment are largely time-invariant, those time-invariant unobserved variables can be differenced out by running regressions in the following form: Y it = b 0 + b 1 X it + b 3 Z it + ε it. (4) However, as pointed out by Zhou (2001), the regression could lack power if the variation in our independent variable (X) comes primarily from the control variables. Fortunately, by using FAS 123R, which has been shown to be a valid shock to managerial incentives in Table 3, we effectively introduce an exogenous source of variation to X. Therefore, running regressions of change in innovation proxies on the change in managerial risk-taking incentive could generate direct evidence on the causality of the risk-taking incentive on innovation. For this purpose, we estimate the following regression: ΔInnovation= α+ β*δincentive+ ΔControls + Industry Fixed Effects+ ϵ. (5) Following Hayes, Lemmon, and Qiu (2012), we compute changes in the dependent and independent variables for each firm by subtracting the mean value during the pre-fas 123R period ( ) from the mean value during the post-fas 123R period ( ). As such, our analysis only includes firms with at least one observation in both the pre- and post-fas 18

20 123R periods so that changes in mean values could be computed. We thus have a total of 950 firm observations. We next estimate regressions of change in innovation on change in managerial incentive and report the results in Table 4. The dependent variables are ΔLn(1+Pat) in columns (1) and (2), ΔLn(1+Cite) in columns (3) and (4), and ΔRD/Assets in columns (5) (8). In all specifications, we include changes in control variables as well. Control variables are as defined in Appendix A. We also include industry fixed effects based on two-digit SIC code. As shown in columns (1) and (3), ΔLn(1+Vega) are positively related to changes in the number of patents and citations per patent, and the results are statistically significant at the 5% level. As we include changes in managerial pay-performance sensitivity ΔLn(1+Delta) as an additional variable in columns (2) and (4) the coefficient estimates on ΔLn(1+Vega) remain positive and significant. In sum, we find strong evidence that reduction in managerial risk-taking incentive (vega) following FAS 123R leads to significant reductions in innovation output, measured by both patent counts and citations per patent. 3.4 Difference-in-Differences (DiD) Approach To further explore how managerial risk-taking incentive affects innovation, we employ an alternative approach difference-in-differences tests. The difference-in-differences approach has some advantages. First, it rules out omitted trends that correlated with risk-taking incentive and firm innovation in both the treatment and control groups. Second, it controls for constant unobserved differences between the treatment and the control group. Following Hayes, Lemmon, and Qiu (2012), we sort our sample into two groups based on the perceived accounting costs of option expensing. We proxy the perceived accounting costs of option expensing as the average value of the pro forma option expenses (deflated by fully diluted 19

21 shares used to calculate earnings per share) the company reported during the pre-fas 123R period of This variable measures the amount by which earnings per share would be reduced if the firm had to recognize compensation expenses based on the fair value of its options and captures the idea that expensing options has a greater impact on the earnings of firms with greater pro forma option expenses. We define firms with pro forma option expenses above the sample median in the pre-fas 123R period as having high accounting impact from FAS 123R and the rest as having low accounting impact from FAS 123R. We conjecture that, if the change in CEOs risk-taking incentive (vega) as a result of the FAS 123R affects firm innovation, then treatment firms with high accounting impact should respond more to the accounting regulation (i.e., should have a larger reduction in innovation activities) than control firms with low accounting impact. We estimate the following difference-in-differences regression model: Innovation = α+ β1*post-123r*highaccimpact + β2*post-123r + β3*highaccimpact+ Controls + Industry Fixed Effects + ϵ. (6) Results are reported in Panel A of Table 5. The dependent variables Ln(1+Pat) are a natural logarithm of one plus the total number of patents filed (and eventually granted) in years t+1, t+2, and t+3, and the results are reported in columns (1) (3), respectively. The dependent variables Ln(1+Cite) are a natural logarithm of one plus the total number of citations received per patent in years t+1, t+2, and t+3, and the results are reported in columns (4) (6). Post-123R is a dummy variable that equals one for observations in the post-fas 123R period ( ) and that equals zero for observations in the pre-fas 123R period ( ). HighAccImpact is a dummy variable that equals one if the firm is highly impacted by the implementation of FAS 123R regulation and zero otherwise. We define highly impacted firms as those with pro forma 20

