Pride and Prestige: Why Some Firms Pay Their CEOs Less∗ Ernst Maug Alexandra Niessen-Ruenzi Evgenia Zhivotova September 5, 2014 Abstract We investigate the impact of firm prestige on CEO compensation and find that CEOs of more prestigious firms earn less. Specifically, total CEO pay is on average 9% lower for firms listed in Fortune’s ranking of America’s most admired companies. We suggest that CEOs derive benefits from working for a company that enjoys public admiration, and that boards extract pay concessions for these benefits. Our identification strategy is based on matching and differences-in-differences regressions. We perform a range of robustness checks and can exclude many alternative explanations, including that prestige just proxies for better corporate governance, or for increased exposure of the pay setting process to media attention. Keywords: CEO compensation, social status, firm prestige JEL classifications: G30, M52 ∗ All authors are at the University of Mannheim, Department of Finance, 68131 Mannheim, Germany. Email: [email protected]; [email protected] (corresponding author, phone +49-(0)621-181-1595); [email protected]. We would like to thank Renee Adams, Pat Akey, Yakov Amihud, Christian Andres (discussant), Laurent Bach (discussant), Morten Bennedsen, Jonathan Berk, Alon Brav, Joan Farre Mensa (discussant), Rüdiger Fahlenbrach, David Feldman, Simon Gervais, Yaniv Grinstein, Cam Harvey, Atif Ikram (discussant), Han E. Kim, Gilberto Loureiro (discussant), Ulrike Malmendier (discussant), Ron Masulis, Daniel Metzger, Kevin Murphy, Clemens Otto, Jerry T. Parwada, Stefan Ruenzi, Luke Taylor, as well as seminar participants at the 2013 Ackerman Conference on Executive Compensation, 2013 American Finance Association (AFA), Copenhagen Business School, Duke University, 2013 European Finance Association (EFA), EIASM Workshop on Corporate Governance, Erasmus Workshop on Executive Compensation and Corporate Governance, ESSEC, ESRC Exeter Business School Workshop, 2012 FMA annual meetings, 2012 French Finance Association, 2012 German Finance Association meetings, the Hanken School of Economics, the New Economic School in Moscow, Oxford University, University of New South Wales, University of St. Gallen, 2013 Swiss Finance Association meetings, University of Mannheim, and the Rothschild-Caesarea Center 9th Annual Academic Conference for helpful comments. We are grateful to Amy Lyman from the Great Place to Work Institute (GPTWI) for generously sharing data on the “Best Companies to Work For” ranking, to Dirk Jenter to share his classification of forced CEO turnover, and to David Reeb for sharing his data on family firms. This research was supported by a grant from the German Science Foundation (grant MA 3317/2-1). All errors are our own. 1 Introduction “To excel in any profession, in which but few arrive at mediocrity, it is the most decisive mark of what is called genius, or superior talents. The public admiration which attends upon such distinguished abilities makes always a part of their reward.” Adam Smith, Wealth of Nations, chapter 10.I. In this paper we propose that CEOs derive benefits from the public prestige of their firms as a factor that allows prestigious firms to reduce the level of CEO compensation. Specifically, we hypothesize that CEOs are willing to trade off part of their monetary compensation against the social status they obtain through the prestige of their firms. There is anecdotal evidence for such a willingness to trade off compensation for the status of the organization for corporate managers, MBA graduates, university professors, and university presidents. For example, Auger et al. (2013) survey MBA graduates and find that corporate reputation accounts for 7.5% of the value of a job contract on offer; other industry sources put the discount for recruiting corporate managers at 10% for reputable brands.1 Our paper shows that CEOs work for about 9% less at prestigious firms, which aligns closely with these more anecdotal estimates. The notion that wages involve compensating differentials not only for the attractiveness of a job, but also for the social status associated with it goes back at least to Adam Smith and Max Weber.2 The implications of a trade-off between monetary compensation and nonmonetary rewards from a job are explored in a later theoretical literature (see Weiss and Fershtman (1998) for a survey). However, we are not aware of any empirical evidence based on field data that documents such a trade-off. CEO compensation provides an excellent environment to study the hypothesized trade-off, because CEOs receive significant media attention and are most likely sensitive to status concerns. We investigate the relationship between firm prestige and CEO pay for the period from 1992 to 2010 and use a firm’s appearance in Fortune’s list of “America’s Most Admired Companies” (FTMA) as a status symbol valued by CEOs.3 This ranking is suitable, because it possesses three characteristics that identify social status according to Heffetz and Frank 1 See Auger et al. (2013), p. 84 for MBA graduates and the HiringSite on the influence of reputable brands on attracting managerial talent. The sources for the claims in the text are provided in Appendix A.1. 2 See the passage from Adam Smith above. Weber (as cited in Weiss and Fershtman (1998), p. 804) develops the notion that status and monetary compensation are distinct categories of rewards and that monetary compensation can partially substitute for social status. 3 By contrast, Malmendier and Tate (2009) and other contributions in the literature study prizes awarded to CEOs (i.e., “superstar CEOs”), not prizes awarded to firms (i.e. “superstar firms”) which is our focus. We discuss the implications of the difference between superstar CEOs and superstar firms below. 1 (2008): Status is (1) positional in that it relies on a ranking of individuals, (2) desirable for the individual, and (3) scarce. We argue that appearing in the FTMA ranking fulfills all three requirements: It is positional and scarce by construction, and desirable for CEOs who arguably value the inclusion in rankings and enjoy the pride of working for a prestigious firm. Our notion is that CEOs care intrinsically about the ranking membership of their firms and that media rankings establish a desirable status symbol. We show that CEOs of publicly listed firms are indeed willing to exchange monetary compensation for non-monetary benefits from social status: CEOs of firms ranked in the top 100 of the FTMA list earn on average 9% less compared to CEOs of companies that are not admired after controlling for a range of other factors. These results are obtained based on panel regressions with a large range of controls and fixed effects; they also hold if we include firm fixed effects. To further establish causality, we rely on several matching estimators, including bias-adjusted matching (Abadie and Imbens (2011)), matched-sample differences-in-differences (Heckman, Ichimura, and Todd (1997)), and difference-in differences regressions. We check for the homogeneity between prestigious firms and control firms in terms of a range of variables. All approaches to identification generate similar and significant estimates. We investigate the relationship between firm prestige and pay also for Fortune’s Best Companies to Work For ranking, which allows us to apply regression discontinuity design. Again, we find a strong negative effect. After establishing the main result of an average prestige effect on CEO compensation, we analyze the channels that drive this effect. To investigate whether the channel for our main result is indeed a preference for status, we match company and CEO-level data with statelevel data from the general social survey (GSS), which provides information about the status preferences of the population surveyed in the state of a firm’s headquarters. We find that the main effect we document is always stronger for firms headquartered in states where the population has stronger status preferences, which supports the social-status hypothesis. We find the opposite effect in “sin” industries such as alcohol, tobacco, and gambling. Publicly visible firms in these industries have to pay their CEOs more. We also investigate whether career concerns can explain our result. According to the career-concerns hypothesis, CEOs believe that the managerial labor market interprets working for a prestigious firm as a signal of CEO talent. CEOs may therefore sacrifice income at a prestigious firm today and expect higher compensation in more lucrative employments in the future.4 We perform several tests on the career-concerns hypothesis and conclude that 4 The importance of career concerns for the managerial labor market was already emphasized by Fama (1980) and developed theoretically by Gibbons and Murphy (1992), Dewatripont, Jewitt, and Tirole (1999), and Holmstrom (1999). 2 the evidence is weak at best and unlikely to be an important channel that drives our main result. Next, we hypothesize that strong and independent boards are better able to extract wage concessions from CEOs for the benefit of working for a prestigious firm, whereas weak boards leave the prestige benefits to CEOs as a rent. In line with this view, we find that the effect of prestige on compensation is concentrated in well-governed firms, particularly in firms with independent compensation committees and small boards. We investigate several other candidate explanations that could imply a negative relationship between firm prestige and CEO compensation. Being ranked may expose firms to public limelight and increase the potential loss of reputation for directors who have awarded excessively high compensation packages before the firm enters the ranking, leading them to reverse executive pay policy. However, the prestige effect is concentrated in firms with belowmedian excess compensation and those with below-median growth in compensation before they enter the ranking. These observations are inconsistent with the limelight explanation. Another possible explanation is that association with a prestigious firm provides incentives, and CEOs of prestigious firms thus may receive less incentive compensation and a high fraction of fixed salary. These CEOs may therefore demand a lower risk premium and lower total pay. Investigating the different components of CEO pay shows that this is not the case. We find no evidence that firm prestige substitutes for explicit incentive provision. In this context, we also consider whether prestigious firms move visible cash compensation to other forms of pay such as perks, but find no evidence for such a shift. Moreover, we would expect perks to be used more in poorly governed firms, but find that the reduction in pay for ranked firms is concentrated in well-governed firms. We also reconsider Max Weber’s notion that wealthy, high-status individuals may deliberately eschew high monetary compensation as a signal of their status. However, such a signal would also be available to CEOs of poorly governed firms and this explanation is difficult to sustain given our results on corporate governance. Finally, we can rule out that the effect we find is driven by family firms, blockholders, or differences in firm complexity. The analysis in this paper contributes to several strands of the literature. The most closely related literature analyzes the impact of winning prestigious business awards on CEOs’ compensation. Malmendier and Tate (2009) find that “superstar CEOs” who win prestigious business awards extract compensation benefits. Our results complement theirs by showing that “superstar firms”, defined as firms that enter prestigious business rankings, pay their CEOs less. While Malmendier and Tate (2009) find that their effect is stronger in poorly-governed firms, the effect we document is stronger in well-governed firms. Overall, the literature on prize-winning CEOs suggests that higher CEO status makes the CEO 3 more powerful and leads to higher compensation levels.5 Our setting is different, because “superstar CEOs” own their prizes, which increases CEOs’ bargaining power. By contrast, firm prestige is a status symbol to which the firm controls access, which strengthens the position of the board of directors relative to the CEO. This paper identifies how non-monetary characteristics of the firm affect CEO compensation. The only other papers we are aware of that make a similar point are Deng and Gao (2013) and Otto (2014). Deng and Gao (2013) find that CEOs are paid more if the firm’s headquarters are located in a polluted or high-crime environment. Otto (2014) shows that boards may extract pay concessions if they realize that the CEO is optimistic and likely to overvalue the firm’s equity-related compensation. None of these papers refers to preferences for firm prestige, which is a separate non-monetary determinant of CEO pay. Huberman, Loch, and Önküler (2004) conduct an experiment in which they show that individuals are willing to give up monetary rewards for being celebrated as a “winner,” even though they have no monetary or other benefit from their winner-status within the experiment or outside. However, to the best of our knowledge, ours is the first paper to provide field evidence on the trade-off between status and monetary rewards. Most of the previous literature on status concerns and pay differentials focuses on incentives based on two different theoretical approaches.6 Tournament theory holds that increased pay disparity can serve to enhance incentives, whereas social comparison theory and the notion of inequity aversion imply that higher pay disparity destroys incentives.7 These contributions do not address the implications of status concerns for the level of compensation, which is our focus. Hayes and Schaefer (2009) analyze the implications of relative income concerns for CEO pay dynamics in a theoretical model in which relative income itself is a status symbol. From this perspective, firm prestige and a high income may be substitutes in their role as signals of CEO status, which is consistent with our main hypothesis. We contribute to the question of what explains the cross-sectional variation in CEO compensation. The vast literature on executive compensation has made significant progress 5 Wade, Porac, Pollock, and Graffin (2006) and Graffin, Wade, Porac, and McNamee (2008) also find that winners of the “CEO of the year” contest as well as other members of their management team derive benefits in the form of higher compensation from these awards. Shemesh (2014) finds that CEOs who win prestigious business awards increase their risk-taking. 6 See Auriol and Renault (2008) and Kosfeld and Neckermann (2011) on status concerns in incentive contracts generally and Giannetti (2011) for an argument how benchmarking against other firms emerges in optimal executive compensation contracts. 7 Ederer and Patacconi (2010) introduce status concerns into a tournament model and show that the usefulness of pay disparities is reduced and may lead to inefficient outcomes. Fredrickson, Davis-Blake, and Sanders (2010) also analyze pay dispersion among top executives and find negative performance implications, whereas Kale, Reis, and Venkateswaran (2009) present evidence that tournament incentives improve firm performance. 4 on this question in recent years, for example by analyzing the market for CEO talent (Eisfeldt and Kuhnen (2010)), or by identifying the importance of corporate governance as a factor that explains pay levels (Hartzell and Starks (2003)).8 More recent papers on pay differentials include Custodio, Ferreira, and Matos (2013) who find a pay premium for CEOs with more general human capital, and Fernandes, Ferreira, Matos, and Murphy (2013) who investigate why US CEOs may earn more than CEOs in other countries. Still, conventional theories and existing empirical examinations of executive compensation seem to be limited in their ability to explain the variation in CEO pay, suggesting that many as yet unidentified factors influence compensation. Finally, the debate on whether contracting about executive compensation conforms to the tenants of the efficient contracting paradigm remains inconclusive, in particular regarding the level of pay, because it is difficult to measure CEOs’ outside options to decide whether CEOs extract rents or whether they simply earn a competitive wage in the managerial labor market. A number of theoretical contributions emphasize the relevance of the participation constraint for compensation contracts, but empirical inferences on this constraint are only indirect.9 Our approach allows us to show that the participation constraint of the CEO is more likely to be binding in well-governed firms, an observation that provides support for efficient contracting theories for this subset of companies. 