The market for CEOs: An empirical analysis

The market for CEOs: An empirical analysis

Journal of Economics and Business 67 (2013) 24–54 Contents lists available at SciVerse ScienceDirect Journal of Economics and Business The market f...

993KB Sizes 0 Downloads 2 Views

Recommend Documents

No documents
Journal of Economics and Business 67 (2013) 24–54

Contents lists available at SciVerse ScienceDirect

Journal of Economics and Business

The market for CEOs: An empirical analysis夽 Varouj A. Aivazian a, Tat-kei Lai b,∗, Mohammad M. Rahaman c a b c

Department of Economics, University of Toronto, Canada Department of Economics, Copenhagen Business School, Denmark Sobey School of Business, St. Mary’s University, Canada

a r t i c l e

i n f o

Article history: Received 15 July 2011 Received in revised form 9 February 2013 Accepted 20 February 2013 JEL classification: G30 J40 Keywords: CEO turnover CEO skills CEO compensation General skills

a b s t r a c t We investigate empirically a market-based explanation for the rise in recent years in external CEO hiring and compensation and find, consistent with the market-based theory, that firms in industries relying on general managerial skills are more likely to hire CEOs externally than firms in industries relying less on such skills. We show that firms relying on internal CEOs have on average higher profits than external-CEO firms and that the difference in profits decreases as general skills become more important in the industry. We relate managerial skills to compensation and show that CEO general skills induce better firm performance and higher compensation. © 2013 Elsevier Inc. All rights reserved.

1. Introduction The executive labor market in the U.S. has changed in several important ways in recent years: First, more companies are looking outside their organization for leadership than in the past.1 Second,

夽 We thank seminar participants at the Northern Finance Association Meetings 2009 and the Eastern Finance Association Meetings 2010 for many helpful comments. Special thanks to Lalitha Naveen (the Guest Editor) and the anonymous referee for many helpful suggestions that significantly improved the paper. Varouj A. Aivazian and Mohammad M. Rahaman are grateful to the Social Sciences and Humanities Research Council of Canada (SSHRC) for financial support. Any remaining errors are solely ours. ∗ Corresponding author. Tel.: +45 3815 2407; fax: +45 3815 2576.

E-mail address: [email protected] (T.-k. Lai). A Financial Week report (August 20, 2007) shows that during the 1970s and 1980s outside hires accounted for 15% and 17% of all CEO replacements, respectively. In the 1990s, they accounted for 25%, and by 2005 40% of CEO hires were outside CEOs. 1

0148-6195/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jeconbus.2013.02.001

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

25

CEOs from outside the company generally receive bigger pay packages than insiders.2 Finally, overall CEO compensation has risen considerably over the last three decades.3 There is disagreement in the academic literature on the causes of these trends. Bebchuk, Fried, and Walker (2002) argue that the escalation in CEO pay levels reflects a dysfunctional governance system that fails to check the power of entrenched CEOs. Spectacular governance failures at firms such as WorldCom, Tyco, and Enron reinforce the view that CEOs expropriate corporate assets at shareholders’ expense. By contrast, Holmström and Kaplan (2001) argue that the U.S. corporate governance system works relatively well and that any defects associated with the CEO salary structure are more than offset by the competitive advantage of the U.S. corporate governance system. Frydman and Jenter (2010) argue that both rent-extraction and market-based effects are important determinants of CEO compensation, and the empirical validity of different theories remains an open question. We investigate the market-driven explanation for the aforementioned trends in the executive labor market. The market-based theories, such as Murphy and Zábojník’s (2004, 2007) and Frydman’s (2006), posit that CEO salaries are determined by competition among firms for executives and are dependent on transferable CEO skills across firms and industries. As a result, increases in external CEO hiring and executive compensation are explained by increases in the importance of general skills, as distinct from firm-specific skills, in the management of corporations. In a similar vein, Gabaix and Landier (2008) show that the best CEO manages the largest firm through matches in the executive labor market, and that CEOs of larger firms receive higher pay because they have higher marginal productivity (skills). Following this logic, Brynjolfsson and Kim (2009) show that the use of information technology increases the size of the firm and the CEO wage. We follow Murphy and Zábojník’s (2007) partial equilibrium model to develop our empirical hypotheses. In the model, the skill matrix of a CEO is categorized into general skills (human capital transferable across firms and industries) and firm-specific skills (human capital specific to a particular firm). In a competitive labor market, these skills are priced in compensation contracts. As a CEO moves between firms, her firm-specific skills are lost and only transferable or general skills are rewarded. As general skills become more important in an industry, the demand for CEOs with such skills (external CEOs) increases and the price for their services rises. Following these essential elements of the market-based model, we develop three testable empirical hypotheses: First, competition among firms for external CEOs results in the allocation to such CEOs of most of the surpluses generated from their recruitment, whereas firms capture most of the surpluses generated by the firm-specific skills of internal CEOs. As a result, an internal-CEO firm has a higher expected profit than an external-CEO firm. Second, as general management skills become more important to the firm, the surplus to the firm from hiring an internal CEO diminishes because firm-specific skills of such CEOs become less valuable to the firm. Thus, it is more likely for a firm to hire an external CEO as general skills become more important. Moreover, as the benefit to a firm from hiring an internal CEO diminishes, the excess in the expected profits of an internal-CEO firm over an external-CEO firm decreases. To test these hypotheses, we use data for firm-level and CEO-level variables from Compustat, and ExecuComp and industry-level data on information and communication technology (ICT) capital stock per worker from the U.S. Bureau of Economic Analysis and the Bureau of Labor Statistics for the period 1992–2006.4 Extant theoretical and empirical studies in labor economics and industrial organization suggest that ICT is positively associated with the demand for workers with more general skills (see, Acemoglu, Aghion, Lelarge, Van Reenen, & Zilibotti, 2007; Bresnahan, 1999; Bresnahan, Brynjolfsson, & Hitt, 2002). In the literature, we also find that industry-level ICT is positively and significantly correlated with other widely-used proxies for CEO general skills such as CEO education, CEO performance fixed effects, and

2 The same issue of Financial Week also reports that, in 2005, outsider CEO hires at S&P 500 companies earned a median pay of $13 million, compared with $5 million for insiders, according to the Corporate Library. 3 Daines, Nair, and Kornhauser (2005) report that “in 1992 the average CEO of an S&P 500 firm earned $2.7 million. By its peak in 2000, average pay for these CEOs had increased to over $14 million, an increase of more than 400%. The increase in CEO pay is more striking in relative terms. Twelve years ago, CEOs at major U.S. corporations were paid 82 times the average earnings of a blue collar worker; in 2004 they were paid more than 400 times the salary of the average blue-collar worker.” 4 We measure ICT as the logarithm of the value of computer and information technology investment per worker. We explain the detailed definition of this variable in Section 4.

26

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Fig. 1. ICT capital stock per worker across industries over the sample years. This figure shows the box plot of the ICT capital stock per worker across various industries over the sample years. In the figure, the vertical axis depicts the log of ICT capital stock per worker (in constant 2000 dollars) and the horizontal axis depicts the sample years. It shows that ICT increases for all industries over the sample periods, but there remains a considerable amount of heterogeneity across industries within a particular year.

CEO general ability index.5 We argue, based on such findings, that an increase in ICT in an industry increases the warranted level of general managerial skills in the industry. Our first set of results supports the first of the aforementioned hypotheses based on the marketbased model. We find that firms relying on internal CEOs have on average higher profits than externalCEO firms; these results are robust to the potential endogeneity of the CEO-hiring decision. As for economic significance, we find that an internal-CEO firm has 1.6% excess return on total assets over an external-CEO firm after removing the variations in firm performance arising from various firm- and CEO-specific factors and industry and year fixed effects. This suggests that 66.82% of the in-sample unconditional performance difference is explained by the external status of the CEO. The second set of results pertains to the relationship between general skills and a firm’s CEO-hiring decision, and relates to the second hypothesis, based on the market-based model. Using a standard Probit regression analysis, we find that firms in industries using more ICT are more likely to hire external CEOs than those in industries with less ICT. The results are robust to a sample selection bias and are economically significant: A 10% increase in ICT from the mean increases the likelihood of external CEO hiring by 7.67%, conditional on all other explanatory variables evaluated at their mean. We also show that firms relying more on ICT are more likely to hire external CEOs from outside their four-digit SIC industries. Second, we perform an industry-level analysis using a standard Poisson model and show that industries using more ICT in any given year are more likely to hire external CEOs from other industries than those using less ICT (i.e., the between-industry effect of ICT) and that, as an industry becomes more reliant on ICT over time, there is an increase in the likelihood of external CEO hiring from other industries (i.e., the within-industry effect of ICT). The third set of results relates to the last hypothesis, on the relationship between the importance of general skills and firm performance. We find that the wedge between the profits of internal- versus external-CEO firms decreases as the importance of general skills in the industry increases. These results are important in view of the fact that (as shown in Fig. 1) the level of ICT in different industries has increased over the sample period, although a considerable amount of heterogeneity exists across industries. The effect of ICT on the significant excess profits of internal- over external-CEO firms is robust to the endogeneity of the CEO hiring decision. As to economic significance, we find that a 10%

5

See, Custódio, Ferreira, & Matos, in press; Frydman, 2006; Graham, Li, & Qiu, 2012.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

27

increase in ICT capital stock per worker from the mean is associated with a 2.15% increase in the postturnover firm performance when the firm hires an external instead of internal CEO. Finally, we use firm performance as a channel to link CEO skills to CEO compensation. We find that the general component of the CEO skills affects firm performance, and that better firm performance in turn leads to higher CEO compensation. An increase over time in the warranted level of CEO general skills translates into higher CEO compensation, all else being equal. Our paper makes two main contributions. First, we provide direct evidence in support of the marketbased explanation of recent trends in the executive labor market. In particular, our results show that external CEO hiring increases with the importance of general skills and that CEO general skills contribute to higher firm performance, leading in turn to higher CEO compensation. Second, we use the exogenous level of the industry ICT stock per worker as the proxy for the importance of general skills. This enables us to empirically identify the effect of CEO skills on firm performance and CEO compensation, without resorting to managerial fixed effects used in the extant literature to measure CEO skills. The rest of the paper is organized as follows. In Section 2, we discuss the literature on ICT and managerial skills. In Section 3, we develop the three empirically testable hypotheses based on the market-based theory of external CEO hiring and CEO compensation. Section 4 describes the data and variables. In Section 5 we discuss the empirical specifications and the evidence for the three testable hypotheses, and Section 6 concludes the paper. 2. Does ICT capture the demand for managerial general skills? A major empirical challenge for a study of the relationship between CEO skills and outcomes such as firm performance, CEO hiring decisions, and CEO compensation is to have an accurate measure of CEO skills. Simple measures such as CEO age, tenure in the firm, and educational background are commonly used as proxies for CEO skills in a univariate analysis. For example, Murphy and Zábojník (2007) argue that the decline in CEO tenure in the firm (a proxy for the CEO’s firm-specific skills) and the increase in the proportion of CEOs with MBA degrees (a proxy for the CEO’s general managerial abilities) in recent years indicate that CEO general skills have become more important than firm-specific skills. Financial economists have sought better skill measures to study the relationship between CEO characteristics and firm outcomes. For example, Bertrand and Schoar (2003) show that a significant level of the heterogeneity in investment, financing, and organizational practices of firms can be explained by the presence of manager fixed effects. This suggests that when a CEO moves between firms, her firmspecific skills may be lost, and managerial fixed effects could be proxying for general skills. Frydman (2006) collects data from biographical sources on executives’ backgrounds including education and career paths to construct an index of executive general human capital. She argues that this index reflects the increasing importance of general human capital, and uses it to explain the increasing wage inequality among top managers within firms and the increasing mobility of senior executives across firms. Graham et al. (2012) construct their skill measure by decomposing executive compensation into firm and manager components, and argue that firms hiring CEOs with larger compensation fixed effects will improve their performance. Rost, Salomo, and Osterloh (2008) use a measure depicting dishonesty of the departing CEO and argue that an increase in this measure makes external CEO hiring more likely. Custódio et al. (in press) use information on a CEO’s industry background, experience as a top executive, and educational training to construct an index of general managerial ability. Note that the skill measures based on CEO background are potentially endogenous and the measures based on managerial fixed-effects cannot proxy for changes of CEO skills over time. By contrast, the industry-level ICT measure is less likely to be affected by idiosyncratic managerial attributes, and ICT varies over time and across industries and captures changes in skill requirements in an industry. A question arises: Do changes in industry-level ICT serve as a measure of demand for general managerial skills for firms? First, the ‘skill-biased technological change’ literature in labor economics and industrial organization provides theoretical explanations and empirical evidence of a positive relationship between ICT and demand for general skills. On the theoretical side, Bresnahan (1999) argues that there is complementarity between computers and workers who possess cognitive and analytical skills; that is, while it is easy to replace workers with computers in routine tasks it is very

