Executive Compensation under Common Ownership ú

Transcription

1 Executive Compensation under Common Ownership ú Heung Jin Kwon 1 1 Department of Economics, University of Chicago November 29, 2016 [Link to Most Recent Version of Paper] Abstract This paper shows that higher common ownership of natural competitors is associated with more use of relative performance evaluation (RPE). First, I use compensation data from Execucomp and find that executives receive more rewards for outperforming peer firms if common ownership concentration increases. Second, I manually collect executive compensation contract details from 2014 proxy statements and show that the likelihood of using RPE increases with common ownership. Further, the size of incentive awards tied with RPE increases with common ownership. These findings suggest that institutional investors with common ownership exert a strong influence on executive compensation in a positive way: less alignment of pay with industry performance. I construct a measure of ownership by common owners and show that institutional investors with common ownership have a stronger influence on compensation than investors without common ownership. Finally, I use the inclusion to S&P 1500 index as an instrumental variable to support a causal connection. These findings signify the potential benefit from governance by large, diversified institutional investors with common ownership. Keywords: Executive Compensation. Common Ownership. Institutional Investors. ú I am deeply grateful to Luigi Zingales, Emir Kamenica, Jacopo Ponticelli, and Doron Ravid for valuable advice, helpful conversations, and continuous support. I thank Jin Yeub Kim, Emre Unlu, and seminar participants at the University of Chicago Stigler Center. kwonheungjin@uchicago.edu

2 1 Introduction There is a growing concern that large, diversified institutional investors own significant shares of natural competitors in product markets. This common ownership weakens incentives to compete because competition, especially in the form of price competition, lowers profits of all firms that the investors with common ownership own. 1 Hence, if the executives of co-owned firms understand these anti-competitive incentives of common owners and are incentivized to maximize the portfolio value of the common owners, the market outcomes will become more monopolistic: higher product prices. There is evidence that common ownership by institutional investors increases product prices in the airline and banking industries (Azar, Schmalz, and Tecu, 2015; Azar, Raina, and Schmalz, 2016). Because most common owners are large, diversified institutional investors who regularly communicate with the executives of portfolio firms, the executives clearly understand the anti-competitive incentives. However, little is known about how executive o cers are incentivized to reduce competition. In this paper, I study the e ects of common ownership on executive compensation, specifically on the use of relative performance evaluation (RPE). I use executive compensation as a testing ground for the following three reasons. First, institutional investors are vocal about both the level of compensation and the alignment of pay with performance (Hartzell and Starks, 2003). Second, executive compensation (particularly how compensation changes with own and peer performance) reflects the environment of strategic competition in the product markets (Aggarwal and Samwick, 1999a; Vrettos, 2013). Lastly, agency theory predicts that executive compensation, in particular the degree of RPE, a ects competitiveness of industries under imperfectly competitive product markets (Fershtman and Judd, 1987; Fumas, 1992; Aggarwal and Samwick, 1999a; Anton et al., 2016). Specifically, if executive compensation is the mechanism between common ownership and less competition, then common ownership 1 For example, Azar (2012) argues the trilemma of achieving shareholder diversification, shareholder value maximization, and product market competition. See also Reynolds and Snapp (1986), Bresnahan and Salop (1986), O Brien and Salop (2000), and Gilo (2000) who theoretically show these anti-competitive incentives from common ownership. 1

3 is expected to reduce the use of RPE in order to soften the incentives to outperform peers. To test this empirical prediction, I use a two-pronged approach to study the e ects of common ownership on executive compensation: implicit and explicit approaches. First, I use the Execucomp database and estimate how annual flow compensation changes with own and peer performance pay-performance elasticities. Then, I test how the ratio of peer payperformance elasticity to own pay-performance elasticity changes with common ownership In doing so, I use Modified Herfindahl Hirschman Index Delta (MHHID) that measures the degree of common ownership in each industry, proposed by O Brien and Salop (2000). However, using the Execucomp database alone is potentially problematic. The implicit test fails to detect RPE with traditionally used industry definitions such as Standard Industrial Classification (SIC) even when firms in the sample explicitly disclose the use of RPE (Gong, Li, and Shin, 2011). Further, estimated pay-performance elasticities may not reflect pay for current or past performance because the majority of performance-based equity awards reported in the Execucomp are designed to be earned upon future performance (De Angelis and Grinstein, 2016). To address these concerns, I complement this implicit approach with an approach that uses a manually collected set of contract details from 2014 proxy disclosures by single-segment firms. Using this data, I study the determinants of the use and size of incentive awards tied with RPE, which is referred to as the explicit test of RPE. 2 Surprisingly, the results from both the implicit and explicit approaches indicate that RPE is positively associated with common ownership. Hence, executive compensation is unlikely to be the mechanism between common ownership and less competitive outcomes in product markets. The implicit approach shows that the ratio of peer to own pay-performance elasticities is negatively correlated with common ownership. This finding implies that executives receive more rewards for outperforming peers under common ownership, thus having more incentives to compete aggressively. This result is robust to alternative industry definitions such as 2 See Gong, Li, and Shin (2011), De Angelis and Grinstein (2011), and De Angelis and Grinstein (2016) for the analysis of RPE use from 2006 proxy disclosures. 2