2 Data and summary statistics 2.1 Data on firm prestige We use Fortune’s Most Admired Companies ranking (FTMA) as our definition of firm prestige. This ranking is based on surveys among corporate directors and financial analysts in the U.S and published annually in widely read U.S. business magazines and newspapers such as Fortune, the New York Times, and the Wall Street Journal. To create an overall ranking of the most admired companies within the U.S., Hay Group (on behalf of Fortune) asks executives, outside directors, and financial analysts to select ten companies out of the Fortune 1000 which they admire most based on eight different attributes using a scale of zero (poor) to ten (excellent). The attributes that determine the ranking comprise the quality of management; quality of products or services; financial soundness; innovativeness; long-term 8 For further evidence on the cross sectional variation in CEO compensation see the surveys by Murphy (1999, 2012) and Frydman and Jenter (2010). 9 See Oyer (2004), Gabaix and Landier (2008), and Terviö (2009). Gabaix and Landier (2008) provide indirect evidence on the participation constraint by making inferences from the compensation-size relationship. 5 investment; ability to attract, develop, and keep talented people; community and environmental responsibility; and use of corporate assets. One advantage of using this ranking as a proxy for prestige in our context is that it is unlikely that a company can actively influence its inclusion in the FTMA ranking on a short term basis. First, the survey questions and variables cannot be easily influenced as they are determined by a third party (i.e., Hay Group on behalf of Fortune). Second, it is almost impossible for a company to find out the names of all executives, directors, and financial analysts who are surveyed and to influence them correspondingly.10 We manually collect FTMA rankings from printed editions of Fortune magazines from 1990 to 2010. Overall, the number of companies included in the ranking varies from 305 to 593. This variation is mainly driven by the number of industries included in the pool. Even though most industries are covered by the FTMA ranking, a large fraction of ranked companies come from industries such as manufacturing, business equipment, and materials. To obtain a first impression of the firms (not) appearing in the FTMA ranking, Appendix A.2 contains a list of firms that appear regularly and never in the FTMA ranking, respectively. 2.2 Compensation data and descriptive statistics Data on CEO compensation are obtained from ExecuComp. ExecuComp contains annual compensation data for the top five executives of each company in the S&P1500. The data comprise the value of total pay (ExecuComp variable tdc1 ), as well as the value of each component of pay: salary, bonus, option grants, and stock grants. We compute “other pay,” as the difference between total pay and the sum of salary, bonus, option grants, and stock grants. The value of “other pay” includes items such as perquisites and contributions to pension and retirement plans. Due to a major change of some ExecuComp variables in 2006, there are several data adjustments regarding the computation of variables like option values and restricted stock that we need to apply. As a measure of the Black Scholes value of stock options granted to a CEO in a given year (options), we use data item opt_blk_valu before 2006 and its post-2006 equivalent, option_awards_fv, afterward. We follow Walker (2009) and adjust the total pay variable tdc1 from its pre-2006 format to the new format: Before 2006 ExecuComp’s data item tdc1 was supposed to capture the total compensation given to the CEO in that year, but in fact it did not measure the ex-ante value of performance shares. Therefore, we first subtract the value of long term incentive plans (ExecuComp variable ltip), which measures the ex-post value of performance shares from tdc1. Then, we multiply the target number 10 Further information on Fortunes’ procedure to classify the most admired companies is provided by the Hay Group (http://www.haygroup.com/Fortune/results/faqs.aspx.). 6 of performance shares granted to the CEO (ExecuComp variable shrtarg) by a firm’s yearending stock price to compute the ex-ante value of performance shares in a given year, which is added to tdc1. For the post-2006 period we use tdc1 as provided in ExecuComp. Similarly, the pre-2006 data item rstkgrnt (restricted stock) indicates the value of nonperformance contingent stock awards but not that of performance shares. For the period 2006 to 2010 a different data item, stock_awards_fv, measures all stock awards (restricted stock plus performance shares). We construct a comparable variable for the pre-2006 period by adding the value of performance shares to data item rstkgrnt.11 Firms in the FTMA ranking are selected from the Fortune 1000 list, which is compiled based on revenues, whereas the ExecuComp universe is based on the S&P 1500 and therefore includes smaller firms. We do not have access to the Fortune 1000 list but restrict the ExecuComp sample to the 1,000 largest firms based on revenues in each year to make sure that our sample only comprises firms that are sufficiently similar to the firms that appear in the FTMA ranking. Overall, our main sample consists of 16,589 firm-year observations during 1992 to 2010 and covers 1,986 unique firms. Descriptive statistics for compensation data as well as for rankings and control variables are provided in Panel A of Appendix A.3. Mean total compensation in our sample is about 5.9 million USD, while median compensation is lower at 3.5 million USD. Overall, descriptive statistics on compensation data are slightly higher than in previous work on CEO compensation based on ExecuComp data (e.g., Peters and Wagner (2014)), because we restrict the sample to the largest 1,000 firms. The table in Appendix A.3 indicates that 9.9% of all firm-year observations are from firms that appear in the FTMA ranking in the respective year. 3 The impact of firm prestige on CEO pay In this section we investigate our main conjecture that firm prestige is negatively related to CEO compensation by adopting alternative empirical strategies. If we could run a perfect experiment, we would compare CEO pay of firms that are ranked as a most admired company by Fortune in a particular year to CEO pay of the same firm in the same year, had it not been ranked. This counter-factual situation is not observed. We therefore have to rely on second-best experiments, in which we compare a firm-CEO combination in which a firm was ranked to another, sufficiently similar firm-CEO pair in which the firm is not ranked. These second-best experiments invariably involve a trade-off: If the comparison is very close we have to focus on a small subset of our sample, whereas analyses based on larger 11 In one robustness check, we restrict our sample to the pre-2006 period. Results are not affected. 7 samples sometimes rely on control firms that differ more from the prestigious firms. Thus, we effectively face a trade-off between establishing causality for a small set of comparable firms and the generalizability of our results to a large sample of firms. We follow several empirical strategies, which resolve this trade-off in different ways. We start with panel regressions in Section 3.1. Section 3.2 presents results based on matching and differences-in-differences analyses. Section 3.4 provides additional evidence on the Fortune Best Companies to Work For ranking, which allows us to apply a regression discontinuity design. 3.1 Panel regressions with firm fixed effects We begin by presenting panel regressions of the logarithm of CEO compensation on Prestige and a range of control variables and fixed effects. This approach maximizes the generalizability of our results by covering the whole sample of ranked firms and all control observations from the largest 1,000 firms in ExecuComp. Including a range of fixed effects (in addition to a large set of control variables) allows us to control for unobserved heterogeneity. However, this approach relies critically on the assumption that the non-ranked firm-CEO matches in our sample are comparable to the ranked observations after removing the influence of these controls through regression analysis. The main dependent variable is the logarithm of total compensation and the independent variable of interest, P restigei,t−1 , is a dummy variable that is equal to one if firm i appears in the top 100 ranks of the FTMA ranking in year t − 1, and zero otherwise. We use the representation of firm prestige as a dummy variable, because it is an easy-to-interpret summary of the data.12 Specifically, we estimate yit = α0 + γP restigei,t−1 + β 0 Xi,t−1 + αk + αt + εit , (1) where yit is the logarithm of total compensation paid by firm i in year t, Xi,t−1 is a vector of control variables, αk denotes industry fixed effects, and αt are year fixed effects. We include several control variables measured as of time t − 1: Core, Holthausen, and Larcker (1999) show that there is a strong relationship between current and lagged CEO pay and we control for the lagged value of total compensation, in most specifications. Lagged pay also controls for some unobserved firm and CEO characteristics. Since better performing managers are paid more, we also include industry adjusted measures of lagged firm performance (stock return and return on assets) as control variables. We use the lagged logarithms of firm sales, market value, and total assets as measures for firm complexity and firm size (e.g., Baker, Jensen, and Murphy (1988)), which are known to influence executive compen12 We provide robustness checks on the chosen cut-off of 100 below. 8 sation (e.g., Murphy (1999)). Controlling for size is critical in compensation regressions, and Prestige may still proxy for higher-order influences of firm size that are not captured by the linear influence of these control variables. We therefore use dummy variables to represent the size deciles based on market value, but do not report the estimates in the table. We need to control for growth opportunities since CEOs in high-growth firms with many growth opportunities may give up some current compensation in return for higher expected future compensation. We therefore use past sales growth and the market-to-book ratio to control for growth opportunities, and follow Core, Holthausen, and Larcker (1999) in using the average of the market-to-book ratio over the last five years. The market-to-book ratio also controls for potential confounding effects between Prestige and the visibility of “glamor firms” (Rau and Vermaelen (1998)). Principal-agent theory suggests that compensation depends critically on firm risk, because CEOs demand a risk premium for accepting higher firm risk (e.g., Lambert, Larcker, and Verrecchia (1991)). We therefore include the standard deviation of a firm’s daily stock returns in the previous year, F irmRisk. We control for CEO ownership to ensure that our main effect is not driven by the negative relationship between CEOs’ equity holdings and compensation (e.g. Core, Holthausen, and Larcker (1999)). Finally, we include the CEO’s tenure, because we expect CEOs with longer tenure to have more firm-specific human capital and eventually lower outside options in the managerial labor market, which reduces their pay. A detailed description of the main variables is provided in Appendix A.3. We include different combinations of year, firm, and industry fixed effects based on the Fama-French 48 industry classification in our regressions. Standard errors are clustered at the firm level. Table 1 presents the estimation results. We find that firm prestige is significantly negatively related to CEO compensation. The coefficient in column (1) is statistically significant at the 1% level and suggests that total compensation is on average 8.6% lower if a CEO works for a prestigious firm. The impact of prestige on compensation is also economically large if compared to the impact of other variables studied in the literature (e.g., Core, Holthausen, and Larcker (1999)). Doubling firm-risk from the median level of 30.3% to 60.6% corresponds to an increase in compensation of 4.5%. Outperforming other firms in the same industry on the stock market by 100% would result in an increase in compensation of only 2.2%.13 A decline in compensation by 8.6% as found in column (1) would also result from a substantial decline in market value by 49% or a decline in sales of 74%.14 13 The median of FirmRisk is 0.303 from the table in Appendix A.3 and the coefficient on FirmRisk is 0.149, so we obtain 0.303 × 0.149 = 0.045. The statement for industry-adjusted returns follows directly from the coefficient of 0.022 in Table 1. 14 The coefficients on ln(Sales) (ln(M V alue)) are 0.064 (0.127). Hence, we obtain a decline in compen- 9 As mentioned above, our first concern is to adequately control for size and we consider three alternative approaches. Our baseline specification in column (1) includes size deciles based on market value. Column (2) includes a squared term for market value in addition to the other size controls, whereas column (3) uses cubic splines of the market value.15 . In all cases, the coefficient on Prestige is hardly affected and remains significant at the 1%-level. Including dummy variables for size deciles is probably the most saturated and conservative approach to controlling for size and we apply it to all subsequent panel regressions. A second concern addresses industry adjustment. All previous regressions (columns (1) to (3)) control for industry fixed effects as well as for size, but the effects may not be homogenous and size effects may vary by industry. Closer inspection of the FTMA ranking reveals some industry clustering, which is not constant over our sample period. We adopt two different approaches to address this effect: In column (4) we interact industry fixed effects with size and in column (5) we use industry times year fixed effects. None of these additional controls has any material effect on our coefficient of interest, which remains of similar magnitude and statistically and economically significant. In column (6) we include firm fixed effects to control for unobserved heterogeneity across firms, which leaves us only with the within-firm variation to estimate our effect. This approach may put the hurdle for identification higher than any of the previous regressions and absorbs all time-invariant unobserved firm characteristics that may be relevant for inclusion in the FTMA ranking. However, even with firm fixed effects, the coefficient estimate is only slightly smaller and the effect remains statistically significant at the 5% level. All regressions except the firm-fixed-effect specification (6) include the lagged dependent variable. However, including lagged dependent variables in a dynamic panel might induce biased estimates. We therefore use a GMM estimation based on Arellano and Bond (1991) in column (7) and again find a large and highly significant effect. Regarding the impact of control variables, we find an impact of lagged pay of around 0.51 in regressions (1) to (5). The lagged dependent variable should capture a large fraction of unobserved, firm-specific factors present in previous years’ pay and should be a good substitute for firm fixed effects. Furthermore, CEOs are paid more after higher stock returns and there is a negative albeit insignificant impact of lagged ROA on total compensation. From the three proxies for firm size, only sales and market value are significant most of the time, whereas total assets are only significant with firm-fixed effects.16 The coefficient sation of 8.6% from ∆ln(Sales) = −1.34 = −0.086/0.064. After the decline, sales therefore have to be exp(−1.34) or 0.26 of what they were before, a decline by 74%. The calculation for ln(M V ) follows from ∆ln(M V ) = −0.086/0.127 = −0.67 and exp(−0.67) = 0.51, a decline of 49%. 15 We use the bspline command in Stata with five knots. 16 The size proxies are positively correlated (the highest correlation is 0.74 between sales and market value). 