Risk-taking Incentives and Returns on R&D Investment

Transcription

1 Risk-taking Incentives and Returns on R&D Investment Bruce K. Billings Florida State University James R. Moon, Jr. Georgia State University Richard M. Morton Florida State University Dana M. Wallace University of Central Florida ABSTRACT: Traditional finance theory suggests that riskier investments should yield higher returns. We assess whether risk-taking incentives associated with executive stock options ( vega ), presumed to increase managers appetite for risk, uniformly yield higher returns to R&D investment. Challenging this claim, anecdotal and empirical evidence suggests that highly-incented managers may take on excessive risk, leading to greater losses, while other theoretical research argues that high levels of option-based compensation may actually encourage risk aversion, resulting in lower profits. Consistent with both scenarios, our results suggest that the return on R&D, as measured by future earnings, patent awards, and market valuation, varies concavely with vega. That is, low to moderate levels of vega are associated with increasing returns on R&D, consistent with vega inducing more profitable investments, but these returns decline as vega increases past an inflection point. Supplemental analysis documents a similar concavity between vega and both the level of R&D and future R&D-driven stock volatility, indicating that the lower R&D returns to higher vega are more likely driven by risk aversion than by excessive risk taking. Overall, our results imply risk-taking incentives (i.e., vega) have their limits. Keywords: Executive compensation; Managerial incentives; Risk-taking; Research and Development

2 1. Introduction This study investigates whether managerial compensation intended to motivate greater risk taking corresponds to greater investment value. Prior research demonstrates a positive relation between risk-taking incentives ( vega ) derived from employee stock options (ESOs) and R&D spending (Coles, Daniel, and Naveen 2006), consistent with ESOs aligning the interests of risk-averse managers with more diversified and risk-seeking shareholders. Interestingly, prior studies have largely ignored whether this option-induced investment corresponds to preferred economic outcomes. We contend that this is the more important issue, whether greater ESOs correspond to greater returns on investment. If ESOs truly encourage value-enhancing investment, we should observe a greater return on that investment (i.e., profitability, innovation, firm value), consistent with the classic risk-return relation. However, recent research and anecdotal evidence suggest that higher levels of ESOs contribute to either excessive risk taking (PWC 2009; Shen and Zhang 2013; Bhagat and Bolton 2014), or, alternatively, risk aversion (Carpenter 2000; Meulbroek 2001; Ross 2004). Either scenario implies some incentive alignment problem and brings into question the presumed link between greater risk-taking incentives and greater returns. 1 We address this issue by examining whether the relations between R&D and future profitability, innovation, and stock returns weaken as vega increases. ESO incentives may lead to excessive risk taking through a combination of overconfidence (Roll 1986; Gervais, Heaton, and Odean 2011), optimism bias (Schrand and 1 The linear relation between vega and R&D documented in Coles et al. (2006) assumes that vega has a uniform effect on risk taking (and under the classical risk-return relation, on returns to investment). That is, vega merely has a multiplier effect on risk-taking behavior (and returns) of the average firm. Thus, in this study, each firm s average return on R&D represents our benchmark for assessing whether greater risk-taking incentives lead to comparably greater returns to shareholders. 1

3 Zechman 2012), and biased risk estimates (Fahlenbrach and Stulz 2011), among other potential contributors. Specifically, overconfidence can lead a manager to select a highly convex payoff contract (Gervais et al. 2011) and engage in more aggressive behavior (Englmaier 2006). In this context, managers more likely underestimate expected risk or overestimate expected returns (Roll 1986). Consistent with this view, many attribute the banking crisis to excessive levels of ESOs, which motivated executives to engage in excessively risky behavior (PWC 2009; Bhagat and Bolton 2014). 2 More broadly, Dong, Wang, and Xie (2010) conclude that greater levels of option compensation can lead managers to make overly risky financing decisions, resulting in a sub-optimal capital structure (i.e., overlevered), and Shen and Zhang (2013) suggest that transient R&D spending by high-vega managers tends to be overly risky (i.e., inefficient) as compared to that by low-vega managers. In contrast, analytical research provides several reasons why the convex payoff structure associated with options may not uniformly provide incentives for greater risk taking. Ross (2004) describes how wealth effects created by ESOs and other compensation may alter the manager s perspective on risk taking (via a shift in the manager s utility function) such that additional risktaking incentives actually heighten risk aversion. Carpenter (2000) argues that magnifying the manager s exposure to total firm risk (higher vega) can lead the manager to moderate that risk. Similarly, Meulbroek (2001) implies that higher vega causes managers to attach lower values to ESOs, which may also lead managers to reduce, rather than increase, risk. More generally, studies like Hayes, Lemmon, and Qiu (2012), who examine ESO awards pre- and post-sfas 123R, and Dittmann and Maug (2007), who analyze ESOs and other 2 In a 2009 PWC survey, 73 percent of senior participants from the financial services industry identify culture and excessive risk-taking as a primary contributor to the banking crisis (p.11). 2

4 compensation in the context of efficient contracting models, challenge the widely held belief that incentive alignment is the driving force for compensating managers with ESOs. 3 Collectively, these studies raise questions about uniformity in the incentive alignment effects of ESOs. 4 Whether due to excessive risk taking or greater risk aversion, the preponderance of evidence indicates that some level of misalignment is possible. To the extent that incentive misalignment exists with respect to investment policy, the discussion above suggests it will be more apparent for managers whose compensation includes greater levels of risk-taking incentives (i.e., vega). In contrast to prior studies that focus on the level of investment in risky R&D (e.g., Coles et al. 2006), we examine the return generated by the company s R&D investment portfolio. We choose to examine return on R&D because it focuses on observable outcomes directly related to risky investment choices. It represents the ultimate measure of investment success or failure from a shareholder s perspective. Specifically, we examine the relations between vega and future earnings attributable to R&D, patent awards attributable to R&D, and the market valuation of R&D investment. We first model each outcome as a linear function of vega. If higher vega corresponds to riskier and more profitable investment, on average, we expect to observe positive relations between vega and our measures of R&D returns. We then allow each relation to vary 3 Hayes et al. (2012) suggests that firms favored ESO compensation over other forms prior to passage of SFAS 123R to reduce expenses and report higher income, consistent with other factors besides incentive alignment driving the heavy use of ESO compensation. Dittmann and Maug (2007) observe that option compensation is rarely predicted as a component of efficient contracting. 4 Further contributing to this concern is survey evidence suggesting that boards of directors and compensation committees may not be giving enough attention to risk oversight and management. For example, 45% of respondents to an October 2008 survey of over 500 senior managers at financial institutions indicated that managers and boards lack risk expertise (KPMG 2008). When asked whether risk management activities are an explicit component in determining management compensation, 56% of firms responding to a survey commissioned by the AICPA selected not at all or minimally (Beasley, Branson, and Hancock 2015). 3

5 non-linearly by modeling the outcome as a quadratic function of vega. 5 That is, we assess whether the expectation of higher returns to R&D continues to be realized uniformly as a manager s risk taking incentives increase. Our results suggest that vega has important implications for all three dimensions of R&D profitability. Using future earnings to measure return, we surprisingly fail to find a positive relation between vega and R&D investment returns in a strictly linear model, indicating that the expected link between risk-taking incentives and the profitability of R&D investment is, at best, non-monotonic. In the quadratic specification, we find a positive (negative) coefficient on the linear (squared) vega-r&d interaction term. This suggests that the profitability of R&D increases with low-to-moderate levels of vega but then declines as vega continues to increase. We find a similar pattern of results using patent awards as an alternative measure of return on R&D. In the linear specification, we again fail to find evidence that vega affects the number of patents generated per dollar of R&D. However, in the quadratic model, we find that the association between patents and R&D is a concave function of vega. Finally, we corroborate these results by showing that investors expectations of future cash flows related to R&D, evidenced by market returns, similarly exhibit a concave relation. In sum, we fail to observe that returns on R&D are uniform across the distribution of vega. Rather, vega has a significantly positive relation with R&D returns in the low-to-moderate vega range, but that relation 5 We model the relation as a quadratic function of vega because it allows for, but does not impose, a concave relation (a second derivative less than zero). Further, if the function is concave, a quadratic specification facilitates the determination of an inflection point (a point at which the first derivative switches from positive to negative). This specification is most appropriate in testing whether there is a reversal of risk taking in R&D for higher levels of vega. Further, prior research examining non-linear effects of stock options also utilizes a quadratic function (Hanlon, Rajgopal, and Shevlin, 2003). 4