28

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

costly, if not impossible, to program a computer to perform non-routine tasks as efficiently as a human. Therefore, he argues, the greater use of computer technology raises the demand for skilled workers. Acemoglu et al. (2007) argue that a higher level of ICT leads to a greater decentralization of decisionmaking rendering firm-specific knowledge less important for top-level managers. Empirically, Autor, Katz, and Krueger (1998) report supportive evidence on the relationship between the demand for skilled workers and the use of ICT at the industry level. Using firm-level data, Bresnahan et al. (2002) find a positive relationship between computer use (in terms of IT capital and total number of PCs) and the hiring of more educated workers. Caroli and Van Reenen (2001) find evidence of complementarity between technological change and organizational change (in terms of decentralization of authority, delayering, and increased multi-tasking) for British and French establishments in the 1980s and 1990s. In sum, a consensus has emerged in the literature that increased ICT intensification simultaneously raises the demand for general-skilled workers. Recent innovations in management tools, induced by increasing computerization, have also increased firms’ reliance on general management skills and ICT (e.g., Alexopoulos & Tombe, 2012; Harland, 1996). For instance, the Management Tools & Trends 2011 Survey6 by Bain & Company finds that Benchmarking, Knowledge Management, and Supply Chain Management are among the widelyused management tools in Europe and North America. These management techniques that emerged as a result of increasing ICT intensification of management processes require high levels of general skills on the part of the manager (Bain & Company, 2011). By contrast, more dated management techniques, such as Total Quality Management (TQM) and Six Sigma, focus on product-specific quality improvements and rely less on general skills and ICT in the management process. Finally, the increasing importance of ICT in modern corporations (Chun, Kim, Morck, & Yeung, 2008) and the importance of general skills in organizational management are emphasized by the popular press and by successful CEOs.7 Based on the foregoing, a strong argument can be made that changes in ICT over time have altered the demand for managerial skills. Computerization has led to a greater demand for workers with cognitive and analytical skills and to a greater decentralization of decision-making within organizations for tasks requiring firm-specific knowledge. These trends suggest that a higher level of ICT use by firms reflects a greater demand for managerial general, as opposed to firm-specific, skills by such firms. Thus, understanding the effect of CEO general skills on firm performance requires an understanding of the process of technological change in the firm’s industry. 3. Economic framework In this section, we explain the economics of the market-based model of CEO compensation and external CEO hiring, and develop three hypotheses for empirical analysis. We describe in Appendix A a more formal theoretical framework based on Murphy and Zábojník (2007) to show how CEO general skills affect the differential profitability of external- and internal-CEO firms.8 In the market-based theoretical framework (described in Appendix A), a firm with a vacant CEO position faces a tradeoff between promoting an internal candidate and hiring an external CEO. The value of a CEO to a firm depends on her general and firm-specific skills. Both the general and firmspecific skills of an internal candidate are relevant to the hiring firm, whereas only general skills of an external candidate are relevant. In the external CEO market, given free entry of firms, competition among firms for external CEOs allows external candidates to capture most of the surplus, leaving near-zero expected profit to firms. On the other hand, a firm hiring an internal CEO only needs to offer a wage equal to her outside option, which is her wage in the external CEO market based on her general skills. Thus, an internal-CEO firm earns a positive expected profit from the surplus generated by the firm-specific skills of the internal CEO. This leads to our first testable hypothesis: 6

http://www.bain.com/publications/business-insights/management-tools-and-trends-2011.aspx. See, The Economist, November, 2009: http://www.economist.com/node/14742618. The theoretical framework and the empirical hypotheses are based on the partial-equilibrium version of the Murphy and Zábojník (2007) model. They also have a general-equilibrium version of the model and its theoretical predictions are similar. 7 8

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Hypothesis 1.

29

An internal-CEO firm has higher expected profits than an external-CEO firm.

If all CEOs have the same ability, then internal CEO hiring is preferred because firm-specific CEO skills generate higher profits. But when the ability of the internal candidate is sufficiently different from what is optimal to the hiring firm, the firm may benefit from hiring an external CEO. Even though only the general skills of the external CEO are useful, the external candidate’s higher ability compensates for the difference. The model postulates that when the internal candidate’s ability falls within a certain range, it is optimal for the firm to promote her; otherwise, it is optimal to hire an external CEO. As general skills increase in importance, CEOs with more general skills become more valuable to the hiring firm. Thus, when the ability of the internal candidate falls outside the optimal range, it is easier for the hiring firm to find a suitable candidate in the external CEO market. At the same time, the surplus that the firm captures by hiring an internal CEO becomes smaller because the firm-specific skills of internal candidates are less valuable to the firm; that is, the advantage in expected profits to internal-CEO firms relative to external-CEO firms diminishes with the increasing importance of general skills. The foregoing discussion follows Murphy and Zábojník (2007) and leads to our second and third testable hypotheses as follows: Hypothesis 2. A firm is more likely to hire an external CEO as general skills become more important. Hypothesis 3. The excess in expected profits of an internal-CEO firm over an external-CEO firm decreases as general skills become more important. Taken together, these hypotheses suggest that an outside CEO improves firm performance, especially as general skills become more important. Firms relying more on general skills increase the demand for external CEOs and competition among firms for such CEO services pushes up CEO compensation. The interaction between warranted CEO skills and firm performance leads to a rise in CEO compensation as the importance of general skills increases. 4. Data and variables 4.1. Data To form the initial sample we begin with the set of firms that was listed in the S&P 500 Index for at least one year between 1992 and 2006. The rationale for focusing on the S&P 500 firms is three-fold: First, S&P 500 firms are broadly representative of the U.S. industrial and service sectors so that empirical regularities identified in this sample can be generalized to other firms in the economy. Second, to be included in the S&P 500, a firm has to perform above a certain threshold, which in turn makes such a sample homogeneous along certain quality (performance) dimensions. Focusing on this quasihomogeneous (in terms of firm quality) sample of firms lessens the possibility of endogeneity due to unobserved firm characteristics that may confound the identification of the regression coefficients. Finally, since some of the CEO characteristics are hand collected, it is practical to focus on the S&P 500 constituents to keep the sample manageable. For firms in the initial sample, we collected firm-specific variables from the Compustat and CEO-specific variables from the ExecuComp database. We also hand collected information from the Marquis Who’s Who Directory, Forbes’ People Tracker, Factiva database, and proxy statements of the firms. The data for the industry-level ICT measure are from the U.S. Bureau of Economic Analysis (BEA) and U.S. Bureau of Labor Statistics (BLS).9 If the CEO in year t of a firm is different from the CEO in year t − 1, we identify this as a CEO-turnover event in our sample. We then apply two additional filters to form our CEO-turnover sample: for each CEO-turnover event the book-value of assets of the firm must be positive and the information on the industry-level ICT measure must be available. These selection criteria result in 983 cases of CEO turnovers.10

9 We use the conversion table published by the U.S. Census Bureau to match a firm’s NAICS-SIC (given in Compustat) with 2-digit BEA industry classifications so that the ICT measure is comparable across industry classifications. 10 Note that sample size varies in our regression analysis depending on the availability of data on all firm- and CEO-specific variables.

30

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

4.2. Main variables 4.2.1. Firm-performance measure We use two performance measures. The first measure is the cumulative returns on a firm’s total assets (CROAt+1,t+2 ). For each CEO turnover in year t, we calculate the Return on Total Assets (ROA) which is the “Net Income” variable divided by the “Total Assets” variable in Compustat. We then cumulate the firm-level ROA over years t + 1 and t + 2 to construct the CROAt+1,t+2 measure. The second measure is the peer-adjusted cumulative returns on a firm’s total assets (CAROAt+1,t+2 ). To calculate the CAROAt+1,t+2 , we first calculate the ROA for each CEO turnover in year t. We then subtract the median ROA of the S&P 500 constituent firms from the firm-level ROA and cumulate the adjusted ROA over years t + 1 and t + 2.11 We use a two-year window to calculate cumulative returns for two reasons: First, enlarging the event window by more than two years to calculate the cumulative abnormal returns makes it difficult to attribute the abnormal performance to the CEO turnover because over a longer horizon other factors may affect a firm’s performance. Second, when a new external CEO has been in the position for a long time, she will acquire firm-specific knowledge so that the firm’s performance reflects the contributions of both general and specific skills. Note that in the empirical specifications of the first and third hypotheses discussed earlier, we are interested to see how the CEO hired in year t affects firm performance in years t + 1 and t + 2. Hence, we require the CEO of a firm in year t to be in the same position throughout years t + 1 and t + 2, and the CEOs in years t − 1 and t − 2 to be the same person. This excludes interim CEO cases, and ensures that the CEO in year t is solely responsible for firm performance in years t + 1 and t + 2. A consequence of this construct is that the sample size for the CROAt+1,t+2 and CAROAt+1,t+2 variables are smaller because, for a given firm, the variable cannot be constructed for the last two years of observations of the firms in the sample. A similar reasoning also applies when we construct the CEO compensation measure below. 4.2.2. CEO compensation Following Custódio et al. (in press) and Frydman and Saks (2010), we use the “TDC1” variable from ExecuComp to construct this variable. This variable measures total CEO pay (in thousand dollars), which consists of salary, bonus, value of restricted stock granted, value of options granted, long-term incentive payout, and other compensations. For a newly-appointed CEO in year t, we define the cumulative CEO wage (CWAGE t+1,t+2 ) as the logarithm of the sum of CEO’s wages in years t + 1 and t + 2. Similar to our second firm-performance measure, we also construct a peer-adjusted CEO wage variable. For a newly appointed CEO in year t, we define the peer-adjusted cumulative CEO wage (CAWAGE t+1,t+2 ) as the logarithm of the sum of CEO’s wages in years t + 1 and t + 2, adjusted by the median CEO wage of S&P 500 constituents CEOs. 4.2.3. External CEO Following Murphy and Zábojník (2007) and Custódio et al. (in press), we define an external CEO hire as follows: a newly appointed CEO is considered an external CEO if her tenure in the firm was less than or equal to one year when she became the CEO; otherwise, she is considered an internal CEO.12 The tenure of the CEO in the current firm is calculated by using the “date of becoming CEO” and “date joined the firm” variables in ExecuComp. 4.2.4. Information & communication technology (ICT) We use data on private assets from the National Income and Product Accounts (NIPA) tables of BEA, and on the number of workers in different industries from the Current Employment Statistics (CES)

11 We also use “Earnings before Interest and Taxes” (EBIT) as the measure of income, and normalized with “Total Sales” instead of “Total Assets.” The main results are similar. 12 The regression results reported below are robust if we use the following alternative definition of an external CEO: an executive who has two years or less tenure in the current firm at the time of CEO appointment.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

31

published by BLS, to construct our ICT measure. For each industry j and year t, we define the ICT measure (in constant 2000 dollars) as the logarithm of computer and information technology investment per worker13 :



ICT jt = log

Value of Computer & Communication Equipment & Softwarejt Total Number of Workersjt



.

Fig. 1 shows a box plot of the ICT measure for all industries over the sample years. It shows that the ICT level increases for all industries over the sample period, but there remains a considerable amount of variation across industries in any year. 4.2.5. Other controls Firm size has been shown to be an important determinant of the trend of increasing CEO compensation (see, Brynjolfsson & Kim, 2009; Gabaix & Landier, 2008). We use the logarithm of firm total assets to control for firm size. Aside from firm size, leverage and asset tangibility are also important determinants of a firm’s external financial access. We use the debt-to-asset ratio and the ratio of net fixed assets over total assets to proxy, respectively, leverage and asset tangibility (e.g., Custódio et al., in press). To control for a firm’s growth opportunities, we use the firm’s Tobin’s Q. The corporate governance structure of the firm is also an important determinant of CEO hiring decisions and of CEO compensation (see, Bebchuk et al., 2002; Holmström & Kaplan, 2001). To control for the state of corporate governance of the firm, we use the Gompers, Ishii, and Metrick (2003) corporate governance score, generally known as the G index. To control for pre-turnover performance, we use the equity return two years prior to the CEO turnover. Following Kale, Reis, and Venkateswaran (2009) and Custódio et al. (in press), we control for firm-level investment (CAPX/SALES, R&D/CAPX, ADVERT /CAPX) and payout (DIVYIELD) policies, agency problems (TOP5MGT ), and business risk (VOLATILITY). We control for the age of the CEO, obtained from ExecuComp, and for educational background. We define an MBA dummy, which is equal to 1 if the CEO holds a Master of Business Administration degree or equivalent. The information is hand collected from the Who’s Who Directory, Forbes’ People Tracker, and the CEO’s biography posted on the firm’s web site. The definitions of these variables are given in Table A.1 in Appendix A. 4.3. Descriptive statistics Table 1 reports the summary statistics of the firm, CEO, and industry-specific variables for the turnover sample. Panel A shows the summary statistics for firm characteristics. Panel B shows that, on average, 28% of the CEOs are recruited externally. Although we do not report the statistics here, in our data we see a clear upward trend in external CEO hires over the sample period: between 1992 and 1996, 26% of new CEOs are recruited externally; between 1997 and 2001, 28% of new CEOs are recruited externally; and between 2002 and 2006, 30% of new CEOs are recruited externally. In panel C, we report the summary statistics of the industry-level ICT capital stock per worker both in levels and logs.14 Table 2 shows summary statistics for ICT, external CEO hiring and CEO wage by various industries. It reveals that the ‘Information & Telecom’ industry, on average, has the highest level of ICT and ‘Accommodation & Food Services’ has the lowest level of ICT. Furthermore, firms in the highest ICT industry are more likely to hire CEOs externally and also from across industries compared to firms in the lowest ICT industry. These firms are also more likely to pay higher compensation to CEOs compared to those in the lowest ICT industry. This is consistent with our hypotheses discussed earlier. Table 3 compares firm- and CEO-specific control variables of the internal-CEO and external-CEO firms. It shows that external-CEO firms’ performance is lower two years following a CEO turnover.

13 See U.S. Department of Commerce (2003) for more details about the construction of a quality-adjusted price index for computer and other equipment. 14 We use the level and log of the ICT variable in our regression estimation and the results are robust to such transformation.

32

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Table 1 Summary statistics of the turnover sample. Mean

Min

Max

0.08 −0.01 8.68 0.25 1.47 0.23 0.24 0.00 0.00 0.34 0.01 0.02 10.00 0.23

−0.46 −0.55 5.07 0.00 0.85 0.00 0.01 0.00 0.00 0.14 0.00 −0.73 4.00 0.00

0.46 0.37 12.70 0.89 9.36 0.80 3.68 3.00 1.96 2.32 0.19 1.73 15.00 29.51

0.14 0.14 1.45 0.23 1.30 0.17 0.63 0.43 0.23 0.31 0.02 0.43 2.54 3.82

767 767 983 963 981 982 962 983 983 946 929 966 938 982

PanelB : CEO − specificvariable 8.76 CWAGE t+1,t+2 −0.21 CAWAGE t+1,t+2 53.17 CEOAGE 0.38 CEOMBA 0.28 EXTCEO 0.03 DIFFIND EXTCEO

8.74 −0.20 53.00 0.00 0.00 0.00

6.41 −2.73 37.00 0.00 0.00 0.00

11.88 2.49 75.00 1.00 1.00 1.00

0.91 0.86 6.56 0.49 0.45 0.18

650 650 983 982 983 983

2.97 2.77 1.09 1.02 0.28

0.16 0.15 −1.84 −1.88 0.15

66.73 66.73 4.20 4.20 0.57

14.53 13.48 1.24 1.25 0.09

983 983 983 983 983

PanelC : Industry − specificvariables 8.39 LEVEL − ICT ijt 7.67 LEVEL − ICT ijt−1 1.27 LOG − ICT ijt 1.16 LOG − ICT ijt−1 0.30 INDHOMOj

SD

N

Median

PanelA : Firm − specificvariables 0.07 CROAt+1,t+2 −0.03 CAROAt+1,t+2 8.83 FIRMSIZE 0.30 TANGIBILITY  1.92 TOBIN SQ 0.25 LEVERAGE 0.47 CAPX/SALES 0.18 R&D/CAPX ADVERT /CAPX 0.08 0.41 VOLATILITY 0.02 DIVYIELD 0.06 PASTRETURN 9.80 GINDEX 1.29 TOP5MGT

This table presents the summary statistics of various variables from the CEO turnover sample. The turnover sample contains the newly-appointed CEOs in S&P 500 firms between 1992 and 2006. Panels A and B show the summary statistics of the firm- and CEO-specific variables. Panel C reports the summary statistics for the industry-level information and communication technology (ICT) and Parrino (1997) industry homogeneity measures. The definitions of theses variables are the same as in Table A.1 in Appendix A.

