Are All CEOs Above Average? An Empirical Analysis of Compensation Peer Groups and Pay Design.

Transcription

1 Are All CEOs Above Average? An Empirical Analysis of Compensation Peer Groups and Pay Design. John Bizjak Portland State University Michael Lemmon University of Utah Thanh Nguyen University of Utah Abstract Critics contend that the use of compensation peer groups has resulted in inflated CEO pay that cannot be justified based on economic fundamentals. We examine this issue using the mandated disclosure of compensation peers that began in Although firms generally select compensation peers based on characteristics that reflect the labor market for managerial talent, we find evidence that the selected peers are chosen in a manner that biases compensation upward. The bias in peer group selection is unrelated to corporate governance and instead appears to be an institutionalized or structural part of the pay setting process. We provide some preliminary evidence that increased disclosure has reduced the biases in peer group choice. JEL classifications: G34, J31, J33 Keywords: executive compensation, benchmarking, peer groups, CEO pay

2 1. Introduction When shareholders question lush pay, they are invariably met with a laundry list of reasons that businesses use to justify such packages. Among that data, no item is more crucial than the peer group, a collection of companies that corporations measure themselves against when calculating compensation. 1 Arguably few economic topics stir as much passion, controversy, and debate as CEO pay levels. As the above quote illustrates, for many firms one of the driving factors in setting both levels of pay and pay structure is the use of comparison or peer groups. The use of peer groups to set pay, however, is not without controversy. One of the biggest concerns is that peer groups can be used to justify inflated pay levels. For example, according to RiskMetrics, the compensation peer group used in 2007 by the hairstyling company Regis Corp, which owns Vidal Sassoon and Supercuts, included Starbucks and H&R Block - firms that are much larger, in different industries, and with significantly higher CEO pay than Regis. 2 In general, critics of the use of peer group benchmarking argue that powerful CEOs and co-opted boards opportunistically choose peer firms in a way that inflates CEO pay. Moreover, the critics also contend that, given the prevalence of benchmarking, the opportunistic choice of peer firms has led to an upward ratcheting of pay levels over time. Alternatively, the use of peers group benchmarking can play an important economic role in the pay process. Properly structured, peer groups provide information to boards of directors for determining the competitive pay level that is necessary to both retain and motivate top executives. As Holmstrom and Kaplan (2003) point out it is hard to see how pay levels can be set in a fair and efficient way without benchmarking. As in any other labor market, the forces of supply and demand are an important determinant of wages for the CEO and other executives. 1 Peer pressure: Inflating executive pay, by Gretchen Morgenson, New York Times, November 26,

3 One effective way to gather information on prevailing market wages is to compare the salary of executives at one firm with those at other firms. In fact the importance of the use of peer groups in determining competitive wages led Holmstrom and Kaplan to argue that we need more effective benchmarking not less of it. Recently the SEC adopted new proxy disclosure rules that require firms to report the peer groups they use to set managerial compensation as long as the use of peer groups is material in determining pay. 3 To provide insight into the role that peer groups play in compensation design we gather data on compensation peers for firms in the Execucomp database in 2006 (2007 proxy year). Using this data we examine how peer groups are structured and the role that peer groups and competitive benchmarking play in the pay setting process. Our analysis provides evidence regarding whether competitive benchmarking is used opportunistically (i.e., to inflate CEO pay) or whether peer groups provide the firm with valuable information to determine the relevant market wage for the CEO. We find that the use of peer groups in setting pay is quite common. For the sample of firms in the Execucomp database that have a fiscal year end in December 2006 or later, nearly 69% (800 firms) report their compensation peer groups. 4 Consistent with the results in Bizjak, Lemmon, and Naveen (2008), we find that the level of CEO pay relative to that of the peer group median firm has a significant effect on subsequent changes in CEO pay. 5 3 The new SEC disclosure requirement became effect for fiscal years on or after December 15, See SEC final rules a, Item 402(b)(2)(xiv). 4 Firms that do not report the use of peer groups appear to be smaller and have lower pay than firms that report peers. In addition, firms that do not report peers have more concentrated ownership by the CEO, more insiders on the board of directors, and in general appear to be more closely held and controlled. These results suggest that firms not reporting the use of peer groups are ones where the CEO is more likely to control the pay process and correspondingly not as likely to rely as much on peers or compensation consultants in setting pay. 5 Bizjak et al did not have the actual peer group used by the firm for setting compensation. They constructed a naïve peer group of firms in the same industry similar in size. Our results suggest that the use of the actual peer group provides additional explanatory power for changes in pay. 2