10 on Firm Risk implies that an increase in volatility increases CEO pay. While sales growth is insignificant, the average market-to-book ratio of the past five years has the predicted negative sign and becomes highly significant with firm fixed effects. After controlling for all other factors, CEOs of firms with many growth opportunities accept lower pay, presumably because they expect to profit from higher compensation in the future.17 Compensation declines with CEO ownership, even though the effect is economically negligible: a CEO who doubles her stake in the firm from 5% to 10% accepts 0.05% lower pay. CEO tenure has the expected negative impact on compensation in line with the firm-specific human capital argument, but this effect is significant only in the GMM specification (7). Taken together, results from Table 1 show that there is a significantly negative impact of firm prestige on CEO pay. CEOs working for prestigious firms earn on average 9% less than CEOs working for non-prestigious firms. This result might seem surprising in the first place, because it suggests that money is taken away from CEOs if a firm is more prestigious. However, a closer look at compensation growth rates for prestigious and non-prestigious firms reveals that total CEO pay increases at both types of firms over time, but growth rates are lower at prestigious firms. Panel A of Appendix A.4 shows a density plot of annual compensation growth rates for ranked and non-ranked firms. There is a higher density for ranked firms for small positive growth rates compared to non-ranked firms. Thus, ranked firms tend to have smaller compensation growth rates on average than non-ranked firms. At the same time, there are no differences in common growth trends between ranked and non-ranked firms before inclusion in the ranking. Panel B of Appendix A.4 shows a density plot of annual compensation growth rates at ranked and non-ranked firms for a time period of five years before ranked firms enter the FTMA ranking. Both plots are very similar in terms of the distribution of growth rates, which suggests that ranked and non-ranked firms have similar trends in compensation growth rates before inclusion in the ranking. 3.2 Matched sample analysis While results in the previous section hold in a large sample of firms and generalizability of our results should be high, there might still be unobserved heterogeneity that can only be taken care of if the set of control firms is narrowed down to a more comparable subsample of firms. In this section we focus on the internal validity of our results and construct a sample of control groups to which we apply several matching estimators. Matching improves on Our results do not change if we include only one of them in our regression. 17 We are grateful to an anonymous referee for pointing out this interpretation. Results for the market-tobook ratio in pay regressions vary in the literature. Core, Holthausen, and Larcker (1999) find a strongly positive effect for an earlier sample period, whereas Otto (2014) shows a negative effect in a more recent sample. 11 the panel regressions above by not assuming the global validity of the regression function. Moreover, the covariates used as controls in Table 1 may interact and then the regression approach may well turn out to be inadequate. The main challenge to applying matching estimators is that assignment to the “treatment” condition - inclusion in the top 100 of the FTMA ranking in a particular year - is by design not random and is most likely based on criteria such as the performance and the growth opportunities of the firm, which are also relevant for setting CEO pay. In Table 2, we compare the values of the control variables for the non-ranked and the prestigious firms. We also compare the values of some additional governance variables, some of which are only available for a subset of observations. In columns (1) to (4), we use the full sample for comparison. We perform standard t-tests (column (4)), but mainly follow Imbens and Wooldridge (2009), who argue that for the purposes of matching, comparisons should be based on the normalized differences, defined as XR − XNR , ∆X = q 2 2 SN + S R R (2) where X R = N1R i,t Xit and X N R = NN1 R i,t Xit are the sample means of the covariates for the ranked observations (“R”) and the non-ranked matching observations (“NR”), re2 2 spectively; SN R and SR are the corresponding estimates of the variances. We report the normalized difference ∆X in column (3). Imbens and Wooldridge (2009) argue that this statistic should not exceed 0.25 if we want to retain good statistical properties for matching estimators. The main advantage of this statistic as compared to a conventional t-test is that it does not depend on sample size. Sample size should not have an impact on how we evaluate the quality of matching. Table 2 shows that prestigious firms are larger, older, less risky, more profitable, and have more growth opportunities as measured by the five-year average of the market-to-book ratio, whereas differences in stock returns and sales growth are statistically significant (t-test), but not significant in terms of the normalized difference ∆X . Unsurprisingly, the statistically and economically largest differences can be observed for the three size proxies. Differences in corporate governance are generally statistically and economically small. In particular, prestigious firms are comparable to non-ranked firms in terms of CEO ownership, CEOchairman duality, the independence of the compensation committee, and the CEO’s pay slice. (See Appendix A.3 for exact variable definitions.) However, prestigious firms do have larger and busier boards, probably because they are larger. Overall, comparisons of ranked and non-ranked firms in the full sample show that there are large differences between the two groups of firms on several observable dimensions. The P P 12 linear regressions in Table 1 only control for the linear impact of these variables (except for firm size), but may not adequately control for higher-order influences. This aspect motivates the use of other identification strategies. We proceed in two steps to find a matching sample without significant differences in those variables that are relevant for inclusion in the ranking and also relevant for compensation. In the first step, we determine which variables determine inclusion in the ranking by running a Probit regression of Prestige on all variables listed in Table 2. In addition, we include stock returns and accounting performance contemporaneously and control for year fixed effects and industry fixed effects. Table 3 shows results from three specifications. The full specification (column (1)) includes all variables that might influence inclusion in the ranking. However, including corporate governance variables reduces sample size significantly, because they are only available for a subset of observations. Given that they are all statistically insignificant, we omit them and estimate a reduced specification on a larger sample in column (2). Dropping the corporate governance variables more than doubles the number of observations in column (2). Finally, we drop all variables from the reduced specification that are insignificant in column (2) and re-estimate the Probit model in column (3). We still observe that the same variables predict inclusion in the ranking. We then use this reduced specification (column (3)) to estimate propensity scores for inclusion. As expected, the variables with the highest explanatory power for ranking inclusion are the proxies for size. The sign on total assets is negative, even though the univariate comparisons in Table 2 show that prestigious firms are larger in terms of total assets. This sign reversal may be due to the high correlation between market value and total assets (ρ=0.73), but may also be interpreted as an influence of the market-to-book ratio, which in itself has no influence. Lagged ROA and lagged stock returns also predict ranking inclusion, whereas contemporaneous ROA and stock returns have no explanatory power. Surprisingly, stock returns are negatively related to Prestige. Sales growth, the market-to-book ratio, and firm age are all insignificant. Following Core, Holthausen, and Larcker (1999), we define excess compensation as the residual from a regression of total compensation on the determinants of pay and use it as a measure for good governance. We hypothesize that firms with lower excess compensation are more likely to be admired by executives from other firms or financial analysts. Excess compensation has a modest negative influence on Prestige and is significant at the 10% level. In Table 2, we construct a matched sample based on propensity scores calculated from the Probit regression in column (3) of Table 3. Rosenbaum and Rubin (1983) show that treated and untreated subjects with the same propensity scores should have identical distributions for all baseline variables to avoid biasing the estimated treatment effects. We first test 13 whether the balancing property is satisfied for our matched sample of treated and control firms. Statistics of the balancing property test are provided in Appendix A.5. We find that in the unmatched sample, there are significant differences between the covariates of ranked and non-ranked firms that we use to estimate the propensity score. The differences in means between treated and control firms are all highly statistically significant for the unmatched sample. All of these biases completely disappear after matching and become insignificant for all covariates of the matched sample firms. The reduction in the standardized percentage bias is large and, for example, as high as 96% for sales.18 In addition to the balancing property tests, we also require a region of common support for the propensity scores. We report the characteristics of the prestigious firms and the matched sample subject to these selection criteria in Table 2, columns (5) to (8). For this matched sample, the highest values for the normalized difference are 0.18 for sales growth and 0.16 for ROA, which reflects a higher average sales growth and higher profitability for prestigious firms compared to matching firms. Comparing the logarithms of market value reveals that prestigious firms are on average about 20% larger than matching firms. All normalized differences are now comfortably below the threshold of 0.25 recommended by Imbens and Wooldridge (2009). We conclude from the balancing tests and the comparisons in Table 2 that matching based on the propensity score estimated from column (3) in Table 3 yields a comparable set of treatment and control firms that allows us to isolate the impact of firm prestige on CEO compensation. In Table 4 we report the results on the impact of prestige on total pay based on five matching estimators, all of which are described in Imbens and Wooldridge (2009). The first one is a standard one-to-one nearest-neighbor estimator based on the propensity score (Rosenbaum and Rubin (1983); Rubin (1973)), whereas the second is a four-to-one nearest neighbor estimator, i.e., we select four matches for each prestigious firm. The third estimator uses all potential matches from the full sample, but weights them based on the distance between the propensity score of the treated observation and those of the matching observation using Kernel weighting (Heckman, Ichimura, and Todd (1998)). The fourth method divides the sample into subsets based on suitable cut-offs for the propensity scores and is referred to as stratification (Rosenbaum and Rubin (1983)). Then we rerun our main regression from Table 1 to obtain estimates for the coefficient on Prestige for each sub-interval, which effectively assumes that all observations within the same interval are homogeneous. Finally, we obtain the stratification matching estimator as a weighted 18 The reduction in standardized percentage bias is calculated as the percentage differences of the sample means in the treated and non-treated sub-samples, expressed as a percentage of the sample standard deviations. 14 average of these coefficients, using the proportions of observations in the combined sample of ranked and non-ranked firms as weights. The fifth and final estimator is the bias-adjusted matching estimator of Abadie and Imbens (2011), which adjusts for the differences in covariates between treated observations and control observations by using regression analysis. The first step estimates an auxiliary regression as in Table 1, but without including Prestige and only on the non-ranked observations. The second step uses the slope coefficients from this regression to adjust the log of total compensation for each control observation for the differences in covariates between the treated observation and its control. The bias-adjusted matching estimator uses this predicted compensation rather than the actual compensation of the matching firm. Bias-adjusted matching effectively performs a local adjustment to the compensation of the matching firm, which improves on the previous matching algorithms, but also on global regressions, because it does not require that the linear regression is valid globally. For each estimator we compute two versions. The first version is the standard estimator applied to levels of total compensation, for which we display the results in columns (1) and (2) in Panel A of Table 4. The second estimator is the matched-sample differences-in-differences (DiD) estimator of Heckman, Ichimura, and Todd (1997), which uses time-series differences of total compensation, for which we display the results in columns (3) and (4) in Panel A of Table 4. Since we look at logarithms of compensation, we can think of first differences as growth rates. For this estimation approach we compare firm-years in which firms enter the top 100 of the FTMA ranking with all other firms. For both estimators, we obtain the smallest estimates (in absolute value) for kernel matching (−0.081 for levels and −0.082 for first differences) and the largest for stratification (−0.126, respectively, −0.135). Statistical significance is at the 5%-level or better except for one-to-one nearest-neighbor matching (p-value of 0.051) and the DiD estimator with bias adjustment. Note from column (6) that these estimators rely on the smallest subset of observations. In terms of the magnitude of the effect, the matching results and the matchedsample DiD results are well in line with the OLS regressions from Table 1. Finally, we repeat the panel regression analysis from Table 1 based on matched samples. Panel B of Table 2 displays the results for two different matched samples. In both cases we include firm fixed effects and year fixed effects, and control for size by including dummy variables for size deciles, that is, the specification corresponds to column (6) of Table 1, where it yields an estimate for Prestige of −0.079. The regression in column (1) in Panel B of Table 2 restricts the sample to the common support of the propensity scores (see above), which reduces sample size by a little over 4%. Column (2) relies on the same one-to-one nearestneighbor matching we use in Panel A, where we lose 80% of the sample. The estimates 15 for Prestige are −0.126 and −0.099, and they are statistically significant at the 1%-level and the 5%-level, respectively. These results are in line with those in Panel A and overall, our matching results are close to the previous multivariate regression results in terms of statistical and economic significance. 3.3 Differences-in-differences regressions As a further identification strategy, we apply a differences-in-differences (DiD) analysis. This analysis is based on first differences of all variables such that differences in levels between prestigious and non-prestigious firms are controlled for. As a result, the sample of contol firms is much more similar to prestigious firms compared to a regression in levels. The central identifying assumption of this approach is that the average change in observables would be the same for prestigious and non-prestigious firms, if prestigious firms did not enter the ranking (“parallel paths assumption”). We already discussed the similarities of growth rates of compensation at the end of Section 3.1 (see also Appendix A.4). We rely on the differenced version of regression (1), but focus only on entries into the ranking and ignore exits from the ranking: Entry ∆yit = ∆α0 + γDi,t−1 + β 0 ∆Xi,t−1 + αk + αt + ∆εikt . (3) Entry In (3), Di,t−1 = 1 whenever Di,t−1 = 1 and Di,t−2 = 0, i.e., whenever firm i enters the Entry ranking in period t − 1. In all other periods Di,t−1 = 0. We observe 1,336 entries into the top 100 rank in our sample. We ignore exits here, because the control group would comprise firms that remain in the ranking and firms that remain non-ranked: The first group probably has lower prestige than a firm that just exits, whereas the second group has more, if exiting firms still benefit from an “afterglow” of their previous prestige. Table 5 shows the results for different approaches to estimating (3). In columns (1) to (3) we consider all entries with three specifications. In all cases we use first differences of the same control variables as in Table 1 and control for size deciles. The specifications differ only in terms of the fixed effects: column (1) uses industry and year fixed effects, column (2) uses year and industry times size fixed effects, and column (3) uses industry times year fixed effects. In all three specifications, the estimate on All Entries is between −0.110 and −0.126 and therefore somewhat larger in absolute value compared to the specifications in levels from Table 1, which center around −0.09. Statistical significance is also larger and always exceeds the 1%-level. In columns (4) to (6) we repeat the analysis for first entries into the top 100. The coefficient estimates hardly change, but statistical significance falls to the 10%-level in 16 two cases, probably because of the lower number of observations on first entries. 