6 diminishes significantly and turns negative for higher values of vega. Thus, our results point to unresolved incentive alignment problems related to managers engaging in risky investment. This concave relation between vega and R&D return is consistent with either of the perverse incentive scenarios described earlier. That is, greater vega may induce excessive risk taking without a commensurate increase in profitability or greater vega may induce greater risk aversion, leading managers to reduce risk taking behavior resulting in lower returns. To disentangle these two possibilities, we conduct two additional analyses. We first examine the relation between vega and future R&D spending, allowing for non-linearity in vega. Under the assumption that greater R&D spending represents riskier investment (Coles et al. 2006), a linear or convex (concave) relation between vega and future R&D levels would be consistent with an increasing (diminishing) appetite for risk. Consistent with Coles et al. (2006), our results reveal a significantly positive, linear association between vega and the level of R&D investment. However, after introducing a vega-squared term to our regression to allow for non-linearity, we find a concave association between vega and R&D spending, consistent with risk aversion. That is, managers with relatively higher levels of vega invest in relatively less R&D. Second, we use total firm risk, measured using future stock return volatility, to assess the effect of vega on the relation between R&D and risk. We again observe a concave relation, consistent with less risk taking once vega reaches an inflection point. In sum, higher vega is associated with lower returns on R&D investments, likely the result of greater risk aversion rather than excessive risk taking. We perform three additional analyses to bolster our conclusions. First, if the concave relations we observe truly represent negative outcomes associated with high vega, we expect these findings to be more pronounced when investor oversight, proxied by institutional ownership, is relatively weaker since research suggests these investors curb investment-related 5

7 risk aversion attributed to equity-based incentives (Panousi and Papanikolaou 2012). We generally find that is the case, as we fail to observe significant concave relations between vega and R&D returns in firms with relatively greater institutional ownership. Second, to substantiate the notion that investment risk is a driving factor for our results, we replace R&D with less-risky capital expenditures in our regressions (Coles et al. 2006). We fail to observe the same pattern of results with this alternative investment measure, consistent with R&D capturing more risksensitive investments. Third, we address the potential for endogeneity to affect our results. After controlling for various endogenous relations, we continue to observe concavity. Overall, our results are consistent with analytical research suggesting that higher vega incentives may not uniformly encourage greater risk taking, bringing into doubt whether higher vega incentives uniformly achieve incentive alignment (Lambert et al. 1991; Carpenter 2000; Meulbroek 2001; Ross 2004). Our study makes several important contributions to the executive compensation and R&D literatures. Most importantly, we are, to our knowledge, the first to provide empirical evidence that high levels of risk-increasing incentives associated with stock options correspond to less risky investment. Our findings relate to Hayes et al. (2012), who suggest that the decline in stock option awards in the wake of SFAS 123R had no effect on risky investment. While their results suggest that some stock option awards are unrelated to incentive alignment, our evidence highlights a potential unintended consequence of ESOs; that is, higher vega incentives associated with these awards may actually contribute to greater risk aversion and lower returns. We also extend Hanlon et al. (2003), who document a positive and non-linear relation between future firm profitability and the fair value of stock option grants. Their study investigates whether 6

8 option awards more likely contribute to incentive alignment or rent extraction. 6 Alternatively, we assess whether the incentive structure of stock options (i.e., vega) affects a specific manager behavior (i.e., R&D investment) that we expect will directly contribute to future returns on R&D. Thus, in contrast to Hanlon et al. (2003), we focus on the effectiveness of the incentive alignment mechanism with respect to investment choice. While they find no evidence that option awards contribute to rent extraction, we conclude that option awards do not uniformly align incentives. We also contribute to research examining the return on and the pricing of R&D. In general, prior research finds that R&D investment, on average, relates to increased firm value (Lev and Sougiannis 1996; Chambers, Jennings, and Thompson 2002; Ciftci and Cready 2011; Curtis, McVay, and Toynbee 2015). Our evidence suggests that the value created by R&D varies largely with managers risk-taking incentives. The market response to R&D increases with vega for firms with low-to-moderate levels of vega but decreases with vega for firms with higher levels of vega, suggesting that high levels of vega observed in practice are not as beneficial in creating value for investors. Finally, we contribute to research on the interaction between influential investors and compensation contract design and effectiveness (Mehran 1992; David, Kochhar, and Levitas 1998; Panousi and Papanikolaou 2012; Cadman and Sunder 2014). Observing that the non-linear effect of vega on R&D return is concentrated in firms with less oversight (i.e., lower institutional ownership) suggests that increased oversight and activism associated with institutional investors 6 Hanlon et al. (2003) view stock options as an investment by the firm and assess whether that investment is associated with future earnings after controlling for other factors that affect performance, including R&D expenditures. They suggest that greater stock option values imply higher quality managers, who enhance future performance. 7

9 acts to facilitate incentive alignment related to R&D investment, particularly when vega incentives might otherwise be less effective. In the next section of the paper, we discuss the related literature and develop our primary hypotheses. We present our research design and results for our primary analyses in Section 3. Section 4 investigates alternative explanations for our primary results. We discuss extended and sensitivity analyses in Section 5, and we conclude in Section Hypothesis development Evidence favoring ESO compensation as a means to align incentives Theory suggests that the convex payoff from ESO-based compensation reduces agency conflicts by better aligning managers interests with shareholders (Jensen and Meckling 1976; Haugen and Senbet 1981; Smith and Stulz 1985; Lambert 1986). Specifically, while share-based compensation has the potential to motivate managers to behave more like investors, managers with large amounts of wealth invested in their organization may exhibit a lower appetite for risk than the firm s investors due to under-diversification. ESO-based compensation overcomes this risk-aversion tendency because managers benefit from stock price gains but are not directly penalized for losses. Despite this theoretical justification, the widespread use of ESOs to compensate executives has been criticized as a form of hidden compensation rather than incentive alignment (Bebchuk and Fried 2003). That is, managers exercise influence over their compensation packages to extract rents. In an effort to test these contrasting views of stock option compensation, Hanlon et al. (2003) investigate the impact of option awards on future profitability. They argue that the fair value of stock option grants proxy for a firm s investment 8

10 in a higher quality manager. If options align incentives, the cost of investing in the manager should be positively related to realized benefits, as evidenced by future earnings. However, if options instead enable managers to engage in rent extraction, then the correlation between ESO values and future earnings will be less positive, zero, or negative. Hanlon et al. s (2003) initial results indicate a negative relation between ESO values and future earnings. However, after introducing a second order term, the squared value of ESO grants, they observe a positive coefficient for the first order term and a negative coefficient for the second order term. They conclude that option grants have a positive effect on future earnings, albeit at a decreasing rate, consistent with ESOs aligning incentives rather than creating opportunities for rent extraction. Relatedly, an extensive body of research examines how specific incentives derived from ESOs, such as the sensitivity of an executive s portfolio to the firm s stock price (i.e., delta) and stock price volatility (i.e., vega), influence managers behavior and corporate policy (see Murphy 1999). For example, Coles et al. (2006) find that higher vega leads to greater R&D spending and lower investment in property, plant, and equipment. Rajgopal and Shevlin (2002) find that oil and gas firms with higher vega engage in riskier exploration activities. Chava and Purnanandam (2010) and Coles et al. (2006) link higher vega to higher leverage. 7 Rego and Wilson (2012) report that vega is associated with more aggressive tax avoidance strategies. Each of these studies supports the premise that the ESO-incentivized manager takes on greater risk, presumably reflecting incentive alignment with shareholders. In another related study, Panousi and Papanikolaou (2012) document that as idiosyncratic risk increases, capital expenditures 7 Chava and Purnanandam (2010) also find that CEO vega incentives are associated with lower cash balances, but Liu and Mauer (2011) find the opposite relation. To our knowledge, the reason for this discrepancy has not been investigated. One possibility is that Liu and Mauer (2011) scale vega by total compensation whereas most studies do not employ this scalar. 9

11 decline, and higher levels of managerial firm ownership exacerbate this relation. Option compensation, however, mitigates the negative risk-investment relation, suggesting ESOs appropriately serve to align shareholder and manager incentives when firm-specific risk is high. 8 Evidence challenging ESO compensation as a means to align incentives The notion that higher levels of ESO compensation are primarily motivated by the desire to achieve incentive alignment is not without challenge. As noted above, some criticize ESOs as a form of hidden compensation rather than a tool to align incentives (Bebchuk and Fried 2003). Supporting this view, stock option plans substantially increased as stock values grew in the 1990s, but declined over the 2000s as stock prices reversed. The decrease in ESO awards following changes in accounting treatment further challenges incentive-alignment as a primary motivation for option-based compensation. Evidence in Hayes et al. (2012) suggests that firms favored ESO compensation over other forms prior to passage of SFAS 123R to reduce expenses and report higher income. 9 They report a substantial decrease in ESOs awarded following implementation of SFAS 123R, and, importantly, this decline appears unrelated to risk-taking behavior. Thus, other factors besides incentive alignment appear to contribute to the use of ESO compensation. Dittmann and Maug (2007) analyze the optimal balance of executive stock, option, and base salary compensation using efficient contracting models. Their results rarely predict option compensation. They conclude that either currently employed contracting models are flawed or observed compensation practice suffers from significant defects. 8 Institutional ownership also mitigates the negative relation between idiosyncratic risk and firm investment. We explore the governance role of institutional investors in Section 5 of our study. 9 Prior to the passage of SFAS 123R, which requires firms to expense the fair value of stock options granted over the service period, firms largely recorded no (or minimal) compensation expense using the then-acceptable intrinsic value method. Under this method, if options were granted with an exercise price equal to market price (i.e., no intrinsic value), no compensation expense was recorded. 10

