Corporate Culture and CEO Turnover January 31st, 2013 Abstract We study the effect of corporate culture on the relationship between firm performance and CEO turnover. Utilizing a measure of cultural dimension developed in Organization Behavior, we quantify corporate culture by assessing corporate documents using a text analysis approach. We employ this quantification to examine the impact of culture on the shareholders’ decision to fire the CEO, especially in the case of poor firm-specific performance. We show that the probability of a CEO change in the case of poor performance is greater for companies with a culture oriented toward control. Irrespective of performance, a culture oriented toward competition or innovation is positively related to the frequency of management turnover, whereas the contrary occurs for a culture oriented toward collaboration. Keywords: Corporate culture, Text analysis, Corporate governance, CEO JEL classification: G14, G21, G34, M14 Acknowledgements: We thank Arnoud Boot, Olivier DeJonghe, Alessandra Ferrari, Giorgio Gobbi, Emmanuel Mamatzakis, Roman Matousek, Phil Molyneux, Nikolas Papanikolaou, Enrico Sette, and Amine Tarazi for helpful comments. We are also thankful to participants at the seminars of the Financial Intermediation of European Studies (FINEST) at the University of Rome III, and the Free University of Bozen. We would like to give special thanks to Anjan Thakor for his continuous support and great suggestions. Franco Fiordelisi also wishes to acknowledge the support of the Fulbright Commission and the Olin Business School of the Washington University in St. Louis, U.S. We alone are responsible for any remaining errors. ___________________________ * Corresponding author: Via S. D’Amico 77, 00145 Rome, Italy, tel. +39 06 57335672; fax. +39 06 57335797; e-mail: fiordelisi@uniroma3.it . 1 1. Introduction The threat of CEO change after poor performance is one of the main instruments available to shareholders to align managers’ interests to their goals. It is widely believed that corporate culture plays an important moderating role in linking corporate culture and past performance. Surprisingly, we are not aware of any large-sample empirical evidence to show whether and how corporate culture influences the link between firm performance and the probability of CEO change. This lack of research is perhaps because the notion of culture is somewhat nebulous and raises several measurement issues in empirical research (Guiso et al., 2006). Nonetheless, recent research has begun to explore the empirical link between culture and various economic phenomena using novel approaches to measuring culture (Fang, 2001; Guiso et al., 2006; Bernhardt et al., 2006; Guiso et al., 2009; Luttemer and Singhal, 2011), but this research has not addressed CEO turnover. What role does corporate culture play in the decision to fire a CEO after bad performance? Is there a specific firm culture that increases (decreases) the probability of changing a CEO after bad performance? These questions are critical for assessing the credibility of the CEO change threat as a corporate governance instrument. The purpose of this paper is to empirically address these questions by focusing on a large sample of US listed companies between 1994 and 2011. Our approach involves obtaining a quantitative measurement of corporate culture by assessing corporate financial statements (e.g., 10-K reports). Text analysis has recently been used in various finance papers (e.g., Antweiler and Murray., 2004; Tetlock, 2007; Li, 2008; Tetlock et al., 2008; Loughran and McDonald, 2011). This method allows us to link the probability of a CEO change to the extent of various corporate culture orientations. Our main result is that corporate culture influences the probability of a CEO 2 change. Specifically, the probability of CEO change increases in the case of a corporate culture oriented toward enhancing competitiveness and emphasizing organizational effectiveness and fast response. In the case of bad performance, the probability of a CEO turnover increases in companies with a high control-oriented culture, i.e., companies focusing on internal improvements in efficiency through the implementation of better processes. These results are consistent with different definitions of CEO change, different time periods (bad performance are 1 and 2 years before the CEO change) and performance indicators (ROA and ROS). The rest of this paper is organized as follows. Section 2 provides a literature review, and Section 3 illustrates our definitions of corporate culture and formulates our research hypotheses. The econometric framework appears in Section 5. Section 6 discusses the empirical results and robustness checks, and Section 7 concludes. 