4 variants of SIC codes four-digit SIC, three-digit SIC, and size-split four-digit SIC of Albuquerque (2009) as well as the text-based fixed industry definitions developed by Hoberg and Phillips (2010) and Hoberg and Phillips (2016). The result is also robust to other measures of compensation, specifically non-equity incentive plan compensation. Using non-equity incentive plan compensation is particularly useful to estimate pay for current or past performance because non-equity incentive plan compensation is disclosed when it is earned upon satisfying performance conditions. The result with non-equity incentive pay confirms the previous finding: more use of RPE in industries with higher common ownership. The explicit approach corroborates the findings from the implicit approach. The likelihood of using RPE (disclosed in proxy statements) increases with common ownership. This result is robust to alternative industry definitions. One may argue that the use of RPE alone does not necessarily give more incentives to compete because the size of awards tied with RPE could small or because the performance benchmarks with which firms compare their performance could be too broad. These two concerns do not weaken my finding. First, the size of plan-based incentive awards evaluated with performance benchmarking increases with common ownership. Further, firms in industries with higher common ownership are more likely to use industry-specific indices such as Philadelphia Semiconductor Index or FTSE NAREIT US Real Estate Index. My findings insinuate that institutional investors with common ownership make executive compensation more e cient by reducing pay for luck more filtering of industry-specific shock that out of executives control. However because common ownership, MHHID, is measured at industry-level, drawing inferences on the role of common owners with MHHID potentially loses the firm-level cross-sectional variations of the holdings by institutional investors with common ownership. Hence, I compute individual investors contributions to common ownership measure in each industry and list the top 5 institutional investors with the contributions to common ownership which I refer to as the top 5 common owners in each industry. Using the sum of ownership by the top 5 common owners, I estimate the 3

5 influence of common owners on the use of RPE at the firm-level. The results confirm that the institutional investors with common ownership are highly influential in RPE use. Specifically, the shares by the top 5 common owners are much more strongly associated with RPE use, compared to total institutional ownership and ownership by top 5 investors without common ownership. Using the ownership by the top 5 common owners the at the firm-level might be subject to an endogeneity issue. Institutional investors might invest more in firms that use RPE, because it might signal better corporate governance. To address the endogeneity issue, I use the addition to the Standard and Poor (S&P) Composite 1500 as an instrumental variable, similar to Aghion, Van Reenen, and Zingales (2013) who use the addition to the S&P 500 as an instrumental variable for institutional ownership. This instrumental variable is particularly useful in identifying the e ects of institutional investors with common ownership because the membership to S&P 1500 primarily increases the ownership by quasi-indexers who are the main source of common ownership. Using this instrumental variable, I find that large institutional investors with common ownership have a significant influence on executive compensation. I provide an additional evidence to support that institutional investors with common ownership reduce pay for luck. Using Bebchuk, Grinstein, and Peyer (2010) s CEO luck measure that is based on the timing of stock option grants, I find that the ownership by common owners is negatively associated with the likelihood of granting options at the lowest price of the month. Compared with other types of institutional investors, institutional investors with common ownership are the most significant to deter the lucky grants. Overall, institutional investors with common ownership reduce pay for luck, which improves the practice of executive compensation. My results imply that the anti-competitive outcomes of common ownership are not likely caused by executive compensation. However, the implications of my work reach even further. At a broader scale, my findings shed a light on the governance by institutional investors with common ownership. Conventional 4

6 wisdom suggests that investors with diversified portfolios do not have strong incentives to improve governance of individual firms because the stake in each firm is small. This claim is challenged by the presence of large, diversified institutional investors who hold significant shares of portfolio firms. By exploring the influence of common owners on compensation, the paper is one of the first papers to explore corporate governance by institutional investors with common ownership. Specifically, my paper di erentiates from the literature on passive investors (e.g. Appel, Gormley, and Keim (2016)) by focusing on institutional investors with common ownership rather than investors with passive investment strategies. By examining the influence of common ownership on executive compensation, this paper expands the current debate on common ownership from anti-competitive product market outcomes to potentially positive governance role by institutional investors with common ownership. The paper is organized as follows. Section 2 discusses related literature and develops empirical hypotheses. Section 3 presents the findings from implicit approach. Section 4 contains the results from explicit approach. Section 5 discusses the influence of institutional ownership by common owners. Section 6 concludes. 2 Related Literature and Hypothesis Development 2.1 Related Literature My paper is broadly related to two strands of literature: common ownership and executive compensation. First, my work is related to the literature on the e ects of common ownership. The theoretical literature in industrial organization predicts that common ownership of natural competitors in oligopolistic market leads to less competitive outcome (Reynolds and Snapp, 1986; Bresnahan and Salop, 1986; O Brien and Salop, 2000; Gilo, 2000). This empirical prediction is supported by the evidences in a growing literature that studies the e ects of common ownership by institutional investors on product markets. For example, Azar, 5