3.4 Additional evidence based on a different ranking An entirely different approach to establish causality is to apply a regression discontinuity design (RDD) on inclusion in the ranking. Unfortunately, the FTMA ranking is not ideally suited to the application of RDD, since the complete list of all ranked firms is published by Fortune and firms ranked above 100 may also benefit from appearing in the ranking, even though they are somewhat less successful and less visible than firms positioned on a higher rank. Therefore, we use Fortune’s list of the Best Companies to Work for (FTBC) in this section to address identification. This ranking is published by the Great Place to Work Institute (GPTWI) and does not suffer from the same shortcoming of the FTMA ranking mentioned above, because ranks outside the top 100 are never revealed to the public. Companies are eligible for consideration in the FTBC ranking if they have more than 500 (from 1998 to 2002) or 1000 employees (since 2003) and if they have been operating for more than 7 years. Unlike the FTMA ranking, the FTBC ranking also includes private firms. The GPTWI applies a two-stage process to construct this ranking. Companies apply in the first stage and Fortune conducts an employee survey, i.e. the “Great Place to Work”Institute’s (GPTWI) “Trust Index,” which measures employees’ trust in their companies’ management, camaraderie, and their pride of working for their employer. The Trust Index alone determines inclusion in the FTBC ranking, but forms only two-thirds of the score for the final ranking. The other third of the final ranking is based on an employer survey, which investigates the company’s workplace culture and human resource policies. The key advantage of the FTBC ranking that makes it suitable for an RDD analysis is that unsuccessful applicants are never disclosed. Hence, firms that fail to rank in the top 100 based on the Trust Index never receive any public recognition, even if they rank just outside the top 100 and receive only a marginally lower trust score compared to firms that score just high enough to enter the ranking. The disadvantage of the FTBC ranking in our context is that it is based on surveys among employees, while the FTMA ranking is based on surveys among other corporate directors and financial analysts and should better reflect firm prestige. By contrast, the FTBC ranking may capture more aspects in addition to firm prestige, for example, the quality of the work environment. Under a confidentiality agreement signed with the GPTWI, we obtain proprietary data of the GPTWI, in particular the list of unsuccessful applicants and the scores for the GPTWI Trust Index.19 We match these firms to the ExecuComp sample. In this step, we lose 19 We are grateful to Amy Lyman for generously sharing the GPTWI data. These data have been used recently by Edmans (2011) and Guiso, Sapienza, and Zingales (2014) for US-based studies and by Edmans, 17 observations for private firms (e.g., law partnerships), small firms, and government agencies. We are left with 1,174 observations, of which 428 are for ranked firms and 746 are for nonranked applicants. Given the way the FTBC ranking is constructed, boards should extract larger pay concessions if their firm is included in the FTBC Top 100 ranking, giving rise to a discontinuity in CEO pay around rank 100. As it is almost impossible for boards to influence the ranks of their firms, the variation in treatment around rank 100 can be considered a randomized experiment that can be used for identification: Firm prestige falls discontinuously at rank 100, while there is no plausible reason to expect that other unobservable characteristics that might drive prestige and compensation show a similar discontinuity.20 To investigate whether applicants and ranked firms are sufficiently similar on observables if they are located close to the top 100 cut-off, we again compute normalized differences and conduct two-sided t-tests for both sub-samples. Results are reported in Appendix A.6. They show that normalized differences between applicants and ranked firms are mostly insignificant and rarely exceed the threshold of 0.25. We first provide a visual impression in Figure 1, which is based on a kernel regression to the right and the left of the cut-off at rank 100. There is a large drop in total compensation around this threshold. Local Wald tests indicate that the differences around the threshold are statistically significant at the 5%-level. While the choice of the kernel function usually has little impact on the results (Lee and Lemieux (2010)), the choice of bandwidth is important. We use the optimized bandwidth following Imbens and Kalyanaraman (2012). In addition, we increase (decrease) the bandwidth by a factor of 1.5 (0.5) to ensure robustness. We still obtain significant results if we increase the bandwidth by a factor of 1.5, but not if we decrease it by 0.5. We expect a decline in statistical significance, because a lower bandwidth reduces the amount of information used in the estimation. In column (3) of Panel B, we report placebo regressions using rank 80 as a threshold. As expected, we do not observe any significant changes around rank 80. In Panel B of Table 6 we also repeat the DiD analysis for the FTBC ranking. Columns (2) to (4) in this table correspond to columns (1) to (3) in Panel A of Table 5. The control variables (not reported) are the same as for the FTMA ranking. The results are qualitatively similar to those for the FTMA ranking, but the point estimate of −0.20 is economically much larger. However, it is much less precisely estimated, because we have less than 10% of the observations for the FTBC ranking compared to the FTMA ranking. Accordingly, statistical Li, and Zhang (2014) in a cross-country study. 20 One additional identifying assumption is that inclusion in the ranking does not in itself have an impact on variables that may influence compensation. 18 significance never exceeds 6%. Discussion of identification strategies. The identification strategies employed in this section differ in how they try to establish causality. Matching is on observables and therefore strong on considering the influence of the covariates included in the analysis, whereas the panel regressions and differences-in-differences regressions also permit us to control for unobserved differences between firms We acknowledge that none of these approaches is based on a fully exogenous shock to firm prestige. However, we take comfort in the fact that all approaches yield relatively similar estimates for the main effect and support our main hypothesis. 4 Channels of the prestige-pay relationship The analysis in the previous section shows that firm prestige has a negative impact on compensation. In this section we provide a more in-depth assessment of the different channels through which firm prestige affects CEO pay. Specifically, we follow up on the status-concerns hypothesis and the career-concerns hypothesis discussed in the Introduction (Section 4.1) and we ask whether the ability of boards to extract pay concessions from CEOs depends on firms’ corporate governance (Section 4.2). 4.1 Preferences for prestige Under the social-status hypothesis, status-related effects should be larger if CEOs live in an environment in which status concerns are stronger. This would happen if the social environment of the CEO affects the CEO’s preferences or if status-seeking CEOs match with prestigious firms and are more likely to locate in more status-conscious environments. Previous papers document the impact of a firm’s local environment on CEO decision making and behavior (Hilary and Hui (2009), Schneider and Spalt (2012)). Status concerns in the general population. We collect data on status concerns across the states of the U.S. from the General Social Survey (GSS). This survey is conducted among the U.S. population and elicits opinions on different topics such as religion, politics, and cultural values. For each state, we compute the fraction of survey respondents who agreed to one of the following survey questions: 1. “I am proud to be working for this organization.” (Pride) 19 2. “I would turn down another job that offered quite a bit more pay in order to stay with this organization.” (Loyalty) 3. “Only if differences in income and social standing are large enough is there an incentive for individual effort.” (Social Standing) We label these questions as “Pride”, “Loyalty”, and “Social Standing”, respectively, for easier reference.21 We interpret agreement with any of these statements as an indication for stronger status concerns in the population. Under the social-status hypothesis, CEOs who work in a state where the population has higher status concerns should be more willing to give up income to work for more prestigious firms. We define dummy variables labeled High Preference based on each of these questions. These variables equal one if the fraction of respondents who affirmed these statements is above the median for the entire United States, and zero otherwise. The dummy variable Low Preference is defined as 1−High Preference. We match these state-level dummy variables to company data by the state in which firms have their headquarters.22 These dummy variables provide us with measures of status concerns within the population around a firm’s headquarters. Table 7 reports our estimation results. We investigate three different interactions of firm prestige with social status concerns and use two different regression specifications in each case.23 Results in Panel A rely on the firm fixed effects estimation of pay levels from column (6) in Table 1, and results in Panel B rely on the differences-in-differences estimation of growth in pay from column (3) in Table 5. Under the social-status hypothesis, we expect negative coefficients on the interactive terms, which should be larger in absolute value for High Preference than for Low Preference. The Wald-test reported at the bottom of the table tests whether these two effects differ significantly from each other. Regressions (1) to (3) in Panel A of Table 7 report economically and statistically much larger coefficient estimates (in absolute value) on the interactions of Prestige with High Preference than with Low Preference, as predicted by the social-status hypothesis. The results for the difference regressions in Panel B of Table 7 replicate these findings for the first two regressions (Pride and Loyalty), but not for the third (Social Standing). All three sample splits based on regional characteristics therefore support the status hypothesis. 21 Specifically, we use GSS variables PROUDORG, STAYORG, and USCLASS. The GSS is conducted every two years for question 3, and we interpolate between adjacent years to obtain annual data and match them to the company data by state and year. Questions 1 and 2 were only asked once, so we assign the values obtained from the survey where the respective question was asked to the entire sample period. 23 Including interaction terms precludes matching in our case, because especially unevenly populated categories would leave too few observations for matching. 22 20 Status concerns across industries. Next, we look at status differences across industries to further investigate the status-concerns hypothesis. CEOs who self-select to work in more admired industries probably also have stronger preferences for status than those who work in less admired industries. We identify firms that enjoy less admiration and appreciation because they provide “sinful” goods and services, such as alcohol, tobacco, and gaming. We would expect that CEOs who self-select to work in these industries are less concerned about prestige and public admiration than those who belong to other industries. We adopt the “sin industry” classification of Hong and Kacperczyk (2009) in column (4) and define Low Preference (High Preference) to be one (zero) for CEOs (not) working for firms in such industries. In column (5) of Table 7, we use a Herfindahl Index to compute the concentration of prestigious firms in each industry. If there are only few prestigious companies in a given industry, boards should be able to extract larger pay concessions from CEOs compared to industries where several prestigious firms exist and CEOs with strong status concerns may find it easier to switch to a prestigious competitor, which reduces the bargaining power of the firm. We therefore expect our results to be stronger in industries with above median concentration of firm prestige. The coefficient estimates are consistent with the status hypotheses for columns (4) and (5) in both specifications in Table 7, but the difference between the interaction terms is statistically significant only for the sin-industry definition of status concerns in Panel B. The positive coefficient on the interaction with Low Preference is remarkable in its own right. This could reflect a compensation premium for sin-industry CEOs. Career concerns. In Table 8 we investigate whether CEOs may be willing to sacrifice current income at a prestigious firm because they believe that it will help them to receive higher pay in a later employment. Note that this expectation would be justified only if CEOs anticipate that the managerial labor market interprets firm prestige as an indicator of CEO talent or enhanced human capital. We hypothesize that career concerns are stronger for younger CEOs compared to older CEOs and weaker for CEOs who are close to retirement (Gibbons and Murphy (1992)). Career concerns should also be weaker for CEOs with a longer tenure on their jobs as the marginal benefit of an additional year of work at a prestigious firm should decrease with tenure. Regressions (1) to (4) in Panel A of Table 8 test these hypotheses by interacting Prestige with a dummy variable for CEO age (columns (1) and (3)), and CEO tenure (columns (2) and (4)), respectively. Columns (1) and (2) display levels regressions and columns (3) and (4) differenced regressions. In all cases, the differences between the estimates for the High Preference and the Low Preference interactions with 21 Prestige point in the direction of the career concerns hypothesis. However, these differences are economically small and always statistically insignificant. In Panel B of Table 8 we test four additional implications of the career-concerns hypothesis. If career concerns are the main driver of our results, we would expect that CEOs use prestigious firms to start their careers and switch to more profitable jobs more frequently than CEOs at non-prestigious firms. Therefore, we would expect that CEO turnover is higher at prestigious firms. In Panel B of Table 8 we regress CEO turnover on Prestige. We classify CEO turnover as either “forced” (column (1)) or “voluntary” (column (2)) following the classification of Jenter and Kanaan (2014).24 The coefficient estimate is negative in both cases, but significant only for forced turnover, indicating that prestigious firms have in fact lower turnover and retain their CEOs slightly longer. This finding opens another potential explanation for our main result: Peters and Wagner (2014) argue that forced executive turnover commands a risk premium. Hence, prestigious firms may pay less, because CEO face lower turnover risk. Based on the calibrations in Peters and Wagner (2014) and our results we would expect a risk premium of about 1.5%, which can therefore explain only a small part of the premium for firm prestige, which we estimate to be in the range of about 9%.25 Columns (3) and (4) provide two additional tests of the career-concerns hypothesis. In column (3), we use the number of a CEO’s outside board appointments as the dependent variable. We obtain a negative but insignificant coefficient on Prestige. Hence, working for a prestigious firm does not increase the number of outside board appointments, which weakens the career concerns hypothesis. In column (4), we restrict the sample to those CEOs who retire after their current employment using the ExecuComp variable reason. We lose about three quarters of the observations for this regression. The coefficient on Prestige is still significantly negative. This suggests that even those CEOs who retire after their current employment are willing to trade off monetary compensation for firm prestige. It may still be the case that CEOs profit from working for a prestigious firm after retirement, because they are hired for consultancy or get other well-paid jobs that we do not 24 These data were used by Jenter and Kanaan (2014) and Peters and Wagner (2014). We are grateful to Dirk Jenter for sharing these data. 25 See Tables 1 to 5 for our estimates. If we assume a 50% loss of pay following forced turnover and observe a baseline rate of 1.9% for forced turnover in out data. Then a 1% reduction in turnover reduces the risk premium for constant relative risk aversion of ρ = 2 by 3.11 percentage points according to Table IA.I of their Internet Appendix of Peters and Wagner (2014). Our estimate in column (1) in Panel B of Table 8 gives a reduction in turnover of 0.5%, hence interpolation would give us 1.56% for the implied premium for firm prestige. Dittmann, Maug, and Spalt (2010) show that realistic compensation models can probably not accommodate risk aversion in excess of ρ=1, which would give an even lower estimate. 