2. Related literature To be considered a valuable corporate governance instrument, CEO change must to be credible in the sense that the shareholders’ decision to fire the CEO is negatively related to firm performance. Early papers (Coughlan and Schmidt, 1995; Warner et al., 1988) find a negative link between firm performance and CEO change. The relationship between performance and CEO change is not a simple and direct one. Since the 1990s, various authors (e.g., Zajac, 1990) have noted that neither the strategic management nor the financial economic literature offer a unified theoretical or empirical framework for topics related to CEO succession. Furthermore, these studies have relied exclusively on archival data, with no attempt to collect or analyze primary data provided by the CEOs themselves. Consequently, the results are not always consistent, and past performance often explains only a very 3 low portion of the turnover phenomenon (Pitcher et al., 2000). More recent papers (e.g., Wiersema and Zhang (2011) recognize that research has advanced our knowledge of the firm performance – CEO turnover linkage, but the relationship continues to appear complex and somewhat ambiguous amid the existence of several variables that may play a moderating role (e.g., CEOs’ influence on boards through direct ownership, as outlined in Easterwood et al., 2012). Jenter and Kanaan (2012) have recently proposed a novel approach that splits firm performance into two components (systematic and firm specific), showing that CEOs are significantly more likely to be dismissed from their jobs after bad industry or bad market performance. Although there appears to be convergent evidence that CEO change is credible, there are no studies assessing the reason for this link, as noted by Jenter and Kanaan (2012, page 4) “more research is needed to identify the root cause of the peer performance effect on CEO turnover”. The main contribution of our paper is that it is the first to provide empirical evidence (based on a large sample) that the relationship between performance and CEO turnover is strongly influenced by corporate culture. Although this finding is certainly logical and intuitive, there have been no studies documenting whether and how corporate culture influences the relationship between firm performance and shareholders’ decision to fire the CEO. We build a unique dataset of all US listed companies between 1994 and 2011 by obtaining a quantitative measurement (at the company level) of corporate culture by assessing financial statements. Our approach is based on text analysis (recently used in such finance papers as Antweiler and Murray, 2004; Tetlock, 2007; Li, 2008; Tetlock et al., 2008; Loughran and McDonald, 2011), which provides us an objective assessment of corporate culture. As suggested by Jenter and Kanaan (2012), we use a two-stage regression 4 approach to assess the sensitivity of CEO turnover to firm-specific performance. Jenter and Kanaan’s (2012) approach enables us to decompose firm performance into a systematic component (caused by peer group performance) and a firm-specific component that should reflect CEO ability. This approach fits our research needs very well. First, this approach is an effective instrumental variable estimation, with peer group performance serving as an instrument for firm performance (Jenter and Kanaan, 2012)1. Second, CEOs should be evaluated based on the firm-specific component of firm performance only and not on the performance of its reference group (e.g., the industry). 3. Theory and Hypotheses Culture is a broad concept and represents the implicit and explicit contracts that govern behavior within an organization (Benabou and Tirole, 2002 and 2011; Tabellini, 2008). Corporate culture is traditionally considered to have an important influence on an organization’s effectiveness (Deal and Kennedy, 1982; Peters and Waterman, 1982; Schein, 1992; Wilkins and Ouchi, 1983), and, in a recent review of the literature, Sackmann (2010) suggests that some culture orientations have a positive effect on performance measures. A first necessary step for our analysis is to define culture in a sufficiently narrow way within this framework so that it is possible to identify its influence on the relationship between CEO turnover and company performance change. We adopt the definition proposed by Cameron et al. (2006) who identify the following four types of corporate cultures (labeled as culture-dimensions): control, compete, collaborate and The tests treat peer performance as a plausibly exogenous instrument for the “luck” that has aided or hampered the CEO’s running of the firm. For peer group performance to be a valid instrument, it is necessary that (i) peer performance be exogenous and (ii) peer performance itself not have an effect on CEO dismissals. 1 5 create. 3.1 Collaboration-oriented culture A collaboration-oriented culture focuses internally on its employees and attempts to develop human competencies and to strengthen organizational culture by building consensus. The goal of this culture is to develop cooperative processes and attain cohesion through consensus and broad employee involvement, e.g., clarifying and reinforcing organizational values, norms, and expectations, developing employees and cross-functional work groups, implementing programs to enhance employee retention, and fostering teamwork and decentralized decision making. Companies with this culture usually succeed because they hire, develop, and retain their human resource base. We posit that as the collaboration-orientation of corporate culture increases, there is a smaller probability of a CEO change. A greater collaboration-orientation of corporate culture implies that boards care about the development of human competencies and consensus on a daily basis rather than organizational effectiveness, resulting in a situation where CEO change is not used as a corporate governance tool and is less likely to occur relative to other companies. Hypothesis 1 (H1): Firms with a greater collaboration-oriented corporate culture have a smaller probability of experiencing CEO change compared to other firms. We also posit that as firm-specific profits decrease in a company with a greater collaboration-orientation culture, there is a smaller probability of a CEO change. In companies with greater collaboration-orientation of corporate culture, CEO change in the case of poor performance is less probable, as shareholders care less about poor 6 performance than in other companies and will not punish the CEO for bad performance. Hypothesis 2 (H2): In the case of poor performance, firms with a greater collaboration-orientation of corporate culture have a lower probability of experiencing CEO change compared to other firms. 