7 Schmalz, and Tecu (2015) empirically assess the e ects of common ownership on airline ticket prices. Using Modified Herfindahl Hirschman Index Delta, they find that common ownership increases the ticket prices. These anti-competitive e ects are further studied in the banking industry (Azar, Raina, and Schmalz, 2016). These findings give rise to active debate on how the anti-competitive incentives of institutional investors lead to actual market outcomes (Elhauge, 2016). Anton et al. (2016) is the most closely related to my paper. Extending Aggarwal and Samwick (1999a), they theoretically show that common ownership leads to less use of RPE. To test the empirical prediction, they use the same Execucomp database with di erent empirical specifications. Their evidences suggest that common ownership leads to less use of RPE, which is exactly opposite to the findings in my paper. More specifically, they estimate the dollar change in pay to the dollar change in firm value (pay-performance sensitivities), whereas I estimate the percentage change in compensation with respect to the percentage change in firm value (pay-performance elasticities). The empirical specifications used in my study have some appealing aspects. First, because pay-performance sensitivites decrease in firm size (see e.g. Schaefer (1998)), the failure of controlling for firm size in sensitivities may lead to biased results. Further, as reported in Table 6, stock return is predominantly used to evaluate executive performance in incentive awards tied with RPE. Recognizing that there are no definite answers on the choice of empirical specifications, I complement the implicit approach with the explicit approach. The explicit approach shows that the RPE use (disclosed in proxy statements) increases with common ownership. The Online Appendix further provides more detailed comparison. Second, the paper is related to the empirical literature on RPE. The study of relative performance evaluation originates from Holmström (1979) s informativeness principle, which shows that compensation contracts should include any signal that gives additional information about how an agent (CEO) exerts e ort (Holmström, 1979, 1982). If a firm s performance is a ected by the CEO s e ort as well as industry-specific factors that simulta- 6

8 neously a ect peer firms, then filtering out peer performance results in greater e ciency by producing a more precise signal about the CEO s e ort. Despite such a clear prediction on RPE, the empirical evidence on the use of RPE in executive contracts is mixed (Antle and Smith, 1986; Barro and Barro, 1990; Gibbons and Murphy, 1990; Jensen and Murphy, 1990; Aggarwal and Samwick, 1999b; Janakiraman, Lambert, and Larcker, 1992; Bertrand and Mullainathan, 2001; Albuquerque, 2009). Such mixed evidences give rise to a new literature of RPE by studying how the use of RPE varies with di erent CEO, firm, and industry characteristics; e.g., CEO s exposure to labor market (Himmelberg and Hubbard, 2000), CEO s ability to hedge the risk from compensation contracts (Garvey and Milbourn, 2003), growth opportunities of the firm (Albuquerque, 2014), the volatility in stock return (Aggarwal and Samwick, 1999b), and industry concentration (Aggarwal and Samwick, 1999a). The paper contributes to the literature by showing that common ownership is one of the key determinants of RPE. In addition, my paper contributes to the literature on the explicit use of RPE. In particular, Gong, Li, and Shin (2011), De Angelis and Grinstein (2011), and De Angelis and Grinstein (2016) use 2006 proxy disclosures and study determinants of RPE use. By using 2014 proxy disclosures as a testing ground, my study gives more recent evidences on RPE use as well as examines how di erent types of institutional investors in terms of portfolio diversification exert influence on RPE use. 2.2 Hypothesis Development If natural competitors in oligopolistic markets are owned by the same set of investors, then optimal executive compensation contract exhibits less RPE. Kraus and Rubin (2010) study the optimal design of executive compensation contracts when two firms are owned by the same investor. In the model, the managers can decide on two types of e ort: cannibalistic e ort that increases own firm s profit at the expense of the rival firm s profit and economy-increasing e ort that does not influence the rival firm. In this case, the investor who owns both firms does not give incentives in the form of RPE because it causes the man- 7

9 agers of both firms exert more cannibalistic e ort and less economy-increasing e ort. The intuition is as follows. Granting incentive plans in the form of RPE gives managers more incentives to outperform rival firms. In imperfectly competitive markets, these additional incentives translate into more aggressive competition and decrease the industry-wide profits. Thus, investors with common ownership whose portfolio values depend on the industry-wide profits rather than individual firms profits should be unwilling to grant RPE awards. The same mechanism drives the theoretical results of Anton et al. (2016). Hence, if executive compensation is the mechanism between common ownership and less competition, then the use of RPE is expected to decrease with common ownership. This hypothesis between common ownership and RPE could be tested in the implicit and the explicit approaches as follows. In the implicit approach, the hypothesis implies that the pay-performance elasticity to peer performance is smaller (less negative) in industries with higher common ownership. In addition, a firm s use of RPE could be measured with the ratio of peer to own pay-performance elasticity compensation ratio (Holmström and Milgrom, 1987). Hence, this compensation ratio should become less negative under common ownership. In the explicit approach, the hypothesis implies that the likelihood of using RPE decreases with common ownership. Further, the size of incentive awards tied with RPE is expected to decrease with common ownership. 3 Implicit Approach In this section, I present the main results of the implicit approach. I document that common ownership is negatively related to the pay-for-performance elasticity to peer firms performance. Further, I show that the use of RPE (defined as the ratio of peer to own pay-performance elasticity) increases with common ownership. The results are robust to di erent industry definitions: four-digit SIC; three-digit SIC; size-split four-digit SIC à la Albuquerque (2009); and the 10-K based fixed industry classification of Hoberg and Phillips (2010) (FIC). In addition, the results are robust to di erent measures of compensation: total 8