4 Given the fact that peer group benchmarking is an important determinant of CEO pay, we turn our attention to examining how firms determine the composition of their peer groups. We estimate logit regressions to identify firm and industry characteristics that drive the choice of peer firms. As explanatory variables we use a number of different factors that we expect to proxy for supply and demand conditions in the labor market for managerial talent. Consistent with the description of the compensation process in Murphy (1999), with the analysis in Bizjak, Lemmon, and Naveen (2008), and with the discussion in corporate proxy statements, we find that firms tend to pick peers that are of similar size and firms in the same industry. In our sample, 62% of peer firms come from the same industry (based on the Fama and French 49 industry definitions), and 36% of peer firms in the same industry have revenues between 50% and 200% of those of the sample firms. Despite the importance of firm size and industry, we also identify a number of other factors that influence the choice of peer firms. For example, firms are also more likely to choose peer firms with similar accounting performance and similar market-to-book ratios. We find, however, that the effects of both relative size and performance are asymmetric. All else equal, firms are more likely to choose larger and better performing peers compared to potential peer firms that are smaller and have worse relative performance. Also consistent with the labor market explanation for peer group selection, more complex organizations such as diversified firms are likely to choose other multi-segment or geographically diversified firms as peers. Firms also tend to draw peers from industries that have higher stock return correlations with the firm s own industry and where the firm s industry purchases a large fraction of its inputs from the peer firm industry, suggesting that economic linkages along the supply chain are also important determinants of peer firm choice. Firms are also more likely to choose peer firms with similar credit ratings, similar geographic or product 3

5 diversity, and firms in the S&P 500 are more likely to choose other S&P 500 firms as peers. Finally, firms are more likely to choose peers from industries which either supply or hire managerial talent from the firm s own industry. While our results indicate that on average peer firms are chosen in a manner that reflects conditions in the managerial labor market, the selection of the peer group is fundamentally a decision of the board of directors. Moreover, since most firms target executive compensation at the peer group median, firms can choose peers largely based on economic considerations, but in a manner that biases the peer group median compensation upward. For example, firms could pick peers in the same industry but choose slightly larger and better performing firms (since pay levels and firm size and performance are correlated) or firms that have higher than average industry pay. Consistent with this view, we find that the actual peer firms are systematically larger, have better performance, and have more highly paid executives than the sample firms. For example, the median firm in the actual peer group is approximately 18% larger and has total pay that is about 7% higher than the sample firm. Interestingly, the sources of bias differ significantly across S&P 500 firms and those outside of the S&P 500. For S&P 500 firms, the median peer firm is similar in size and has similar pay levels to the sample firm. In non-s&p 500 firms, however, the actual peer firms are 27% larger and have total pay that is 15% higher compared to the sample firms. To provide additional insight into the sources of bias in peer group choice, we employ our logit model to identify a benchmark peer group using propensity-score matching and compare the characteristics of this benchmark (i.e., the propensity-score matched peer group) to those of the actual peers. We find significant deviations between the benchmark peers and the 4

6 actual peers. For S&P 500 firms, the median firm in the benchmark peer group is 7.5% smaller and has total pay that is 9.3% lower compared to the sample firm. Firms in the S&P 500 appear to choose peers in a manner that justifies their high pay, but not in a way that would lead to substantial pay raises, on average. For non-s&p 500 firms, the median firm in the benchmark peer group is 19% larger and has total pay that is 11% higher compared to the sample firm. For non-s&p 500 firms, the difference in compensation between the actual peers and the sample firms reflects two sources of bias in peer group choice. The first is the bias that arises from the fact that the sample firms systematically choose peers that are larger and have better performance. The second is the bias that arises from choosing more highly compensated peers, after holding differences in size and performance equal. Smaller firms perhaps have a greater ability to opportunistically select peers because they have a greater number of potential peer firms both inside and outside the industry that are larger and have higher levels of compensation compared to S&P 500 firms. In contrast, S&P 500 firms are already some of the largest firms in an industry with some of the highest levels of compensation and may find it more difficult to opportunistically select larger and more highly compensated peers. Moreover, S&P 500 firms are more visible and may attract more scrutiny from outsiders, which limits the extent of opportunism in these firms. Another important feature of how peer group benchmarking is used in practice is the targeting of pay levels to a specific percentile (usually the median) of the pay distribution of the peer group. This feature of the benchmarking process suggests that it may be relatively easy to manipulate the relevant characteristics of the peer group in a manner that inflates pay with only small changes in peer group composition. We explore this possibility by ranking the actual peer firms for each sample firm in terms of total compensation from low to high and selecting the 5