This unconditional difference is consistent with our first hypothesis that internal-CEO firms show, on average, higher profits compared to external-CEO firms. Panel A of the table also shows that externalCEO firms are smaller in size and asset tangibility compared to internal-CEO firms. Such firms also have higher intangibles (R&D), lower managerial equity ownership and stock return prior to the CEO turnover. Since we use the turnover sample for our regression analysis, the differences shown in Panel A warrant a robust regression analysis to control for a potential sample-selection bias. In panel B, we do not observe any systematic difference between external- and internal-CEO-specific characteristics. Finally, panel C of the table shows that industry-level ICT is higher for an external-CEO firm than for an internal-CEO firm.

4.4. Cross-validation analysis Before discussing our regression analysis, we present a cross-validation analysis to justify the use of ICT variables as a proxy for the importance of general skills. We first examine how the ICT variables are correlated with other widely-used measures of managerial general skills. Ever since Bertrand and Schoar (2003), managerial fixed-effects have been widely used to proxy for unobserved managerial general skills. We follow Graham et al. (2012) and estimate performance fixed-effects for CEOs in

Table 2 Industry-level ICT, external CEO hiring, and CEO wage: summary statistics. Description

21

Mining, quarrying, and oil and gas exploration Utilities Construction Manufacturing Wholesale trade Retail trade Transportation and warehousing Information and telecommunication Finance and insurance Real estate and retail and leasing Professional, scientific, and technical services Administrative services Health-care and social assistance Arts, entertainment, and recreation Accommodation and food services

22 23 31 42 44 48 51 52 53 54 56 62 71 72

ICT (Mean)

ICT (Median)

CEO Turnover

Internal CEO

External CEO

Same SIC EXT

Diff. SIC EXT

Internal (%)

External (%)

Same SIC EXT (%)

Diff. SIC EXT (%)

CWAGE (Mean)

4.31

4.01

31

23

8

8

0

74.19%

25.81%

25.81%

0.00%

8.41

11.40 0.66 2.46 3.16 0.69 5.92

10.09 0.76 2.76 2.85 0.70 6.47

79 9 471 18 79 26

58 9 337 13 56 21

21 0 134 5 23 5

18 0 112 5 21 4

3 0 22 0 2 1

73.42% 100.00% 71.55% 72.22% 70.89% 80.77%

26.58% 0.00% 28.45% 27.78% 29.11% 19.23%

22.78% 0.00% 23.78% 27.78% 26.58% 15.38%

3.80% 0.00% 4.67% 0.00% 2.53% 3.85%

8.53 8.34 8.73 8.54 8.55 8.54

50.10

50.41

92

59

33

28

5

64.13%

35.87%

30.43%

5.43%

9.09

8.36 10.74

8.86 10.74

125 2

96 2

29 0

28 0

1 0

76.80% 100.00%

23.20% 0.00%

22.40% 0.00%

0.80% 0.00%

9.12 10.10

8.19

8.52

10

6

4

4

0

60.00%

40.00%

40.00%

0.00%

8.49

1.25 0.47

1.13 0.47

8 12

5 8

3 4

3 4

0 0

62.50% 66.67%

37.50% 33.33%

37.50% 33.33%

0.00% 0.00%

8.37 8.67

0.60

0.60

2

2

0

0

0

100.00%

0.00%

0.00%

0.00%

8.43

0.25

0.20

19

14

5

5

0

73.68%

26.32%

26.32%

0.00%

8.74

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

BEA Ind.

This table provides summary statistics for ICT and CWAGE for 2-digit Bureau of Economic Analysis (BEA) industries. It also shows the total number and percentage of external and internal CEO hiring for these industries. The definitions of all relevant variables are given in Table A.1 in Appendix A.

33

34

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Table 3 Summary statistics: differences between external and internal CEOs. Median difference

Mean difference

2 -stat

t-Stat

N

Difference

PanelA : Firm − specificvariables −0.039*** CROAt+1,t+2 −0.040*** CAROAt+1,t+2 −0.255** FIRMSIZE −0.053*** TANGIBILITY −0.119 TOBIN  SQ −0.002 LEVERAGE −0.04 CAPX/SALES R&D/CAPX 0.114*** ADVERT /CAPX 0.012 0.027 VOLATILITY −0.001 DIVYIELD −0.076** PASTRETURN 0.231 GINDEX −0.540** TOP5MGT

[−3.34] [−3.58] [−2.47] [−3.22] [−1.28] [−0.17] [−0.89] [3.75] [0.72] [1.21] [−0.33] [−2.44] [1.24] [−1.99]

767 767 983 963 981 982 962 983 983 946 929 966 938 982

−0.023** −0.020*** −0.112 −0.054*** 0.079 −0.006 −0.048** 0.000 0.000 0.017 −0.002 −0.058* 0.000 −0.058**

[−2.46] [−3.86] [−0.75] [−2.75] [1.15] [−0.40] [−2.23] [0.00] [0.00] [1.21] [−0.96] [−1.70] [0.00] [−1.96]

767 767 983 963 981 982 962 983 983 946 929 966 938 982

PanelB : CEO − specificvariables CWAGE t+1,t+2 −0.013 −0.078 CAWAGE t+1,t+2 −0.272 CEOAGE 0.003 CEOMBA

[−0.16] [−1.03] [−0.58] [0.08]

650 650 983 982

−0.054 −0.135 0.000 0.000

[−0.48] [−1.32] [0.00] [0.00]

650 650 983 982

PanelC : Industry − specificvariables 1.609* LEVEL − ICT ijt LEVEL − ICT ijt−1 1.330* LOG − ICT ijt 0.119* 0.120* LOG − ICT ijt−1 −0.011* INDHOMOj

[1.65] [1.66] [1.68] [1.65] [−1.81]

983 983 983 983 983

0.028 0.000 0.009 0.000 −0.017***

[0.39] [0.00] [0.39] [0.00] [−22.19]

983 983 983 983 983

Difference

N

This table reports the differences in means and medians of the external CEO sample and the internal CEO sample. Column (1) of the table reports the difference in means and column (2) reports the t statistic of the hypothesis that the means are the same. Column (3) reports the number of observations used in the mean difference test. Columns (4)–(6) report the median difference test with column (5) reporting the 2 statistic of the hypothesis that the medians are the same. The definitions of all variables are the same as in Table A.1 in Appendix A. * Significance at the 10% level. ** Significance at the 5% level. *** Significance at the 1% level.

our sample.15 Table 4 shows the correlations between our estimated CEO performance fixed-effects obtained from a firm-performance equation and the ICT variables. The correlation is positive and statistically significant at the 1% level. Custódio et al. (in press) use a CEO’s lifetime work experience to construct a proxy of general skills that are transferable across firms and industries. Table 4 shows that ICT variables are positively correlated with their general ability index, and the correlation is significant at the 1% level. Finally, CEO education is another widely-used measure of CEO general skills and we show in Table 4 that our ICT measures are positively and significantly correlated with the CEO MBA dummy. These correlation statistics suggest that our ICT measures are indeed related to the other common measures of managerial general skills used in the literature. However, our ICT measures improve on the existing measures of managerial skills, at least on two fronts: First, managerial fixedeffects are, by construction, fixed for a particular manager and hence cannot capture the learningby-doing human capital accumulation by managers. By contrast, the ICT measures that we use vary across time and industries, and thus capture any changes in skill requirements in an industry. Second, CEO-level proxies for skills are more prone to endogeneity and selection bias. For instance, whether

15 To be precise, we regress the “Return on Total Assets” (Net Income/Total Assets) on a set of control variables, including Total Assets, Market-to-Book Ratio, the tenure of the CEO. We obtain the managerial fixed-effects following the procedures in Graham et al. (2012).

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

35

Table 4 Cross-validation check for the ICT variables. [1] [1] LEVEL − ICT jt [2] LEVEL − ICT jt−1 [3] LOG − ICT jt [4] LOG − ICT jt−1 [5] CEO − MBA [6] CEO − GABILITY [7] CEO − FIXEDEFFECT [8] EXT − FRACTION [9] CEO − WAGE

1.00 0.98 [0.00] 0.79 [0.00] 0.79 [0.00] 0.01 [0.13] 0.18 [0.00] 0.01 [0.00] 0.14 [0.00] 0.11 [0.00]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

1.00 0.78 [0.00] 0.79 [0.00] 0.01 [0.26] 0.19 [0.00] 0.01 [0.00] 0.11 [0.00] 0.10 [0.00]

1.00 1.00 [0.00] 0.06 [0.00] 0.29 [0.00] 0.08 [0.00] 0.23 [0.00] 0.17 [0.00]

1.00 0.05 [0.00] 0.31 [0.00] 0.08 [0.00] 0.23 [0.00] 0.17 [0.00]

1.00 0.10 [0.00] 0.04 [0.00] 0.07 [0.03] 0.06 [0.00]

1.00 0.00 [0.80] 0.63 [0.00] 0.24 [0.00]

1.00 −0.01 [0.77] 0.33 [0.00]

1.00 0.10 [0.01]

1.00

This table shows the correlation between the ICT measure with other widely used proxies for managerial general skills and also with CEO compensation in the turnover sample. The managerial fixed-effects CEO − FIXEDEFFECT is from Graham et al. (2012). CEO − GABILITY is from Custódio et al. (in press). EXT − FRACTION is the fraction of externally hired CEOs in a given year in the turnover sample. The definitions of all variables are the same as in Table A.1 in Appendix A. P-values of significance are given in the bracket.

or not a CEO has an MBA degree may not be random in a given sample of CEOs which, in turn, can confound the identification of CEO education (skills) on firm outcomes. The ICT measures are at the industry level and are less likely to be affected by idiosyncratic managerial attributes. Next, we investigate whether our proxy for the importance of general skills is related to CEO hiring and CEO wages. Our economic framework above posits that external CEO hiring and CEO wages increase with the importance of general skills. Table 4 shows the correlations between the industrylevel ICT, executive compensation, and the fraction of external CEOs relative to total new CEOs during the sample years. The correlation analysis supports our hypotheses that CEO wage and external CEO hiring are significantly and positively correlated with our ICT variables. Although correlation does not imply causation, we are able to utilize our proxy to test the hypotheses discussed earlier. 5. Empirical specifications and results 5.1. External CEO hiring and firm performance Our first empirical hypothesis asserts that an internal-CEO firm has, on average, higher economic profits than an external-CEO firm. To test this hypothesis, we estimate the following linear model:  CROAijt+1,t+2 = ˛ + ˇ · EXTCEOijt + Xijt ı + j + t + εijt ,

(1)

where i, j, t are the indices for CEO-firm pair, industry, and year, respectively; CROAijt+1,t+2 is the cumulative return on total assets in years t + 1 and t + 2 for firm i in industry j with a CEO turnover in year t; EXTCEOijt is the external CEO hire dummy; Xijt contains the firm-level and CEO-level control variables; j and t are the industry fixed effects and year fixed effects, respectively; and εijt is the error term. If the internal-CEO firms have higher profits relative to the external-CEO firms, we should expect that the external CEO hire dummy has a negative coefficient, i.e., ˇ < 0. Note that we control for industry and year fixed effects in the cross-section of the CEO turnover sample so that the identification is provided mainly by the changes in EXTCEOijt between firms in an industry over time. Also, we have a few repeated turnovers at the firm level, but there is a significant number of repeated turnovers within the same industry in our sample. Thus, to allow for a serial correlation among the error terms

36

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Table 5 External CEO hiring and firm performance. OLS

OLS with control function

CROAt+1,t+2 [1] EXTCEOijt

CAROAt+1,t+2 [2]

−0.016** [−2.52]

VOLUNTARY − EXTCEOijt

[3]

[4]

−0.017** [−2.72] −0.022** [−2.52]

CROAt+1,t+2 [5]

CAROAt+1,t+2 [6]

−0.017** [−2.59] −0.026** [−2.80]

[7]

[8]

−0.017** [−2.56] −0.022** [−2.44]

−0.026** [−2.77]

Firm&CEOcharacteristics −0.010 FIRMSIZE [−1.39] 0.244*** TANGIBILITY [10.54] 0.078*** TOBIN  SQ [24.56] 0.076* LEVERAGE [1.86] −0.042 CAPX/SALES [−1.66] −0.074*** R&D/CAPX [−3.39] ADVERT /CAPX 0.053*** [5.22] −0.108** VOLATILITY [−2.95] DIVYIELD 0.672*** [3.23] PASTRETURN −0.029*** [−3.08] GINDEX 0.000 [0.08] 0.002 TOP5MGT [1.50] 0.001 CEOAGE [1.39] 0.008 CEOMBA [0.68] 0.181 Constant [1.70]

−0.006 [−0.90] 0.203*** [10.19] 0.083*** [19.86] 0.086** [2.37] −0.057** [−2.36] −0.091*** [−3.78] 0.062* [1.83] −0.102** [−2.94] 0.409* [1.97] −0.029** [−2.33] 0.002** [2.34] 0.002 [1.53] 0.002** [2.73] 0.006 [0.55] 0.045 [1.13]

−0.009 [−1.45] 0.227*** [6.72] 0.077*** [36.37] 0.098** [2.48] −0.050** [−2.19] −0.064*** [−3.04] 0.050*** [5.43] −0.117*** [−3.41] 0.670*** [3.28] −0.027** [−2.76] −0.001 [−0.32] 0.002* [1.87] 0.001 [0.66] 0.005 [0.32] −0.098 [−0.86]

−0.004 [−0.77] 0.193*** [6.88] 0.080*** [32.43] 0.103** [2.86] −0.066** [−2.53] −0.074** [−2.65] 0.060 [1.77] −0.117*** [−4.09] 0.411* [1.97] −0.025* [−2.07] 0.001 [1.47] 0.003* [1.81] 0.002* [1.96] 0.002 [0.14] −0.250*** [−6.50]

−0.044** [−2.65] −0.120 [−0.50] 0.059*** [3.19] 0.286* [2.11] −0.001 [−0.03] 0.011 [0.14] 0.109*** [3.57] −0.171** [−2.78] 1.400*** [3.59] −0.130** [−2.81] 0.010 [1.48] −0.007 [−1.04] −0.000 [−0.02] 0.023 [1.25] 0.910* [2.10]

−0.003 [−0.20] 0.246 [1.26] 0.086*** [5.82] 0.071 [0.90] −0.063 [−1.44] −0.102 [−1.74] 0.058* [1.96] −0.097** [−2.40] 0.376 [1.47] −0.023 [−1.03] 0.001 [0.21] 0.003 [0.63] 0.002* [1.99] 0.006 [0.59] −0.035 [−0.09]

−0.028* [−2.08] −0.050 [−0.27] 0.055*** [3.25] 0.191* [1.96] −0.012 [−0.45] 0.016 [0.23] 0.076*** [3.45] −0.143** [−2.73] 0.912*** [3.15] −0.067** [−2.73] 0.005 [1.05] −0.004 [−0.66] −0.000 [−0.60] 0.009 [0.60] 0.383 [1.09]

−0.009 [−0.49] 0.127 [0.56] 0.075*** [4.28] 0.126 [1.48] −0.057 [−1.22] −0.057 [−0.81] 0.065** [2.27] −0.123*** [−3.13] 0.462 [1.67] −0.033 [−1.45] 0.002 [0.55] 0.002 [0.26] 0.001 [1.13] 0.003 [0.25] −0.127 [−0.30]

Controlfor Industry fixed effects Year fixed effects

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

R2 Num. of obs.