10 annual flow compensation based on the grant-date fair values of restricted shares and options (TDC1 in Execucomp) and non-equity incentive plan compensation. 3.1 Data I obtain compensation data from Execucomp, stock return and CPI index from CRSP monthly database, and financial data from Compustat North America. Executive Compensation I use ExecuComp database for executive compensation from 1993 to ExecuComp provides annual panel data of executive compensation in S&P 1500 firms as well as some additional firms. 3 As a baseline compensation measure, I use total annual flow compensation based on the grant-date fair values of equity awards, which is the sum of salary, bonus, nonequity incentive plan compensation, grant-date fair values of stock and option awards, excess earnings on deferred compensation, and other annual compensation. 4 Because institutional investors influence compensation through actions taken by boards, it is suitable to use current flow compensation over which boards exert direct control. 5 In addition to total flow compensation, I study how non-equity incentive plan compensation changes with own and peer performance. Performance-based equity incentive plan which consists of a huge portion of total compensation often vests upon future performance conditions. 6 Hence, equity compensation may be considered as adjustments to executive incentives, not remuneration for performance (Core and Guay, 1999). In this regard, nonequity incentive plan is useful because executive compensation disclosure rules require the non-equity incentive pay to be disclosed when it is earned. Following the literature, I adjust all compensation measures for inflation in constant 2006 dollars by the value of the CPI 3 Usually, the additional firms include previous constituents of S&P 1500 that are still trading in stock market. 4 In the Online Appendix, I present additional results with total annual flow compensation based on realized pay including gains from exercising options (TDC2 in Execucomp). 5 By using flow compensation, I cannot draw implications on the overall incentives implied by portfolio held by executives. 6 For example, future performance conditions such as 3-year stock return since the grant date are frequently used as vesting conditions for performance-based restricted share units. 9

11 index of the fiscal year-end month. Table 1 presents summary statistics for compensation. An average (median) executive gets a real total annual flow compensation of $2.173 million ($1.029). 70.5% of executives after 2006 receive non-equity incentive plan compensation. For executives who receive nonequity incentive pay, 24.9% (22.4%) of total compensation comes from non-equity incentive plan. Industry Definitions In order to measure peer performance as well as common ownership of competitors, well-defined industries are necessary. As the baseline specification, I use four-digit SIC codes from Compustat North America instead of the CRSP SIC codes used in Anton et al. (2016). Previous studies report significant di erences in SIC codes assigned by Compustat and CRSP (Guenther and Rosman, 1994; Kahle and Walkling, 1996). By using the SIC codes from Compustat that covers more firms than CRSP, I can assign industries to every firm in Compustat without dropping non-crsp firms or replacing missing SIC codes of non- CRSP firms with Compustat SIC codes. 7 Because the majority of ExecuComp firms operate in multiple segments, I use coarser three-digit SIC classifications to check the robustness. I also use Albuquerque (2009) s size-split SIC codes which is shown to have more power in detecting RPE. Specifically, I divide each four-digit SIC codes into size quartiles in each fiscal year. 8 In addition to industry definitions based on SIC codes, I use the text-based fixed industry definition developed by Hoberg and Phillips (2010) and Hoberg and Phillips (2016). Using the product descriptions in annual reports (10-K), Hoberg and Phillips (2010) assign Compustat firms to 25, 50, 100, 200, 300, 400, and 500 industries from 1996 to This classification of industries could be more useful than SIC codes to identify product market 7 Guenther and Rosman (1994) and Kahle and Walkling (1996) report significant di erences in assigning SIC codes by Compustat and CRSP in terms of procedure and data source, which justifies the use of Compustat SIC codes to keep consistency of industries assigned to firms. However, using CRSP SIC codes with or without replacing missing SIC codes with Compustat does not qualitatively change the main results. 8 Size-split industry definitions with two-digit and three-digit SIC codes result in qualitatively similar results. 10

12 competitors and detect RPE use by the implicit approach, which is shown in Jayaraman, Milbourn, and Seo (2015) I use the finest classifications, FIC-400 and FIC-500, which assign firms to 400 and 500 industries respectively, so industries reflect close product market competitors. Performance Measure Following Garvey and Milbourn (2003) and Albuquerque (2009), I use annual stock returns obtained from the CRSP monthly to measure own and peer firms performance. I measure own performance as the continuously compounded annual stock return, assuming that dividends are reinvested. To measure peer performance, I use the annual return on value-weighted portfolios of firms in the same industry excluding own firm. 9 There are advantages to using stock returns instead of other potential performance metrics (e.g. accounting measures such as return on assets or improvement in shareholder wealth used in Jensen and Murphy (1990) and Aggarwal and Samwick (1999a)) Compared with accounting measures such as return on assets, stock returns are less susceptible to earnings management reported in Bergstresser and Philippon (2006). Compared with the change in shareholders wealth recently used in Anton et al. (2016), stock returns have three advantages. Theoretically, which measure is more appropriate depends on whether the marginal product of managerial e orts is invariant to firm size (Hall and Liebman, 1998; Baker and Hall, 2004). When executive o cers actions a ects shareholder value in dollars, then payperformance sensitivites are appropriate. On the other hand, if the managerial e ort a ects shareholder return, then pay-performance elasticities are appropriate. 10 With that in mind, stock return is a more appropriate measure of performance driven by strategic decisions on competition because competitiveness in industries a ects profitability. 9 Using the return on equal-weighted portfolio yields qualitatively similar results. 10 Baker and Hall (2004) give some examples of both cases. For example, whether to buy a corporate jet or sell non-core assets can be deemed as the decision that influences shareholder value in dollars. On the other hand, reorganizing firm or developing corporate strategies could be considered as the action that changes shareholder return. 11