22 observe in the ExecuComp data. Using data from BoardEx, we follow CEOs’ careers and check whether CEOs receive more profitable jobs after their current position. We are able to track 345 CEOs and extract information on their next job’s total compensation. We find that both, CEOs from prestigious and CEOs from non-prestigious firms, earn less after their current position. The difference between the change in compensation for CEOs of prestigious and non-prestigious firms is not statistically significant. However, given the small number of observations for which we obtain this information, we decided not to exploit these data any further. With the data we have access to we cannot trace career benefits after their employment with a prestigious firm. Overall there is at best weak support for the career-concerns hypothesis as a key explanation for our main result. We conclude that CEOs do not anticipate that the market for managerial labor regards the FTMA ranking as additional evidence for CEO talent or as a proxy for human capital. This is probably a rational belief, given that appointments are made by boards, who should only reward firm prestige when appointing a CEO from another firm if they believe that media rankings contain information in addition to observable performance and information that can be generated in the recruiting process. Hence, it seems plausible that the market for managerial labor disregards media rankings of the firm and that CEOs anticipate that the career value of these rankings is negligible. 4.2 Prestige and corporate governance In this section we analyze two governance-related aspects of our main effect. We first ask if the prestige effect is concentrated in well-governed firms. We then investigate if firms pay more attention to CEO compensation after being ranked due to increased public scrutiny. 4.2.1 Outside options and corporate governance The effect of prestige on compensation should be particularly pronounced in well-governed firms, because it requires that boards extract pay concessions from CEOs and do not leave the non-monetary benefits from prestige as a rent to CEOs. We hypothesize that extracting these concessions requires strong governance, in particular of the pay-setting process. This hypothesis is independent of where preferences for firm prestige come from. Such preferences could come from concerns about social status or from career concerns and need to be known by the board. Moreover, the board must be in a position to take this information into account when setting CEOs’ compensation. We address the hypothesis that pay concessions for prestigious firms are high in well governed firms in Panel A of Table 9. We use seven different measures of good corporate 23 governance used in the prior literature, and for each measure we define a dummy variable Good Governance to indicate a high corporate governance standard. For binary variables, it equals one if firms conform to the corresponding standard, and zero otherwise. For other variables we divide the sample at the median and Good Governance equals one for the firms with above-median governance standards. Poor Governance is defined as 1−Good Governance. For the sake of brevity, we only report differences-in-differences regressions with the same specifications and controls as in Panel B of Table 7.26 We use the following governance measures. In column (1), we follow Ferris, Jagannathan, and Pritchard (2003) and use the independence of the compensation committee to measure good corporate governance. Good Governance equals one if the majority of members of the compensation committee is independent, and zero otherwise. In our context, this is probably the best governance proxy, because we are concerned with executive compensation. Based on prior work, (e.g., Rosenstein and Wyatt (1990)), we expect compensation contracts to be more efficient if the compensation committee is independent. It has been shown that large boards and busy boards are weak monitors (e.g., Yermack (1996) and Fich and Shivdasani (2006)). Therefore, we use the number of board members as a measure for board size and the average number of outside board seats of board members as a measure for busyness in columns (2) and (3). We conjecture that large and busy boards are less likely to set contracts efficiently and to pay their CEOs less if the firm is prestigious. In column (4), we investigate whether our results are stronger for firms with lower levels of excess CEO compensation, which we use as another indicator of good governance. In column (5) we split the sample according to whether a larger or smaller fraction of CEO pay is categorized by ExecuComp as “other pay.” The reasoning is that this category may be less visible and mask higher compensation and perks. Accordingly, it can be viewed as a measure of the quality of corporate governance. The literature argues that CEO duality weakens the ability of the board to monitor the CEO and leads to poor corporate governance (Shivdasani and Yermack (1999) and Ryan and Wiggins (2004)). In column (6), Poor Governance equals one if the CEO is also the chair of the board, otherwise Good Governance equals one. Some authors argue that the proportion of the CEO’s compensation relative to the compensation of the top executives (“CEO pay slice”) is a measure of CEO entrenchment (Bebchuk, Cremers, and Peyer (2011)) and column (7) defines governance as good if the CEO’s pay slice is below the sample median. Finally, in column (8), we investigate CEO entrenchment and the relevance of external corporate governance by using the Gompers, Ishii, and Metrick (2003) governance index as a measure of good governance. 26 Our results (untabulated) also hold if we use a panel regression with firm fixed effects instead. 24 The governance hypothesis predicts that the effect of Prestige is concentrated in wellgoverned firms, i.e., the coefficient on the interaction with Good Governance is larger in absolute value than for Poor Governance. For all seven governance measures we find coefficient estimates consistent with this prediction. Differences in estimates are also statistically significant for the first two measures: If companies have independent compensation committees (small boards), they pay their CEOs 13.7% (28.4%) less. Both figures are statistically significant at the 1% level. The interaction of Prestige with Poor Governance is statistically insignificant in most cases, whereas the interaction with Good Governance is statistically significant in most cases, which supports the view that only well-governed prestigious firms force CEOs to make pay concessions. Our results on prestige and corporate governance also rule out Max Weber’s notion that wealthy, high-status individuals may deliberately give up high monetary compensation as a signal of their status. Such a signal would also be available to CEOs of poorly governed firms and this explanation is therefore difficult to uphold given our results on independent compensation committees and small boards. 4.2.2 Governance and outrage: The limelight hypothesis Another governance-related interpretation of our results suggests that firms are exposed to more publicity and enter the public limelight when they are ranked. Consequently, prestigious firms are more visible and under higher public scrutiny and may pay their CEOs less because they are more likely to face public outrage (Bebchuk and Fried (2004)). In that case, firm prestige would lead to lower CEO pay because of public scrutiny and anticipated public outrage. Boards of prestigious firms may just be more reluctant to grant CEOs large compensation packages and may be inclined to reverse previously awarded excessive compensation. The limelight hypothesis emphasizes the impact of directors’ reputational concerns and we treat it as distinct from the governance-related hypotheses developed above. Theoretical work by Levit and Malenko (2013) shows that reputational mechanisms have an ambiguous impact on corporate governance outcomes. In our context, directors may want to develop a reputation for not being restrictive on CEO pay if such a stance enhances their employment opportunities. A first indication that the limelight hypothesis is unlikely to explain our results is already provided in column (4) of Table 9, where we show that the prestige effect is concentrated in firms with below-median excess compensation. This finding is inconsistent with the notion that ranking membership alerts boards to pay closer attention to excessive compensation. Nevertheless, it might still be the case that boards revert previously awarded compensation to avoid public outrage, even though compensation increases may not show up as excess 25 compensation. Panel B of Table 9 addresses these concerns and builds on the same baseline specifications as Panel A. For the first test we reason that the limelight effect is strongest immediately after the firm enters the ranking and should not be present two or three years later, since we condition on last year’s growth in pay. In columns (2) and (3) we lag Prestige by two and three years, respectively, and still obtain a significantly negative coefficient on firm prestige, which is of similar magnitude to the effect in column (1). According to the limelight hypothesis, the effect of Prestige should be stronger in firms with higher growth of compensation in the years leading up to its inclusion in the ranking. In columns (4) to (6), we interact P restige with two dummy variables indicating whether a firm had above or below median growth in total compensation over the past two, three, or five years before it entered the ranking. Our results show that the effects tend to be stronger in firms with lower increases in total compensation than in firms with higher increases in total compensation. The Wald test shows that these differences are significant. Hence, reduced pay in ranked firms does not reverse previously higher growth in compensation. Finally, in column (7), we include firm visibility, measured by the number of analysts following a particular firm, as an additional control variable. Firm visibility reduces CEO pay by 3.8%, although this effect is not significant. However, the impact of firm prestige on CEO pay remains unaffected. Taken together, the results in Table 9 contradict the idea that firms entering the limelight attempt to revert previously awarded large compensation increases. The results support the conclusion that the prestige effect is strongest in well-governed firms, in which boards have enough power to extract the prestige premium from the CEO. 4.3 Prestige as a substitute for incentive provision Conventional principal-agent theory suggests that risk-averse CEOs are paid more if they obtain a larger portion of their pay in the form of risky, typically equity-based, compensation because they value one dollar of stock or stock options less than one dollar of fixed compensation.27 The economic literature on social status emphasizes the incentives that may be generated by the desire to hold on to status symbols (Auriol and Renault (2008)). However, firm prestige would provide such additional incentives only to the extent that the benefits from firm prestige are not compensated through lower compensation. Otherwise, if pay differentials fully compensate for the benefits from prestige, the CEOs’ overall utility and the threat of dismissal remain unaffected by inclusion in the ranking. 27 See Lambert, Larcker, and Verrecchia (1991) and Hall and Murphy (2002) for a calibration-based analysis of executives’ valuation of stock and stock options. 26 Hence, if we assume that the adjustment of CEO pay for prestige is incomplete, firm prestige provides additional incentives to CEOs, and these status-induced incentives may substitute for incentive pay and lead to lower equity-related compensation. Prestigious firms would then have to pay a lower risk premium to their CEOs, which would in turn reduce total CEO compensation. According to this substitution hypothesis, total compensation falls because fixed compensation increases by less than equity-based compensation falls, the difference being the risk premium. We expect exactly the opposite for the components of pay based on the social-status hypothesis. Status concerns work through the outside option of the CEO and the board’s objective to retain the CEO, and retention objectives should be mainly reflected in fixed compensation, which is valued higher by CEOs and therefore more efficient. We therefore expect that fixed compensation is lower at prestigious firms based on the social-status hypothesis. To shed light on this question, we investigate the impact of firm prestige on different components of pay. The results are presented in Table 10. Each column uses one component of pay as a dependent variable. We include the same controls as before but do not report them. Results in column (1) show that firm prestige has a significant and negative impact on salary, although the point estimate of the coefficient is small. The influence of firm prestige on bonus payments is also highly significant. We do not find a significant impact of firm prestige on stock option grants (columns (3)); the influence on restricted stock (column (4)) is marginally significant at the 10% level. Hence, deferred equity compensation is not significantly reduced in prestigious firms, although the coefficient estimates are negative. Other pay is also insignificantly lower in prestigious firms than in non-prestigious firms (column (5)). This also shows that prestigious firms do not engage in more camouflage (Bebchuk and Fried (2004)) by moving compensation from visible cash components to some less visible form of other pay. The results on pay-for-performance sensitivity (PPS) are insignificant.28 Taken together, the results in Table 10 contradict the substitution hypothesis. In particular, the substitution hypothesis makes the opposite predictions on cash compensation. However, all results are consistent with the hypothesis that CEOs value firm prestige, which affects their outside option and leads to lower cash compensation without necessarily affecting other components of pay. 28 We compute pay-for-performance sensitivity (PPS) as in Edmans, Gabaix, and Landier (2009). As some of the variables required to compute PPS are not available after the major relaunch of ExecuComp in 2006, the data on PPS ends in 2006. The number of observations entering the regression in column (6) is therefore lower than the number of observations in columns (1) to (5). 