0.50 653

0.52 421

0.44 653

0.45 421

0.41 653

0.52 421

0.44 653

0.45 421

This table reports results from regression analysis of the effect of hiring an external CEO on a firm’s performance. The first dependent variable is the financial performance of a firm measured by cumulative return on assets, CROAt+1,t+2 , two years following the CEO turnover. The second dependent variable is adjusted cumulative return on assets, CAROAt+1,t+2 , two years following the CEO turnover. The definitions of all variables are the same as in Table A.1 in Appendix A. Standard errors are clustered at the industry level. * Significance at the 10% level. ** Significance at the 5% level. *** Significance at the 1% level.

in the same industry, we cluster the standard errors by industry and report the robust t-statistics in parenthesis. Table 5 reports the OLS regression results from Eq. (1). Column (1) shows that after removing the performance effects due to firm- and CEO-level characteristics, the firms with new external CEOs have cumulative returns on total assets that are 1.6% lower relative to firms with new internal CEOs. The

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

37

result is similar when we use CAROAt+1,t+2 as our dependent variable in column (3). To get a sense of the economic significance of this performance difference due to external CEO hiring, we first use the in-sample data (653 turnover cases used in the regression estimation) and find that the CROAt+1,t+2 of a typical (median) external-CEO firm is 6.80% and that of an internal-CEO firm is 9.23%. This suggests that 66.82% ([−0.016/(0.068 − 0.0923)] × 100) of the in-sample unconditional performance difference is explained by the external status of the CEO. Second, we find that the size (book value of assets) of the median in-sample external-CEO firm is $4 billion at the time of the CEO turnover. Thus, a 1.6% performance difference between external- and internal-CEO firms translates into $64 million bookvalue loss over two years following the turnover to an external-CEO firm compared to an internal-CEO firm. Finally, we find that the size of the median in-sample firm is $5 billion at the time of the turnover, and that a 1.6% performance decline over two years translates into $80 million book-value loss to the firm if it hires an external CEO.16 Among the other control variables in the regression, our results show that asset tangibility, future-growth opportunity (Tobin’s Q), leverage, return volatility, investment and payout policies, corporate governance (GINDEX), and past performance are significantly related to post-turnover firm performance. Furthermore, CEO age is significantly positively related to firm performance in some specifications. These results are similar to those reported in the literature (e.g., Lang & Stulz, 1994).17 Huson, Malatesta, and Parrino (2004) find that in the case of a forced CEO turnover, there is a strong positive market reaction when an external CEO is appointed. They show that a turnover-induced market reaction is positively correlated with the subsequent accounting-based measures of firm performance. If the interaction between the nature of the CEO turnover (voluntary versus forced) and the type of new-CEO hire (external versus internal) has a bearing on the subsequent firm performance, consistent with our economic framework, we should observe a smaller reduction in post-turnover firm performance in the case of a forced-external CEO hire and a greater reduction in firm performance in the case of a voluntary-external CEO hire. To investigate this we estimate the following regression models:  CROAijt+1,t+2 = ˛v + ˇv (EXTCEOijt × VOLUNTARY ijt ) + Xijt ıv + j + t + εijt

(2)

where VOLUNTARY ijt is a dummy variable; it equals 1 if the new CEO is hired after a voluntary turnover and 0 if the new CEO is hired after a forced CEO turnover. Consistent with our economic framework, we expect that ˇv < ˇ. We hand-collected information on the reasons for the CEO turnover for our sample firms from newspaper articles around the time of the turnover and could clearly identify 176 cases of forced and 351 cases of voluntary turnovers. Table 5 reports the results from regression model (2). It shows that, consistent with the results of Huson et al. (2004) and our economic framework, voluntary-external CEOs have a greater negative impact on post-turnover firm performance than the average external CEO. We also find that ˇ − ˇv ) > 0 and the difference is statistically significant at the 5% level. The foregoing regression results show that an internal-CEO firm has higher profits than an externalCEO firm. However, there is a potential endogeneity problem in the above OLS regression; the decision to hire an external versus an internal CEO may depend on unobservables not controlled for in our regression, and such unobservables may also affect the post-CEO turnover performance.18 Intuitively, endogeneity arises when part of the error term in the baseline regression in (1) is correlated with the external-CEO hiring dummy, EXTCEO. A common way to deal with the endogeneity problem is by

16 Note that these are likely the upper bounds in potential value loss due to external CEOs. This is because it is impossible to control for all possible determinants of firm performance in a regression and that the negative effects of some uncontrolled (but observed) determinants on firm performance may show up via the external CEO dummy variable. 17 All our results in Table 5 remain valid without the industry and year fixed effects; that is, when the identification is provided mainly by the changes in EXTCEOijt between industries over time. Besides, the regression results are similar when we use “Earnings before Interest and Taxes” (EBIT) as a measure of income or use “Total Sales” to normalize the income measure. 18 There is another potential limitation: a sample selection problem (Heckman, 1979) may arise because we focus only on the turnover sample here, which may not be a random sample, so that firms with CEO turnover may be systematically different from firms without CEO turnover. We control for this problem by the Heckman two-stage method and still find a similar pattern. The results are not reported here.

38

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

using the Instrumental Variable (IV) Approach. Under this approach, we need to find an instrument that is correlated with the CEO hiring decision of the firm, but is unrelated to the outcome variable, which is the post-CEO turnover firm performance. One might think of factors affecting the CEO hiring decision, e.g., firms with poor performance relative to the industry (Parrino, 1997), but such factors would likely affect firm performance. An alternative approach to deal with the endogeneity problem is the Control Function Approach, which we will use here. The essence of this approach is to construct a control function, CF, so that conditional on CF the disturbance terms in Eq. (1) are uncorrelated with the external CEO- versus internal CEO-hiring decision of the firm.19 The feasibility of the Control Function Approach in any economic setting depends on whether we are able to recover the control function CF so that it can be conditioned upon when the parameters are estimated. To this end, we follow Heckman and Robb (1985) and implement the Control Function Approach as follows. We impose the ‘Conditional Independence Assumption’ (CIA): we assume that the CEO hiring decision is random when we condition on the observable factors (i.e., selection on observables). We first estimate a Probit model by regressing the external CEO hire dummy on the control variables X and obtain the propensity scores. Then, in the second stage, we use a polynomial in the estimated propensity scores to flexibly model our control function and estimate the following regression model:  CROAijt+1,t+2 = ˛ + ˇ · EXTCEOijt + Xijt ı + j + t +

2 

k pˆ kijt +  εijt ,

(3)

k=1

where pˆ kijt is a polynomial in the estimated propensity scores from the Probit model in the first stage. Under CIA, when we condition on the control function, the resulting disturbance term ε˜ ijt is uncorrelated with EXT ijt .20 In the actual regressions, we use a quadratic polynomial of the propensity score as the control function.21 Columns (5)–(8) of Table 5 report the control function regression estimates from the regression model (1). The results show that the external CEO-hire dummy still has a statistically significant effect on the excess economic profits of firms. 5.2. ICT and external CEO hiring In this section, we test our second hypothesis that the propensity of hiring an external CEO increases as general skills become more important. Table 6 reports summary statistics describing the unconditional relationship between ICT, external CEO hiring, and CEO compensation. For a given year, we identify an industry as high-ICT if the ICT of the industry is the highest among all industries in that year. Similarly, we identify an industry as low-ICT if its ICT is the lowest among all industries in that year. We then report the differences in external CEO hiring and CEO wage between high-ICT and lowICT industries. Table 6 shows that firms in high-ICT industries are 10.87% more likely to hire CEOs externally and 5.43% more likely to hire CEOs from outside their industries. Furthermore, high-ICT industries pay higher compensation to external CEOs compared to low-ICT industries. Next, we estimate the following Probit model to provide firm-level evidence of the effect of ICT on external CEO hiring:  ı + j + t ). P(EXTCEOijt = 1) = (˛ + ˇ · ICT jt + Xijt

(4)

The dependent variable, EXTCEOijt , is a dummy variable indicating whether the CEO hired in period t by firm i is an external CEO. The other variables are the same as defined earlier. We are interested

19 Note that the Control Function and IV Approaches require different sets of identifying assumptions. See Heckman and Navarro-Lozano (2004) for a discussion. 20 We check the validity of CIA by performing a “balancing test.” In particular, we compare the means of the control variables for the internal- and the external-CEO firms when the observations are re-weighted by the propensity scores obtained in the first-stage regressions. The results, not reported here, show that re-weighted samples of internal- and external-CEO firms are not significantly different. 21 Our results are virtually unchanged if we also use a polynomial of a higher degree than two.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

39

Table 6 ICT, external CEO hiring, and CEO wage: descriptive statistics. Panel A: ICT and external CEO HIRING Internal CEO

External CEO

Same SIC EXTCEO

Diff. SIC EXTCEO

High-ICT industries Low-ICT industries

59 15

33 5

28 5

5 0

92 20

Total turnover

74

38

33

5

112

Difference (High-Low)

44

28

23

5

72

High-ICT industries (%) Low-ICT industries (%)

64.13% 75.00%

35.87% 25.00%

30.43% 25.00%

5.43% 0.00%

100.00% 100.00%

Total turnover (%)

66.07%

33.93%

29.46%

4.46%

100.00%

% Difference (High-Low)

10.87%

Total Turnover

ICT (Mean)

CWAGE (Mean)

50.10 0.24

8.96 8.27

49.86***

0.69**

5.43%

Panel B: ICT, CEO type, and compensation External-CEO Wage (Mean)

Internal-CEO Wage (Mean)

Difference

High-ICT industries Low-ICT industries

9.14 7.20

8.91 8.88

0.23 −1.68**

Difference

1.93***

0.02

This table reports summary statistics on external CEO hiring and CEO wage for high-ICT and low-ICT industries. For a given year, we identify an industry as high-ICT if the ICT of the industry is the highest among all industries in that year. Similarly, we identify an industry as low-ICT if its ICT is the lowest among all industries in that year. The definitions of all variables are the same as in Table A.1. Standard errors are clustered at the industry level. * Significance at the 10% level. ** ***

Significance at the 5% level. Significance at the 1% level.

primarily in the ˇ coefficient in regression model (4) above; we expect ˇ > 0, that is, the likelihood of hiring an external CEO increases with the importance of general skills. Table 7 reports the results from the above Probit model. In columns (1) and (2), we report the results of the pooled cross-section regressions using the current-year ICT measure and the one-year lagged ICT measure, controlling for firm and CEO characteristics. Moreover, since we also control for year and industry fixed effects in these regressions, the identification is provided mainly by the changes in ICT within an industry over time. The results show that the two ICT measures are positively associated with the likelihood of external CEO hiring, and the results are statistically significant at the 1% level for the current year ICT and at the 10% level for the one-year lagged ICT measure. These results are also economically significant: in column (1) a 10% increase in ICT capital stock per worker from its mean (about $1600–1700) is associated with a 7.67% increase in the likelihood of external CEO hiring (when all other explanatory variables are evaluated at their mean). Regression results related to firmand CEO-characteristics variables are also plausible. Among the firm characteristics, the results show that poor corporate governance of the firm increases the propensity for external CEO hiring; larger firms and firms with more tangible assets and greater growth opportunities in place are less likely to hire an external CEO. Firms with higher managerial ownership are also less likely to hire a CEO externally. These results are consistent with the findings in Huson, Parrino, and Starks (2001) and Hermalin (2005). An important issue related to identification of the effect of ICT on the external hiring decision is that other industry-specific factors simultaneously varying with ICT could make it difficult to disentangle the effect of ICT-induced general skills from those of other unobservables not controlled for in the above Probit regression. To address this issue, we first use an industry homogeneity measure to control for an industry-specific human capital effect on external CEO hiring. Parrino (1997) shows that industry

40

Table 7 ICT and external CEO hiring. Two-way fixed effects model [1] ICT jt

[2]

ICT jt−1

[3]

0.376* [1.95] 1.703** [2.18] −1.839** [−2.00]

INDHOMOjt ICT jt × INDHOMOjt ICT jt−1 × INDHOMOjt−1 Firm&CEOcharacteristics FIRMSIZE −0.099*** [−3.40] −1.420*** TANGIBILITY [−2.96] TOBIN  SQ −0.108*** [−2.71] LEVERAGE 0.490* [1.71] CAPX/SALES 0.163 [1.38] R&D/CAPX 0.372*** [6.12] 0.090 ADVERT /CAPX [0.67] −0.188 VOLATILITY [−0.71] DIVYIELD 1.446 [1.21] −0.153 PASTRETURN [−1.32] 0.031** GINDEX