13 Empirically, the elasticity approach is better at explaining the cross-sectional variations in compensation compared with the sensitivity approach (Murphy, 1999). Further, payperformance elasticities measured with stock returns are relatively invariant to firm size, whereas pay-performance sensitivities measured with shareholder wealth decrease in firm size (Gibbons and Murphy, 1992; Schaefer, 1998). Behaviorally, stock return is the most commonly used performance metric in compensation contracts. Table 6 shows that 218 out of 241 RPE firms use stock return as their performance metrics while no firm uses the dollar increase in the market value of the firm as a metric. 11 The importance of aligning executive pay with stock return or total shareholder return is underscored by Institutional Shareholder Services (ISS) s US Executive Compensation Policies. 12 Common Ownership Measure I measure common ownership with Modified Herfindahl Hirschman Index Delta, as proposed by O Brien and Salop (2000) and used by Azar, Schmalz, and Tecu (2015), Azar, Raina, and Schmalz (2016), and Anton et al. (2016). In O Brien and Salop (2000), firm j maximizes the sum of shareholders portfolio values weighted by their control shares: max x j Mÿ ÿ N ij ik fi k, i=1 k=1 where ij is the measure of control rights of firm j by owner i, ik is the cash flow rights of firm k by owner i, and fi k is the profit in the Cournot model. The model gives the markup of price over marginal cost as follows: Nÿ j=1 P MC j P = 1 C HHI + Nÿ ÿ j=1 k =j q D i ij ik s j s k q, (1) i ij ij 11 When I consider performance metrics in performance-based awards (PBA), 268 out of 850 firms (31.5%) that explicitly use PBA use stock return as performance metrics. Restricting the sample to financial firms, 78 out of 139 PBA firms in financial sector (56.1%) use stock return to evaluate executive performance, implying that financial institutions use stock return as performance metrics more often than other firms. 12 See ISS s US executive compensation policies: (Accessed October 26, 2016) 12

14 where is the elasticity of demand and s j is the firm j s market share. The increase in markup due to common ownership, q q N q i j=1 k =j s j s ij ik k q i ij ij, can be interpreted as the impact of common ownership on product market competition, which is denoted as MHHID. I closely follow Azar, Schmalz, and Tecu (2015) to construct MHHID. I use Thomson- Reuters 13F database that is based on quarterly reportings by the financial institutions with over $100 million in qualifying assets under management. 13 The database also provides the number of voting shares. I use the proportion of shares to total outstanding shares to proxy the cash flow rights s. I use the proportion of voting shares to total outstanding shares to proxy the measure of control rights s. I construct MHHID to measure the influence of common ownership throughout each fiscal year. At the end of each quarter, I average each institutional investor s holdings of each firm (as a percentage of shares outstanding) over the previous 12 months, which measures the cash flow right s in Equation (1). I also average each institutional investor s holdings of sole-voting shares (as a percentage of shares outstanding) over the previous 12 months, which proxies the control right s in Equation (1). Using ownership averaged over the previous 12 months enables me to measure the overall influence of institutional investors during that period. Following Azar, Schmalz, and Tecu (2015), I restrict the sample to those with at least 0.5% of shares outstanding. 14 This construction of and excludes the influence of short-term and small-scale investments that are likely to have limited influence on boards. I construct market share and Herfindahl Hirschman Index (HHI) with sales (in constant 2006 dollars) from Compustat North America Quarterly. Similar to how I handle institutional ownership, I construct a quarterly series on sales over the previous 12 months at the end of each quarter. I then compute market share over the previous 12 months to compute MHHID at the end of each quarter as well as HHI. The resulting data provide HHI and MHHID of each industry over the previous 12 months at the end of each quarter. 13 The 13F database give institution-level holdings. Thus, the estimated common ownership measure is based on the assumption that investment companies aggregate each fund s voting power at the fund family level, which is reported by (Davis and Kim, 2007). 14 Constructing MHHID without this restriction does not qualitatively change the results. 13