22 option expenses in the pre-fas 123R period above the sample median. Control variables are as defined in Appendix A. We include industry fixed effects based on two-digit SIC codes. Table 3 shows that the adoption of FAS 123R significantly curtails executive option grants, which in turn reduces managerial risk-taking incentive (vega). Consequently, vega will be reduced to a larger extent in firms that are more affected by FAS 123R than in firms that are less affected by the accounting rule change. Hence, if reduction in vega leads to reduction in innovation, we expect a greater decrease of innovation activities in the HighAccImpact group than in the LowAccImpact group. The difference between the treatment and control group is captured by the coefficient estimate on the interaction term Post-123R*HighAccImpact in Table 5. We expect the coefficient β1 to be significantly negative. As shown in Panel A of Table 5, we find significant and negative coefficient estimates on Post-123R*HighAccImpact in all regressions, suggesting that firms that are more impacted by accounting regulation FAS 123R experience significantly greater reduction in both innovation quantity and quality than those that are less impacted by the regulation. The results lend support to a causal effect of managerial risk-taking incentive on firm innovation. 3.5 Does FAS 123R Affect Innovation Input R&D Investments? The effect of the FAS 123R regulation on innovation activities could result from two distinct channels. One is the managerial risk-taking incentive channel, by which investments in risky innovative projects are dampened due to reductions in option-based compensation as a result of the accounting regulation. Consequently, this leads to reductions in innovation output. An alternative channel is the capital budgeting channel, by which managers reduce R&D investments after FAS 123R simply because of capital budgeting constraints that result from the 21

23 higher costs of option expensing under FAS 123R. To differentiate the two channels, we investigate the effect of FAS 123R on one of the main innovation input R&D investments. If capital budgeting is the channel by which FAS 123R affects innovation quantity and quality, we expect to find significant reductions in R&D investments after the adoption of FAS 123R. As shown in columns (5) and (6) of Table 4, neither ΔLn(1+Vega) nor ΔLn(1+Delta) is significantly associated with ΔRD/Assets in the OLS regressions. In columns (7) (8), we estimate the Heckman models to account for potential bias in reporting missing values in R&D and find that both ΔLn(1+Vega) and ΔLn(1+Delta) are positively related to ΔRD/Assets; however, the results are only marginally significant at the 10% level. Overall, we find very weak evidence that change in vega around FAS 123R causes any significant change in innovation input. In Panel B of Table 5, we run the DiD test using RD/Assets in years t+1, t+2, and t+3 as the dependent variables in columns (1) (3). The coefficient estimates on the dummy, Post-123R, are positive and statistically significant, suggesting that control firms experience a significant increase in their R&D investments following the adoption of FAS 123R. The coefficient estimates on the interaction terms Post-123R*HighAccImpact are all negative but are only statistically significant in year t+3. As we rerun the DiD test using the Heckman model framework, we find that none of the coefficient estimates on Post-123R*HighAccImpact is statistically significant. The overall results indicate that firms that are more impacted by the accounting regulation experience changes in R&D investments that are not different from those that are less affected by the regulation. The findings above are inconsistent with the capital budgeting channel. Instead, our findings demonstrate that the FAS 123R regulation has a materially adverse effect on managerial 22

24 risk-taking incentive, through which firm innovation is curtailed. Although managers did not cut R&D investments after the regulation, they might nevertheless have chosen to invest in projects that are less risky and also less innovative, resulting in a lower innovation output level. We believe that patenting activity captures a firm s innovation better than R&D because patenting is an innovation output variable that encompasses the successful usage of all (both observable and unobservable) innovation input. In a subsequent section, we will explore how managerial risk-taking incentive affects the scope of firm innovation. 3.6 Robustness Tests In this section, we conduct a series of additional tests to determine the robustness of our findings, including using an alternative definition of treatment group, estimating regressions with different models, and exploring subsample analysis Alternative Definition of Treatment Group First, we examine our findings using an alternative definition of treatment group in the DiD analysis. Bakke et al. (2015) point out that FAS 123R would not affect the firms that did not grant options in employee compensations before the implementation of FAS 123R. Therefore, we define a dummy variable, Withoption, that equals one for firms with option expenses in 2003 or 2004, and zero otherwise. The firms without option expenses in either year will be classified as the control group. The DiD results are reported in Appendix Table B1. In Panel A, where we examine the effect on innovation output quantity and quality, we find that the coefficient estimates on the interaction term Post-123R*Withoption are negative and statistically significant at the 1% level in five out of the six models. In Panel B, where the dependent 23