12 In addition to challenging the link between incentive-alignment and ESOs in general, whether higher levels of ESO compensation uniformly contribute to incentive alignment has also been questioned. Some evidence links ESOs to excessive risk taking, while alternative theories identify scenarios in which ESOs could contribute to risk aversion. First, a combination of overconfidence (Roll 1986; Gervais Heaton, and Odean 2011) and optimism bias (Schrand and Zechman 2012) may motivate managers to engage in overly risky decision-making. Overconfidence leads to more aggressive behavior (Englmaier 2006), which may in turn lead to less than value-maximizing decisions if managers underestimate expected risk or overestimate expected returns (Roll 1986; Fahlenbrach and Stulz 2011). As a consequence, managers may take on higher risk projects that fail to yield returns commensurate with the expected project risk and the manager s level of risk-taking incentives. 10 Given that Gervais et al. (2011) show that substantial overconfidence can lead a manager to select a highly convex payoff contract that exposes him or her to excessive risk, we suspect this combination of overconfidence and optimism bias is more pronounced in high-vega managers. In addition to the effects of overconfidence and optimism, several studies suggest that the asymmetric reward structure itself encourages excessive risk. For example, Bhagat and Bolton (2014) attribute the banking crisis to excessive levels of ESOs, which motivated executives to engage in overly risky behavior. They report that executives with high levels of stock incentives at banks considered too big to fail (p.316) experienced positive net payoffs 10 Some companies appear to recognize the potential for compensation incentives to lead to excessive risk taking. For example, in Prudential Financial Inc. s 2015 proxy statement, they indicate that [t]he Company s general risk management controls also serve to preclude decision-makers from taking excessive risk to earn the incentives provided under our compensation plans (p. 21; Whether companies more generally anticipate and oversee the potential for excessive risk taking and limit such behavior is an empirical question. 11

13 from insider trading and cash compensation relative to the net losses that investors experienced via share price declines leading up to and including the crisis period. They conclude that incentives generated by executive compensation programs are positively correlated with excessive risk-taking by banks (p. 335). In a different context, Dong, Wang, and Xie (2010) conclude that greater levels of option compensation can lead to managers making overly risky financing decisions, resulting in a sub-optimal capital structure (i.e., overlevered). Finally, Shen and Zhang (2013) find that, following large increases in R&D, firms with high levels of vega underperform relative to those firms with lower levels of vega, consistent with self-interested managers acting to increase the value of their compensation. 11 In contrast, some have argued that higher levels of ESOs may be counterproductive; that is, they may actually contribute to greater risk-aversion. As Ross (2004) notes, wealth effects from high levels of existing ESOs and other compensation may alter the manager s perspective on risk taking. By moving the manager to a different portion of his or her utility function, additional risk-taking incentives can actually yield more risk aversion. Meulbroek (2001) investigates the tension between incentive alignment and the manager s lack of diversification, noting that the manager is exposed to total firm risk while diversified investors are only exposed to systematic firm risk. She shows that managers value their options at less than fair market value. Thus, as vega increases, the gap between managers and investors expected returns widens. At some point, the managers expected return fails to 11 Note that Shen and Zhang (2013) focus on a small sample of approximately 900 firm-year observations exhibiting substantial increases in R&D spending in a given year. They report lower future profitability for their portfolio of high vega firms relative to low vega firms and, in a test of market efficiency, find higher future returns for low vega relative to high vega firms. Instead of focusing on these transient R&D expenditures, our analyses examine the relation between vega and the overall R&D investment profile. Most importantly, we allow for non-linearity in the relation between vega and the return on R&D. 12

14 compensate them for their risk exposure (which is greater than an investor s), leading to risk averse behavior. Similarly, Carpenter (2000) suggests that option compensation does not necessarily lead managers to be more risk seeking. Her model demonstrates that increasing the proportion of options in a manager s total portfolio value can increase the manager s exposure to the underlying assets risk. This exposure, in turn, incentivizes the manager to decrease the volatility of the underlying assets. In other words, increasing a manager s sensitivity to asset risk can make him or her seek less risk. 12 Measuring investment riskiness and predictions We focus our analysis of risky investment on R&D. Prior literature typically associates R&D investment with greater risk and future earnings variability (Chambers et al. 2002; Kothari, Laguerre, and Leone 2002; Ho, Xu, and Yap 2004; Ciftci and Cready 2011). More germane to our study, Coles et al. (2006) examine the relation between vega and the level of R&D investment under the assumption that higher R&D levels imply greater risk taking. Accordingly, they interpret their evidence of greater R&D spending as consistent with incentive alignment. However, the discussion in the prior section describes potential scenarios that could lead to some level of incentive misalignment. That is, managers with particularly high levels of risk-taking incentives may select excessively risky or very low-risk R&D projects relative to managers with more moderate levels of risk-taking incentives. To capture this possibility, we examine average returns to firms R&D investment portfolios. We emphasize that the intent of incentive alignment is not simply to encourage higher levels of R&D, but rather to realize comparably greater returns on R&D investment. In fact, we argue that higher return per dollar invested 12 As Meulbroek (2001, p. 7) points out, if stock-based compensation were purely designed to align incentives, there would be no natural stopping point, and managers compensation would be 100 percent equity-based. 13

15 represents a sounder measure of investment riskiness than the number of dollars invested in R&D projects. If vega incentivizes managers to invest in more risky R&D with higher expected returns, we should find a positive relation between vega and indicators of future return on R&D. 13 However, if at some point increasing levels of vega motivate managers to engage in projects that earn lower returns, either due to excessive risk taking or efforts to limit their own risk exposure, we expect a non-linear relation between the return on R&D and vega. Therefore, our initial analysis examines this relation to assess whether the return on R&D is a non-linear, concave function of vega. Our proxies for the return on, or value of, R&D include the extent to which R&D corresponds with greater future earnings, patent awards, and current period stock returns, which presumably capture expected future cash flows associated with R&D. While we expect each proxy to capture a common factor of return on R&D investment, each does so in a somewhat unique fashion. First, future earnings reflect realized profits from R&D investment, yielding arguably the most direct measure of R&D success or failure. Second, patent awards represent the ability of the company to protect future returns to R&D investment from a regression toward the mean that results from competition. Thus, we expect the frequency of patent awards to be adversely affected by either excessive risk taking, which contributes to greater failure rates, or greater risk aversion, which reduces innovation. Finally, Lev and Sougiannis (1996) suggest that 13 In contrast to the link between expected and realized equity returns, which is admittedly tenuous (Elton 1999), we expect realized returns on R&D to be a reasonable proxy for expected investment returns. Expected equity returns rely little on individual firm characteristics pertaining to investments. Rather, most models employ broad assessments of firm risk and long-run growth. While the process by which firms estimate expected returns for R&D investments is unobservable and varies among firms, these expectations almost certainly reflect specific estimates pertaining to expected future sales, expense reductions, asset growth, etc. 14

16 the market capitalizes R&D expenditures. Extending this notion, we posit that investors condition their pricing of R&D on the nature of the investment portfolio. 14 The degree to which investors value R&D expenditures reflects their expectations of future cash flows associated with those investments. Combining traditional theory suggesting ESOs align incentives by encouraging risk taking with other research suggesting excessive risk-taking incentives may lead to less profitable R&D investments, we predict that future earnings, patent awards, and R&D pricing are all concave functions of vega. We express these expectations with respect to the three aspects of return on R&D discussed above in the following hypotheses (stated in the alternative form): H1: The relation between future earnings and R&D investment is a concave function vega. H2: The relation between patent awards and R&D investment is a concave function of vega. H3: The relation between stock returns and R&D investment is a concave function of vega. 3. Research design and results related to primary analysis Sample, data, and variable measurement Our sample begins with estimates of vega and delta for each firm s CEO, which we obtain from Dr. Lalitha Naveen. 15 Vega (Delta) measures the sensitivity of CEO s equityholdings to a one-percent change in stock volatility (price). 16 These estimates are derived from 14 We do not suggest that investors know the riskiness and expected return of specific R&D investments. Rather, we contend that investors have sufficient information to price a firm s R&D investment, whether that arises from knowledge of managers incentives, knowledge of product lines in development, or some other source. 15 We graciously thank Dr. Naveen for providing vega and delta estimates and explanations on her website ( Details of these calculations can be found in Coles, Daniel, and Naveen (2013). 16 Detailed variable definitions can be found in Appendix A. 15