27 5 Alternative explanations and robustness checks In this section we consider further alternative explanations for our result of a negative impact of firm prestige on CEO compensation and conduct several robustness checks. Results are reported in Table 11. If prestigious firms are more likely to be family or heir-controlled firms, our main result might be driven by the fact that CEOs of family firms earn less than those of non-family firms (Anderson and Reeb (2003)). In column (1) of Table 11, Panel A we add Family Firm as an additional control variable. We define as family firms those firms where family ownership exceeds 5% and use the data and definition from Anderson, Reeb, and Zhao (2012).29 The impact of Family Firm is highly significant, consistent with the results in Anderson, Reeb, and Zhao (2012), but the coefficient estimate on Prestige is unaffected. Malmendier and Tate (2009) show that firms whose CEOs win prestigious business awards have a lower fraction of blockholders. To make sure that our results are not driven by differences in blockholder ownership, we include a dummy variable for the presence of at least one blockholder as a control variable in column (2) of Table 11 and interact it with Prestige. We lose more than 80% of our sample, but the coefficient on Prestige is still significant and negative, although the point estimate becomes somewhat large. In column (3) of Table 11, we include a measure of firm complexity as an additional control variable. One might argue that prestigious firms are easier to manage because they are less complex, which would then lead to lower CEO compensation in these firms. To address this concern, we follow Berry, Bizjak, Lemmon, and Naveen (2006) and compute a firm’s segement-sales based Herfindahl Index as a proxy for firm complexity. Including firm complexity as an additional control variable does not affect the coefficient on Prestige. An additional concern may be that CEOs who care more about prestige could also be more likely to end up working for prestigious firms and more willing to give up compensation for the status associated with firm prestige than an average CEO would be. We already introduced this possibility in Section 4 above and consider it to be a legitimate interpretation of our results. It means that the CEOs who do not work for prestigious firms may not be willing to give up pay for prestige if given the opportunity. Finally, we run several robustness checks regarding the econometric specification of the main regression. We report the results in Panel B of Table 11. In column (1) we cluster standard errors by year and in column (2) we cluster by firm and year. In both cases we obtain slightly larger standard errors, but the point estimate on the impact of firm prestige on CEO compensation remains significant at the 1%-level. In column (3), we restrict the sample 29 We are grateful to David Reeb for generously sharing these data. 28 to the time period before 2006 because in that year ExecuComp changed the composition of some data items so that several adjustments need to be made if data is used beyond 2005 (see Section 2). Although we lose about 30% of the observations, we still find a significantly negative impact of firm prestige on total compensation. In order to control for the influence of outliers on our results, we truncate the sample and drop the observations with the largest and the smallest values for total compensation. This should alleviate concerns that our results are driven by CEOs with one-dollar salaries (Loureiro, Makhija, and Zhang (2011)) or by erroneous data points. In column (4) of Table 11, we truncate the overall distribution of CEO pay at the top and bottom 1% separately for each year in our sample. Our results for Prestige are unaffected. Throughout the paper, we define Prestige to be equal to one if a firm is ranked in the top 100. However, our research design should be robust to the choice of the cut-off. In column (5) we re-run our standard regressions and define Prestige to be one for all ranked firms. In columns (6) and (7), we define Prestige to be one for the top 75 or top 125 firms, respectively. We still observe significantly negative coefficients for these alternative definitions of prestige. 6 Conclusion We investigate the relationship between CEO compensation and firm prestige and find that CEOs of firms that are ranked in the Fortune 100 Most Admired Companies ranking earn on average 9% less than CEOs of non-ranked companies after controlling for other factors that affect compensation. We argue that CEOs value the social status they obtain from working for firms that appear in highly visible media rankings. Accordingly, boards can extract pay concessions for these benefits. Our second finding is that the negative impact of firm prestige on compensation is concentrated in well-governed firms, an effect that shows up particularly strongly if we measure the quality of governance by the independence of the compensation committee. We conclude from this observation that only sufficiently strong boards can extract pay concessions from CEOs and that weaker boards leave the social benefits from working for a prestigious firm to the CEOs as a rent. 29 References Abadie, A., and G. W. Imbens, 2011, “Bias-Corrected Matching Estimators for Average Treatment Effects,” Journal of Business & Economic Statistics, 29(1), 1–11. Anderson, R. C., and D. M. Reeb, 2003, “Founding-Family Ownership and Firm Performance: Evidence from the S&P 500,” Journal of Finance, 58(3), 1301–1328. Anderson, R. C., D. M. Reeb, and W. Zhao, 2012, “Family-Controlled Firms and Informed Trading: Evidence from Short Sales,” The Journal of Finance, 67(1), 351–385. Arellano, M., and S. Bond, 1991, “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations,” Review of Economic Studies, 58(194), 277–297. Auger, P., T. M. Devinney, G. R. Dowling, C. Eckert, and D. Lin, 2013, “How Much Does a Company’s Reputation Matter in Recruiting,” MIT Sloan Management Review, 54(3). Auriol, E., and R. Renault, 2008, “Status and incentives,” The RAND Journal of Economics, 39(1), 305–326. Baker, G. P., M. C. Jensen, and K. J. Murphy, 1988, “Compensation and Incentives: Practice vs. Theory,” Journal of Finance, 43(3), 593–616. Bebchuk, L. A., M. Cremers, and U. Peyer, 2011, “The CEO pay slice,” Journal of Financial Economics, 102(1), 199–221. Bebchuk, L. A., and J. M. Fried, 2004, Pay Without Performance - The Unfulfilled Promise of Executive Compensation. Harvard University Press. Berry, T. K., J. M. Bizjak, M. L. Lemmon, and L. Naveen, 2006, “Organizational complexity and CEO labor markets: Evidence from diversified firms,” Journal of Corporate Finance, 12(4), 797–817. Core, J. E., R. W. Holthausen, and D. F. Larcker, 1999, “Corporate governance, chief executive officer compensation, and firm performance,” Journal of Financial Economics, 51(3), 371–406. Custodio, C., M. A. Ferreira, and P. Matos, 2013, “Generalists versus specialists: Lifetime work experience and chief executive officer pay,” Journal of Financial Economics, 108, 471–492. Deng, X., and H. Gao, 2013, “Nonmonetary Benefits, Quality of Life, and Executive Compensation,” Journal of Financial and Quantitative Analysis, 48(1), 197–218. Dewatripont, M., I. Jewitt, and J. Tirole, 1999, “The Economics of Career Concerns, Part II: Application to Missions and Accountability of Government Agencies,” The Review of Economic Studies, 66(1), 199–217. 30 Dittmann, I., E. Maug, and O. Spalt, 2010, “Sticks or Carrots? Optimal CEO Compensation when Managers Are Loss Averse,” The Journal of Finance, 65(6), 2015–2050. Ederer, F., and A. Patacconi, 2010, “Interpersonal comparison, status and ambition in organizations,” Journal of Economic Behavior and Organization, 75(2), 348–363. Edmans, A., 2011, “Does the stock market fully value intangibles? Employee satisfaction and equity prices,” Journal of Financial Economics, 101(3), 621–640. Edmans, A., X. Gabaix, and A. Landier, 2009, “A Multiplicative Model of Optimal CEO Incentives in Market Equilibrium,” Review of Financial Studies, 22(12), 4881–4917. Edmans, A., L. Li, and C. Zhang, 2014, “Employee Satisfaction, Labor Market Flexibility, and Stock Returns Around The World,” National Bureau of Economic Research Working Paper Series, No. 20300. Eisfeldt, A. L., and C. M. Kuhnen, 2010, “CEO Turnover in a Competitive Assignment Framework,” Working Paper, Northwestern University. Fama, E. F., 1980, “Agency Problems and the Theory of the Firm,” Journal of Political Economy, 88(2), 288–307. Fernandes, N., M. A. Ferreira, P. Matos, and K. J. Murphy, 2013, “Are U.S. CEOs Paid More? New International Evidence,” Review of Financial Studies, 26(2), 323–367. Ferris, S. P., M. Jagannathan, and A. C. Pritchard, 2003, “Too Busy to Mind the Business? Monitoring by Directors with Multiple Board Appointments,” The Journal of Finance, 58(3), 1087–1112. Fich, E. M., and A. Shivdasani, 2006, “Are Busy Boards Effective Monitors?,” Journal of Finance, 61(2), 689–724. Fredrickson, J. W., A. Davis-Blake, and W. M. G. Sanders, 2010, “Sharing the wealth: social comparisons and pay dispersion in the CEO’s top team,” Strategic Management Journal, 31(10), 1031–1053. Frydman, C., and D. Jenter, 2010, “CEO Compensation,” Annual Review of Financial Economics, 2, 75–102. Gabaix, X., and A. Landier, 2008, “Why Has CEO Pay Increased so Much?,” Quarterly Journal of Economics, 123(1), 49–100. Giannetti, M., 2011, “Serial CEO Incentives and the Structure of Managerial Contracts,” Journal of Financial Intermediation, 20, 633–662. Gibbons, R., and K. J. Murphy, 1992, “Optimal Incentive Contracts in the Presence of Career Concerns: Theory and Evidence,” Journal of Political Economy, 100(3), 468–505. Gompers, P., J. Ishii, and A. Metrick, 2003, “Corporate Governance and Equity Prices,” The Quarterly Journal of Economics, 118(1), 107–156. 31 Graffin, S. D., J. B. Wade, J. F. Porac, and R. C. McNamee, 2008, “The Impact of CEO Status Diffusion on the Economic Outcomes of Other Senior Managers,” Organization Science, 19(3), 457–474. Guiso, L., P. Sapienza, and L. Zingales, 2014, “The Value of Corporate Culture,” Journal of Financial Economics (forthcoming), (0). Hall, B. J., and K. J. Murphy, 2002, “Stock Options for Undiversified Executives,” Journal of Accounting and Economics, 33(2), 3–42. Hartzell, J. C., and L. T. Starks, 2003, “Institutional Investors and Executive Compensation,” The Journal of Finance, 58(6), 2351–2374. Hayes, R. M., and S. Schaefer, 2009, “CEO pay and the Lake Wobegon Effect,” Journal of Financial Economics, 94(2), 280–290. Heckman, J. J., H. Ichimura, and P. Todd, 1998, “Matching As An Econometric Evaluation Estimator,” The Review of Economic Studies, 65(2), 261–294. Heckman, J. J., H. Ichimura, and P. E. Todd, 1997, “Matching As An Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme,” The Review of Economic Studies, 64(4), 605–654. Heffetz, O., and R. Frank, 2008, Preferences for Status: Evidence and Economic Implications, in: Jess Benhabib, Alberto Bisin, Matthew Jackson: Handbook of Social Economics,Elsevier, pp. 69–91. Hilary, G., and K. W. Hui, 2009, “Does religion matter in corporate decision making in America?,” Journal of Financial Economics, 93(3), 455–473. Holmstrom, B., 1999, “Managerial Incentive Problems: A Dynamic Perspective,” The Review of Economic Studies, 66(1), 169–182. Hong, H., and M. Kacperczyk, 2009, “The price of sin: The effects of social norms on markets,” Journal of Financial Economics, 93(1), 15–36. Huberman, B. A., C. H. Loch, and A. Önküler, 2004, “Status As a Valued Resource,” Social Psychology Quarterly, 67(1), 103–114. Imbens, G., and K. Kalyanaraman, 2012, “Optimal Bandwidth Choice for the Regression Discontinuity Estimator,” The Review of Economic Studies, 79(3), 933–959. Imbens, G. W., and J. Wooldridge, 2009, “Recent Developments in the Econometrics of Program Evaluation,” Journal of Economic Literature, 47(1), 5–86. Jenter, D., and F. Kanaan, 2014, “CEO Turnover and Relative Performance Evaluation,” Journal of Finance (forthcoming), (1992). Johnston, C., September 3, 2004, “Market forces bring wide pay variations,” The Times Higher Education. 32 Kale, J. R., E. Reis, and A. Venkateswaran, 2009, “Rank-Order Tournaments and Incentive Alignment: The Effect on Firm Performance,” The Journal of Finance, 64(3), 1479–1512. Kosfeld, M., and S. Neckermann, 2011, “Getting More Work for Nothing? Symbolic Awards and Worker Performance,” American Economic Journal: Microeconomics, 3, 86–99. Lambert, R. A., D. F. Larcker, and R. Verrecchia, 1991, “Portfolio Considerations in Valuing Executive Compensation,” Journal of Accounting Research, 29(1), 129–149. Lee, D. S., and T. Lemieux, 2010, “Regression Discontinuity Designs in Economics,” Journal of Economic Literature, 48, 281–355. Leef, G. C., and J. Sanders, 1999, “The Professors are Not Underpaid,” Pope Center for Higher Education Policy. Levit, D., and N. Malenko, 2013, “The Labor Market for Directors, Reputational Concerns, and Externalities in Corporate Governance,” Reputational Concerns, and Externalities in Corporate Governance (November 16, 2012). Loureiro, G., A. K. Makhija, and D. Zhang, 2011, “Why Do Some CEOs Work for a One Dollar Salary?,” Working Paper, Ohio State University. Malmendier, U., and G. Tate, 2009, “Superstar CEOs,” Quarterly Journal of Economics, 124(4), 1593–1638. Murphy, K., 2012, Executive Compensation: Where We are, and How We Got There, in: Constantinides, George and Harris, Milton and Stulz, Rene, Handbook of the Economics of Finance (Forthcoming), Elsevier Science North Holland, Amsterdam. Murphy, K. J., 1999, Executive Compensation, in O. Ashenfelter, and D. Cards, eds.: Handbook of Labor Economics,North-Holland, Amsterdam, pp. 2485–2563. Otto, C. A., 2014, “CEO optimism and incentive compensation,” Journal of Financial Economics, (0). Oyer, P., 2004, “Why Do Firms Use Incentives that have No Incentive Effects?,” Journal of Finance, 59(4), 1619–1650. Peters, F. S., and A. F. Wagner, 2014, “The Executive Turnover Risk Premium,” The Journal of Finance, 69(4), 1529–1563. Rau, P. R., and T. Vermaelen, 1998, “Glamour, Value and the Post-acquisition Performance of Acquiring Firms,” Journal of Financial Economics, 49(2), 223–53. Rosenbaum, P. R., and Donald, 1985, “Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score,” The American Statistician, 39(1), 33–38. Rosenbaum, P. R., and D. B. Rubin, 1983, “The central role of the propensity score in observational studies for causal effects,” Biometrika, 70(1), 41–55. 33 Rosenstein, S., and J. G. Wyatt, 1990, “Outside directors, board independence, and shareholder wealth,” Journal of Financial Economics, 26(2), 175–191. Rubin, D. B., 1973, “Matching to Remove Bias in Observational Studies,” Biometrics, 29(1), 159–183. Ryan, H. E., and R. A. Wiggins, 2004, “Who is in whose pocket? Director compensation, board independence, and barriers to effective monitoring,” Journal of Financial Economics, 73(3), 497–524. Schneider, C., and O. G. Spalt, 2012, “Conglomerate Investment, Skewness, and the CEO Long Shot Bias,” Working Paper, Tilburg University. Shemesh, J., 2014, “CEO Social Status and Risk Taking,” Working Paper, University of Southern California. Shivdasani, A., and D. Yermack, 1999, “CEO Involvement in the Selection of New Board Members: An Empirical Analysis,” The Journal of Finance, 54(5), 1829–1853. Terviö, M., 2009, “The Difference That CEOs Make: An Assignment Model Approach,” American Economic Review, 99(3), 642–668. Wade, J. B., J. F. Porac, T. G. Pollock, and S. D. Graffin, 2006, “The Burden of Celebrity: The Impact of CEO Certification Contests on CEO Pay and Performance,” Academy of Management Journal, 49(4), 643–660. Walker, D. I., 2009, “Evolving Executive Equity Compensation and the Limits of Optimal Contracting,” Law and Economics Research Paper Series. Weiss, Y., and C. Fershtman, 1998, “Social status and economic performance: A survey,” European Economic Review, 42(3-5), 801–820. Yermack, D., 1996, “Higher market valuation of companies with a small board of directors,” Journal of Financial Economics, pp. 185 – 211. 34 A Appendix A.1 Anecdotal Evidence This appendix summarizes several sources on the relationship between prestige and compensation. • Corporate Managers According to the Hiring Site, “it will cost companies a 21 percent increase in compensation to lure candidates who feel a company’s brand is unattractive, compared to an 11 percent premium for candidates who feel a brand is attractive.” This comparison is the basis for the 10% discount for managers mentioned in the Introduction. Source: http://thehiringsite.careerbuilder.com/2013/10/01/candidates-will-accept-lowersalary-good-brands/ • MBA Students Auger et al. (2013) examine how 303 MBA and executive MBA students chose among 16 pairs of experimentally designed job offers.They offer these students hypothetical employment contracts which include specific salary offers, allowing the authors to estimate each individual participant’s willingness to pay for corporate reputation as well as other features of the job offer. Results show that corporate reputation accounts for 7.5% of the value of the job contracts on offer. • University Professors Leef and Sanders (1999) report on compensation of full professors at all major public and private schools in the US adjusted for cost-of-living and find that out of 86 research institutions, Harvard (Stanford, Yale, MIT) ranks only 75 (77, 43, 80) in this list. Furthermore, Johnston (September 3, 2004) report for the UK that “Oxford and Cambridge are able to lure staff with lower salaries for the privilege of working at institutions with great reputations”. • University Presidents Statistics in the Chronicle of Higher Education show that among the 10 most highlypaid university presidents of private colleges in the U.S. in 2011, only five lead colleges ranked in the top 100 colleges by Forbes. For example, the president of Northeastern University (Forbes rank #236) earns $3.1 million, compared to the presidents of Harvard, Princeton (both $0.9 million), or Stanford ($1.1 million). Source: http://chronicle.com/article/Executive-Compensation-at/143541/). 