15 Table 1 gives summary statistics for MHHID and HHI. An average (median) industry has MHHID as 1,568 (1,412) and HHI as 3,212 (2,702). Considering common ownership into account greatly increases market power implied by modified Herfindahl Hirschman Index, which underscores the significance of common ownership problem. 3.2 Empirical Specification I use a standard empirical specification to measure pay-performance elasticities used in the literature. I estimate the following panel regression model: log(executive Pay ijt )=k ij + jt Own Perf jt + jt Peer Perf jt + Control Vars ijt + Á ijt, (2) where Executive Pay ijt is the compensation to executive i of firm j at time t. The k ij is the individual fixed e ects for each executive-firm pair. Own Perf jt and Peer Perf jt measure performance of firm j and its peer group respectively. Given the specification, the jt and jt measure the percentage change in pay with respect to the percentage change in the shareholder wealth of own and peer firms. So the jt and jt in equation (2) measure the pay-performance elasticities. I use control variables to capture variations in compensation that are not related to performance measures. In order to estimate the e ects of common ownership on pay-performance elasticities, I specify pay-performance sensitivities as linear functions of common ownership measure: jt = F (MHHID jt ) (3) jt = F (MHHID jt ), (4) where F (MHHID jt ) is the sample cumulative distribution function of common ownership measure in the sample Using sample cumulative distribution function serves as a normalization of common ownership measure to make easy interpretation of regression results. The results are qualitatively unchanged if I use common ownership measure. 14

16 Combining equations (2),(3), and (4), the baseline regression model is log(executive Pay ijt )=k ij + 0 Own Perf jt + 1 F (MHHID jt )Own Perf jt (5) + 0 Peer Perf jt + 1 F (MHHID jt )Peer Perf jt + Control Vars ijt + Á ijt. (6) Following the literature, I control for industry concentration with HHI (Aggarwal and Samwick, 1999a); firm size with the logarithm of sales (Rosen, 1982); idiosyncratic risk with residual variance from regressing individual stock return on the return on value-weighted industry portfolio over the previous 36 months (Core and Guay, 1999; Aggarwal and Samwick, 1999a); growth options with market-to-book value of assets (Core and Guay, 1999; Albuquerque, 2014); regulation dummy with the indicator variable which equals one for the firms in the gas and electric services industries (Albuquerque, 2009); CEO dummy with the indicator variable which equals one if the executive is the CEO 16 ; and executive tenure with the logarithm of tenure defined by the years since the executive joins the firm (Bertrand and Mullainathan, 2001). 17 In addition, I control for common ownership with F (MHHID) to analyze the e ect of common ownership on the level of compensation. In the Online Appendix, I provide a detailed description about the variable construction. I control for year, industry, and individual fixed e ects. Year fixed e ects are used to control for unobservable time series variations in compensation such as business cycles (Murphy, 1999) and changes in regulation of compensation disclosure (Gipper, 2016). Industry fixed e ects capture unobservable variations in compensation due to industry-specific factors such as di erent executive labor markets (Himmelberg and Hubbard, 2000). I use individual fixed e ects as in Albuquerque (2009) to control for unobservable factors of each executive-firm 16 Following Albuquerque (2006), I use the date at which an executive became CEO to identify CEOs in each fiscal year. I drop observations with CEO tenure the di erence between the date an executive became CEO and the data date as in the number of years less than one. Doing so excludes observations without full-year service. 17 I use the logarithm of tenure to account for potential diminishing gains from experience as measured by tenure. If the date which an executive joins the firm is missing, then the number of prior years that the executive appears in ExecuComp is used as in Carter, Franco, and Giné (2016). Using CEO tenure as in Albuquerque (2009) does not change the results. 15

17 pair. Examples of these unobservable factors include risk aversion, individual-specific value of outside options, and tendency to use specific compensation contracts such as a one-dollar salary. Controlling for individual fixed e ects, I exploit the time series variations in common ownership measure within each individual. 18 I cluster all regressions at the firm level because compensation is likely to be correlated within the firm (Albuquerque, 2009; Frydman and Saks, 2010). Following Holmström and Milgrom (1987), the use of RPE implied by the panel regression model is the ratio of peer to own pay-performance elasticity the compensation ratio: jt = F (MHHID jt ) jt F (MHHID jt ). A decrease in compensation ratio which means a bigger negative weight on peer performance relative to own performance could be interpreted as a more use of RPE. Thus, if common ownership reduces the use of RPE for less competition, then the compensation ratio jt / jt becomes less negative, which gives less rewards to executive o cers from outperforming peers. Formally, the influence of common ownership on RPE is tested with the null hypothesis H 0 = ˆ( jt / jt ) ˆF(MHHID jt ) Æ 0, (7) against the alternative, H a : ˆ( jt / jt ) ˆF(MHHID jt ) > 0. Rejecting the null hypothesis is consistent with the hypothesis that executive compensation is the mechanism between common ownership and a decrease in product market competition. I test the hypothesis at the median value of common ownership: F (MHHID jt )= Using individual fixed e ects has two potential caveats. Given the short panel length of each executive, the estimation could lose some statistical power. Indeed, untabulated results with firm fixed e ects that make the panel longer deliver more significant results than those reported in the paper. Second, it is possible that the estimated coe cients purely come from time-series variations rather than cross-sectional variations in common ownership. Untabulated results without individual fixed e ects as in Anton et al. (2016) with my specification deliver qualitatively similar results: common ownership increases RPE. 16