[4]

1.331*** [4.19]

[5]

[6]

5.185*** [4.23] 0.890*** [2.94] 1.317* [1.69]

[7]

−0.107*** [−3.71] −1.489*** [−3.48] −0.115** [−2.49] 0.462 [1.49] 0.160 [1.46] 0.375*** [6.50] 0.074 [0.59] −0.223 [−0.84] 1.386 [1.10] −0.146 [−1.31] 0.029**

−0.109*** [−3.95] −1.495*** [−3.55] −0.112** [−2.53] 0.466 [1.46] 0.164 [1.46] 0.379*** [6.43] 0.082 [0.63] −0.218 [−0.81] 1.288 [1.04] −0.133 [−1.13] 0.030**

[8]

6.129*** [5.55] 3.652*** [4.23] 2.033*** [3.48] −2.772*** [−6.06]

−1.501 [−1.39] −0.102*** [−3.65] −1.423*** [−3.02] −0.107*** [−2.72] 0.490* [1.68] 0.162 [1.33] 0.374*** [6.02] 0.095 [0.69] −0.186 [−0.69] 1.361 [1.15] −0.143 [−1.17] 0.031**

Four-way fixed effects model [9]

[10]

0.453** [2.17] 0.574*** [3.46] 1.746*** [3.26]

[11]

0.482** [2.17] 1.090 [1.06] −3.411*** [−3.23]

−2.761*** [−6.19] −0.131*** [−6.74] −1.578*** [−3.33] −0.086*** [−3.18] 0.650 [1.58] 0.257 [1.21] 0.372*** [5.25] 0.054 [0.39] −0.312 [−1.26] 0.352 [0.16] −0.157 [−1.05] 0.022

−0.131*** [−6.74] −1.578*** [−3.33] −0.086*** [−3.18] 0.650 [1.58] 0.257 [1.21] 0.372*** [5.25] 0.054 [0.39] −0.312 [−1.26] 0.352 [0.16] −0.157 [−1.05] 0.022

−0.141*** [−6.84] −1.659*** [−3.96] −0.089*** [−2.93] 0.607 [1.46] 0.251 [1.26] 0.369*** [5.69] 0.048 [0.36] −0.342 [−1.42] 0.472 [0.21] −0.155 [−1.10] 0.022

−0.141*** [−6.83] −1.657*** [−3.97] −0.089*** [−2.94] 0.608 [1.47] 0.249 [1.26] 0.371*** [5.68] 0.048 [0.36] −0.346 [−1.43] 0.466 [0.21] −0.152 [−1.08] 0.023

[12]

1.562*** [3.42] 1.576*** [3.30] 0.694 [0.75]

−3.366*** [−3.05] −0.162*** [−3.67] −0.515 [−0.46] −0.212*** [−3.92] 0.643* [1.78] −0.393* [−1.73] 0.877*** [2.63] −0.407 [−1.22] −0.314 [−1.03] −11.189** [−2.28] −0.016 [−0.12] 0.005

−0.162*** [−3.67] −0.515 [−0.46] −0.212*** [−3.92] 0.643* [1.78] −0.393* [−1.73] 0.877*** [2.63] −0.407 [−1.22] −0.314 [−1.03] −11.189** [−2.28] −0.016 [−0.12] 0.005

−0.165*** [−3.94] −0.597 [−0.55] −0.200*** [−3.33] 0.530 [1.43] −0.369 [−1.64] 0.817** [2.30] −0.460 [−1.33] −0.320 [−1.03] −10.936** [−2.24] −0.018 [−0.14] 0.007

−0.165*** [−3.92] −0.586 [−0.54] −0.198*** [−3.32] 0.530 [1.45] −0.376* [−1.69] 0.815** [2.27] −0.470 [−1.36] −0.321 [−1.03] −10.972** [−2.25] −0.012 [−0.10] 0.006

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

0.631*** [3.04]

Three-way fixed effects model

[2.23] −0.037** [−2.27] −0.003 [−1.19] 0.060 [0.56] −0.184 [−0.27]

[2.09] −0.040*** [−2.60] −0.003 [−1.14] 0.058 [0.56] −1.276 [−1.43]

[2.19] −0.040** [−2.55] −0.003 [−1.07] 0.063 [0.58] −0.490 [−0.64]

[1.04] −0.053*** [−3.59] −0.002 [−0.33] 0.076 [0.63] −8.638*** [−3.96]

[1.04] −0.053*** [−3.59] −0.002 [−0.33] 0.076 [0.63] −5.603*** [−3.80]

[1.05] −0.054*** [−3.44] −0.001 [−0.30] 0.078 [0.65] −9.159*** [−4.62]

[1.05] −0.055*** [−3.44] −0.001 [−0.31] 0.077 [0.65] 1.055*** [3.12]

[0.30] −0.001 [−0.03] −0.006 [−0.38] −0.050 [−0.41] −0.495 [−0.41]

[0.30] −0.001 [−0.03] −0.006 [−0.38] −0.050 [−0.41] −0.458 [−0.39]

[0.42] −0.001 [−0.03] −0.006 [−0.33] −0.069 [−0.51] −0.772 [−0.73]

[0.41] −0.001 [−0.05] −0.006 [−0.34] −0.070 [−0.52] −0.646 [−0.62]

Controlfor Industry fixed effects Year fixed effects Industry -year fixed effects Turnover-reason fixed effects

Yes Yes No No

Yes Yes No No

Yes Yes No No

Yes Yes No No

Yes Yes Yes No

Yes Yes Yes No

Yes Yes Yes No

Yes Yes Yes No

Yes Yes Yes Yes

Yes Yes Yes Yes

Yes Yes Yes Yes

Yes Yes Yes Yes

Psedo-R2 Log likelihood Num. of obs.

0.06 −470.82 837

0.06 −471.45 837

0.07 −469.16 837

0.06 −470.23 837

0.10 −409.11 729

0.10 −409.11 729

0.10 −407.56 729

0.10 −407.53 729

0.22 −164.34 329

0.22 −164.34 329

0.22 −163.40 329

0.22 −163.38 329

CEOAGE CEOMBA Constant

This table reports results from Probit regressions of external CEO hiring on industry-level ICT. The dependent variable is the external CEO hire dummy, EXTCEO, which is equal to 1 if the new CEO is an external hire. The definitions of all variables are the same as in Table A.1. Standard errors are clustered at the industry level. * Significance at the 10% level. ** Significance at the 5% level. ***

Significance at the 1% level.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

[2.17] −0.037** [−2.26] −0.003 [−1.25] 0.056 [0.54] −0.753 [−0.97]

TOP5MGT

41

42

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

homogeneity is an important determinant of whether the new CEO is hired externally. If an external CEO is hired from an homogenous industry and ICT varies simultaneously with industry homogeneity, then ICT may capture the effect of CEO skills that is more likely to be bounded by CEOs’ industry experiences, i.e., industry-specific human capital.22 Following Parrino (1997), we construct an industry homogeneity measure at the two-digit SIC level and control for interaction of this measure with the ICT variable to separate the ICT-induced general skill effect from the industry-specific human capital effect. We find that the industry homogeneity measure is positively correlated with our industry-level ICT measures, and that the correlation statistics are significant at the 1% level.23 When we estimate the regression model (4) using industry homogeneity and its interaction with our ICT measures as additional controls, we find, consistent with the results of Parrino (1997), in columns (3) and (4) of Table 7 that industry homogeneity is positively and significantly related to the likelihood of external CEO hiring.24 Furthermore, our ICT measures still have positive and statistically significant effects (independent of industry-specific human capital effect) on the likelihood of external CEO hiring. Beside industry-specific human capital, there might be other unobservables that may simultaneously vary with ICT and thus confound the identification of the effect of ICT on the external CEO hiring. To alleviate this concern, we estimate a three-way fixed-effects Probit model. In addition to industry- and year-fixed effects, we also introduce industry-year fixed effects in regression model (4). The three-way fixed-effects Probit model controls for the unobserved heterogeneity that varies across industries and over time and simultaneously across industry and time. Columns (5)–(8) of Table 7 report results from the three-way fixed-effects Probit model and show that ICT variables are still significant at the 1% level and have a positive effect on external CEO-hiring decision. For additional robustness, we also estimate the regression model (4) after controlling for the reasons for the departure of the incumbent CEO as these factors are shown to be related to whether or not the new CEO is hired externally (e.g., Huson et al., 2004; Parrino, 1997). We report these results in a four-way fixed-effects Probit model in columns (9)–(12). The results are similar to what we reported earlier.25 Finally, we investigate the effect of ICT on the likelihood of across-industry external CEO hiring. We gathered information on whether the external CEO was hired from outside the four-digit SIC industry of the firm from a CEO’s tenure path documented in the ExecuComp and newspaper articles around the time of CEO turnover. We find that only 3.5% (34 out of 983) of the sample CEOs are hired from outside the industry and the rest are hired from within the same industry. We then estimate the effect of ICT on the likelihood of a firm hiring an external CEO from a different four-digit SIC industry than its own. Table 8 reports our results. In columns (1)–(4), we estimate a Probit model where the dependent variable is a dummy variable (DIFFIND EXTCEO) indicating whether the new external CEO is hired from outside the four-digit SIC of the firm. We find that our ICT measure is positively and significantly related to the likelihood of across-industry external CEO hiring. Furthermore, the marginal effect of ICT on DIFFIND EXTCEO is higher than its marginal effect on EXTCEO. Next, we perform an industry-level analysis using a Poisson regression model, where the dependent variable is the number of CEOs hired from a different industry in a given year. Columns (5)–(8) report results from a population-averaged Poisson model, where the ICT coefficient captures the between-industry effect of ICT on the likelihood of across-industry external CEO hiring. Columns (9)–(12) report results from a fixed-effect Poisson model, where the ICT coefficient captures the within-industry effect of ICT on the likelihood of acrossindustry external CEO hiring. The industry-level analysis highlights two points: First, industries using more ICT in any given year are more likely to hire external CEOs from other industries than those using less ICT, i.e., the between-industry effect of ICT. Second, as an industry becomes more reliant on ICT over time, there is an increase in the likelihood of external CEO hiring from other industries, i.e., the

22

We thank the anonymous referee for providing this insight. To be precise, the correlation between the Parrino (1997) industry-homogeneity measure and our ICT measures is 11%. 24 Note that the Parrino (1997) industry-homogeneity measure is at the two-digit SIC level whereas the industry-fixed effects are at the two-digit BEA-industry level. There is no direct concordance between these two classifications. This allows us to separately estimate the INDHONOj coefficient in the presence of industry fixed effects. 25 Note that the data on turnover reasons are limited because for some turnover cases, we simply did not have any information about the reason for the turnover. For this reason, the data points drop significantly in columns (9)–(12). 23

Table 8 ICT and across-industry external CEO hiring. Firm-level PROBIT model

Industry-level POISSON model Population-averaged POISSON

[1]

[2]

2.411*** [3.30]

ICT jt−1

[3]

0.697*** [5.16] −0.083 [−0.13] −1.834 [−1.19]

INDHOMOjt ICT jt × INDHOMOjt ICT jt−1 × INDHOMOjt−1 Firm&CEOcharacteristics : 0.057** FIRMSIZE [2.16] 0.036 TANGIBILITY [0.04] 0.081 TOBINQ [1.36] LEVERAGE 1.332*** [3.39] CAPX/SALES −0.176 [−0.82] 0.090 R&D/CAPX [0.50] 0.147 ADVERT /CAPX [0.40] −0.465* VOLATILITY [−1.94] 1.180 DIVYIELD [0.77] −0.726*** PASTRETURN [−2.80]

[4]

2.839*** [3.27]

[5]

[6]

0.122*** [5.86] 0.943*** [5.17] −0.037 [−0.06]

[7]

0.117*** [5.98] −1.353 [−1.42] −0.011 [−1.64]

−2.488* [−1.71] 0.086** [2.53] 0.036 [0.04] 0.074 [1.51] 1.451*** [3.49] −0.194 [−0.79] 0.108 [0.67] 0.291 [0.82] −0.471** [−2.03] 1.156 [0.76] −0.686*** [−2.84]

0.053* [1.78] 0.141 [0.14] 0.090 [1.39] 1.290*** [3.29] −0.182 [−0.76] 0.092 [0.53] 0.155 [0.42] −0.470* [−1.91] 1.273 [0.82] −0.722*** [−2.73]

0.079** [2.48] 0.130 [0.12] 0.084 [1.60] 1.390*** [3.47] −0.201 [−0.74] 0.112 [0.72] 0.295 [0.84] −0.479** [−2.00] 1.213 [0.79] −0.683*** [−2.75]

[8]

0.099*** [3.14]

[9]

[10]

0.101*** [2.93] 0.117*** [4.05] −0.729 [−0.84]

[11]

0.104*** [3.24] −2.196* [−1.79] −0.012 [−1.49]

−0.011* [−1.90] 0.033 [0.87] 1.157*** [2.71] 0.009 [0.20] −0.556 [−1.44] −0.179 [−1.22] −0.266* [−1.67] −0.122 [−0.48] 0.248 [1.09] −1.137 [−0.39] −0.104 [−1.05]

0.046 [1.23] 1.188*** [2.81] 0.017 [0.37] −0.654* [−1.69] −0.186 [−1.26] −0.252 [−1.52] −0.139 [−0.55] 0.144 [0.63] −1.388 [−0.48] −0.081 [−0.79]

0.068 [1.15] 0.563 [1.08] 0.052 [0.94] −0.012 [−0.02] 0.088 [0.42] −0.449** [−2.21] 0.209 [0.58] 0.312 [1.02] −0.801 [−0.30] −0.084 [−0.80]

0.039 [0.76] 0.697 [1.40] 0.049 [0.95] 0.017 [0.03] 0.026 [0.13] −0.442** [−2.28] 0.202 [0.60] 0.405 [1.54] −1.906 [−0.72] −0.067 [−0.65]

[12]

0.068* [1.77] 0.079** [2.04] −1.875 [−1.50]

−0.010 [−1.42] 0.022 [0.52] 0.580 [1.12] 0.038 [0.77] −0.545 [−1.34] −0.055 [−0.30] −0.246 [−1.26] −0.387 [−1.19] 0.120 [0.47] 0.502 [0.16] −0.077 [−0.71]

0.031 [0.72] 0.657 [1.26] 0.041 [0.84] −0.660 [−1.62] −0.072 [−0.39] −0.232 [−1.17] −0.398 [−1.27] 0.033 [0.13] 0.076 [0.02] −0.052 [−0.49]

0.116 [1.56] 0.187 [0.33] 0.060 [1.05] −0.053 [−0.09] 0.227 [1.01] −0.504** [−2.17] 0.241 [0.58] 0.128 [0.36] 0.158 [0.06] −0.058 [−0.56]

0.101 [1.37] 0.252 [0.44] 0.057 [1.05] −0.145 [−0.26] 0.193 [0.85] −0.477** [−2.04] 0.183 [0.43] 0.103 [0.29] −0.321 [−0.11] −0.056 [−0.55]

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

ICT jt

Fixed-effect POISSON

43

44

Table 8 (Continued) Firm-level PROBIT model

Industry-level POISSON model Population-averaged POISSON

GINDEX TOP5MGT CEOAGE CEOMBA Constant Controlfor : Industry fixed effects Year fixed effects Industry-year fixed effects Pseudo-R2 Log likelihood 2 Num. of obs.