35 A.2 Firms in (not in) FTMA ranking This table presents the 20 firms that appear in the FTMA ranking most often (column (1)) as well as 20 randomly selected firms that never appear in the FTMA ranking (column (2)). 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. Frequently ranked in FTMA (1) Never ranked in FTMA (2) Abbott Laboratories Alcoa Inc. AMR Corp. Apple Inc. AT&T Inc. Berkshire Hathaway Inc. Boeing Co. Caterpillar Inc. Chevron Corp. Coca-Cola Co. Colgate-Palmolive Co. Deere & Co. Dow Chemical Co. E. I. du Pont de Nemours and Co. Eli Lilly and Co. Exxon Mobil Corp. Hewlett-Packard Co. International Business Machines Corp. J. C. Penney Company Inc. Johnson & Johnson Kroger Co. PepsiCo Inc. Procter & Gamble Co. Safeway Inc. Target Corp. Altera Corp. Aon Corp. Avery Dennison Corp. Bed Bath & Beyond Inc. Big Lots Inc. Carnival Corp. CBS Corp. CenturyLink Inc. Cincinnati Financial Corp. Citrix Systems Inc. DTE Energy Co. Ecolab Inc. Freeport-McMoran Copper Frontier Communications Hasbro Inc. Hershey Co. Linear Technology Corp. Molex Inc. Molson Coors Brewing Co. Monster Worldwide Inc. Nabors Industries Ltd. Rockwell Automation Inc. Rowan Companies Inc. Schlumberger Ltd. Sigma-Aldrich Corp. 36 A.3 Variable Description and Summary Statistics Panel A of this table contains detailed descriptions on the construction of all variables used in the empirical analysis. Panel B reports summary statistics. Panel A Total Payi,t Options i,t Salaryi,t Bonusi,t Restr. Stocki,t Other Payi,t Prestigei,t FTBCi,t ROAadj i,t Returnadj i,t Salesi,t Market Valuei,t Total Assetsi,t Sales Growthi,t Firm Riski,t Firm Agei,t MTBi,t CEO Ownershipi,t CEO Tenurei,t CEO/Chair G-Index Ind. Committee CEO Pay Slice Busy Board Definition Log of total pay in year t (ln(tdc1 + 1)). Source: ExeCucomp (item tdc1). Log of Black-Scholes value of options granted in year t (thousand $’s) Source: ExeCucomp (items blk_valu & option_awards_f v). Log of salary in year t (ln(salary + 1)). Source: Execucomp Log of bonus in year t (ln(bonus + 1)). Source: ExecuComp Log of restricted stock in year t (ln(rstk + 1)). Source: ExecuComp Ln(T otalCompensationt - Salary − Bonust - Optionst - Stockt ) Dummy variable equal to one if firm i appears in Fortune’s Most Admired Company ranking in year t, and zero otherwise. Source: Fortune. Dummy variable equal to one if firm i appears in Fortune’s best companies to work for ranking in year t, and zero otherwise. Source: Fortune. Return on assets of firm i less return on assets of all firms in the same FF48 industry. Source: ExeCucomp (item roa). Annual stock return of firm i less equal weighted stock return of all firms in the same FF48 industry. Source: CRSP/Compustat. Log of firm sales (plus one) in millions. Source: CRSP/Compustat Log of market value of firm i in year t (Ln(M V alue + 1). Source: CRSP/Compustat. Log of total assets of firm i in year t. Source: CRSP/Compustat. Sales growth of firm (%). Source: ExeCucomp (item salechg). Annualized standard deviation of firm i’s stock return. Source: CRSP. Log of difference between current year and the year a firm first appears in the CRSP stock database. Source: CRSP. Market to book ratio of firm (%). MTB is measured as assets (Compustat data # 6) less book value of equity (Compustat data #60) plus market value of equity (Compustat data #199 · # 25) scaled by total assets. We drop observations with MTB > 10. Source: Compustat. CEO ownership in percent (ExeCucomp variable shrownpc). Source: ExecuComp. Log of years CEO is in her position at the current firm. Source: ExecuComp. Dummy variable equal to one if CEO of a firm is also a chair of the board, and zero otherwise. Source: Risk Metrics. Corporate governance index as in Gompers, Ishii, and Metrick (2003). Percentage of independent members of a compensation committee. Source: Risk Metrics Fraction of CEO’s total compensation relative to all other top five directors. Source: ExecuComp Number of outside appointments of all board members of a firm. Source: Risk Metrics. 37 Panel A: cont’d Large Board Excess Payi,t Forced Turnoveri,t Voluntary Turnoveri,t VFAi,t Panel B: Summary Statistics Prestigei,t Total Payi,t (in ’000) Optionsi,t (in ’000) Salaryi,t (in ’000) Bonusi,t (in ’000) Restr. Stocki,t (in ’000) Other Payi,t (in ’000) ROAadj. i,t Returnadj. i,t Salesi,t Market Valuei,t Total Assetsi,t Sales Growthi,t Firm Riski,t Firm Agei,t MTBi,t CEO Ownershipi,t CEO Tenurei,t CEO/Chairi,t Ind. Committeei,t CEO Pay Slicei,t Busy Boardi,t Large Boardi,t Excess Payi,t Definition Number of board members of a firm. Source: Risk Metrics. Total compensation less predicted compensation derived from our main regression (see Table 4). Dummy variable equal to one if CEO is fired at a firm, and zero otherwise. Source: Jenter and Kanaan (2014), Peters and Wagner (2014). Dummy variable equal to one if CEO voluntarily leaves a firm, and zero otherwise. Source: Jenter and Kanaan (2014), Peters and Wagner (2014). Dummy variable equal to one if firm i is covered by at least one analyst according to I/B/E/S variable numrec, and zero otherwise. Source: I/B/E/S. Mean (1) S.D. (2) Median (3) 1st Perc. (4) 99th Perc. (5) Obs. (6) 0.099 5,886.17 2,417.70 778.29 753.19 1,249.25 702.15 0.003 0.015 8.026 7.979 8.265 0.118 0.356 32.92 1.808 0.499 6.699 0.667 0.935 0.372 8.464 10.157 0.014 0.299 11,766.97 8,728.80 354.31 1,989.21 6,360.47 2,029.36 0.078 0.896 0.961 1.373 1.314 0.774 0.218 20.06 1.273 2.957 6.984 0.478 0.160 0.125 7.321 2.660 0.728 0.000 3,492.27 746.57 750.00 345.15 0.000 125.91 0.005 -0.004 7.874 7.974 8.096 0.075 0.303 31.00 1.414 0.000 5.000 1.000 1.000 0.373 7.000 10.000 0.014 0.000 311.34 0.000 28.00 0.000 0.000 0.000 -0.275 -1.034 6.394 4.639 5.876 -0.388 0.101 4.00 0.817 0.000 0.000 0.000 0.333 0.049 0.000 5.000 -1.956 1.000 39,935.90 24,683.79 1,972.77 6,501.32 12,844.75 7,423.26 0.200 1.544 10.203 11.165 11.604 1.090 1.175 82.00 7.044 11.700 33.000 1.000 1.000 0.725 31.000 18.000 1.966 16,589 16,589 16,446 16,589 16,589 16,445 16,256 16,549 16,470 16,589 16,527 16,486 16,538 15,425 16,589 16,527 16,589 16,101 12,000 10,337 16,589 12,000 12,000 15,377 38 A.4 Compensation Growth Rates The following figure displays a kernel plot of compensation growth rates for firms that enter the FTMA ranking in the previous year, and firms that do not enter the FTMA ranking, respectively. The following figure displays a kernel plot of compensation growth rates for firms that never enter the FTMA ranking, and firms that will enter the FTMA ranking in five years, respectively. 39 A.5 Balancing Property of Propensity Score Analysis This table presents test-statistics on the balancing property of our propensity score based matched sample analysis. Means of treated firms (column (1)) and control firms (column (2)) are displayed for all covariates entering the propensity score estimation. Results are displayed for covariates before and after matching. The standardized percentage bias is shown before and after matching, as well as the achieved percentage reduction of the absolute bias. The standardised % bias is the % differences of the sample means in treated and non-treated sub-samples as a percentage of the square root of the average of the sample variances in the two sub-samples (Rosenbaum and Donald (1985)). t-statistics are conducted for equality of means in the two sub-samples. Variable Mean Treated (1) Mean Control (2) % bias t-test p>|t| (5) (6) ROAadj. unmatched ROAadj. matched 0.066 0.066 0.041 0.068 35 -2.3 93.4 12.35 -0.66 0.000 0.510 Returnadj. unmatched Returnadj. matched -0.064 -0.064 0.030 -0.046 -13.2 -2.5 80.9 -3.96 -0.85 0.000 0.398 Sales unmatched Sales matched 8.945 8.944 7.834 8.897 126.2 5.4 95.7 43.17 1.5 0.000 0.134 Market Value unmatched Market Value matched 9.330 9.327 7.780 9.272 128.1 4.5 96.4 45.3 1.24 0.000 0.214 Total Assets unmatched Total Assets matched 9.180 9.179 8.103 9.114 84.8 5.1 94 29.33 1.4 0.000 0.160 Excess Pay unmatched Excess Pay matched -0.074 -0.073 0.023 -0.032 -12.8 -5.4 58 -4.8 -1.35 0.000 0.176 (3) 40 % reduct bias (4) A.6 Closeness of firms around FTBC Cut-off This table presents mean firm and CEO characteristics for ranked firms (column (1)) and for applicant firms (column (2)) according to the FTBC ranking, respectively. Column (3) reports normalized differences ∆X from equation 2 between ranked firms and applicants. Column (4) reports t-statistics based on two-sided t-tests between the two subsamples. Stock Return ROA Sales Market Value Total Assets Sales Growth Firm Risk Firm Age Market to Book (5yr avg) CEO Ownership CEO Tenure CEO/Chair IndCC CPS Busy Board Large Board Excess Pay G-Index Prestige (1) No Prestige (2) ∆X (3) t-stat (4) 0.080 0.084 8.661 9.099 8.685 0.073 0.336 3.251 2.625 0.240 1.608 0.741 0.965 0.368 12.121 10.534 -0.060 9.633 0.007 0.062 8.241 8.782 8.662 0.128 0.371 3.174 2.361 0.485 1.675 0.644 0.944 0.359 10.561 10.206 -0.030 9.031 0.107 0.260 0.326 0.174 0.012 -0.199 -0.155 0.079 0.135 -0.103 -0.053 0.147 0.109 0.052 0.161 0.097 -0.035 0.173 0.75 3.53 4.04 2.91 1.64 -1.47 -2.69 1.91 0.63 -0.35 -0.65 1.33 1.02 1.23 1.79 2.37 -0.44 1.54 41 Table 1: Average Effect of Firm Prestige on CEO Pay. This table presents panel and GMM regressions with Total Pay as the dependent variable. The main independent variable, F irm P restige, is a dummy variable indicating a top 100 rank in the Fortune Most Admired Companies ranking. All control variables are described in Section A.3. Column (1) and columns (4) to (7) include size deciles based on market value (decile 1=omitted category). Column (2) includes squared market value as an additional control variable, column (3) includes cubic size splines. Column (6) includes firm fixed effects. Column (7) reports results from a dynamic GMM Arellano and Bond (1991) regression. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Prestigei,t−1 ROAadj. i,t−1 Baseline (1) Size2 (2) Splines (3) Ind×Sz (4) Ind×Yr (5) Firm FE (6) GMM (7) –0.086 –0.088 –0.095 –0.085 –0.092 –0.079 –0.207 (–3.22) (–3.33) (–3.60) (–3.13) (–3.31) (–2.03) (–2.64) –0.097 –0.094 –0.160 –0.099 –0.025 0.104 0.904 (–0.80) (–0.78) (–1.31) (–0.80) (–0.18) (0.72) (0.80) Returnadj. i,t−1 0.022 0.021 0.023 0.021 0.023 0.010 –0.060 (2.26) (2.21) (2.24) (2.26) (2.14) (1.17) (–0.77) Salesi,t−1 0.064 0.063 0.055 0.065 0.063 0.100 0.016 (4.14) (4.12) (3.76) (4.06) (3.89) (2.03) (0.10) 0.127 0.120 –0.063 0.086 0.111 0.323 0.519 (4.01) (2.18) (–2.37) (1.68) (3.35) (6.25) (1.83) Total Assetsi,t−1 0.018 0.018 0.011 0.014 0.033 –0.170 0.007 (0.91) (0.88) (0.55) (0.68) (1.57) (–3.78) (0.03) Sales Growthi,t−1 0.004 0.004 0.004 0.004 0.004 0.007 0.192 (1.08) (1.08) (1.30) (1.10) (1.06) (0.98) (0.85) Firm Riski,t−1 0.149 0.147 0.139 0.157 0.153 0.066 0.782 (3.20) (3.17) (3.03) (3.33) (3.02) (1.23) (1.99) Firm Agei,t−1 –0.004 –0.004 –0.003 –0.009 –0.003 –0.101 0.031 (–0.33) (–0.35) (–0.26) (–0.73) (–0.27) (–1.12) (0.90) MTB5y i,t –0.020 –0.022 –0.026 –0.027 –0.015 –0.049 –0.124 (–1.40) (–1.52) (–1.77) (–1.78) (–0.99) (–2.05) (–1.93) CEO Ownershipi,t−1 –0.007 –0.007 –0.007 –0.007 –0.008 –0.007 –0.007 (–1.98) (–1.97) (–1.93) (–1.92) (–2.10) (–1.48) (–1.92) CEO Tenurei,t –0.010 –0.010 –0.011 –0.009 –0.010 –0.000 –0.034 (–1.09) (–1.09) (–1.32) (–1.07) (–1.08) (–0.02) (–2.72) 0.513 0.514 0.510 0.510 0.515 0.273 (15.85) (15.87) (15.68) (15.65) (15.84) (2.33) yes no no no yes no 0.532 14,421 0.000 0.259 0.569 yes yes yes no no no – 14,421 Market Valuei,t−1 Payi,t−1 Market Value2i,t−1 0.002 (0.72) AR(1) test (p-value) AR(2) test (p-value) Hansen test (p-value) Size deciles Year FE Industry FE Industry × Size FE Industry × Year FE Firm FE Adj. R2 Observations yes yes yes no no no 0.530 14,421 no yes yes no no no 0.530 14,421 no yes yes no no no 0.541 14,421 42 yes yes no yes no no 0.530 14,421 yes yes no no no yes 0.610 14,599 Table 2: Normalized Differences. This table presents mean firm and CEO characteristics for prestigious firms (columns (1) and (5)) and for non-prestigious firms (columns (2) and (6)), respectively. Standard errors are clustered at the firm level. Results in columns (1) to (4) are based on the full sample of firms, results in columns (5) to (8) are based on a matched sample of prestigious firms and their nearest neighbor among the non-prestigious firms according to a propensity score analysis. Columns (1) and (5) ((2) and (6)) report the sample means for the prestigious (nonprestigious) firms. Columns (3) and (7) report the normalized differences ∆X from equation 2 and columns (4) and (8) report two-sample t-statistics. Full Sample ROAadj. Stock Returnadj. Sales Market Value Total Assets Sales Growth Firm Risk Firm Age MTB5y i,t CEO Ownership CEO Tenure CEO/Chair Ind. Committee CEO Pay Slice Busy Board Large Board Excess Pay G-Index Matched Sample Prestige (1) No Prestige (2) ∆X (3) t-stat (4) Prestige (5) No Prestige (6) ∆X (7) t-stat (8) 0.071 0.017 8.916 9.354 9.171 0.112 0.292 3.450 2.353 0.399 1.727 0.705 0.943 0.367 12.440 11.139 -0.047 9.664 0.038 0.096 7.762 7.688 8.092 0.362 0.360 3.247 1.805 0.469 1.656 0.663 0.933 0.372 7.934 10.032 0.020 9.645 0.324 -0.072 0.899 0.969 0.593 -0.009 -0.252 0.204 0.320 -0.018 0.055 0.064 0.046 -0.027 0.407 0.301 -0.065 0.005 8.67 -6.97 22.41 22.88 15.50 -3.05 -8.58 4.52 6.93 -0.49 1.05 1.78 1.15 -1.54 9.30 7.76 -2.24 0.1 0.061 0.034 8.784 9.079 9.025 0.137 0.301 3.330 2.133 0.251 1.711 0.693 0.936 0.375 12.613 10.958 0.007 9.879 0.043 -0.048 8.747 8.898 9.032 0.064 0.334 3.426 2.096 0.304 1.621 0.676 0.935 0.375 11.561 10.843 -0.066 10.143 0.164 0.129 0.031 0.115 -0.003 0.181 -0.119 -0.102 0.022 -0.022 0.071 0.026 0.006 -0.003 0.087 0.032 0.068 -0.075 4.85 -1.07 1.68 4.19 0.29 3.15 -2.49 -4.74 1.23 0.92 2.26 0.74 0.04 -0.27 2.02 0.34 1.40 0.7 43 Table 3: Determinants of Prestige. This table presents results from probit regressions with P restige, as dependent variable. Prestige is a dummy variable indicating a top 100 rank in the Fortune Most Admired Companies ranking. Column (1) presents results including all control variables, column(2) includes a reduced set of control variables where governance variables are dropped, and column (3) only includes those control variables that are statistically significant in column (2). Standard errors are clustered at the firm level. z-statistics are provided in parentheses. Payi,t−1 ROAi,t ROAi,t−1 Stock Returni,t Stock Returni,t−1 Salesi,t−1 Market Valuei,t−1 Total Assetsi,t−1 Sales Growthi,t−1 MTB5y i,t Firm Agei,t−1 Excess Payi,t−1 CEO/Chairi,t−1 Ind. Committeei,t−1 CEO Pay Slicei,t−1 Busy Boardi,t−1 Large Boardi,t−1 G-Indext−1 Year FE Industry FE Pseudo R2 Observations Probit All controls (1) Probit Reduced set of controls (2) –0.015 (–0.23) –0.036 (–0.06) 1.174 (1.86) –0.051 (–0.69) –0.176 (–2.88) 0.550 (5.54) 0.779 (7.87) –0.331 (–2.83) –0.249 (–1.66) –0.031 (–0.65) –0.100 (–1.32) –0.032 (–1.76) 0.109 (1.23) –0.100 (–0.48) –0.342 (–0.99) –0.009 (–1.54) 0.012 (0.59) 0.018 (0.97) yes yes 0.369 6,061 –0.026 (–0.83) –0.145 (–0.44) 0.933 (2.32) –0.031 (–1.06) –0.130 (–5.09) 0.553 (7.39) 0.729 (10.17) –0.320 (–3.82) –0.196 (–1.57) –0.004 (–0.11) –0.050 (–0.90) –0.043 (–1.82) –0.124 (–5.21) 0.567 (7.65) 0.657 (9.85) –0.291 (–3.67) yes yes 0.354 14,980 yes yes 0.351 15,129 44 Probit Propensity score (3) 0.873 (1.78) –0.040 (–1.73) Table 4: Matching Results. Panel A of this table presents average treatment effects of entering the top 100 of the Fortune Most Admired Companies ranking on CEO compensation. Matching is based on a propensity score derived from column (3) in Table 3. A common support is required. Matching algorithms are as indicated in the table and described in Section 3.2. Columns (1) and (2) show the results for matched-sample estimators and levels of pay. Columns (3) and (4) show the results for matched-sample differences-in-differences estimators. Panel B presents regression results based on the firm-fixed-effect specification (6) in Table 1. Results in column (1) are based on a sample including all firms within the region of common support of their propensity scores. Results in column (2) are based on a sample including all prestigious firms and their nearest non-prestigious neighbor based on propensity scores. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Panel A Nearest Neighbor 1:1 Nearest Neighbor 4:1 Kernel Stratification Bias-adjusted Panel B Prestigei,t−1 ROAadj. i,t−1 Level of Pay Coefficient t-stat ∆ Pay Coefficient t-stat (1) (2) (3) (4) Number of Treated Obs. (5) -0.097 -0.087 -0.081 -0.126 -0.110 -1.95 -2.08 -2.11 -3.67 -1.97 -0.103 -0.091 -0.082 -0.135 -0.010 -2.03 -2.17 -2.17 -4.00 -1.40 1,431 1,431 1,431 1,431 1,431 Common Support (1) Nearest Neighbor 1:1 (2) –0.126 –0.099 (–3.04) (–1.96) 0.263 0.453 (1.68) (1.11) Returnadj. i,t−1 0.008 –0.022 (0.90) (–1.08) Salesi,t−1 0.132 0.070 (2.51) (0.40) Market Valuei,t−1 0.570 0.611 (14.39) (6.92) Total Assetsi,t−1 –0.016 0.021 (–0.33) (0.17) Sales Growthi,t−1 0.000 0.174 (0.03) (1.35) Firm Riski,t−1 0.344 0.741 (7.05) (5.03) Firm Agei,t−1 0.584 0.393 (7.47) (1.29) MTB5y i,t –0.022 –0.117 (–0.84) (–2.69) CEO Ownershipi,t−1 –0.008 –0.016 (–1.72) (–2.03) 0.003 0.027 CEO Tenurei,t Adj. R2 Observations Size deciles Year FE Firm FE Number of Control Obs. (0.25) (0.87) 0.590 13,800 yes yes yes 0.479 2,807 yes yes yes 45 (6) 1,431 5,724 15,128 13,852 1,431 Table 5: Differences-in-Differences Results. This table presents differences-in-differences regressions with Total Pay as the dependent variable. In columns (1) to (3), the main independent variable, All Entries, is a dummy variable indicating that a firm has achieved a top 100 ranking in the Fortune Most Admired Companies ranking in the previous year. In columns (4) to (6), the main independent variable, F irst Entry, is a dummy variable indicating the first time a firm obtains a top 100 ranking in the Fortune Most Admired Companies ranking in our sample period. All control variables are described in Section A.3. They are included as first differences in percent except for CEO tenure, which is included as absolute change from t-1 to t. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. All entries (n=1,336) All Entriesi,t−1 First entries (n=312) (1) (2) (3) –0.120 –0.110 –0.126 (–4.20) (–3.79) (–4.23) First Entryi,t−1 ∆ Controls included Size deciles Industry FE Year FE Industry ×Size FE Industry ×Year FE Adj. R2 Observations yes yes yes yes no no 0.530 13,064 yes yes no yes yes no 0.531 13,064 yes yes no no no yes 0.533 13,064 46 (4) (5) (6) –0.094 –0.086 –0.111 (–1.83) (–1.69) (–2.13) yes yes yes yes no no 0.529 13,064 yes yes no yes yes no 0.530 13,064 yes yes no no no yes 0.532 13,064 Table 6: Fortune’s Best Companies to Work For Ranking. Panel A of this table presents differences-in-differences results for the FTBC ranking. The sample comprises all firms that have been ranked or have applied to be ranked, but were not included in the ranking. Panel B presents results for a regression discontinuity design with a kernel regression using a triangular kernel and the optimized Imbens and Kalyanaraman (2012) bandwidth (bw=1). We modify the optimized bandwidth by factors of 1.5 and 0.5 to ensure robustness of our results. A sharp regression discontinuity design is assumed, where the treatment variable (FTBC ranking) jumps from one to zero at rank 100. We run this analysis for all firms ranked between 90 and 110 (column (1)), and 80 and 120 (column (2)). Column (3) displays a placebo test for the hypothetical cut-off set to 80 and firms in the interval between 70 and 90. z-statistics are provided in parentheses. Panel A: RDD P restigei,t−1 (bw=1) P restigei,t−1 (bw= 1.5) P restigei,t−1 (bw=0.5) Observations Cut-off 100 Window Size [90;110] (1) Cut-off 100 Window Size [80;120] (2) Cut-off 80 Placebo [70;90] (3) –6,055.01 (–2.45) –4,253.76 (–2.21) – 5,517.60 (–1.44) 202 –6,346.66 (–2.38) –4,853.92 (–2.41) –4,926.28 (–1.14) 401 –2,570.29 (–0.88) –2,538.39 (–0.96) –165.83 (0.03) 202 Panel B: Diff-in-Diff ∆ Pay (1) ∆ Pay (2) ∆ Pay (3) ∆ Pay (4) Best Companyi,t−1 –0.201 (–1.93) yes no yes yes no no 0.333 1,104 –0.200 (–1.87) yes yes yes yes no no 0.331 1,104 –0.217 (–1.89) yes yes no yes yes no 0.353 1,104 –0.238 (–1.63) yes yes no no no yes 0.234 1,104 ∆ Controls included Size deciles Industry FE Year FE Industry ×Size FE Industry ×Year FE Adj. R2 Observations 47 Table 7: Status Concerns. This table presents regressions with Total Pay as the dependent variable in Panel A and ∆T otal P ay as the dependent variable in Panel B. The main independent variable, Prestige, is a dummy variable equal to one if a firm belongs to the top 100 in the FTMA ranking, and zero otherwise. Control variables are included in levels in Panel A and in first differences in Panel B, respectively. Results in Panel A are based on the same specification as column (6) in Table 1. Results in Panel B are based on the same specification as column (3) in Table 5. In each regression, we interact P restige with a dummy variable indicating high, respectively, low preferences for prestige. In columns (1) to (3), we split the sample into US regions with high and low status concerns according to the median GSS variables PROUDORG, STAYORG, and USCLASS. In column (4), we set Low Preference equal to one if a firm operates in a sin industry as defined by Hong and Kacperczyk (2009). In column (5), we split the sample into subsamples based on median prestige dispersion measured by the Herfindahl index based on sales. Level regressions in Panel A include firm fixed effects. Differences-in-differences regressions in Panel B include year times industry fixed effects. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Panel A: Firm FE (2) Social Standing (3) Sin Industry (4) Prestige Dispersion (5) –0.216 –0.232 –0.180 –0.087 –0.120 (–2.56) (–2.49) (–2.55) (–2.31) (–1.63) –0.024 –0.034 –0.035 0.068 –0.060 (–0.71) (–0.98) (–0.88) (0.40) (–1.62) Wald test (High-Low) p-value Controls included Size deciles Firm FE Year FE Adj. R2 Observations 4.50 0.034 yes yes yes yes 0.589 10,549 3.95 0.047 yes yes yes yes 0.590 10,487 3.22 0.073 yes yes yes yes 0.591 10,623 0.80 0.370 yes yes yes yes 0.616 14,421 0.53 0.468 yes yes yes yes 0.616 14,421 Panel B: Diff-in-Diff Pride Loyalty (1) (2) Social Standing (3) Sin Industry (4) Prestige Dispersion (5) –0.226 –0.209 –0.149 –0.131 –0.136 (–3.72) (–3.29) (–2.96) (–4.31) (–2.58) –0.064 –0.099 –0.131 0.165 –0.119 (–2.16) (–2.88) (–3.29) (3.10) (–3.66) 6.65 0.010 yes yes yes 0.533 9,729 2.65 0.104 yes yes yes 0.532 9,671 0.09 0.760 yes yes yes 0.533 9,798 25.58 0.000 yes yes yes 0.533 13,064 0.07 0.787 yes yes yes 0.533 13,064 Prestige × High Preference Prestige × Low Preference Prestige × High Preference Prestige × Low Preference Wald test (High-Low) p-value ∆ Controls included Sizedeciles Industry × Year FE Adj. R2 Observations Pride Loyalty (1) 48 Table 8: Career Concerns. This table presents regressions with Total Pay as the dependent variable in Panel A and∆T otal P ay as the dependent variable in Panel B. The main independent variable, Prestige, is a dummy variable equal to one if a firm belongs to the top 100 in the FTMA ranking, and zero otherwise. Control variables are included in levels in Panel A and in first differences in Panel B, respectively. Results in columns (1) and (2) are based on the same specification as column (6) in Table 1. Results in columns (3) and (4) are based on the same specification as column (3) in Table 5. In regressions (1) to (4) we interact P restige with a dummy variable indicating high, respectively, low preferences for prestige. In columns (1) and (3) ((2) and (4)) we assign High Preference to CEOs with below median age (tenure). Panel B shows regressions of turnover on Prestige. Fixed effects are indicated in the table. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Panel A Firm FE Diff-in-Diff CEO Age (1) CEO Tenure (2) CEO Age (3) CEO Tenure (4) Prestige × High Preference –0.105 –0.084 –0.163 –0.135 (–1.62) (–2.06) (–3.14) (–3.89) Prestige × Low Preference –0.064 –0.077 –0.103 –0.116 (–1.97) (–1.33) (–3.25) (–2.60) 0.41 0.520 yes yes yes yes no 0.626 13,929 0.01 0.910 yes yes yes yes no 0.610 14,599 1.14 0.286 yes yes no no yes 0.537 12,562 0.14 0.709 yes yes no no yes 0.533 13,064 Forced Turnover (1) Voluntary Turnover (2) Outside Appointm. (3) Retirement (4) –0.005 –0.004 –0.416 –0.099 (–2.33) (–0.68) (–1.25) (–2.36) yes yes yes no yes 0.105 9,060 yes yes yes no yes 0.041 9,060 yes yes no yes no 0.494 11,086 yes yes no yes no 0.659 4,219 Wald test (High-Low) p-value (∆) Controls included Size deciles Firm FE Year FE Industry × Year FE Adj. R2 Observations Panel B Prestigei,t−1 Controls included Size deciles Industry FE Industry × Year FE Year FE Pseudo/ Adj. R2 Observations 49 Table 9: Corporate Governance and the Impact of Firm Prestige on CEO Pay. Both panels present regressions with ∆T otal P ay as the dependent variable and specifications are based on column (3) in Table 5 with year times industry fixed effects and control variables are in first differences. The main independent variable, P restige, is a dummy variable equal to one if a firm belongs to the 100 most admired companies according to the FTMA ranking, and zero otherwise. In Panel A, we interact P restige with a dummy variable for good and one for poor corporate governance. In column (1), Good Governance is equal to one if the compensation committee of the company is independent, and zero otherwise. In column (2), Good Governance is equal to one if the number of board members is smaller than (larger or equal to) the median number of board members in the sample, and zero otherwise. In column (3), Good Governance is equal to one if the number of outside appointments of all board members is smaller than (larger or equal to) the median number of outside appointments in the sample, and zero otherwise. In column (4), Good Governance is equal to one if excess pay at a firm is smaller than (larger or equal to) median excess pay in the sample, and zero otherwise. In column (5), Good Governance is equal to one if the CEO is not at the same time chairman of the board, and zero otherwise. In column (6), Good Governance is equal to one if the CEO pay slice (CPS) is smaller than (larger or equal to) the median CPS, and zero otherwise. In column (7), Good Governance is equal to one if the Gompers, Ishii, and Metrick (2003) G-Index of a firm is smaller than (larger or equal to) the median G-Index of (9) in the sample, and zero otherwise. The opposite definitions hold for Poor Governance, which is defined as 1−Good Governance. In Panel B we test the limelight hypothesis. In column (1) (respectively, (2), (3)), we lag the main variable, P restige, by one (two, three) years. In column (4) ((5), (6)), we interact P restigei,t−1 with two variables indicating whether the change in total compensation over two (three, five) years before the firm entered the ranking was above or below the sample median change. In column (7), we control for firm visibility measured by the number of analysts following a firm. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Panel A: Ind. Large Busy Excess Other CEO Corporate Governance Comm. (1) Board (2) Board (3) Pay (4) Pay (5) Prestige × Good Governance –0.137 –0.284 –0.136 –0.109 (–3.77) (–3.81) (–2.45) (–2.71) –0.037 –0.064 –0.116 (–0.52) (–2.07) (–3.25) 5.17 0.000 yes yes yes 0.514 9,363 8.43 0.000 yes yes yes 0.524 10,699 0.12 0.731 yes yes yes 0.525 10,699 Prestige × Poor Governance Wald test p-value ∆ Controls included Size deciles Industry × Year FE Adj. R2 Observations 50 Chair (6) CEOPay Slice (7) GIndex (8) –0.141 –0.165 –0.100 –0.140 (–2.32) (–2.74) (–3.09) (–3.18) –0.023 –0.099 –0.115 –0.039 –0.124 (–0.94) (–3.41) (–3.16) (–0.57) (–3.01) 2.99 0.084 yes yes yes 0.729 12,562 0.42 0.516 yes yes yes 0.583 12,760 0.65 0.419 yes yes yes 0.514 9,875 0.60 0.440 yes yes yes 0.588 13,064 0.08 0.775 yes yes yes 0.524 9,086 Panel B: Limelight P restigei,t−1 Higher Lags of Prestige BaseLag Lag line 2 Yrs 3 Yrs (1) (2) (3) Changes in Pay before Rank Trend Trend Trend [t-1;t-2] [t-1;t-3] [t-1;t-5] (4) (5) (6) –0.126 (7) –0.122 (–4.23) P restigei,t−2 Visibility (–4.10) –0.102 (–3.78) P restigei,t−3 –0.093 (–3.03) P restigei,t−1 × ∆P ay low P restigei,t−1 × ∆P ay high –0.124 –0.142 –0.187 (–1.68) (–2.02) (–2.49) 0.015 0.041 0.034 (0.35) (0.91) (0.70) V F Ai,t−1 –0.053 (–2.60) Controls included yes yes yes yes Size deciles yes yes yes yes Industry × Year FE yes yes yes yes Adj. R2 0.533 0.532 0.523 0.321 Observations 13,064 13,010 11,833 14,346 Wald test against H0 : P restige × ∆P ay high = P restige × ∆P ay low p-value 0.03 51 yes yes yes 0.310 12,836 yes yes yes 0.298 10,103 0.00 0.00 yes yes yes 0.533 13,064 Table 10: Firm Prestige as a Substitute for Incentives. This table presents tests for various components of pay. The main independent variable, P restige, is a dummy variable equal to one if a firm belongs to the 100 most admired companies according to the FTMA ranking, and zero otherwise. All dependent variables are included as first differences. In column (1), we use the logarithm of (salary+1), in column (2) the logarithm of (bonus+1), in column (3) we use the logarithm of the Black-Scholes option value, in column (4) we use the logarithm of restricted stock (rstk+1), in column (5) we use the logarithm of all other types of pay (=tdc1-options-stocksalary-bonus+1), and in column (6) we use pay-for-performance sensitivity computed as in Edmans, Gabaix, and Landier (2009). The regession specification is based on column (3) in Table 5 with year times industry fixed effects and control variables are in first differences. Standard errors are clustered at the firm level. t-statistics are provided in parentheses. Prestigei,t−1 ∆ Controls included Size deciles Industry × Year FE Adj. R2 Observations ∆ Salary ∆ Bonus ∆ Options (3) ∆ Restr. Stock (4) ∆ Other Pay (5) ∆ Pay-Perf. Sens. (6) (1) (2) –0.005 (–2.64) –0.027 -0.035 –0.041 –0.018 –0.0023 (–2.06) (–0.91) (–1.75) (–1.35) (–0.02) yes yes yes 0.075 12,958 yes yes yes 0.329 7,966 yes yes yes 0.066 7,763 yes yes yes 0.210 4,591 yes yes yes 0.136 12,080 yes yes yes 0.285 2,175 52 Table 11: Alternative Explanations and Robustness Checks. This table presents tests for alternative explanations and specifications. The main independent variable, P restige, is a dummy variable equal to one if a firm belongs to the 100 most admired companies according to the FTMA ranking, and zero otherwise. Panel A shows results for alternative explanations and presents regressions with ∆T otal P ay as the dependent variable and specifications are based on column (3) in Table 5 with year times industry fixed effects and control variables are in first differences. In column (1) we add a dummy variable indicating family firms as additional control variable. In column (2), we include the number of blockholders as an additional control variable and interact it with firm prestige. In column (3) we include a firm’s industry concentration as an additional control variable. Panel B shows specification tests and presents regressions with ∆T otal P ay as the dependent variable and specifications are based on column (3) in Table 5 with year times industry fixed effects and control variables are in first differences. In columns (1) and (2), we cluster standard errors by year and by year and firm, respectively. In column (3), we restrict the data to all years before 2006. In column (4) we exclude the top and bottom 1% of total compensation in each year. If not indicated differently, standard errors are clustered at the firm level. t-statistics are provided in parentheses. Panel A: Alternative Explanations Family Firm (1) Blockholder (2) Firm Complexity (3) Prestigei,t−1 –0.129 –0.291 –0.126 (–2.92) (–3.02) (–4.22) Family Firm –0.143 (–4.55) Prestige × BLK Holder 0.052 BLK Holderi,t –0.056 (0.50) (–0.81) Industry Concentration –0.014 (–0.55) ∆ Controls included Size deciles Industry ×Year FE Adj. R2 Observations Panel B: Robustness Prestigei,t−1 ∆ Controls included Size deciles Industry × Year FE Adj. R2 Observations yes yes yes 0.495 6,177 yes yes yes 0.466 1,639 Year Clst (1) TwoWay Cls (2) Year < 2006 (3) Drop 1% Outl. (4) –0.126 –0.126 –0.138 (–3.93) (–3.60) (–4.31) yes yes yes 0.533 13,064 yes yes yes 0.533 13,064 yes yes yes 0.523 10,386 53 yes yes yes 0.533 13,064 All Ranked (5) Top 125 (6) Top 75 (7) –0.100 –0.052 –0.103 –0.153 (–4.18) (–2.52 ) (–4.01) (–4.41 ) yes yes yes 0.232 12,570 yes yes yes 0.561 13,064 yes yes yes 0.561 13,064 yes yes yes 0.561 13,064 Figure 1: Discontinuity at Top 100. This graph presents the result of two local polynomial regressions based on total compensation. The figure displays non-parametric estimates for total compensation around the cut-off, which is set to 100 in the FTBC ranking. The sample comprises all firms ranked between 90 and 110. 54
© Copyright 2024