7 peer firm that is one firm below the median firm in the pay distribution (we refer to this as the sub-median peer firm). The results show that simply moving down one firm in the peer group pay distribution eliminates any evidence of upward biases in pay. We also explore the extent to which the biases in peer group choice vary systematically with the quality of corporate governance. Using a variety of governance measures we find little evidence that the propensity to choose peer firms in an opportunistic manner varies systematically with corporate governance. Instead, the findings suggest that the opportunism in peer group choice seems to be an institutionalized part of the pay process. In other words, there appears to be a systematic inclination on the part of firms to select peers opportunistically that is unrelated to differences in corporate governance. Finally, we provide some evidence on whether the new reporting requirements have affected how peer groups are chosen. We do this by examining how firms change the composition of their peer group in the year following the new disclosure rule. Firms making large changes to their peer groups in 2007 appear to do so in a manner that reduce the biases in size and compensation that were evident in 2006 (the first year in which the new regulations became effective). This finding provides some evidence that the new regulation has prompted firms to be less opportunistic in the selection of firms to include in the compensation peer group. Our work has direct implications for explaining the dramatic rise in CEO pay that has occurred over the last twenty years. Recent research has tied the increase in CEO pay to changes in the need for managerial talent. Murphy and Zabojnik (2008) argue that CEO pay has increased because of a change in the skill set necessary to run a company. They argue that equilibrium wages have increased over time due to a need for more general managerial skills by the CEO and competition among employers for these skills. Gabaix & Landier (2008) develop a 6

8 model that ties the increase in CEO pay to a shift in the size and organizational complexity of companies in the economy. We offer an additional explanation. To the extent that firms systematically choose peer groups with higher pay and given that firms generally target the median level of pay or higher, these twin effects can over time lead to a ratcheting of compensation. As has been colorfully articulated executive pay is often set with methods imported from Lake Wobegon, Garrison Keillor s imaginary place where all children are above average. 6 Our results suggest that the opportunistic use of peer group benchmarking is an additional factor that has contributed to the dramatic observed increase in CEO pay that has occurred over the last 20 years. Our work is also related to several largely contemporaneous papers that also examine the effect of peer group benchmarking on CEO pay. Similar to our results, Faulkender and Yang (2009) also find some evidence that peer groups are chosen in a way that inflates CEO pay. Albuquerque, De Franco, and Verdi (2009) and Cadman, Carter, and Semida (2009) argue that peer groups are chosen largely based on labor market characteristics, and that there is only limited evidence that firms strategically choose peer firms in order to inflate CEO pay. We differ from these studies by examining a more extensive set of labor market factors that are associated with peer firm choice, by exploring how biases in peer group choice differ across S&P 500 firms and other firms, and by providing evidence on changes in peer group composition in response to the regulation. The remainder of the paper is organized as follows. Section 2 describes the data used in the analysis. Section 3 provides a discussion of how peer groups are used by boards to set compensation and lays out our general research questions and hypotheses. Section 4 examines 6 Executive Cash Machine: How a Pliable System Inflates Pay Levels by Adam Bryant, New York Times, November

9 how peer groups affect managerial compensation. Section 5 examines how firms choose peer groups and provides evidence on whether peers are chosen opportunistically or to provide information on managerial labor markets. Section 6 provides further analysis of opportunistic peer group selection. Section 7 examines the effectiveness of the regulation by analyzing changes the composition of peer groups from 2006 to Section 8 concludes with a brief a summary and discussion. 2. Data Collection and Summary Statistics We begin with the list of firms with compensation information reported in Execucomp with fiscal year ends between December 2006 and May For each of these firms, we examine the corporate proxy statement for 2007 and gather information on the peer groups reported by the firm that are used to set executive compensation along with the name of any compensation consultants the firm employs. Wherever possible, we collect compensation data for all firms in the peer groups. 8 When peer group firms are not part of the Execucomp database we collect the compensation data from corporate proxy statements. 9 We collect data on the governance characteristics of the firm, including board size, board structure, whether or not the CEO is also Chair of the Board, and other items, from the IRRC database, and we collect financial data from the CRSP and COMPUSTAT databases. 7 Firms were required to report information on their compensation peer groups for filings at companies with fiscal year ends on or after December 15 th Prior to 2006 some firms did voluntarily report the use of their peer groups, but it was rare. For example, Faulkender et al (2008) report that only 83 firms in the S&P 500 reported their peer group in We cannot obtain data for a small fraction of peer firms that are either foreign or private firms. 9 Note that the reporting format for compensation changed in 2006, and that the reporting changes affect the way in which total compensation is reported in Execucomp. In most of our analysis, we focus on compensation data for 2005 (2006 proxy year), and thus rely on data prior to the reporting changes. In some analysis, we do examine changes in compensation from 2005 to In these cases, the comparisons may partially reflect the reporting changes rather than real changes in compensation. The changes in total compensation as reported by Execucomp are described at 8