[2] **

[3] **

[4] **

[5] **

[6] *

Fixed-effect POISSON

[7]

*

[11] *

[12]

0.042 [2.05] 0.003 [0.41] 0.008 [0.57] 0.065 [1.17] −11.384*** [−3.92]

0.044 [2.18] 0.003 [0.40] 0.004 [0.30] 0.061 [1.19] −3.249*** [−4.00]

0.039 [1.80] −0.015 [−0.80] −0.012** [−1.99] −0.242 [−1.64] −4.173*** [−10.76]

0.040 [1.82] −0.020 [−0.97] −0.013** [−2.07] −0.284* [−1.93] −4.216*** [−10.79]

0.041 [1.65] −0.008 [−0.34] −0.015* [−1.89] 0.015 [0.08] −4.234*** [−8.29]

0.040 [1.74] −0.015 [−0.62] −0.013* [−1.88] −0.018 [−0.10] −4.388*** [−9.08]

0.040 [1.77] −0.029 [−1.27] −0.009 [−1.27] −0.195 [−1.20] −4.071*** [−6.55]

0.042 [1.89] −0.034 [−1.43] −0.009 [−1.31] −0.226 [−1.41] −4.150*** [−6.45]

0.051 [1.82] −0.002 [−0.09] −0.017* [−1.86] 0.009 [0.04] −4.225*** [−5.82]

0.054** [2.00] −0.007 [−0.23] −0.016* [−1.75] −0.027 [−0.14] −4.247*** [−5.79]

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

No No No

No No No

No No No

No No No

Yes No No

Yes No No

Yes No No

Yes No No

0.13 −93.08

0.13 −96.29

0.13 −92.85

0.13 −95.94

729

729

729

729

148

−44.31 87.31 148

−43.51 85.07 148

−41.97 83.06 148

−41.68 82.20 148

148

*

[10]

0.042 [2.10] 0.004 [0.58] 0.003 [0.22] 0.075 [1.50] −4.672*** [−7.81]

148

*

[9]

0.040 [1.97] 0.004 [0.54] 0.007 [0.55] 0.079 [1.51] −11.520*** [−3.43]

148

*

[8]

*

This table reports regression results of across-industry external CEO hiring on industry-level ICT. The dependent variable in columns (1)–(4) is the across-industry external-CEO hire dummy, DIFFIND EXTCEO, which is equal to 1 if the new CEO is hired from outside the four-digit SIC industry of the firm. The dependent variable in columns (5)–(12) is a non-negative count variable; it indicates the total number of external CEOs hired from outside the industry in a given year. The definitions of all variables are the same as in Table A.1. Standard errors are clustered at the industry level. * ** ***

Significance at the 10% level. Significance at the 5% level. Significance at the 1% level.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

[1]

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

45

within-industry effect of ICT. To the extent that ICT proxies for the importance of general skills, our results provide evidence in support of the second hypothesis of the market-driven theory.26 5.3. ICT, external CEO hiring, and firm performance The third empirical hypothesis we test is the following: the excess profits of an internal-CEO firm over an external-CEO firm decrease as general skills become more important. We estimate an OLS regression model similar to (4), but we also include the interaction between the ICT measure and the external CEO hire dummy:  ı + j + t + εijt , CROAijt+1,t+2 = ˛ + ˇ1 EXTCEOijt + ˇ2 ICT jt + ˇ3 (ICT jt × EXTCEOijt ) + Xijt

(5)

where the definitions of variables are the same as previously. If the excess profits of an internal-CEO firm over an external-CEO firm decrease as general skills become more important, we should expect the coefficient of the interaction term (ICT jt × EXTCEOijt ) to be positive, i.e., ˇ3 > 0. Columns (1)–(4) of panel A in Table 9 report the OLS regression results for regression model (5) using CROAt+1,t+2 as the dependent variable, and columns (5)–(8) report the OLS regression results using CAROAt+1,t+2 as the dependent variable. It shows that the coefficients of the interaction term (ICT jt × EXTCEOijt ) are positive and statistically significant at the 5% level. These coefficients capture the effects of ICT filtered through the external CEO dummy variable. We argued previously that an external CEO can contribute to a higher performance relative to an internal CEO when general skills are more important. The positive and statistically significant coefficients of the interaction terms in panel A of Table 9 provide evidence that with an increase in the importance of general skills, the superior performance of firms with internal CEOs relative to those with external CEOs decreases. Among the other control variables in the regression, the future growth opportunities of firms, proxied by Tobin’s Q, and dividend yield are significantly correlated with post-turnover firm performance at the 1% level and have positive coefficients. Furthermore, return volatility and past equity returns have negative effects on post-turnover performance and are also significant at the 1% level.27 As for economic significance: a 10% increase in ICT capital stock per worker from the mean is associated with a 2.15% increase in the post-turnover firm performance in column (1) when the firm hires an external CEO instead of an internal CEO (conditional on all other explanatory variables being evaluated at their mean). To address the problem of endogeneity of the external CEO hiring decision, we also use the Control Function Approach. The identification strategies are the same as described in Section 5.1. Panel B of Table 9 shows the corresponding regression results. Similar to the OLS results, we also find positive and significant coefficients on the interaction terms. Note that in estimating regression model (5) using OLS and control function, we also control for industry homogeneity suggested by Parrino (1997). Thus, the effect of ICT on post-turnover firm performance via the external CEO hiring is independent of the variations in external CEO hiring in homogeneous industries. Together with additional firm- and CEO-level controls and industry and year fixed effects, results suggest that ICT significantly affects post-turnover firm performance when an external, as opposed to an internal, CEO is hired. In short, the regression results presented above show that an internal-CEO firm makes greater profits than an external-CEO firm. However, when interacted with the prevalence of information and communication technology, a proxy for the importance of general skills, the external CEO does have a positive effect on firm performance relative to an internal CEO. To the extent that ICT can be put to the best possible use with general skills, the interaction effects provide evidence that the excess economic profit of an internal-CEO firm over an external-CEO firm decreases with the importance of general skills.

26 27

The results are similar when we exclude the high-tech industry in the regression. We do not report these results here, but they are available upon request.

46

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Table 9 ICT, external CEO hiring, and firm performance. CAROAt+1,t+2

CROAt+1,t+2 [1] PanelA : OLS EXTCEOijt × ICT jt

0.016** [2.74]

EXTCEOijt × ICT jt−1 EXTCEOijt ICT jt

[2]

−0.038*** [−3.65] −0.057 [−1.74]

[3] 0.015** [2.73]

0.016*** [3.10] −0.033*** [−3.01]

−0.031** [−2.58] −0.292*** [−3.77]

−0.103 [−1.74]

ICT jt−1

[4]

INDHOMOj

0.075 [0.48] 0.378** [2.23]

ICT jt × INDHOMOj

[5]

[6]

0.018*** [3.14] 0.014** [2.29] −0.031** [−2.35]

−0.041*** [−3.59] −0.046* [−1.84]

−0.224** [−2.74] 0.135 [0.99]

[7]

[8]

0.018*** [3.19] 0.019*** [3.66] −0.036*** [−3.17]

−0.035** [−2.78] −0.273*** [−3.66]

−0.073 [−1.72] 0.078 [0.52] 0.373** [2.39]

0.017** [2.76] −0.034** [−2.54]

−0.193** [−2.78] 0.135 [1.05]

0.002 [0.01]

0.050 [0.27]

0.169 [1.15]

0.334** [2.30] 0.033 [0.26]

0.078 [0.46]

0.113 [0.65]

0.295** [2.41]

0.335* [2.11] 0.179* [1.83]

Control for Firm and CEO characteristics Industry fixed effects Year fixed effects

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

R2 Num. of Obs.

0.24 653

0.28 653

0.29 653

0.29 653

0.25 653

0.27 653

0.28 653

0.28 653

ICT jt−1 × INDHOMOj Constant

PanelB : OLSwithControlFunction 0.018*** EXTCEOijt × ICT jt [3.68] EXTCEOijt × ICT jt−1 EXTCEOijt ICT jt

−0.034*** [−3.17] −0.161*** [−4.00]

0.016** [2.91] 0.017*** [3.24] −0.033** [−2.86]

−0.031** [−2.50] −0.294*** [−3.80]

−0.106* [−1.82]

ICT jt−1 INDHOMOj

0.078 [0.48] 0.371** [2.24]

ICT jt × INDHOMOj

0.020*** [4.25] 0.015** [2.44] −0.031** [−2.28]

−0.038*** [−3.33] −0.143*** [−3.17]

−0.224** [−2.79] 0.138 [0.97]

0.018*** [3.41] 0.019*** [3.89] −0.036*** [−3.03]

−0.035** [−2.70] −0.275*** [−3.67]

−0.075* [−1.78] 0.080 [0.52] 0.367** [2.40]

0.018** [2.96] −0.034** [−2.47]

−0.194** [−2.82] 0.137 [1.02]

0.870** [2.21]

0.790* [1.99]

0.958** [2.39]

0.328* [2.12] 0.800* [2.05]

Control for Firm and CEO characteristics Industry fixed effects Year fixed effects

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

R2 Num. of obs.

0.29 653

0.28 653

0.30 653

0.29 653

0.28 653

0.27 653

0.29 653

0.28 653

ICT jt−1 × INDHOMOj Constant

0.777* [2.11]

0.698* [1.88]

0.801** [2.15]

0.328** [2.32] 0.626 [1.72]

This table reports results from regression analysis of the effect of CEO general skills on a firm’s performance. The dependent variables are the performance of a firm measured by cumulative return on assets, CROAt+1,t+2 and S&P adjusted cumulative return on assets, CAROAt+1,t+2 two years following CEO turnover. The definitions of all variables are the same as in Table A.1. Standard errors are clustered at the industry level. * ** ***

Significance at the 10% level. Significance at the 5% level. Significance at the 1% level.

Table 10 ICT, external CEO hiring, firm performance, and CEO compensation. OLS regression results

IV regression results

CWAGE t+1,t+2 [1] EXTCEOijt × ICT jt

[2]

0.122**

ICT jt−1

−0.123 (−0.56) −0.703**

INDHOMOj

(−2.27) 0.188

ICT jt × INDHOMOj

(2.23) 0.023 (0.21)

−0.118 (−1.09) −0.911** (−2.79)

−0.140 (−0.56) −0.630** (−2.50)

Firm&CEOcharacteristics 1.328*** CROAt+1,t+2 (6.91) 0.363*** FIRMSIZE (23.45) −0.612*** TANGIBILITY (−3.77) 0.140*** TOBIN  S : Q (4.39) LEVERAGE 0.029 (0.08) −0.076 CAPX/SALES (−1.02) R&D/CAPX −0.217 (−1.26) ADVERT /CAPX −0.020 (−0.25)

1.347*** (6.81) 0.366*** (22.57) −0.616*** (−3.74) 0.138*** (4.31) 0.010 (0.03) −0.073 (−0.92) −0.228 (−1.28) −0.010 (−0.11)

1.346*** (6.58) 0.365*** (21.26) −0.550*** (−3.12) 0.140*** (4.27) 0.000 (0.00) −0.082 (−1.10) −0.222 (−1.30) −0.017 (−0.22)

[8]

(2.51) −0.101 (−1.02)

[9]

−0.101 (−0.39) −0.012 (−0.02) −0.821

(2.48) −0.111 (−1.08)

CAWAGE t+1,t+2 [11]

(2.46) 0.023 (0.24)

0.020 (0.19) −0.340 (−0.96)

−0.737** (−2.53) 0.043

(−0.71)

[14]

(2.46) 0.029 (0.30)

−0.050 (−0.22) −0.680*** (−3.03)

[15]

[16]

0.178***

(3.34) 0.115**

−0.083 (−0.40)

0.261 (0.56) −0.017 (−0.02)

[13] 0.169***

(2.48) 0.117**

0.013 (0.12) −0.330 (−1.23)

[12]

0.117**

(2.49) 0.179**

−0.126 (−1.12) −0.600 (−1.13)

[10]

0.120**

(2.40) 0.172**

(0.28) ICT jt−1 × INDHOMOj

CWAGE t+1,t+2 [7] 0.184**

(2.45) 0.117**

0.012 (0.10) −0.443 (−1.34)

[6]

(3.03) 0.164***

(3.45) −0.032 −0.022 (−0.29) (−0.21) −0.612** (−2.04) 0.122 (0.34)

0.174***

−0.044 (−0.36) 0.044 (0.07)

−0.185 (−0.19) −1.692

(0.07)

(3.19) −0.032 (−0.28)

0.788 (1.31) −0.311 (−0.40)

(−1.31)

0.069

−0.977

−0.048

−1.737

(0.11)

(−0.88)

(−0.08)

(−1.51)

1.369*** (6.53) 0.367*** (21.21) −0.556*** (−3.14) 0.138*** (4.20) −0.023 (−0.07) −0.079 (−1.01) −0.234 (−1.33) −0.007 (−0.08)