18 3.3 Results In Table 2, I present the main results of the implicit test. Column (1) of Panel A provides the panel regression results without control variables. Pay-performance elasticity to own performance is positive and statistically significant, whereas common ownership does not have a meaningful impact on own pay-performance elasticity. This implies that executives are rewarded for their own firm s performance regardless of common ownership level. Payperformance elasticity to peer performance is statistically indistinguishable from zero in the absence of common ownership. The negative coe cient on the interaction terms between peer performance and common ownership indicates that executives are more punished from peer performance under higher common ownership. Panel B provides the result on the hypothesis test. The test statistic, ˆ( jt / jt ) ˆF(MHHID jt, is negative and statistically significant. ) In sum, the result indicates that RPE implied by compensation increases with common ownership, contradicting the prediction that common ownership reduces RPE to induce less competition. In column (2), I add control variables to the previous specification. With controls, the interaction between peer performance and common ownership remains negative at qualitatively similar statistical significance and levels. The result from the compensation ratio test in Panel B confirms the previous result: the use of RPE is positively related to common ownership. The coe cients on control variables indicate that executives in firms with higher common ownership, bigger firm size, bigger idiosyncratic risk and better growth options receive more compensation. The coe cient on common ownership implies that the level of compensation is positively associated with common ownership. This finding is consistent with evidence reported by Anton et al. (2016). This result is counterintuitive in my case. If common ownership increases the use of RPE, then compensation contracts bear less risks by filtering out industry-specific shock from performance evaluation. The decrease in risks will result in less expected compensation. This finding could potentially stem from endogeneity between compensation and 17

19 common ownership. For example, compensation is higher in industries with more growth opportunities. At the same time, more growth options are associated with higher institutional ownership. If the market-to-book ratio does not properly control for growth options, this may lead to an upward bias on the coe cient on common ownership. The result on RPE is not only statistically significant but also has quantitatively meaningful. To quantitatively interpret the result, I examine how peer pay-performance elasticity changes as common ownership changes from the lowest to highest in column (2) of Table 2 In industries with lowest degree of common ownership (F (MHHID jt )=0), the peer payperformance sensitivity is a mere 0.007, which is statistically indistinguishable from zero. An increase in the peer performance by one standard deviation (0.324) decreases executive pay by = 0.23% under separate ownership. On the other hand, in industries with highest degree of common ownership (F (MHHID jt )=1), the peer pay-performance elasticity becomes An increase in the peer performance by one standard deviation decreases executive pay by 2.4% under common ownership. Thus, executives under common ownership are more severely punished from underperforming peers. In columns (3) and (4), I test if the results survive in the CEO and non-ceo executive subsamples, respectively. With the CEO subsample, the result becomes less statistically significant; the interaction term between peer performance and common ownership as well as compensation ratio test loses some statistical significance but retains its sign. Given that the sample size is significantly smaller, the loss of significance is not very surprising. Column (4) shows that the use of RPE increases with common ownership in the non-ceo subsample. In summary, the evidences show that firms with higher common ownership use more RPE. Executives under common ownership are rewarded more if their firm outperforms its peers. Also, the use of RPE implied by the ratio of peer to own pay-performance elasticity increases with common ownership. The result is inconsistent with the hypothesis that firms under common ownership should use less RPE in order to lessen competition. 18

20 3.4 Robustness The executive compensation literature has shown that the test of RPE using pay-performance sensitivities from Execucomp is sensitive to industry definitions (Albuquerque, 2009). Columns (1) through (4) in Table 3 show that the previous result is robust to di erent industry definitions: three-digit SIC, three-digit SIC divided by size quartiles of Albuquerque (2009), FIC-400, and FIC-500 developed by Hoberg and Phillips (2010), respectively. 19 In all specifications, the interaction term between peer performance and common ownership measure is negative and statistically significant. Compensation ratio tests presented in Panel B corroborate the finding: the use of RPE is positively correlated with common ownership. The significance levels presented with the alternative industry definitions are stronger than the baseline results with four-digit SIC code. The di erences in the statistical significance could be interpreted as a result of the fact that the alternative industry definitions are more representative of product market competitors. As the majority of firms in ExecuComp operate in multiple segments, using three-digit SIC code might reduce the likelihood of assigning firms in incorrect industries. In addition, the size-split SIC codes and the text-based FIC are shown to have better capacities in detecting RPE according to Albuquerque (2009) and Jayaraman, Milbourn, and Seo (2015), which might explain the improvement of statistical significance in the table. I further test the e ects of common ownership on pay-performance elasticities after controlling for other factors such as industry concentration as used in Aggarwal and Samwick (1999a), idiosyncratic risk in Holmström and Milgrom (1987) and Aggarwal and Samwick (1999b), and stock return beta with respect to industry portfolio in Holmström and Milgrom (1987) and Aggarwal and Samwick (1999b). 20 Specifically, I add more interaction 19 Untabulated results with FIC-100, FIC-200, and FIC-300 give qualitatively similar results to FIC-400 and FIC-500: peer pay-performance elasticity decreases in common ownership at the 5% significance level. Results with FIC-25 and FIC-50 retain the sign on the interaction term between peer performance and common ownership while losing significance level, possibly due to too coarse industry definition. 20 I construct stock return beta by regressing individual stock return on the return on value-weighted industry portfolio over the previous 36 months. 19