10 Out of 1,161 firms in the ExecuComp database for which we obtained proxy data for the year 2007 (2006 fiscal year), we find that 800 firms report the peer group they use for setting executive pay. The remaining 361 firms do not report the use of peer groups. For the S&P 500 firms we find that 307 report the compensation peer group. 198 of the S&P midcap firms in the sample report their compensation peer group, and 202 of the S&P small cap firms report the composition of their peer group. 93 of the firms not currently contained in those indexes but covered by ExecuComp report their compensation peer group. Table 1 reports summary statistics comparing the characteristics of firms that report peer groups to those that do not report using a peer group. Firms that do not report compensation peer groups are about half as large as firms reporting peer groups. In addition, these firms tend to have higher CEO ownership, longer CEO tenure and slightly more insiders on the board of directors. These firms appear to be more closely held and are firms where the CEO potentially has a greater control over compensation. For these types of firms compensation peers may be less crucial in determining compensation. Finally, firms that do not report peers have lower levels of pay. The median salary and bonus and total compensation are also lower in firms that do not report compensation peers. 3. The Compensation Process and Peer Group Benchmarking For many publicly traded companies compensation is set by the firm s compensation committee, which is comprised of members of the board of directors. In many cases, the compensation committee selects a peer or comparison group of firms that it uses to gather information on pay practices and pay levels. Commonly, but not always, the company will use an outside compensation consultant to construct a peer group and provide information on pay 9

11 levels and pay practices within the industry and at the peer firms. In most firms, salary, bonuses, option pay, and total compensation are in some form anchored to the peer group. Firms typically target the various components of pay at the median pay level of the comparator group. It is not uncommon, however, for firms to target pay above the median (e.g., at the 75 th percentile). 10 While the compensation committee is primarily responsible for selecting compensation consultants and in deciding on the compensation peer group it is not unusual for managers to participate in the selection of peers. Boards often rely on managerial input on the company s business strategy, corporate goals and economic condition. Managers also provide information on the hiring needs and labor market conditions the firm faces. Typically peer groups are formed following the selection of an initial set of peers by a compensation consultant and input from management and the board. Often adjustments are made to the peer group after consultation with management. Understanding how firms choose compensation peers provides insight into the role they play in determining managerial compensation. 3.1 Peer groups as a gauge of managerial labor markets If managerial ability is an important factor in determining firm performance then the ability to retain and motivate the management team is critical to a firm s success. The nature of the managerial labor market plays an important role in assessing the amount of compensation that is necessary to retain valuable human capital. Potentially one of the most important ways that a board can get information on the labor market for its executives is to look at compensation practices in firms that it competes with for managerial talent. If peer group benchmarking is used to evaluate the reservation wage necessary to attract and retain qualified managers, then we 10 According to RiskMetrics Group 99.5% of firms in the S&P 1500 targeted pay at our above the median of their peer group. 10

12 expect peer firms to be selected based on factors related to supply and demand conditions in the managerial labor market. Consistent with this view, corporate proxy statements frequently mention that the purpose of compensation peer groups is to provide information on the labor market for executives. For example, the 2007 proxy statement of ATC Technologies Inc. states that the primary criteria for the selection of compensation peers is to provide a broad view of the executive labor market against which we compete for executive-level talent. In what follows, we discuss a number of characteristics that should proxy for commonalities in the managerial labor market between the firm and its selected peers. Industry. A natural source for compensation peers are firms in the same industry. Firms that provide similar products and that are competitors for customers and revenue are also likely to be firms where a company will look to when trying to recruit executives. Corporate proxy statements also indicate that industry is an important factor when picking compensation peers. For example, the 2007 proxy statement for Fossil Inc. states that their compensation program is designed to be competitive with the companies in the industry in which we must vie for talent. Firm Size: Firm size is an indicator of organizational complexity and scope. Consequently we expect firms to select as compensation peers other firms that are similar in size. A reading of corporate proxy statements supports the notion that firm size plays an important role in peer selection. For example, Fossil Inc. states that the peer group is comprised such that the median revenue size of the peer group is at or close to our annual revenue. While size and industry are important factors in the selection of peer groups, firms often go outside their own industry when picking peers. For example, the 2007 proxy statement of Biogen states The named peer group is reviewed annually by the Committee for 11

13 appropriateness, considering such factors as size (e.g., revenue and market capitalization), complexity (e.g., multiple marketed products), geographic scope of operations (e.g., global versus domestic only presence), etc. There are a number of reasons a firm may not pick peers from the same industry. Large firms that dominate a particular industry may choose not to include other peers within the industry if the firm is significantly larger than its industry competitors. The same could also be said for the smaller firms within an industry. Other relevant labor market factors include: Firm Performance: Firm performance may be used to identify firms for inclusion in the compensation peer group. Firms with similar market-to-book ratios may share similarities in profit models or organizational structure (Smith and Watts (1992)), and firms with similar profitability may be exposed to similar demand shocks. Customers and Suppliers: A firm s customers and suppliers are familiar with the firm s operational structure and serve as a natural source for recruiting executives. For example, executives at auto parts suppliers may have a thorough understanding of automobile manufacturing and provide for a pool of talent that automobile manufacturers can hire from. Pharmaceutical companies might hire from hospitals where they sell their products. To the degree that a firm s customers and suppliers provide a resource for managerial talent we would expect that these firms would be appropriate for selection as part of the compensation peer group. Capital Markets: Companies that are competitors for equity or other types of financial capital may also serve as relevant peers. For example, the 2007 proxy statement for Avon Products Inc states that peers are selected based on the fact that the Company competes with these organizations for employees, customers and shareholders (bold added). For firms that are 12