0.798** (2.73) 0.334*** (24.92) −0.384 (−1.51) 0.220*** (4.04) 0.336 (0.97) −0.229* (−1.99) −0.380*** (−5.65) −0.420* (−1.89)

0.855** (2.79) 0.343*** (24.71) −0.399 (−1.51) 0.215*** (3.87) 0.282 (0.79) −0.219* (−1.79) −0.412*** (−5.33) −0.400 (−1.64)

0.846** (2.71) 0.332*** (20.50) −0.355 (−1.39) 0.215*** (3.64) 0.280 (0.87) −0.237* (−1.99) −0.388*** (−6.14) −0.424* (−1.88)

0.908** (2.84) 0.339*** (22.19) −0.377 (−1.45) 0.211*** (3.49) 0.218 (0.66) −0.227* (−1.79) −0.418*** (−5.75) −0.409 (−1.71)

1.817*** (2.74) 0.363*** (27.19) −0.620*** (−4.15) 0.110** (2.13) 0.043 (0.13) −0.065 (−1.04) −0.169 (−0.94) −0.045 (−0.51)

1.756** (2.44) 0.366*** (25.61) −0.624*** (−4.10) 0.113** (2.05) 0.023 (0.07) −0.064 (−0.96) −0.186 (−0.99) −0.033 (−0.34)

1.940*** (2.82) 0.365*** (23.99) −0.552*** (−3.40) 0.103* (1.88) 0.004 (0.01) −0.071 (−1.14) −0.167 (−0.96) −0.046 (−0.53)

1.885** (2.49) 0.367*** (23.38) −0.558*** (−3.41) 0.106* (1.80) −0.018 (−0.06) −0.070 (−1.05) −0.185 (−1.02) −0.036 (−0.38)

4.747*** (3.43) 0.346*** (14.24) −0.431** (−1.96) −0.025 (−0.22) 0.426 (1.31) −0.140 (−1.32) −0.029 (−0.20) −0.579** (−2.22)

4.578*** (3.32) 0.352*** (13.17) −0.442* (−1.94) −0.016 (−0.14) 0.377 (1.12) −0.138 (−1.29) −0.072 (−0.51) −0.560** (−2.07)

5.061*** (3.32) 0.342*** (12.49) −0.359 (−1.56) −0.047 (−0.38) 0.305 (1.09) −0.155 (−1.36) −0.029 (−0.21) −0.594** (−2.19)

4.977*** (3.12) 0.346*** (11.87) −0.379 (−1.61) −0.043 (−0.32) 0.254 (0.87) −0.149 (−1.31) −0.060 (−0.42) −0.590** (−2.14)

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

ICT jt

(2.23) 0.018 (0.17)

[5] 0.179**

(2.25) 0.119**

0.005 (0.04) −0.372 (−1.36)

[4]

0.119**

(2.25) EXTCEOijt × ICT jt−1 EXTCEOijt

CAWAGE t+1,t+2 [3]

47

48

Table 10 (Continued) OLS regression results

IV regression results

CWAGE t+1,t+2

DIVYIELD PASTRETURN GINDEX TOP5MGT CEOAGE CEOMBA constant Controlfor Industry fixed effects Year fixed effects R2 Num. of obs.

CWAGE t+1,t+2

CAWAGE t+1,t+2

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

−0.096 (−0.60) 2.540 (1.27) 0.199*** (4.22) 0.018 (1.56) −0.016 (−1.46) −0.000 (−0.00) 0.021 (0.46) 5.417*** (16.99)

−0.095 (−0.59) 2.505 (1.21) 0.195*** (4.22) 0.018 (1.55) −0.017 (−1.52) −0.000 (−0.03) 0.020 (0.40) 5.211*** (14.10)

−0.093 (−0.59) 2.646 (1.31) 0.196*** (3.81) 0.019 (1.60) −0.016 (−1.42) 0.000 (0.00) 0.017 (0.36) 6.957*** (17.78)

−0.093 (−0.59) 2.619 (1.26) 0.193*** (3.81) 0.019 (1.59) −0.016 (−1.51) −0.000 (−0.02) 0.015 (0.31) 6.463*** (14.19)

−0.169 (−0.86) 7.807*** (4.88) 0.343*** (4.98) −0.000 (−0.00) −0.012 (−1.30) −0.002 (−0.16) 0.115 (1.57) −2.084* (−1.92)

−0.162 (−0.81) 7.653*** (5.02) 0.332*** (5.44) 0.000 (0.01) −0.014 (−1.41) −0.002 (−0.18) 0.109 (1.45) −2.713** (−2.74)

−0.180 (−0.97) 7.985*** (4.91) 0.348*** (4.68) −0.000 (−0.01) −0.013 (−1.26) −0.002 (−0.15) 0.113 (1.58) −1.081 (−0.88)

−0.177 (−0.97) 7.857*** (5.02) 0.341*** (4.90) 0.000 (0.01) −0.015 (−1.39) −0.002 (−0.16) 0.107 (1.48) −2.539** (−2.23)

−0.024 (−0.15) 2.380 (1.34) 0.203*** (3.93) 0.018* (1.72) −0.017 (−1.60) −0.000 (−0.08) 0.016 (0.37) 6.577*** (16.95)

−0.031 (−0.19) 2.362 (1.29) 0.198*** (3.75) 0.018* (1.72) −0.017* (−1.65) −0.000 (−0.09) 0.014 (0.32) 6.125*** (16.02)

−0.012 (−0.07) 2.495 (1.38) 0.203*** (3.65) 0.019* (1.79) −0.017 (−1.57) −0.000 (−0.09) 0.010 (0.24) 6.850*** (16.87)

−0.019 (−0.12) 2.480 (1.33) 0.198*** (3.45) 0.019* (1.78) −0.018* (−1.65) −0.000 (−0.09) 0.009 (0.20) 6.372*** (15.40)

0.272 (1.10) 6.576*** (4.00) 0.433*** (5.46) 0.001 (0.05) −0.017** (−2.17) −0.006 (−0.52) 0.090 (1.59) −2.035** (−2.02)

0.257 (1.02) 6.466*** (4.21) 0.419*** (5.59) 0.002 (0.06) −0.018** (−2.21) −0.006 (−0.50) 0.085 (1.45) −3.349*** (−3.56)

0.275 (1.13) 6.911*** (3.95) 0.447*** (5.32) 0.001 (0.04) −0.020** (−2.04) −0.006 (−0.51) 0.082 (1.57) −1.792* (−1.68)

0.265 (1.05) 6.793*** (4.08) 0.441*** (5.22) 0.002 (0.07) −0.021** (−2.17) −0.006 (−0.50) 0.076 (1.44) −3.049*** (−2.82)

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

0.47 550

0.47 550

0.47 550

0.47 550

0.24 550

0.24 550

0.24 550

0.24 550

0.47 549

0.47 549

0.47 549

0.47 549

0.13 549

0.14 549

0.12 549

0.13 549

This table reports results from OLS estimates and IV estimates of CEO compensation. The dependent variables are the cumulative CEO compensation (in log), CWAGE t+1,t+2 and S&P adjusted cumulative CEO compensation (log), CAWAGE t+1,t+2 two years following the CEO turnover. The definitions of all variables are the same as in Table A.1. Standard errors are clustered at the industry level. * ** ***

Significance at the 10% level. Significance at the 5% level. Significance at the 1% level.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

VOLATILITY

CAWAGE t+1,t+2

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

49

5.4. ICT, firm performance, and CEO compensation In this section, we investigate whether the positive effect on firm performance of an external CEO via ICT can also explain the rise in executive compensation. In a performance-based compensation scheme, better firm performance implies higher compensation for the CEO. When CEO general skills induce better firm performance, we should also see a rise in CEO compensation. We hypothesize that when CEO general skills lead to better firm performance, superior performance translates into higher CEO wages, all else being equal. We already show in columns (1) and (2) of Table 4 that ICT is positively and significantly correlated with CEO compensation. Table 6 also shows that the unconditional difference of CEO compensation between high-ICT industries and low-ICT industries is positive and significant at the 5% level. Panel B of Table 6 also shows that external CEOs in high-ICT industries fetch bigger pay packages than CEOs (both external and internal) in low-ICT industries. However, correlation does not imply causality. To this end, we estimate the following regression model by OLS:  CWAGE ijt+1,t+2 = ˛ + ˇ · (ICT jt × EXTCEOijt ) + CROAijt+1,t+2 + Xijt ı + εijt

(6)

where CWAGE ijt+1,t+2 is the cumulative total CEO compensation in years t + 1 and t + 2 for a CEO hired in year t; the other variables are the same as defined earlier; and Xijt contains other explanatory variables, including industry and year fixed effects. In the regression model (6) above, the coefficient of the interaction term, ˇ, captures the effect of the CEO general skills on the CEO wage after firm performance has been accounted for. We separately control for CROAijt+1,t+2 in the wage equation to isolate the skill effect (ˇ) from the performance effect (). We also estimate the same model using CAWAGE t+1,t+2 as an alternative dependent variable, where CAWAGE t+1,t+2 is defined as the peeradjusted cumulative total CEO compensation in years t + 1 and t + 2 for a CEO hired in year t; we subtract the median S&P 500 constituent CEO wage from the sample CEO wage to calculate the peeradjusted CEO compensation. Table 10 reports the regression results for the above models. In columns (1)–(8), Table 10 reports the OLS regression results and show that CEO wage increases with the current performance of a firm, and more importantly, that the interaction coefficients between external-CEO hiring and ICT are positive and statistically significant at the 5% level. These results suggest that industry-level ICT, our proxy for general skill requirement, has a significant effect on CEO wages when an external CEO, as opposed to an internal CEO, is hired. The interaction effect is also economically significant: a 10% increase in the industry-level ICT from the mean is associated with a 1.4% increase in CEO compensation in column (3) when all other explanatory variables are evaluated at their mean. As we take into account other control variables, results in columns (1)–(8) show that larger firms and firms with better future growth opportunities and higher past stock returns pay higher wages to CEOs. One limitation with the regression specification (6) is the endogeneity in the firm performance variable. To control for the potential endogeneity bias, we estimate (6) using the instrumental variable method. For identification, we use the past accounting performance of a firm as an instrument for CROAijt+1,t+2 and CAROAijt+1,t+2 . In particular, we use the cumulative (excess) returns on a firm’s total assets in years t − 2 and t − 1 (CROAijt−2,t−1 and CAROAijt−2,t−1 ) as instruments since past accounting performance is likely to be correlated with current and future performance, but it is not obvious why accounting performance in periods t − 1 and t − 2 should be related to CEO wages in years t + 1 and t + 2 when the CEO is hired in year t. We estimate regression model (6) by using 2SLS, and the results are reported in columns (9)–(16) in Table 10. We find that the coefficients of the interaction terms are still positive and significant at the 5% level. In short, the empirical evidence indicates that the general component of the CEO skills affects firm performance, and that better firm performance in turn leads to higher CEO compensation. 6. Conclusion In recent years, an increasing number of corporate boards have looked outside their company ranks to fill CEO vacancies. At the same time, the executive labor market has witnessed a dramatic rise in the overall level of executive compensation. This paper focuses on the market-based theory of executive

50

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

compensation and external CEO hiring and assesses empirically the relationship between CEO general skills and trends in the executive labor market. We argue, on the basis of a market-based theory, that when general skills become more important relative to firm-specific skills, firms are more likely to hire external CEOs. As firms compete in the executive labor market, CEOs capture most of the surplus and, as a result, executive compensation increases. The theoretical framework also implies that when general skills become more important, the excess in expected profits of an internal-CEO firm over an external-CEO firm decreases. Our paper provides direct evidence in support of the market-based theory. Using ICT as a proxy for the importance of general skills, we show the following: First, firms in industries with more ICT capital stock per worker hire more external CEOs compared to those in industries with less ICT capital stock per worker. Furthermore, as the level of ICT increases in an industry, firms in that industry are more likely to hire external as opposed to internal CEOs. Second, internal-CEO firms have higher profits compared to external-CEO firms, but the difference in profits decreases as ICT levels increase in the industry. Finally, we relate CEO skills to CEO compensation and show that CEO general skills induce better firm performance, which in turn leads to greater CEO compensation. Taken together, our results suggest that external CEOs bring in valuable social capital (general skills) to the company. Appendix A. Theoretical framework In this section, we briefly explain the partial equilibrium CEO-hiring model of Murphy and Zábojník (2007) and show that when general skills become more important, the excess of a firm’s expected profits from hiring an internal CEO over an external CEO decreases.28 Formally, we follow Murphy and Zábojník (2007) and assume that each firm hires one CEO and n production workers with the following profit function: (n, a, s) = f (n)sa − wn − wM (a)

(7)

where f ( · ) is the production function with f  ( · ) > 0 and f  ( · ) < 0, a is the ability of the CEO, wM (a) is the market wage for a CEO with ability a, w is the market wage for production workers, and s captures whether the CEO is internally promoted or externally hired: s = 1 if the CEO is internally promoted and s =  < 1 if she is externally hired.29 Assume that there is free entry of firms and that a CEO can always set up a firm that is best matched with her ability a. Therefore, a CEO with ability a must be paid at least (a) = f [n∗ (a)]a − wn∗ (a), where n∗ (a) is the optimal size of the firm set up by the CEO. Since a CEO can choose to be a production worker and receive a wage w, the market wage for a CEO with ability a is given by: wM (a) = max{w,

(a)} = max{w, f [n∗ (a)]a − wn∗ (a)}.

(8)

The firm’s optimal hiring decision is as follows. If the firm hires an external CEO with ability a∗ that is the best match given its size, i.e., n = n∗ (a∗ ), then the firm’s profit will be exactly zero: E = 0.30 If the firm has an internal employee with ability aˆ such that I = (n, aˆ , 1) ≥ 0, then the firm will promote her as the CEO. When a firm hires an internal CEO, it can make full use of the CEO’s ability for production, but if the CEO works for another firm, only  aˆ of her ability is valued by the market. Therefore, a firm’s profit is positive when an internal CEO is hired. These results are shown graphically in Fig. 2.31 We restrict our attention to the case when the relative importance of general skills is L . In this case, the firm earns a positive profit if the internal candidate’s ability aˆ is between aL (L ) and aH (H ). In other words, the firm will hire an internal CEO if aˆ ∈ [aL (L ), aH (H )]; otherwise, it will hire an external CEO (Table A.1).