17 Execucomp, which covers the S&P 1500, and are available beginning in For patent awards, we use the patent data from Kogan, Papanikolaou, Seru, and Stoffman (2012). 17 We collect remaining variables from standard sources. Specifically, we obtain required annual financial statement information from the Compustat Annual Fundamentals file and returns data from the Center for Research in Security Prices (CRSP) monthly and daily files. Given our interest in returns on R&D investment, we restrict our sample to firms with positive, non-missing values for R&D (Compustat data item XRD), which further reduces the sample to 7,641 observations. 18 Because we use lagged values of vega throughout the analysis, our sample period begins in For our test of H1, we require three years of future earnings. We use earnings data through 2012, thus our sample period ends with R&D from For H2, we relate R&D to patent awards in the current and subsequent year. Patent data is available through 2010, so we similarly end our patent sample in For our test of H3, we use stock return data through Thus, our total sample varies between 7,641 and 9,981, depending on the test. Table 1 provides descriptive statistics for our sample. All continuous, unlogged variables are winsorized at the 1st and 99th percentiles. Consistent with most prior research (Chava and Purnanandam 2010; Dong et al. 2010), we use the natural logarithm of vega and delta in our empirical analyses as our measures of CEO wealth risk and price sensitivity (Vega and Delta, respectively) to control for extreme skewness in the untransformed distributions. The median of unlogged Vega (Delta) is (207.86). Thus, a one percent increase in implied volatility (stock 17 We graciously thank Dr. Noah Stoffman for making the patent data from Kogan et al. (2012) publicly available on his website ( 18 The one exception to this restriction is when we analyze the relation between R&D and vega, using R&D as the dependent variable. Dropping firms with zero reported R&D would yield a censored distribution, violating a requirement of OLS. 16

18 price) increases the median CEO s wealth by approximately $51,850 ($207,860). These compare favorably with Coles et al. (2006) who report median vega and delta estimates of $34,000 and $206,000, respectively, for an earlier time period. We scale R&D investment by lagged total assets (R&D) and multiply by 100. Mean (median) R&D is 7.07 (4.56) percent of assets. For our analysis related to the return on R&D investment, we calculate future return on assets (ROA) using industry-adjusted average earnings over the three years following the R&D expenditure. Specifically, we use earnings before interest, taxes, depreciation, and amortization (i.e., EBITDA) plus R&D and advertising expense, scaled by total assets in year t-1 (ROA3). We industry-adjust future earnings to increase the variability in future earnings and to reduce the likelihood that industry differences in R&D profitability explain our results. Further, we expect the impact of risk-taking incentives specific to individual CEOs will more likely correspond to abnormal earnings in future periods. The mean (median) ROA3 in Table 1 is (0.0213), suggesting a fairly symmetric distribution of abnormal future earnings. The use of Execucomp and additional data requirements result in a sample of fairly large firms. The mean (median) unlogged market value of equity (MVE) is 8.7 billion (1.4 billion). Table 2 reports Spearman (Pearson) correlations among select variables above (below) the diagonal for the positive R&D sample. Italics indicate statistically (and economically) insignificant correlations (p>0.05). As expected, the correlation between Vega and Delta is positive. Both constructs are also positively related to firm size. These two measures of executive compensation incentives typically move in tandem, highlighting the importance of controlling for Delta in evaluating the role of Vega. We observe relatively weak correlations between Vega and R&D; the Spearman (Pearson) correlation is (-0.034). We also observe that ROA3 is positively related to R&D, Vega, and Delta. 17

19 Test of H1 Our hypotheses tests examine alternative measures of return on R&D investment, with the expectation that the rate of return on R&D will be a concave function of vega. To test our first hypothesis using future earnings as a proxy for return, we estimate the following model: 2 RRRRRR3 ii,tt = α + RR&DD ii,tt ββ 1 + ββ 2 VVVVVVVV ii,tt 1 + ββ 3 VVVVVVVV ii,tt 1 2 ββ 6 MMMMMM ii,tt + ββ 7 VVVVVVVV ii,tt 1 + ββ 8 VVVVVVVV ii,tt ββ 4 DDDDDDDDDD ii,tt 1 + ββ 5 DDDDDDDDDD ii,tt ββ 9 DDDDDDDDDD ii,tt 1 + ββ 10 DDDDDDDDDD ii,tt 1 ββ 11 MMMMMM ii,tt + ββ 12 MMMMMM ii,tt + ββ 13 LLLLLL ii,tt + ββ 14 AAAAAA ii,tt + ββ 15 TTTTTTTT ii,tt + ββ 16 RRRRRR0 ii,tt + εε iiii (1) + The terms in parentheses capture the future earnings generated by a dollar of R&D conditional on the lagged incentive structure and firm size. We refer to this composite weighting as the return on R&D. 19 Our parameters of interest are β2 and β3. β2 captures how R&D s relation to future earnings varies with the level of Vega. If Vega leads to more risky investment with greater expected returns, we expect β 2 > H1 further predicts that the return on R&D is concave in Vega, meaning the second derivative of the return on R&D is negative. Thus, we expect β3 < For control purposes, we similarly include Delta and Delta 2, although we make no prediction of a non-linear relation with Delta. Prior research indicates that firm size can create economies of 19 Pakes and Schankerman (1984) suggest that R&D generally contributes to the firm s revenue stream up to two and a half years following the expenditure. We use a three-year window to fully capture this range. 20 Some studies suggest the potential for endogeneity between executive compensation and firm performance. Research finds little support for the notion that future pay-offs influence current compensation (Holthausen, Larcker, and Sloan 1995; Rajgopal and Shevlin 2002); nevertheless, we address the potential for this endogeneity in two ways. First, we include firm fixed effects, which addresses endogenous relations between compensation and performance that are persistent in nature (Larcker and Rusticus 2010). Second, since current performance likely correlates with future performance and with compensation contract design, we include current earnings (ROA0) as an independent variable in equation (1). Prior studies also suggest endogeneity between vega and R&D (Coles et al. 2006), which is a not a concern in this model since both variables enter as right-hand-side variables. We do address endogeneity when considering the relation between R&D and vega in Section Curtis et al. (2015) document a significant decline in the return on R&D over time. However, this decline is steepest from 1980 to 1994, and, more importantly, relatively flat over the latter years of their sample period, which corresponds to the years we study. 18

20 scale that allow larger firms to better exploit R&D opportunities and generate larger return on investment (Ciftci and Cready 2011). In addition, firm size is positively correlated with Vega and Delta. Accordingly, we include MVE as a determinant of the return on R&D in the model. Thus, we estimate the return on R&D as a function of Vega, Delta, and MVE. We also include the main effects of Vega and Delta to capture other ways these incentive proxies influence manager behavior and impact future earnings. Finally, we include several firm characteristics that likely relate to future earnings. Larger firms are also less likely to incur losses, so we include the main effect of MVE. Firms with a higher market-to-book ratio (MTB) have greater expected economic rents or growth opportunities, which likely generate future earnings. Greater leverage (Lev) implies fewer growth prospects and potentially some level of financial distress. Advertising (Adv) and tangible assets (Tang) represent other investment outlays that contribute to future earnings. We also control for current ROA (ROA0), using an earnings measure consistent with ROA3. We include firm and year fixed effects when estimating equation (1), which allows R&D, Vega, Delta, and our other control variables to explain the within-firm variation in future earnings. 22 We report results from estimating equation (1) in Table 3. Initial results reported in the first column, which exclude the Vega and Delta interaction terms from the model, confirm Ciftci and Cready s (2011) result that the return on R&D increases in firm size. Specifically, the coefficient on the interaction between R&D and MVE is significantly positive (β6 = 0.001, p < 0.05). As shown in the middle column, we fail to find that the rate of return on R&D increases 22 By including firm fixed effects, our coefficients capture within-firm variation. In other words, the average withinfirm relation between future profitability and R&D conditional on vega represents our benchmark for assessing whether the return on R&D lacks uniformity across the range of vega. 19