14 dominant in an industry or that have few direct competitors, another potential source of peer firms comes from other companies that investors might view as substitutes in their portfolios. For example the top ten firms in the S&P 500 have few organizations in the same industry that are similar in size and subsequently operating characteristics. While these firms may not be direct competitors, one thing they have in common is the need to raise and manage enormous sums of capital. These firms may look to each other as peers or to other firms where they compete for equity capital. Diversified Firms: Diversified companies are often more complex organizations that require a specific set of managerial skills. Because of this, we expect that diversified firms will be more likely to look to other diversified firms for executive talent and to be more likely to include these firms in their peer group. Diversification may be measured across either product lines or across geographic regions. 3.2 Peer groups and opportunism While the use of compensation peer groups can provide valuable information to boards for determining appropriate compensation, the nature of the process by which peer firms are chosen leaves scope for the possibility that the peer group will be chosen in order to justify higher pay. There are number of ways that managers can potentially influence the peer group selection process in a manner that enhances compensation. 11 Since it is well documented that pay is correlated with firm size, executives can justify or seek higher pay levels if they select firms in the peer group that are larger. Besides putting larger firms in the peer group, firms can 11 The 2007 proxy statement of Schnitzer Inc. discuss the role that the CEO and management plays in the selection of peers: The CEO, with the assistance of Towers Perrin, analyzes survey data and makes recommendations to the Committee regarding compensation for the executive officers. The CEO participates in Committee meetings at the Committee s request to provide background information regarding the Company s strategic objectives and his evaluation of the performance of and compensation recommendations for the other executive officers. With respect to his own compensation, the CEO responds to requests from the Committee. 13

15 directly pick peers with higher levels of pay. For example, ValueVision Media, a firm that operates shopping network ShopNBC, with a market capitalization of $143 million includes Amazon and Ebay in its peer group. While all three companies provide product deliveries online to customers, both Amazon and Ebay are considerably larger than ValueVision and have CEOs with much higher levels of compensation. It seems unlikely that candidates for the CEO position at Amazon, Ebay or firms of similar size to these two companies would also be candidates for the CEO position at ValueVision. Moreover, the rules that companies use to justify inclusion of a firm into a peer group can be amorphous. According to the 2007 proxy statement of Best Buy Inc, the firm selects peers because of Admiration within their industry and those that have a track record of innovation. While both qualities might be admirable, this type of selection criteria can lead to opportunistic behavior on the part of management. Finally, the fact that compensation is generally benchmarked to the median firm in the peer group suggests that it may be relatively easy to manipulate the relevant characteristics of the peer group in a manner that inflates pay with only small changes in peer group composition. If managers are opportunistically selecting firms in the peer group in order to inflate pay we expect that managers will systematically select peers firms that are larger and have higher levels of compensation. Moreover, if managers are using peer groups opportunistically, we anticipate that the composition of the peer group will deviate from the peer group that would be chosen purely based on the economic criteria discussed above that describe the relevant managerial labor market. 3.3 Preliminary evidence on the selection of peer firms 14

16 Table 2 reports information on the size of the peer group and the fraction of peer firms in the same industry. The average (median) size of the peer group is around 16 (14) firms. S&P 500 firms include more firms in their peer groups compared to firms not in the S&P 500, but the differences are small. The majority of firms in the peer group come from the same industry. Using the Fama-French (1997) 49-industry classification, we find that on average (median) 62% (70%) of firms in the peer group are in the same industry as the firm. 12 Small cap and Mid cap firms select more firms from the same industry compared to firms in the S&P 500. To get a feel for how size also affects the selection of peer firms, we examine the fraction of peer firms in the same industry that also have sales revenue between 50% and 200% of that of the firm which is a common criteria stated in proxy statements. Using this taxonomy we find that on average 36% of peer firms are in the same industry-size classification. Firms outside of the S&P 500 tend to have more of their peers chosen from the same industry-size group. Table 3 compares size, accounting performance, and compensation between the firm and its peer group. We report results for the entire sample in Panel A, and for S&P 500 and non- S&P 500 firms in Panels B and C, respectively. For the full sample, Panel A, we find that peer firms are close in size (i.e., sales revenue) to the firm on average, but that the median peer firm tends to be larger. Selected peers also tend to have better accounting performance. In terms of pay, we find that differences in the level of pay between the firm and its peer group are generally not significant. What stands out, however, are the differences in the characteristics of peer firms that are outside the firms industry. The analysis in Panel A of Table 3 shows clear evidence that when firms go outside their industry to select peers, the peer firms tend to be larger and have higher levels of compensation at least at the median. 12 Faulkender and Yang (2008) report an average (median) size of peer groups to be 18 (16). Using two-digit SIC code, they find that the average (median) fraction of firms in the peer group that are in the same industry as the sample firm is 46% (44%). 15