25 variables are R&D investments in subsequent years, we find no significant difference in the effect of the FAS 123R implementation on innovation input between the treatment and control firms. These results confirm our findings as reported in Table Poisson and Negative Binomial Models Patent and citation data are non-negative integer values, and a large proportion of the observations are zero. In addition to taking logarithms, alternative estimation methods deal with such data, including the Poisson model and the Negative Binomial model (NB2 model). In Appendix Table B2, we estimate the following model using a Poisson regression: # Patent or # Citation = α + β1*post-123r*highaccimpact + β2*post-123r + β3*highaccimpact+ Controls + Industry Fixed Effects+ ϵ. (7) The dependent variable # Patent is the total number of patents filed (and eventually granted). The dependent variable # Citation is the total number of citations received per patent. As shown in Appendix Table B2, the coefficient estimates on interaction terms are negative and significant at the 1% level in five out of the six models, confirming the robustness of our earlier findings. We also estimate the DiD test using Negative Binomial regression and the results are reported in Appendix Table B3. The coefficient estimates on interaction terms are all negative and statistically significant for all models, except column (6) Does the Financial Crisis Drive the Results? We next investigate whether our results are driven by the financial crisis, during which firms experienced significant reductions in corporate investments. It is likely that, because of the financial crisis, which began in 2007, firms curtailed executive option grants and 24

26 innovation activities because of shrinking investment opportunities and declining corporate profitability. As such, the positive correlation between managerial incentive and innovation activities could be driven by the financial crisis. To address this concern, we limit our sample to the pre-crisis period from 2003 to 2007 (excluding 2005) and re-run our difference-in-differences regressions. The results are reported in Appendix Table B4. Though the sample becomes smaller after restricting it to the pre-crisis period, our results remain robust. As with our findings in Panel A of Table 5, treatment firms experience a significantly larger drop in both patent counts and citations per patent than control firms after the FAS 123R implementation. Thus, our finding is unlikely driven by the financial crisis The Effect of Performance-Vesting (p-v) Equity Grants According to Bettis et al. (2015), about 34% U.S. companies granted performance-vesting (p-v) equity awards to top executives during In contrast to the traditional time-based vesting, p-v awards specify a vesting schedule where the number of shares being vested is conditional on the achievement of one or more performance condition, including accounting performance, stock price, or other metrics (market share, sales growth, or customer satisfaction). Hence p-v equity awards create additional convexity in executive compensations with respect to stock performance. Bettis et al. (2015) show that when p-v grants depend on stock price, in comparison to non-p-v grants, vega from restricted stock with p-v provisions become non-zero and vega from options with p-v provisions increase by 31%. In contrast, when p-v grants depend on accounting measures, in comparison to non-p-v grants, vega derived from both options and p-v grants is 18% smaller than vega derived from options alone. Bettis et al. (2015) attributes this to the fact that 25

27 p-v grant schedule provides managers an incentive to smooth accounting performance. One concern of our study is that we might mis-measure the true risk-taking incentives without taking into account the p-v equity awards in executive compensation. In particular, if firms increase p-v awards of restricted stocks after the adoption of FAS 123R, we might overstate the decline in vega after the accounting regulation since we did not take into account the vega imbedded in equity awards with p-v provisions. Nevertheless Bettis et al. (2015) document a significant reduction in total vega after the implementation of FAS 123R by counting both options and p-v grants, suggesting a decline in the managerial risk-taking incentive. To address the concern above, we investigate the robustness of our results by focusing on a subsample of firms that do not have any p-v provisions in a sample year during For such firms, vega derived from option grants is an accurate measure of CEOs risk-taking incentive. We use the Incentive Lab database to identify the firms that do not have any p-v provisions during our sample period. Incentive Lab collects detailed information on all short-term and long-term equity-based awards and cash awards from proxy statements for the largest 750 firms, measured by stock market capitalization every year starting in In each year, we identify firms without any p-v provisions and then merge them with our innovation data, managerial incentive variables from ExecuComp, and control variables from Compustat. Our final sample contains 943 firm-year observations during the period (excluding 2005). We estimate the same Difference-in-Difference tests as those in Table 5 and the results are reported in the Appendix Table B5. In Panel A of Appendix Table B5, despite the much smaller sample, the coefficient estimates on interaction term on Post-123R*HighAccImpact are negative and statistically 3 Given the complex nature of the p-v equity awards, we are not capable of computing vega for those awards. 26