21 linearly with Vega. While the coefficient on Vega*R&D is positive (β2 = 0.013), it is statistically insignificant at conventional levels. Among the control variables, MTB, Adv, and ROA0 are positively related to future earnings, while Lev is negatively related, as expected. 23 Interestingly, we fail to observe a significant relation between R&D* Delta and ROA3. Results in the last column for the full model, including the squared terms, provide our test of H1. As predicted, the return on R&D varies non-linearly with Vega. We find that β2 is now significantly positive (β2 = 0.211, p < 0.05), and β3 is significantly negative (β3 = , p < 0.05). This suggests that the accounting return to R&D is a concave function of Vega. That is, Vega enhances the profitability of R&D, but as Vega increases to relatively high levels, the rate of return on R&D begins to decline. 24 Test of H2 Our tests of H2 involve a more direct outcome of R&D investment, the number of patents subsequently granted to the firm. This analysis builds on established research linking R&D success to patent awards (Hausman, Hall, and Griliches 1984; Hall, Griliches, and Hausman 1986; Griliches 1990). Prior research also suggests that the relation is nearly contemporaneous (i.e., one or two lags of R&D spending effectively explain patent awards), consistent with firms filing for patents early in the R&D process (Hall, Griliches, and Hausman 1986). Accordingly, 23 In untabulated results, omitting ROA0 from equation (1) has little effect on the relations between ROA3 and R&D*Vega (coefficient = 0.264; p-value < 0.05) and R&D*Vega 2 (coefficient = ; p-value < 0.05), suggesting that potential omitted variable bias regarding the relation between current profitability and vega is not severe. 24 As a robustness test, we re-estimate equation (1) with the inclusion of additional terms. First, we include MVE 2 and the interaction between MVE 2 and R&D since MVE 2 is arguably correlated with Vega 2 and Delta 2. Second, we include the fair value of stock option grants and fair value of stock option grants squared as control variables given evidence in Hanlon et al. (2003). Including these additional terms does not affect inferences drawn from reported results. 20

22 we regress Patent (current and year-ahead patent awards, scaled by total assets in year t-1, times 100) on current R&D conditional on vega to test H2 using the following model: PPPPPPPPPPPP ii,tt = α + RR&DD ii,tt (ββ 1 + ββ 2 VVVVVVVV ii,tt 1 + ββ 3 VVVVVVVV ii,tt 1 + ββ 4 DDDDDDDDDD ii,tt 1 + ββ 5 DDDDDDDDDD ii,tt ββ 6 MMMMMM ii,tt ) + ββ 7 VVVVVVVV ii,tt 1 + ββ 8 VVVVVVVV ii,tt 1 + ββ 9 DDDDDDDDDD ii,tt 1 + ββ 10 DDDDDDDDDD ii,tt 1 + ββ 11 MMMMMM ii,tt + ββ 12 MMMMMM ii,tt + ββ 13 LLLLLL ii,tt + ββ 14 AAAAAA ii,tt + ββ 15 TTTTTTTT ii,tt + ββ 16 RRRRRR0 ii,tt + εε iiii (2) As reported in Table 4 (middle column), we fail to find a significant relation between Patent and R&D conditional on Vega in the linear specification (β2 = 0.787). Rather, we find that the relation between patent issuance and lagged R&D is a concave function of Vega. 26 Specifically, in the full model, β2 is significantly positive (β2 = 4.516, p < 0.05), and β3 is significantly negative (β3 = , p < 0.05). Consistent with our previous results, these findings suggest that low-to-moderate vega yields more successful R&D investments (as measured by subsequent patent issuance) on a per unit basis than does higher vega. 27 Test of H3 Our tests of H3 involve investor pricing of R&D investment conditional on vega with the expectation that investors will perceive relatively lower future cash flows to R&D induced by higher vega, consistent with our tests of future earnings and patents. We estimate a model similar 25 Consistent with R&D, we scale patents by assets at t-1 so that we model the number of patents per dollar of R&D expense. Using the raw count of patent awards (or natural log of patent awards) would express patent issuance as a function of the R&D to asset ratio rather than dollars invested. 26 We also estimate Equation (2) using patents issued in year t+1 and the sum of years t+1 and t+2 and find qualitatively similar results. 27 As with equation (1), excluding ROA0 from equation (2) has little effect on the coefficient estimates relating future patents to R&D*Vega (coefficient = 4.566; p-value < 0.05) and R&D*Vega 2 (coefficient = ; p-value < 0.05), indicating that potential omitted variable bias regarding the relation between current profitability and vega is not severe. 21

23 to equation (1) but replace the dependent variable ROA3 with abnormal stock returns in year t (AR0). 2 AAAA0 ii,tt = α + RR&DD ii,tt (ββ 1 + ββ 2 VVVVVVVV ii,tt 1 + ββ 3 VVVVVVVV ii,tt 1 2 ββ 6 MMMMMM ii,tt ) + ββ 7 VVVVVVVV ii,tt 1 + ββ 8 VVVVVVVV ii,tt ββ 4 DDDDDDDDDD ii,tt 1 + ββ 5 DDDDDDDDDD ii,tt ββ 9 DDDDDDDDDD ii,tt 1 + ββ 10 DDDDDDDDDD ii,tt 1 + ββ 11 MMMMMM ii,tt + ββ 12 MMMMMM ii,tt + ββ 13 LLLLLL ii,tt + ββ 14 AAAAAA ii,tt + ββ 15 TTTTTTTT ii,tt + ββ 16 RRRRRR0 ii,tt + εε iiii (3) + The dependent variable, AR0, is defined as the fiscal year buy-and-hold stock return for firm i less the buy-and-hold return of a matching size and book-to-market portfolio. 28 We include industry-adjusted earnings in year t (ROA0) in the model as a proxy for firm-specific news during the period. We estimate this model using OLS and include industry, rather than firm, and year fixed effects. Unlike equations (1) and (2), we estimate equation (3) without firm fixed effects since there is no reason to expect that abnormal returns contain a fixed firm component. Once again, we are primarily interested in β2 and β3. The second model in Table 5 suggests that the pricing of R&D declines linearly with Vega (β2 = ). However, results reported for the full model again suggest a concave pricing of R&D with respect to Vega, though significance is weaker than that of our R&D profitability (Table 3) and patents (Table 4) results. We find a positive value for β2 (β2 = 0.216, p < 0.10) and a negative value for β3 (β3 = , p < 0.01). This market-based result generally corroborates our earlier evidence based on ROA3 and Patent and, in support of H3, suggests that investors understand and price the non-linear effect of vega on the value of R&D investments. In summary, it appears that R&D investments by high-vega managers provide relatively less value to investors than those made by more moderately incentivized managers. 28 We obtain portfolio breakpoints and returns from Ken French s website ( 22

24 Although we observe relatively lower returns to R&D investment for higher vega firms, it is not necessarily the case that the activity is sub-optimal. Without knowledge of the next best alternative use of those resources, we cannot definitively conclude what is in the best interest of the firm. Also, declining short-run rates of return may not capture all of the benefits R&D provides (i.e., future sales growth, new market entry, etc.). However, to the extent that the returns on R&D for our low-to-moderate vega firms represent the expected return, on average, per unit of vega, our results indicate that higher vega firms returns are not keeping pace. While the evidence in Table 5 suggests that investors discount the value of R&D investment by high vega firms, prices may not immediately reflect the implications of vegainduced R&D. For example, prior research argues that information asymmetries are greater for firms with higher levels of R&D (Aboody and Lev 2000). This suggests that investors may not immediately discern the consequences of vega for future performance. Thus, we assess whether the lower contemporaneous market response may merely be the result of a pricing delay. We do so by re-estimating equation (3) after replacing AR0 with AR3, the cumulative abnormal return over years t+1 through t+3. We also include ROA3 in the model to control for earnings news for years t+1 through t+3. Results fail to indicate a concave relation between Vega and future R&D return (i.e., initial underreaction) or any indication that the results in the first column reverse (i.e., initial overreaction). In untabulated analysis, neither β2 nor β3 from equation (3) are significant at conventional levels. Thus, our evidence suggests that the market pricing of the implications of vega for the return on R&D is complete in the current year. Illustrating the concave relation between Vega and R&D returns Each of the prior analyses suggests that the relation between Vega and value created by R&D investment reverses at an inflection point. To illustrate this inflection point, we also 23

25 include a visual representation of the non-linear relations between our three returns measures (future earnings, patent awards, and current stock returns) and R&D investment conditional on vega in Figure 1. We form 50 portfolios based on the rank of Vega and obtain the mean vega of each portfolio (Vega p). For each portfolio, we calculate the β2 coefficient for Vega*R&D times Vegap plus the β3 coefficient for Vega 2 *R&D times Vega 2 p for each return equation (i.e., (1), (2), and (3)). The resulting estimates capture the predicted return per dollar of R&D for each of the 50 portfolios. 29 We then standardize these estimates to have a mean of zero and standard deviation of one so that the resulting estimates are comparable across the different return metrics (ROA3, Patents, and AR0). These standardized values are plotted in Figure 1. The plots for future ROA and future patents suggest the same basic conclusion low-to-moderate levels of vega correspond to increasing returns to R&D investments. However, higher levels of vega yield declining returns to R&D. Interestingly, investor pricing of R&D appears to rise briefly and gradually decline for increasing vega, generally consistent with our conclusion that higher levels of vega do not uniformly correspond to value-increasing investments in R&D. 4. Discriminating between excessive risk taking and greater risk aversion Observing a concave relation between vega and all three indicators of R&D returns could reflect either of the perverse incentive scenarios described earlier. That is, greater vega may correspond to excessively risky investment, leading to a higher frequency of losses, or greater vega may induce more risk aversion, leading to a more conservative investment policy and lower returns. We next employ tests intended to determine which explanation is more descriptive. 29 The resulting equation is β 2*mean Vega + β 3*mean Vega 2. For example, mean Vega for portfolio 25 = and β 2 = and β 3 = in Table 3 for ROA3. The predicted return per dollar of R&D for the 25th portfolio equals 0.211* (-0.031)* =