17 Panels B and C show that these differences are concentrated in non-s&p 500 firms. S&P 500 firms tend to pick peers that are similar in size and with similar levels of compensation. This result holds even for peer firms outside the firm s industry although at the median, peer firms outside the industry tend to be slightly larger. In contrast, non-s&p 500 firms tend to pick peers that are larger with higher levels of compensation. In particular, peer firms that are not in the same industry are bigger, both on average and at the median, and have higher levels of salary and bonus and higher median total compensation. While not definitive, the results suggest that, smaller firms in particular, tend to pick peers in a manner that could lead to unjustified pay increases. 4. The Effect of Peer Groups on CEO Pay To provide some motivation for our analysis on how firms construct peer groups we begin with an analysis of how the use of peer groups affects compensation. Our analysis follows Bizjak et al (2008) and is designed to mimic the way in which firms use peer groups to benchmark compensation as described in Murphy (1999). We begin with a simple empirical model to estimate factors that are related to changes in CEO pay. The results are reported in Table 4. The dependent variable is the change in the log of total compensation between 2005 and Independent variables include log sales, change in log sales (2005 to 2006) and firm volatility. Also included are measures of current and prior stock price and accounting performance. In model 1, The R-squared of the regression is 4.5%. Changes in pay are positively related to current stock and accounting performance, although only 13 In the regressions we use ln(1+ total compensation) because eight observations in our data report zero total compensation. Excluding these observations has no effect on the inferences from the analysis. 16

18 the coefficient on stock returns is statistically significant. The remaining variables in the regression are not statistically significant. To assess the effect of peer group benchmarking on pay changes we introduce a naïve peer group (firms of similar size and the same industry) into the empirical model. The naïve peer group is based on industry and size and is constructed as in Bizjak et al (2008). For each firm we compute the difference in log compensation between the median firm in the peer group and the sample firm based on compensation data in This distance measure thus captures how much the manager is paid relative to the median firm in the peer group. The coefficient estimate on this variable measures how the firm adjusts the manager s compensation in 2006 as a function of the manager s pay relative to the peer group. We include this specification to provide a comparison to the results in Bizjak et al (2008). Including the naïve peer group distance measure in model 2 significantly increases the R- squared of the regression to 24.4% and the coefficient estimate on the peer group distance variable indicates that a manager with pay 1% below the peer group median pay level receives an increase in pay that is approximately 0.5% larger compared to a manager with pay equal to the peer group median. In model 3, we substitute the difference in pay between the median firm in the actual peer group and the sample firm for the pay difference measured relative to the naïve peer group. The R-squared of the regression increases to 33.0% and the coefficient estimate on the peer group distance measure indicates that a manager with pay 1% below the actual peer group median receives an increase in pay that is 0.6% larger compared to a manager with pay equal to the peer group median. The results are consistent with Bizjak et al (2008) and indicate that their naïve measure of peer group benchmarking provides a reasonable approximation of the effects of peer groups on compensation. More importantly, the results clearly show that the use 17

19 of peer group benchmarking has a first order effect on CEO pay, suggesting that analyzing the peer group selection process in more detail is important for furthering our understanding of how executive pay is determined. 5. The Selection of Firms for the Peer Group To provide insight into how companies select peer groups, we estimate a multivariate logit regression to identify the factors that firms use when selecting peer firms. The dependent variable is one if the firm is selected as a member of the compensation peer group at a particular firm and zero otherwise. Based on the discussion in Section 2, the independent variables in the regression attempt to capture firm and industry traits that should reflect supply and demand conditions in the managerial labor market. The independent variables include an indicator equal to one if the firm and the potential peer share the same Fama-French 49 industry classification. To assess how firm size affects peer firm selection we include two relative size variables. The first is equal to the difference in log sales between the potential peer and the firm when this difference is positive and is set equal to zero otherwise. The second is equal to the difference in log sales between the potential peer and the firm, when the distance is negative, and zero otherwise. This specification allows for asymmetry in how relative size affects the choice of peer firms. 14 We form similar asymmetric measures for relative accounting performance and for market-to-book. Allowing for asymmetry in the size and performance measures allows us to examine whether firms have a tendency to pick peer firms that are larger and with better performance. 14 Note that by using the difference in log sales we are essentially examining differences in relative size in percentage terms, which is consistent with how firms describe peer group choice in the proxy statements. 18