28 Murphy and Zábojník (2007) also have a general equilibrium version of the model, and its theoretical predictions are basically the same. 29 The parameter  ∈ (0, 1) measures the importance of general skills relative to specific skills, and the higher  is, the more important general skills are. 30 This can be seen from (7) and (8): E = [n∗ (a∗ ), a∗ , ] = f [n∗ (a∗ )]a∗ − wn∗ (a∗ ) − wM (a∗ ) = 0. 31 This figure is based on Figure 6 of Murphy and Zábojník (2007).

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

51

Fig. 2. Importance of general skills and firm performance. This figure is adopted from Murphy and Zábojník (2007). It shows the difference between the expected profits of hiring an internal and an external CEO for a given level of general skills. When the importance of general skills is L , the difference between the expected profits of hiring an internal and an external CEO is given by the area ABL CL . When the importance of general skills is H , the difference between the expected profits of hiring an internal and an external CEO is given by the area ABH CH . Together, these imply that when general skills become relatively important ( increases), firms with an externally hired CEO perform better than firms with an internally hired CEO in the sense that the area (ABL CL and ABH CH ) becomes smaller as  increases.

Proposition 1. The excess of a firm’s expected profits from hiring an internal CEO over hiring an external CEO decreases with the importance of general skills. A.1. Graphical exposition of Proposition 1 From Fig. 2, we can see that when general skills become more important ( increases), the market wage of a CEO goes up because (a) shifts up, and the threshold value aL stays the same, but aH becomes smaller. The expected difference between the profits of hiring an internal or an external CEO when the importance of general skills is L is the area ABL CL , and when the importance of general skills is H it is the area ABH CH . The area becomes smaller when  increases. The intuition for the above result is as follows. A firm with a vacant CEO position faces the tradeoff between promoting an internal candidate and hiring an external CEO. Internal promotion capitalizes on the CEO’s firm-specific knowledge, but this comes at the risk of not getting the best possible CEO for the job. When general skills become relatively more important (relative to firm-specific skills), the costs of external CEO hiring (foregone internal CEO’s firm-specific knowledge) decrease and in turn increase the likelihood of firms with vacant CEO positions hiring external CEOs as opposed to internal CEOs. Besides, competition among firms to hire the best possible CEO enables a CEO to extract maximum rent from her employer, which in turn drives down the firm’s profit to zero, but the CEO’s general skills become fully priced into the compensation contract. Furthermore, the firm-specific knowledge of the internal CEO is less valuable so that the advantages of firms with internal CEOs relative to firms with external CEOs diminish. A.2. Formal proof of Proposition 1 Assume that the internal candidate’s ability a follows a certain distribution with pdf (a) and a support [0, a] for some maximum ability level a. Denote E( I ) and E( E ) as the expected profits of

52

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Table A.1 Variable definitions. Variable Firmcharacteristics CROAt+1,t+2

CAROAt+1,t+2

FIRMSIZE ijt TANGIBILITY ijt TOBIN  SQijt

LEVERAGE ijt (CAPX/SALES)ijt (R&D/CAPX)ijt (ADVERT /CAPX)ijt VOLATILITY ijt

DIVYIELDijt PASTRETURN ijt GINDEX ijt TOP5MGT ijt CEOcharacteristics CWAGE t+1,t+2

CAWAGE t+1,t+2

CEOAGE ijt CEOMBAijt EXTCEOijt DIFFIND EXTCEOijt VOLUNTARY ijt GABILITY ijt CEOFE ijt Industrycharacteristics LEVEL − ICT ijt

LEVEL − ICT ijt−1 LOG − ICT ijt LOG − ICT ijt−1 INDHOMOj

Definition The “Cumulative Return on Total Assets” (CROAt+1,t+2 ) is calculated by summing the “Return on Total Assets” (ROA) over two years following the CEO turnover. The ROA is calculated from the COMPUSTAT data items as: ebitda/at. The “Cumulative Adjusted Return on Total Assets” (CAROAt+1,t+2 ) is calculated by summing the “S&P Adjusted Return on Total Assets” (AROA) over two years following the CEO turnover. AROA is calculated by subtracting the median ROA of the S&P constituent firms from the sample firm ROA. The proxy for firm size (total assets) calculated from the COMPUSTAT data item as: Log(at) “Asset Tangibility” is defined as Net Fixed Assets/Total Assets and is calculated using the COMPUSTAT data items as: ppent/at. Tobin’s Q is calculated following Duchin (2010) and COMPUSTAT data items as: market value of assets (book assets (at) +market value of common equity (csho × prcc) – common equity (ceq) – deferred taxes (txdb))/ (0.9 × book value of assets (at) + 0.1 × market value of assets). The book leverage is calculated using COMPUSTAT data items as: (debt in current liabilities (dlc) +long-term debt (dltt))/book assets (at). The “Capital Expenditure over Sales” is from (Kale et al., 2009) and calculated using COMPUSTAT data items as: capx/sale. The “Research &Development Expenditure” is from (Kale et al., 2009) and calculated using COMPUSTAT data items as: xrd/capx. The “Advertising Expenditure” is from (Kale et al., 2009) and calculated using COMPUSTAT data items as: xad/capx. The “Return Volatility” is defined as the standard deviation of stock return over the past ten years before the CEO turnover. Stock return is calculated using COMPUSTAT data items as: ((prcc it /prcc it−1 ) − 1). The “Dividend Yield” is from (Kale et al., 2009) and defined as dividend per share by ex-date by close price of the fiscal year. It is calculated using COMPUSTAT data item as: dvpsp f . The stock return two years prior to the CEO turnover event. The corporate governance index from (Gompers et al., 2003). The fraction of shares owned by the top five executives in the firm and is collected from the Executive-COMPUSTAT. The “Cumulative CEO Wages” is calculated by summing the “CEO Wage” (WAGE) over two years following the CEO turnover. The WAGE calculated from the Executive-COMPUSTAT data item tcd1 following Custódio et al. (in press). The “Cumulative Adjusted CEO Wage” is calculated by summing up the “S&P Adjusted CEO Wage” (AWAGE) over two years following the CEO turnover. AWAGE is calculated by subtracting the median WAGE of the S&P constituent firms from the sample firm WAGE. The WAGE calculated from the Executive-COMPUSTAT data item tcd1 following Custódio et al. (in press). The age of the CEO from the Executive-COMPUSTAT and hand collected by authors. Dummy variable indicating whether the CEO has an MBA degree and is hand collected by authors. Dummy variable indicating whether the CEO is an externally hired CEO and is calculated by the authors. Dummy variable indicating whether the external CEO is hired from outside the 4-digit SIC industry of the firm. Dummy variable indicating whether the previous CEO turnover was voluntary as opposed to forced. The general CEO ability index from Custódio et al. (in press). CEO performance fixed effects estimated from Graham et al. (2012). The “Information and Communication Technology” (ICT) measure is based on private assets data from the National Income and Product Accounts (NIPA) tables of BEA, and on the number of workers in different industries from the Current Employment Statistics (CES) published by BLS. For each industry j and year t, we construct the level ICT measure as the ICT capital stock per worker: ICT jt = (Stock of Computer & Communication Equipment & Softwarejt )/(Total Number of Workersjt ). The value of the ICT assets are in constant 2000 dollars. See U.S. Department of Commerce (2003) for more details about the construction of a quality-adjusted price index for computer and other equipment. The one year lag of the ICT ijt measure. The logarithm of the ICT ijt measure. The logarithm of the lag of the ICT ijt measure. Industry homogeneity measure from Parrino (1997).

This table provides the definitions of firm-, CEO- and industry-specific variables used in the paper.

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

53

hiring an internal CEO and an external CEO respectively, and = E( I ) − E( E ). We want to show that ∂ /∂ < 0. a From (1) and (2), we can see that E( E ) = 0. On the other hand, E( I ) = a H (n, ˛, 1)(˛)d˛ =

 aH aL

[f (n)˛ − wn − wM (˛)](˛)d˛. Therefore, the difference in profits is:



L

aH

[f (n)˛ − wn − wM (˛)](˛)d˛.

=

(9)

aL

By the Leibniz Integral Rule,32 we have:



∂ = ∂ ∂



aH ()

[f (n)˛ − wn − wM (˛)](˛)d˛



aL aH ()



= aL

 ∂wM (˛) ∂aH () (˛)d˛ + f (n)aH () − wn − wM [aH ()] [aH ()] . ∂ ∂

(10)

The first term in the last line above is negative because ∂wM ( · )/∂ > 0. For the second term, the expression in the curly bracket is positive because we are integrating over the region [aL , aH ] in which the firm makes a positive profit. Since ∂aH ()/∂ < 0, the second term is negative. Thus, ∂ /∂ < 0. . References Acemoglu, D., Aghion, P., Lelarge, C., Van Reenen, J., & Zilibotti, F. (2007). Technology, information, and the decentralization of the firm. Quarterly Journal of Economics, 122, 1759–1799. Alexopoulos, M., & Tombe, T. (2012). Management matters. Journal of Monetary Economics, 59, 269–285. Autor, D., Katz, L., & Krueger, A. (1998). Computing inequality: Have computers changed the labor market? Quarterly Journal of Economics, 113, 1169–1213. Bain & Company, 2011. Management Tools & Trends 2011 Survey. Bebchuk, L., Fried, J., & Walker, D. (2002). Managerial power and rent extraction in the design of executive compensation. University of Chicago Law Review, 69, 751–761. Bertrand, M., & Schoar, A. (2003). Managing with style: The effects of managers on firm policies. Quarterly Journal of Economics, 118, 1169–1208. Bresnahan, T. (1999). Computerisation and wage dispersion: An analytical reinterpretation. Economic Journal, 109, 390–415. Bresnahan, T., Brynjolfsson, E., & Hitt, L. (2002). Information technology, workplace organization, and the demand for skilled labor: Firm-level evidence. Quarterly Journal of Economics, 117, 339–376. Brynjolfsson, E., & Kim, H. (2009). CEO compensation and information technology. Working paper, Massachusetts Institute of Technology. Caroli, E., & Van Reenen, J. (2001). Skill-biased organizational change? Evidence from a panel of British and French establishments. Quarterly Journal of Economics, 116, 1449–1492. Chun, H., Kim, J., Morck, R., & Yeung, B. (2008). Creative destruction and firm-specific performance heterogeneity. Journal of Financial Economics, 89, 109–135. Custódio, C., Ferreira, M., & Matos, P. Generalists versus specialists: Lifetime work experience and CEO pay. Journal of Financial Economics, in press. Daines, R., Nair, V., & Kornhauser, L. (2005). The good, the bad and the lucky: CEO pay and skill. Working paper No. CLB-06-005, New York University. Duchin, R. (2010). Cash holdings and corporate diversification. Journal of Finance, 65, 955–992. Frydman, C. (2006). Rising through the ranks: The evolution of the market for corporate executives, 1936–2003. Working paper, MIT Sloan School of Management. Frydman, C., & Jenter, D. (2010). CEO compensation. Annual Review of Financial Economics, 2, 75–102. Frydman, C., & Saks, R. (2010). Executive compensation: A new view from a long-term perspective, 1936–2005. Review of Financial Studies, 23, 2099–2138. Gabaix, L., & Landier, A. (2008). Why has CEO pay increased so much? Quarterly Journal of Economics, 123, 49–99. Gompers, P., Ishii, J., & Metrick, A. (2003). Corporate governance and equity prices. Quarterly Journal of Economics, 118, 107–155. Graham, J., Li, S., & Qiu, J. (2012). Managerial attributes and executive compensation. Review of Financial Studies, 25, 144–186. Harland, C. (1996). Supply chain management, purchasing and supply management, logistics, vertical integration, materials management and supply chain dynamics. In N. Slack (Ed.), Blackwell Encyclopedic Dictionary of Operations Management. UK: Blackwell.

32

The Leibniz Integral Rule says that:

∂ ∂˛

 b(˛) a(˛)

f (x, ˛)dx =

 b(˛) ∂f (x,˛) a(˛)

∂˛

dx + f [b(˛), ˛] ∂b(˛) − f [a(˛), ˛] ∂a(˛) . ∂˛ ∂˛

54

V.A. Aivazian et al. / Journal of Economics and Business 67 (2013) 24–54

Heckman, J. (1979). Sample selection bias as a specification error. Econometrica, 47, 153–161. Heckman, J., & Navarro-Lozano, S. (2004). Using matching, instrumental variables, and control functions to estimate economic choice models. Review of Economics and Statistics, 86, 30–57. Heckman, J., & Robb, R. (1985). Alternative methods for evaluating the impact of interventions: An overview. Journal of Econometrics, 30, 239–267. Hermalin, B. (2005). Trends in corporate governance. Journal of Finance, 60, 2351–2384. Holmström, B., & Kaplan, S. (2001). Corporate governance and merger activity in the United States: Making sense of the 1980 and 1990. Journal of Economic Perspectives, 15(2), 121–144. Huson, M., Parrino, R., & Starks, L. (2001). Internal monitoring mechanisms and CEO turnover: A long-term perspective. Journal of Finance, 56, 2265–2297. Huson, M., Malatesta, P., & Parrino, R. (2004). Managerial succession and firm performance. Journal of Financial Economics, 74, 237–275. Kale, J., Reis, E., & Venkateswaran, A. (2009). Rank-order tournament and incentive alignment: The effect on firm performance. Journal of Finance, 64, 1479–1512. Lang, L., & Stulz, R. (1994). Tobin’s q, corporate diversification, and firm performance. Journal of Political Economy, 102, 1248–1280. Murphy, K., & Zábojník, J. (2004). CEO pay and appointments: A market-based explanation for recent trends. In American economic review papers and proceedings (p. p94). Murphy, K., & Zábojník, J. (2007). Managerial capital and the market for CEOs. Working paper, Marshall School of Business, University of Southern California and Department of Economics, Queen’s University. Parrino, R. (1997). CEO turnover and outside succession: A Cross-sectional analysis. Journal of Financial Economics, 46, 165–197. Rost, K., Salomo, S., & Osterloh, M. (2008). CEO appointments and the loss of firm-specific knowledge: Putting integrity back into hiring decisions. Corporate Ownership and Control, 5, 86–98. U.S. Department of Commerce (2003). Fixed assets and consumer durable goods in the United States: 1925–1997. U.S. Department of Commerce, Economics and Statistics Administration, Bureau of Economic Analysis, Washington, DC.