26 Level of R&D and Vega Our first discriminating test examines the relation between vega and the level of R&D spending, following the assumption in Coles et al. (2006) that higher R&D spending indicates greater risk taking. If some higher vega managers are those who, through overconfidence, have accepted a highly convex pay structure, leading to more aggressive behavior and excessive risk taking, we should observe either a linear or a non-linear and convex relation between the level of R&D spending and vega. Alternatively, higher vega incentives may be associated with managers who have higher levels of existing ESOs and are therefore less diversified. Or, greater wealth effects associated with existing ESOs may have shifted managers to a portion of their utility function where additional risk-taking incentives actually yield more risk aversion. If diminishing returns on R&D for higher vega are the result of greater risk aversion, we should observe a nonlinear and concave relation between the level of R&D spending and vega. To investigate the relation between R&D spending and vega, we estimate a multivariate model similar to that in Coles et al. (2006): 2 2 RR&DD ii,tt = + ββ 1 VVVVVVVV ii,tt 1 + ββ 2 VVVVVVVV ii,tt 1 + ββ 3 DDDDDDDDDD ii,tt 1 + ββ 4 DDDDDDDDDD ii,tt 1 + ββ 5 MMMMMM ii,tt + ββ 6 MMMMMM ii,tt + ββ 7 LLLLLL ii,tt + ββ 8 AAAAAA ii,tt + ββ 9 TTTTTTTT ii,tt + ββ 10 TTTTTTTTTTTT ii,tt + ββ 11 CCCCCChCCCCCCCC ii,tt + ββ 12 GGGGGGGGGGh ii,tt + ββ 13 SSSSSSSSSSSSSS ii,tt + ββ 14 SSSSSSSS ii,tt + εε iiii (4) In addition to the controls included in previous models we add the natural log of one plus the tenure of the CEO (Tenure), the natural log of total cash compensation (CashComp), sales growth (Growth), cash surplus (Surplus), and the natural log of total sales (Sale). In addition to identified control variables, we include year and firm fixed effects in equation (4) as in earlier analyses. Based on evidence in Coles et al. (2006), we expect ββ 1 > 0. However, our particular interest is in whether higher vega is associated with an increasing or diminishing rate of investment in R&D, indicated by ββ 2 > 0 or ββ 2 < 0, respectively. 25

27 Panel A of Table 6 displays the results of estimating equation (4) using OLS. Note that unlike prior analyses, we include observations with zero R&D expense to avoid truncating the distribution. 30 The first column of coefficients excludes the squared terms, Vega 2 and Delta 2 to more closely mirror analyses in Coles et al. (2006). We observe a significant positive coefficient on Vega, consistent with Coles et al. (2006) and the notion that greater sensitivity to stock volatility encourages riskier investment. With the squared terms included, the coefficient on Vega remains positive (0.195) and significant (p = 0.001), and the coefficient on Vega 2 is negative (-0.023) and significant (p < 0.05), indicating that as vega increases beyond some inflection point, R&D spending declines. Thus, higher vega does not appear to incentivize managers to overinvest in risky R&D, rather higher vega results in a declining rate of investment relative to more moderate vega levels. We find no evidence of non-linearity in the relation between R&D and delta, as Delta 2 is insignificant. Prior studies express concerns that the relation between R&D and vega is endogenous (Coles et al. 2006; Shen and Zhang 2013). That is, they suggest that this relation may be susceptible to reverse causality in that greater vega incentives are more likely for firms that invest heavily in R&D (Coles et al. 2006). To address this concern and assess the robustness of our earlier results, we examine the relation between R&D and vega within a five-equation simultaneous system with R&D, Vega, Vega 2, Delta, and Delta 2 as our endogenous variables. 31 We estimate the following system using two-stage least squares (2SLS): 30 As in prior research, we set missing values of R&D to zero for the purposes of this analysis (Chambers et al. 2002; Coles et al. 2006; Ciftci and Cready 2011). Note that we find similar evidence using a Tobit estimator, which corrects for bias associated with censored distributions. 31 Our five-equation system represents a system nonlinear in the endogenous variables. As Wooldridge (2002, p. 230) shows, nonlinearity in the endogenous variables requires that, for identification purposes, the squared endogenous variables be modeled as separate endogenous variables in the system. 26

28 R&D = f{vega, Vega 2, Delta, Delta 2, Tenure, CashComp, MTB, Lev, Surplus, Sale, AR0} Vega = f{r&d, Delta, CashComp, Sale, MTB, Lev, CapEx, RetVol0} Vega 2 = f{r&d 2, Delta 2, CashComp 2, Sale 2, MTB 2, Lev 2, CapEx 2, RetVol0 2 } Delta = f{r&d, Vega, CapEx, Tenure, Sale, MTB, Surplus, Lev, RetVol0} Delta 2 = f{r&d 2, Vega 2, CapEx 2, Tenure 2, Sale 2, MTB 2, Surplus 2, Lev 2, RetVol0 2 } (5) For the R&D, Vega, and Delta equations, we identify our instruments based on Coles et al. (2006). For the Vega 2 and Delta 2 equations, we square the variables from the original related equations and use them as natural instruments (Wooldridge 2002, p. 233). Panel B of Table 6 displays the results for the key variables from the R&D and Vega models within equation (5), which confirm our earlier findings. That is, the coefficient on Vega in the R&D equation is positive (0.262) and significant (p < 0.000), while the coefficient on Vega 2 is negative (-0.030) and significant (p < 0.000). Results for the other variables are generally consistent with expectations and results in Coles et al. (2006). For example, in the Vega equation, the coefficient on R&D is positive while the coefficient on CapEx is negative, consistent with Coles et al. (2006), who argue that vega encourages managers to invest more in risky R&D and less in capital expenditures. Our results extend Coles et al. (2006) by suggesting that this result is not uniform across the distribution of observed values of vega. 32 To summarize, risk-taking incentives introduced through stock option compensation significantly affect both the return on and level of R&D investment. These results indicate that low-to-moderate levels of vega encourage greater returns on and investments in R&D than do higher levels of vega. In turn, the lack of uniformity in the effect of vega on risk taking suggests 32 Larcker and Rusticus (2010) highlight issues related to and limitations of 2SLS that potentially lead to coefficient estimates which are more biased than those obtained through OLS. We estimate this system of simultaneous equations merely to corroborate our Panel A estimates and to follow prior research. We do not suggest that our 2SLS results are more appropriate or reliable than those obtained through OLS. 27

29 that higher levels of vega may not be as effective as more moderate levels at aligning investor and manager preferences related to R&D policy. Total firm risk and R&D conditional on Vega A more direct assessment of whether lower return on R&D spending for higher vega firms is indicative of excessive risk taking or greater risk aversion would be to examine the risk profile of managers R&D portfolios. Unfortunately, that is unobservable. However, as a secondary discriminating test, we examine how the magnitude of vega affects the relation between R&D investment and stock return volatility. If higher vega managers are engaging in excessive risk taking, we expect to observe either a positive and linear or non-linear and convex relation between stock return volatility and R&D spending conditional on vega. Alternatively, if higher vega induces greater risk aversion for some managers, we should observe a non-linear and concave relation between stock return volatility and R&D spending conditional on vega. Similar to Coles et al. (2006), we investigate the relation between Vega and return volatility (RetVol) using the following model: 2 RRRRRRRRRRRR ii,tt+3 = α + RR&DD ii,tt ββ 1 + ββ 2 VVVVVVVV ii,tt 1 + ββ 3 VVVVVVVV ii,tt 1 2 ββ 6 MMMMMM ii,tt + ββ 7 VVVVVVVV ii,tt 1 + ββ 8 VVVVVVVV ii,tt ββ 4 DDDDDDDDDD ii,tt 1 + ββ 5 DDDDDDDDDD ii,tt ββ 9 DDDDDDDDDD ii,tt 1 + ββ 10 DDDDDDDDaa ii,tt 1 ββ 11 MMMMMM ii,tt + ββ 12 MMMMMM ii,tt + ββ 13 LLLLLL ii,tt + ββ 14 AAAAAA ii,tt + ββ 15 TTTTTTTT ii,tt + ββ 16 RRRRRR3 ii,tt + ββ 17 RRRRRRRRRRRR ii,tt + εε iiii (6) + 28