20 To capture commonalities across industries, we include the correlation of returns between the firm s industry and the potential peer firm s industry, where the correlation is measured using industry daily returns over the period We also include a measure of whether the potential peer firm is in an industry with significant customer or supplier relationships with the firm s own industry. To measure supply chain relationships, we use the commodity flow data from the input-output (IO) tables provided by the Bureau of Economic Analysis (BEA). 16 For each firm and potential peer, we compute the fraction of output that the firm s industry sells to the potential peer s industry and the fraction of input that the firm s industry purchases from the potential peer s industry (Lemelin (1982) and Fan and Lang (2000)). To capture labor flows between industries we compute the fraction of external hires in the firm s industry that come from or leave for the potential peer firm s industry. We compute this measure using data from the five year period 2001 through To proxy for other potential labor market linkages, we include a dummy variable equal to one if the firm and the potential peer are both in the S&P 500. We conjecture that S&P 500 firms are more likely to compete with other firms in the S&P 500 for executive talent. We include a dummy equal to one if the firm and potential peer share the same credit rating to proxy for common capital market characteristics. Finally, to capture commonalities in business complexity, we include dummy variables equal to one if both the firm and the potential peer report multiple business segments or multiple geographic segments. The variable definitions are provided in the appendix. Finally, the regression also includes firm fixed effects to account for 15 We obtain data on industry returns from Ken French s website 16 The website for the I/O data is 17 To gather information on the firms/industries that the company either hired or lost talent to, we examined executive turnover over the last 5 years using the Execucomp data. Note that this measure does not capture labor movements across firms not included in the Execucomp data, and is therefore a noisy measure of labor flows. 19

21 the differences across firms in the unconditional probabilities of a firm being chosen as a peer firm that arise because the size of the peer group differs across firms. We report two different specifications for robustness. In the first specification, the set of potential peer firms include all sample firms and all of the actual peer firms disclosed. In the second specification, we do not put any restrictions on the potential pool of peers and include all firms in the Compustat Universe. Both specifications include firm fixed-effects to allow for differences in the unconditional probabilities of a given firm being chosen as a peer that reflect differences in peer group size across firms. The results of both models are presented in Table 5. The p-values based on robust standard errors clustered at the firm level are reported in parentheses below the coefficient estimates and the marginal effects are reported in brackets. 18 Focusing on the first specification in Table 5, all of the coefficient estimates in the regression are statistically significant and nearly all of them have the expected sign. Moreover, many of the effects are economically large. The results indicate that firms in the same industry are 39% more likely to be chosen as peer firms, and when firms do go outside the industry they tend to select firms from industries that have higher stock return correlations with their own industry. With respect to firm size, the likelihood of a given firm being chosen as a compensation peer declines as the absolute difference in log sales between the firm and the potential peer gets larger. The effect, however, is asymmetric. Conditional on the potential peer firm being larger (smaller) than the firm of interest, the marginal effect indicates that a one unit increase (decrease) in relative size decreases the probability of the firm being chosen as a peer by 18 The marginal effects reported in brackets in Table 5 are calculated as the partial derivative of the event probability and are computed at the following values of covariates: all dummy variable (1, 13-16) set to 1; all distance variables (3-8) set to 0; the remaining variables (2, 9-12) set at the respective means. 20

22 15% (35%). 19 Thus, although firms are less likely to choose firms as peers that are either larger or smaller than themselves, when they do choose peers that are different in size, they are more likely to choose larger firms as peers and less likely to choose smaller firms. Similar results hold for the variables measuring relative performance and market-to-book. Firms are more likely to choose peers that have better performance and higher market-to-book ratios compared to potential peers with lower relative performance. Firms are not more likely to choose customers as peer firms, but are more likely to choose suppliers. In addition, peers are more likely to come from industries which either hire or supply CEO candidates to the firm s industry. Other notable findings are that firms in the S&P 500 are more likely to choose peers that are also in the S&P 500, and diversified firms are more likely to choose other diversified firms. Firms are also more likely to select peers that share the same credit rating. Overall the results from both model specifications are consistent with the discussion above on the firm, financial, and industry characteristics that characterize the labor market for executive talent. The analysis suggests that, for the most part, when firms pick peer groups the goal is to identify other firms that share common labor market characteristics. The logit analysis, however, also suggests some potentially opportunistic behavior in the selection of peers. In particular the asymmetric effects of firm size and firm performance indicate that firms tend to favor larger and better performing firms when choosing the peer group. In the next section we investigate more directly whether there are opportunistic biases in the peer group selection process. 19 Note that for the coefficient α 4, the independent variable (difference in log size between the peer and the firm conditional on the firm being larger than the potential peer) is negative, and thus a positive coefficient estimate indicates that the probability that a particular firm is chosen as a peer is decreasing as the firm gets larger than the potential peer. 21