3.2 Competition-oriented culture A competition-oriented culture focuses on the organization’s external effectiveness by pursuing enhanced competitiveness and emphasizing organizational effectiveness, fast response, and customer focus. These companies usually attach the highest priority to customers and shareholders and judge success on the basis of indicators such as market share, revenues, meeting budget targets, and profitability growth. We posit that as the competition-orientation of corporate culture increases, there is a greater probability of a CEO change. A greater competition-orientation of corporate culture implies that boards care on a daily basis about the competitiveness and organizational effectiveness; thus, CEO change (i.e., a last-resort measure) is more likely than in other companies. Hypothesis 3 (H3): Firms with a greater competition-orientation of corporate culture have a higher probability of experiencing CEO change compared to other firms. Second, we assume that as the firm-specific profits decrease in a company with a greater competition-orientation culture, there is a greater probability of a CEO change. In companies with greater competition-orientation of corporate culture, a CEO change (i.e., a last-resort measure) in the case of poor performance is more probable because shareholders will punish the CEO for poor performance more than in other companies where competition is less important (H4). Hypothesis 4 (H4): In the case of poor performance, firms with a greater competition-orientation of corporate culture have a higher probability of experiencing CEO change compared to other firms. 7 3.3 Control-oriented culture A control-oriented culture refers to one that focuses on creating value through internal improvements in efficiency, the implementation of better processes (e.g., by the extensive use of processes, systems and technology) and quality enhancements (such as statistical process control and other quality control processes). Companies that have this culture usually make extensive use of standardized procedures and emphasize rule reinforcement and uniformity. We posit that as the control-orientation of corporate culture increases, there is a greater probability of a CEO change. A greater control-orientation of corporate culture implies that boards care on a daily basis about the implementation of better production processes and thus that CEO change (i.e., a last-resort measure) is more probable than in other companies. Hypothesis 5 (H5): Firms with a greater control-orientation of corporate culture have a higher probability of experiencing CEO change compared to other firms. Second, we believe that as the firm-specific profits decrease in a company with a greater control-orientation culture, there is a greater probability of a CEO change. In companies with greater control-orientation of corporate culture, the CEO change (i.e., a last-resort measure) in the case of poor performance is more likely because shareholders will punish the CEO for poor performance more than in other companies where control is less important. Hypothesis 6 (H6): In the case of poor performance, firms with a greater control-orientation of corporate culture have a higher probability of experiencing CEO change compared to other firms. 3.4 Creation-oriented culture 8 A creation-oriented culture focuses on creating future opportunities in the marketplace through innovation in the organization’s products and services of the organization. The organization encourages entrepreneurship, vision and constant change, e.g., allowing for freedom of thought and action among employees so that rule breaking and reaching beyond barriers are common characteristics of the organization’s culture. These companies usually aim to include innovative product-line extensions, radical new process breakthroughs, innovations in distribution and logistics that redefine entire industries and to develop new technologies. We posit that as the creation-orientation of corporate culture increases, there is a smaller probability of a CEO change. A greater creation-orientation implies that boards care about products and service innovations on a daily basis more than they do organizational effectiveness and thus that CEO change is not used as corporate governance tool and is less likely than in other companies. Hypothesis 7 (H7): Firms with a greater creation-oriented corporate culture have a smaller probability of experiencing CEO change compared to other firms. We also assume that as firm-specific profits decrease in a company with a greater creation-orientation of corporate culture, there is a smaller probability of a CEO change. In companies with a greater creation-orientation, the probability that the CEO will be fired in the case of poor performance (i.e., a last-resort measure) is lower than in other companies, as shareholders care more about vision and constant change than about performance. Hypothesis 8 (H8): In the case of poor performance, firms with a greater creation-orientation of corporate culture have a lower probability of experiencing CEO change compared to other firms. 