30 RetVol is the average monthly stock return volatility for firm i in years t+1 to t As in equation (4), we include firm and year fixed effects in equation (6) along with various controls that are likely related to stock return volatility. Interestingly and contrary to expectations, results in the second column of Table 7, which excludes the Vega 2 terms, indicate that as vega increases, the relation between return volatility and R&D decreases (coefficient of for Vega*R&D). Alternatively, when we include the Vega 2 terms in the last column, results indicate that while R&D conditional on vega is positively related to risk (Vega*R&D coefficient = 0.010; p < 0.05), that relation decreases for higher vega (Vega 2 *R&D coefficient = ; p < 0.001). These results fail to support concerns that higher vega incentives may lead to excessive risk taking. Rather, they support our general conclusion that managers exposed to greater levels of vega reduce the risk-return profile of R&D relative to managers with more moderate vega (i.e., our results are more consistent with risk aversion than excessive risk taking). This, in turn, suggests that a heavy reliance on stock options may not necessarily align investor and manager risk preferences with respect to R&D investment. 5. Extended and sensitivity analyses Investor oversight Our earlier results suggesting that investors recognize and price the lower-yielding R&D associated with greater vega implies a general inability on the part of stakeholders to curtail this behavior in managers. Weak oversight may be a contributing factor. 34 Observing some loss in 33 We also compute RetVol using daily stock return volatility and obtain virtually identical results. 34 As discussed earlier, other contributing factors may be the lack of adequate attentiveness to risk or inadequate risk expertise by the board of directors. Our inability to adequately measure these characteristics prevents us from pursuing these explanations further. 29

31 value, investors may take a more active role in monitoring managerial investment decisions. Since monitoring requires expertise and is costly, small investors are unlikely as effective as institutional investors, who have incentives to develop the requisite expertise to more effectively monitor management. Accordingly, we next consider whether different levels of investor activism can explain cross-sectional differences in the future performance of vega-induced R&D. Institutional investors, such as pension funds and mutual funds, control over half of publicly traded shares in the United States (Blume and Kleim 2012). By virtue of their size, institutional owners are likely to exercise more oversight and influence managements investment behavior to avoid lower risk/lower returns that are not commensurate with greater stock option compensation. Furthermore, Gillan and Starks (2000) report that shareholder proposals sponsored by institutions garner substantially more support than those sponsored by individuals, and Bushee (1998) finds that when institutional ownership is high, managers are less likely to myopically cut R&D to maintain earnings growth. Finally, Panousi and Papanikolaou (2012) find that managerial risk aversion in the presence of high idiosyncratic risk is mitigated when institutional investors own large fractions of the firm. If our pricing results for higher vega firms are indeed indicative of non-value-maximizing investment, sophisticated institutional owners who are more skilled at monitoring and disciplining managers are more likely to intervene and ensure that managers focus on long-term value rather than short-term growth. We conjecture that greater attention and activism by institutional investors relative to individual investors will limit any undesirable R&D investment by higher vega managers. Thus, we predict that the concave relation between vega and the rate of return on and level of R&D investment is mitigated for firms with greater levels of institutional oversight. To evaluate this conjecture, we use the percentage of shares owned by institutions (InstHold) as a proxy for 30

32 institutional oversight. We partition our sample based on institutional ownership by identifying observations with InstHold above and below yearly sample medians. We then re-estimate equations (1), (2), and (3) within each partition. We present the results of this estimation in Table 8. Consistent with our expectation, the concave relations between Vega and R&D profitability (column 1) and number of patents (column 3) that we document earlier, respectively, appear to be focused in the low (below median) InstHold sample, though we fail to find this pattern of results using AR0 as the dependent variable. These results suggest that R&D investment decisions by firms with greater shareholder activism are less sensitive to risk-taking incentives of the CEO. One possible explanation for this finding is that more active shareholder monitoring substitutes for risk-taking incentives to influence managers investment decisions. In untabulated analysis, we observe slightly higher mean Vega for firms in the high InstHold sample, perhaps indicating that shareholder monitoring interacts with and curbs any negative effects of vega, as opposed to eliminating the need for that incentive mechanism. Capital expenditures A key premise throughout our analysis is that R&D, a risky form of investment, affects managers payoffs derived from options. To increase the likelihood that our results are explained by investment risk and not the nature of investment decisions more generally, we also examine the returns to capital expenditures conditional on the level of risk-taking incentive. Specifically, we replace R&D with capital expenditures deflated by total assets (CapEx), a less risky form of investment, in our returns equations (1) through (3). Table 9 provides the results from this 31

33 analysis. In short, we find no evidence to suggest that the concave relation between returns to R&D and lagged vega is a systematic relation reflected in other forms of less risky investment Conclusion This study demonstrates that risk-taking incentives embedded in executive stock options have a diminishing effect on the actual economic benefits that derive from risky investment. We show that as vega increases, the return on R&D (i.e., future earnings, patents awarded, and market valuation) initially increases. However, as vega increases past an inflection point, the relations between vega and returns on R&D become negative. These results are consistent with high levels of vega leading managers to either engage in excessive risk taking, leading to greater losses, or greater risk aversion, resulting in less risky investing. Subsequent analysis suggests that managers with higher vega engage in less risk taking, indicating that they offset personal exposure to risk through lower discretionary investment in R&D projects. In addition, we document a concave relation between future stock return volatility and R&D conditional on vega. Our results support theory suggesting that greater stock option compensation does not always incentivize risk-averse, less-diversified managers to increase the risk profile of the firm s investments (Lambert et al. 1991; Carpenter 2000; Meulbroek 2001; Ross 2004). Namely, while moderate option-based incentives can increase the efficiency of a firm s investment policy, greater levels of options may encourage less profitable investment and a lower level of less risky investment. 35 We also examine the relation between the level of CapEx and vega analogous to equation (4) but also fail to observe either concavity or convexity. 32

34 We also show that the effects of vega on the profitability of R&D are concentrated among firms with relatively low institutional holdings, suggesting that a certain level of investor involvement can mitigate any negative consequences of excessive option compensation. Thus, while executive stock options are still a viable means of aligning manager and shareholder incentives, our study points to potential unintended consequences in that excessive awards could have less desirable effects on investment policy. 33

35 Appendix A. Variable definitions Variable Name Adv AR Description Total advertising expense (Compustat item XAD) for firm i in year t divided by total assets (Compustat item AT) for firm i in year t-1. Abnormal returns, computed as the difference between buy-and-hold returns for firm i and a matched size and book-to-market portfolio. We accumulate returns over the 12 months corresponding to the firm s fiscal year. We use 5x5 portfolio returns and size/book-to-market breakpoints available from Dr. Kenneth French s website. AR0 (AR3) is the abnormal return corresponding to fiscal year t (t+1 to t+3). CapEx Total capital expenditures (Compustat item CAPX) for firm i in year t divided by total assets in year t-1. CashComp Delta InstHold The natural logarithm of total cash compensation (salary + bonus) for the CEO of firm i in year t. The natural logarithm of delta in year t-1. Delta estimates are obtained from the data provided by Coles, Daniel, and Naveen (2013), which are computed based on the methods described in Core and Guay (2002) and Coles et al. (2006). In general, delta is the sensitivity of CEO stock and option holdings to a 1% change in price of the underlying stock, measured in thousands. Delta 2 is the square of Delta. Percentage of common shares outstanding owned by institutional investors for firm i as of the end of fiscal year t (obtained from Thomson-Reuters). Lev MTB Financial leverage ratio, measured as total long term debt (Compustat items DLTT plus DLC) divided by total assets for firm i in year t The natural logarithm of the ratio of market value of equity (Compustat items CSHO*PRCC_F) to book value of equity (Compustat item CEQ) for firm i at time t. MVE The natural logarithm of the market value of equity for firm i at time t. Patent R&D Total number of patents granted to firm i over years t to t+1, divided by total assets in year t-1 times 100. Total R&D (Compustat item XRD) for firm i in year t divided by total assets in year t-1 times 100. Ret 12-month return for firm i in year t. RetVol The standard deviation of monthly stock returns. RetVol0 (RetVol3) is return volatility corresponding to fiscal year t (t+1 to t+3). 34