23 6. Peer group selection and the bias in CEO pay. To provide additional insight into whether firms choose peers opportunistically, we use the coefficient estimates from Model 1 in logit regressions estimated in Table 5 to identify a benchmark peer group and compare the characteristics of this benchmark (we refer to the benchmark as the matched peer firms) to those of the actual peers and to the sample firm. To form this benchmark peer group, we compute the propensity score (i.e. the predicted probability of being chosen as a peer firm) for each firm in the actual peer group based on its characteristics from the logit model. 20 We then match each firm in the actual peer group firm to a potential peer firm that that has the closest propensity score to the actual peer firm. For each sample firm the set of potential peer firms includes all of the sample firms and their selected peers. 21 The matching is done without replacement. To compare the actual and benchmark peers, we focus on differences in sales revenue, accounting performance, and compensation. Specifically for each sample firm we compare the characteristics of the median firm (in terms of total compensation) within each peer group (the actual and benchmark peer group) with the sample firm. We also report the differences in characteristics between the median compensation firms in each peer group. The results are presented in Table 6. Panel A reports results for the whole sample, and Panels B and C report the results for S&P 500 and non-s&p 500 firms, respectively. For the sake of brevity the table reports medians, however, inferences based on means are similar. For the full sample (Panel A), the median peer firm in the actual peer group is about 18% larger than 20 For a description of Propensity Score Matching see Wooldridge (2002, Ch. 18). 21 In our main analysis we also include the remaining firms in the actual peer group in the set of potential peers (other than the peer firm being matched). The results are similar if we instead exclude all firms in the actual peer group before matching. 22

24 the sample firm. The median propensity score matched peer firm is also larger than the sample firm, but the size bias is smaller. This size bias arises from the asymmetric effect of size on the likelihood that a given peer will be chosen as documented in the logit model. Similarly, firms tend to choose peer firms with better accounting performance. This difference in performance is not reflected in the matched peers, however. In Panel A, the size and performance differences between the actual peers and the matched peers are statistically significant, although the magnitudes of the differences are small. The remaining rows in the table report differences in compensation both in logs and in dollars. The median peer firm has total compensation that is higher than the sample firm by 6.7%, or equivalently $249,000. In contrast, the differences in pay levels between the median matched firms and the sample firms are small. Overall, the pay at the median actual peer firm exceeds the pay at the median benchmark peer firm by about 4%, or approximately $153,000. Thus, even accounting for the fact that the actual peers are larger and have better performance than the sample firms, it appears that firms choose peers with systematically higher compensation than themselves. Panels B and C present the results separately for S&P 500 and non-s&p 500 firms and reveal some notable differences between the two groups. For the S&P 500 firms (Panel B), the actual peer firms are similar in size, have better performance, and have similar levels of compensation compared to the sample firms. In contrast, the matched peers exhibit similar accounting performance to the sample firms, but are 7.5% smaller in terms of total revenues and have significantly lower compensation. For example, the median total compensation of the matched peer firms is 9.3% smaller compared to that of the sample firms. The results for the non-s&p 500 firms contrast sharply with those reported for S&P 500 firms. The actual and matched peer firms are significantly larger than the sample firms, and the actual peer firms also 23

25 exhibit better accounting performance. The actual peer firms have total compensation that is 15% higher than the sample firms, while the matched firms exhibit compensation that is 10% higher. The 15% difference in compensation between the actual peers and the sample firms reflects two sources of bias in peer group choice. The first is the bias that arises from the fact that the sample firms systematically choose peers that are larger and have better performance. The second is the bias that arises from choosing more highly compensated peers, after holding differences in size and performance equal. In dollar terms, for non-s&p 500 firms, the median actual peer has total compensation that is $408,000 larger than the compensation of the sample firm, and the median matched firm has total compensation that is $273,000 higher than the sample firm. The median difference in compensation between the actual and matched peers is $54,000 dollars. For non-s&p 500 firms, these biases in pay are economically meaningful. One potential difficulty with our analysis arises because we are unable to collect compensation data for all of the peer firms when the peers are either foreign firms or private firms. If the peers for which we cannot obtain data are systematically smaller and have lower pay compared to the peers for which data is available our estimates on the extent of biases in peer group selection reported above will be upwardly biased. To address the possibility we repeat the analysis in Table 6 for the subsample of firms for which we have complete data on peer group composition (253 firms). The results (not reported) are nearly identical to those reported above for both S&P 500 and non-s&p 500 firms. To summarize, it appears that S&P 500 firms choose their peer firms in a significantly different manner compared to the non-s&p 500 firms. S&P 500 firms choose peer firms that are similar in size and performance to themselves, and also choose peers that have similar levels of compensation. Compared to the matched peers, S&P 500 firms appear to choose peers based on 24