21 terms, assuming that the pay-performance elasticities jt and jt in Equation (2) are given by jt = F (MHHID jt )+ 2 F (HHI jt )+ 3 Idiosyncratic Risk jt + 4 Beta jt (8) jt = F (MHHID jt )+ 2 F (HHI jt )+ 3 Idiosyncratic Risk jt + 4 Beta jt. (9) Columns (1) through (5) of Table 4 confirm the evidences that the use of RPE is positively associated with common ownership for four-digit SIC, three-digit SIC, size-split four-digit SIC, FIC-400, and FIC-500 industry definitions, respectively. The coe cients on the interaction term between peer performance and common ownership remain negative, which is corroborated with compensation ratio test. The coe cients on peer performance interacted with control variables imply that the pay-performance elasticity to peer performance is negatively related to idiosyncratic risk and stock return beta. It is notable that industry concentration is no longer statistically significant as in Aggarwal and Samwick (1999a), which imply that the e ects of common ownership might be more crucial factor in executive pay than the strategic incentives from competition in industry. 21 The finding is further confirmed with compensation ratio tests presented in Panel B, which are evaluated at the median: F (MHHID)=0.5, F(HHI)=0.5, Idiosyncratic Risk =0.013, Beta =0.95). One remaining concern is whether total flow compensation based on grant-date fair values of equity incentive plans is appropriate to estimate the change in compensation in response to performance. As Panel E in Table 6 points out, a majority of equity-based RPE awards rewards future performance such as three-year stock returns since granted. 22 Thus, the value of equity awards in TDC1 might not measure rewards for performance but reflect the changes in executive equity portfolios to keep their incentives (Core and Guay, 1999). To address these concerns, I use non-equity incentive awards that are reported since Non-equity 21 Note, however, that my results are not readily comparable with Aggarwal and Samwick (1999a) dueto the huge di erences in the sample as well as empirical specifications. 22 De Angelis and Grinstein (2016) also give similar evidences from 2006 proxy disclosures. Only 17% of S&P 500 firms use one-year performance horizon. 20

22 incentive plan compensation could be useful to detect pay-performance elasticities because it is reported if performance conditions are satisfied. 23 Because 29% of executives do not receive non-equity incentive awards, I estimate a Tobit regression model with one-sided limit at zero. Table 5 reports the results on non-equity incentive award. The results are broadly consistent with the previous findings. Columns (1) through (3) report the results with four-digit SIC, three-digit SIC, and size-split four-digit SIC codes. 24 All three specifications present evidences for the strong positive correlation between the use of RPE and common ownership: the pay-performance elasticity to peer performance is negatively correlated with common ownership. Compensation ratio tests reported in Panel B confirm the previous findings. The statistical significance of the results with non-equity incentive awards exceeds that with total flow compensation. This di erence might be due to the fact that non-equity incentive awards are more representative of pay for performance. In columns (4) through (6), I confirm the findings by running the panel regressions with the logarithm of non-equity incentive plan compensation as the dependent variable with four-digit SIC, three-digit SIC, and size-split four-digit SIC codes. Hence, the results presented in columns (1) through (3) are not driven by whether executives receive non-equity incentive plan compensation but driven by how the level of non-equity incentive changes with performance. In sum, the results from the implicit approach suggest that the use of RPE is positively related to common ownership. The results are robust with respect to di erent industry definitions and compensation measures. I also test if the e ects of common ownership are present after controlling for other determinants of pay-performance elasticities such as industry concentration, idiosyncratic risk, or stock beta with respect to industry portfolio. The finding suggests that compensation is not the mechanism between common ownership and decreased product market competition. Despite the convincing results, there are concerns 23 I don t include bonus because it is likely to be a discretionary award. Using the sum of non-equity incentive awards and bonus provide qualitatively similar results. 24 The results with FIC-400 and FIC-500 are omitted for the sake of brevity. Using FIC industry specifications yields qualitatively similar results. 21

23 about the implicit approach. As Bebchuk and Fried (2004) suggest, incentives from compensation work only if executives recognize the incentives. Thus, the implied incentives from realized pay might not be as e cient as contractual details which executives have a full understanding of. Another concern is the endogeneity problem between compensation and performance (Frydman and Jenter, 2010); the correlation between executive pay and performance may appear because performance a ects compensation through incentive plan, or because compensation a ects performance through executive decision-making. Thus, even if I control for the endogeneity of common ownership measure, the resulting outcome does not necessarily imply that firms under common ownership are more likely to use contracts that give more or less incentives to compete. To address these concerns, I study the e ects of common ownership on the incentive awards explicitly disclosed in proxy disclosures. 4 Explicit Approach In this section, I present my main results with the explicit approach. I document that common ownership is positively associated with the use of RPE and the size of awards tied with RPE. In addition, I test how common ownership is related to performance benchmarks with which firms compare their performance. Specifically, I find that common ownership is positively associated with the use of industry index. These findings corroborate the previous results in the implicit approach. 4.1 Data RPE from Proxy Disclosures Since the introduction of Compensation Discussion and Analysis in 2006, firms have been required to provide details on performance-based awards. For the incentive awards evaluated with performance benchmarking, the firms disclose performance benchmarks, metrics, and targets that are used to determine the size and vesting conditions of awards. To test the e ects of common ownership on the use of RPE, I collect the executive contract details 22