36 Variable Name ROA Description Average earnings for firm i in year t (ROA0) or years t+1 to t+3 (ROA3). Earnings are measured as income before interest, taxes, depreciation and amortization (Compustat item EBITDA) plus R&D and advertising, divided by total assets in year t-1. Earnings are adjusted by the industry-year mean. Sale The natural logarithm of total revenue (Compustat item REVT) for firm i in year t. Growth Sales growth for firm i from time t-1 to t. Surplus Tang Tenure Vega Cash surplus, measured for firm i as (cash flow from operations depreciation + R&D) in year t / total assets in year t-1. In Compustat, (OANCF DP + XRD) / AT. Tangible assets, measured as the sum of net property plant and equipment, inventory, investments and advances (equity and other) (Compustat items PPENT plus INVT plus IVAEQ plus IVAO) for firm i in year t divided by total assets for firm i in year t-1. CEO tenure, measured as the days between the date the CEO became CEO (per Execucomp) and the date of the fiscal year-end in year t, scaled by 365. The natural logarithm of vega in year t-1. Vega estimates are obtained from the data provided by Coles, Daniel, and Naveen (2013), which are computed based on the methods described in Core and Guay (2002) and Coles et al. (2006). In general, vega is the sensitivity of CEO options to a 1% change in implied volatility, measured in thousands as of end of fiscal year t. Vega 2 is the square of Vega. 35

37 References Aboody, D., and B. Lev Information asymmetry, R&D, and insider gains. Journal of Finance 55 (6): Bhagat, S., and B. Bolton Financial crisis and bank executive incentive compensation. Journal of Corporate Finance 25: Beasley, M., B. Branson, and B. Hancock Report on the current state of enterprise risk oversight: Update on trends and opportunities. North Carolina State University, Bebchuk, L. A., and J. M. Fried Executive compensation as an agency problem. Journal of Economic Perspectives 17 (3): Blume, M., and D. Kleim Institutional investors and stock market liquidity: Trends and relationships. Working paper, University of Pennsylvania. Bushee, B The influence of institutional investors on myopic R&D investment behavior. The Accounting Review 73 (3): Cadman, B., and J. Sunder Investor horizon and CEO horizon incentives. The Accounting Review 89 (4): Carpenter, J., Does option compensation increase managerial risk appetite? Journal of Finance 55 (5): Chambers, D., R. Jennings, and R. Thompson Excess returns to R&D-intensive firms. Review of Accounting Studies 7 (2): Chava, S., and A. Purnanandam CEOs versus CFOs: Incentives and corporate policies. Journal of Financial Economics 97 (2): Ciftci, M., and W. Cready Scale effects of R&D as reflected in earnings and returns. Journal of Accounting and Economics 52 (1): Coles, J., N. Daniel, and L. Naveen Managerial incentives and risk-taking. Journal of Financial Economics 79(2): Coles, J., N. Daniel, and L. Naveen Calculation of compensation incentives and firmrelated wealth using Execucomp: Data, program, and explanation. Working paper, Temple University. Core, J., and W. Guay Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of Accounting Research 40 (3): Curtis, A., S. McVay, and S. Toynbee The changing implications of research and development expenditures for future profitability. Working paper, University of Washington. David, P., R. Kocchar, and R. Levitas The effect of institutional investors on the level and mix of CEO compensation. Academy of Management Journal 41 (2): Dittmann, I., and E. Maug Lower salaries and no options? On the optimal structure of executive pay. Journal of Finance 62 (1):

38 Dong, Z., C. Wang, and F. Xie Do executive stock options induce excessive risk-taking? Journal of Banking & Finance 34 (10): Elton, E Presidential address: Expected return, realized return, and asset pricing tests. Journal of Finance 54 (4): Englmaier, F A strategic rationale for having overconfident managers. Working paper, Harvard University. Fahlenbrach, R., and R. M. Stulz Bank CEO incentives and the credit crisis. Journal of Financial Economics 99 (1): Financial Accounting Standards Board (FASB) Statement of Financial Accounting Standards No. 123R: Share-Based Payment. Norwalk, CT: FASB. Gervais, S., J. B. Heaton, and T. Odean Overconfidence compensation contracts, and capital budgeting. Journal of Finance 66 (5): Gillan, S., and L. Starks, L. Corporate governance proposals and shareholder activism: The role of institutional investors. Journal of Financial Economics 57 (2): Griliches, Z Patent statistics as economic indicators: A survey. Journal of Economic Literature 28 (4): Hall B., Z. Griliches, and J. Hausman Patents and R and D: Is there a lag? International Economic Review 27 (2): Hanlon, M., S. Rajgopal, and T. Shevlin Are executive stock options associated with future earnings? Journal of Accounting and Economics 36 (1): Haugen, R., and L. Senbet Resolving the agency problems of external capital through options. Journal of Finance 36 (3): Hausman J., B. Hall, and Z. Griliches Econometric models for count data with an application to the patents R&D relationship. Econometrica 52 (4): Hayes, R., M. Lemmon, and M. Qiu Stock options and managerial incentives for risktaking: Evidence from FAS 123R. Journal of Financial Economics 105 (1), Ho, Y., Z. Xu, and C. Yap R&D investment and systematic risk. Accounting & Finance 44 (3): Holthausen, R. W., D. F. Larcker, and R. G. Sloan Annual bonus schemes and the manipulation of earnings. Journal of Accounting and Economics 19 (1): Jensen, M., and W. Meckling Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3 (4): Kogan, L., D. Papanikolaou, A. Seru, and N. Stoffman Technological innovation, resource allocation, and growth. Working paper, Massachusetts Institute of Technology. Kothari, S., T. Laguerre, and A. Leone Capitalization versus expensing: Evidence on the uncertainty of future earnings from capital expenditures versus R&D outlays. Review of Accounting Studies 7 (4), KPMG Never again? Risk management in banking beyond the credit crisis

39 Lambert, R Executive effort and selection of risky projects. The Rand Journal of Economics 17 (1): Lambert, R., D. Larcker, D, and R. Verrecchia Portfolio considerations in valuing executive compensation. Journal of Accounting Research 29 (1): Larcker, D., and T. Rusticus On the use of instrumental variables in accounting research. Journal of Accounting and Economics 49 (3): Lev, B., and T. Sougiannis The capitalization, amortization, and value-relevance of R&D. Journal of Accounting and Economics 21 (1): Liu, Y., and D. Mauer Corporate cash holdings and CEO compensation incentives. Journal of Financial Economics 102 (1): Mehran, H Executive incentive plans, corporate control, and capital structure. Journal of Financial and Quantitative Analysis 27 (4): Meulbroek, L The efficiency of equity-linked compensation: Understanding the full cost of awarding executive stock options. Financial Management 30 (2): Murphy, K Executive compensation. Handbook of Labor Economics 3 (B): Pakes, A. and M. Schankerman The rate of obsolescence of patents, research gestation lags, and the private rate of return to research resources. In R&D, Patents, and Productivity edited by Z. Griliches, Chicago: University of Chicago Press. Panousi, V. and D. Papanikolaou Investment, idiosyncratic risk, and ownership. Journal of Finance 67 (3): PWC Reward: A new paradigm? Rajgopal, S., and T. Shevlin Empirical evidence on the relation between stock option compensation and risk-taking. Journal of Accounting and Economics 33 (2): Rego, S., and R. Wilson Equity risk incentives and corporate tax aggressiveness. Journal of Accounting Research 50 (3): Roll, R The hubris hypothesis of corporate takeovers. Journal of Business 59 (2): Ross, S Compensation, incentives, and the duality of risk aversion and riskiness. Journal of Accounting Research 59 (1): Schrand, C., and S. L. C. Zechman Executive overconfidence and the slippery slope to financial reporting. Journal of Accounting and Economics. 53 (1): Shen, C., and H. Zhang CEO risk incentives and firm performance following R&D increases. Journal of Banking & Finance 37 (4): Smith, C., and R. Stulz The determinants of firms' hedging policies. Journal of Financial and Quantitative Analysis 20 (4): Wooldridge, J. M Econometric analysis of cross sectional and panel data. Cambridge, MA: MIT Press. 38

40 Figure 1 Relation between R&D and Future Earnings, Patent Awards, and Returns as a Function of Vega Future ROA Future Patents Stock Returns Figure 1 plots the predicted returns on R&D attributed to vega based on vega coefficients estimated in the ROA3, Patents, and AR0 equations (equations (1) to (3)) across 50 Vega portfolios. Specifically, we first sort the sample into 50 portfolios based on the rank of Vega and compute the mean value of Vega for each portfolio (Vega p). The return on R&D based on future ROA (line marked with ) for each portfolio is computed by summing the products of the coefficients on Vega*R&D and Vega 2 *R&D (both from the third column in Table 3) and Vega p and Vega p2, respectively. The return on R&D based on future patents (line marked with ) for each portfolio is computed by summing the products of the coefficients on Vega*R&D and Vega 2 *R&D (both from the third column in Table 4) and Vega p and Vega p2, respectively. The return on R&D based on current returns (line marked with ) for each portfolio is computed by summing the products of the coefficients on Vega*R&D and Vega 2 *R&D (both from the third column in Table 5) and Vega p and Vega p2, respectively. We standardize each of these three sets of return on R&D measures to have a mean of 0 and standard deviation of 1 so they are comparable. Footnote 30 details these calculations for a sample portfolio. 39