9 Table 1 summarizes the attributes of the corporate culture orientations proposed by Cameron et al. (2006). <<< INSERT TABLE 1 >>> 4. Empirical Approach This section describes the data we use in our analysis, the econometric approach and the testable hypotheses. 4.1 Data To answer our research questions, we build a unique dataset by collecting information from three different sources to have a comprehensive and careful profile of each company. First, we obtained information on top executive officers (and specifically, on CEOs’) from the Execucomp database. Data were available for the period 1992-2011, resulting in 209,840 year-observations. We excluded all cases of inconsistent or missing data (i.e., if the CEO annual flag was in conflict with the dates when the interested manager joined or left the company or if the information about the identity of the CEO was available only for a specific year but not for the previous one, rendering it impossible to know whether there was a turnover). Second, we obtained financial data by extracting 247.796 simplified balance sheets from Compustat between 1990 and 2011. As in the previous case, we removed all companies with missing data. Third, we obtained 128.489 company filings by downloading the 10-K reports (available from 1994 to 2011) from the SEC’s Edgar database, and for each of these 10 filings, we ran a text analysis to estimate each cultural dimension identified by Cameron et al. (2006) (513,956 texts analyzed overall). Our final sample includes all US listed companies between 1994 and 2011 for which it was possible to a) collect information on top managers, b) determine accounting-based performance, and c) measure the relevant cultural dimensions. Information relative to the same company and drawn from different databases was matched using the GVKEY code. As a result, we have a unique dataset of 19.453 yearobservations, combining managerial, accounting and cultural information. A CEO change occurs at a frequency of approximately 10% (see Table 2). We distinguished financial from non-financial companies on the basis of the SPINDEX code and the industry group definitions: observations for financial and non-financial companies represent, respectively, approximately 15% and 85% of our sample. We considered the following industry groups as financial groups: Asset Management and Custody banks; Consumer finance; Diversified banks; Diversified REITs; Industrial REITs; Insurance brokers; Investment banking and brokerage; Life and health insurance; Mortgage REITs; Multi-line insurance; Office REITs; Other diversified financial services; Property and casualty insurance; Real investment trust; Regional banks; Reinsurance; Residential REITs; Retail REITs; Specialized finance; Specialized REITs; and Thrifts and mortgage finance. Companies without an assigned industry group were excluded from the sample. <<< INSERT TABLE 2 >>> 4.2. Corporate culture estimation 11 We now describe our corporate culture variables. First, we outline our text analysis approach to estimate Cameron et al.’s (2006) corporate culture dimensions and then present our variables to measure culture homogeneity and heterogeneity. To quantitatively measure Cameron et al.’s (2006) four dimensions of corporate culture, we use text analysis. Text analysis is a technique used to examine, in a systematic and objective manner, the characteristics specific to a text (Stone et al., 1966). The idea underlying our approach is based on the assumption that the words and expressions used by the members of an organization (labeled “vocabulary”) represent the outcome of the culture they develop over time (Levinson, 2003). We posit that the distinctive features of any organization are mirrored in its written documents. Text analysis methodology (Stone et al., 1966) is instrumental in measuring the semantic content of firms’ publicly available official documents (namely, annual reports to shareholders). This technique provides us with measures that are less prone to the subjectivity of our opinion as researchers in interpreting the data. Recently, the text analysis approach has been exploited in various finance and management papers (e.g., Antweiler and Murray, 2004; Tetlock, 2007; Li, 2008; Tetlock et al., 2008; Hoberg and Phillips, 2010; Hoberg and Hanley 2010; Loughran and McDonald, 2011). To estimate Cameron et al.’s (2006) culture dimensions (i.e., collaborate, compete, control, and create, as defined in Table 1), we identify a large set of synonyms for each of these aspects. Following Carretta et al. (2011), synonyms are selected using a two-step procedure that minimizes subjectivity in the selection process. First, we selected synonyms suggested by authors (Cameron et al., 2006) to identify each culture dimension. Second, all words selected in the first step have been included within in the Harvard-IV-4 Psycho Social Dictionary to identify other 12 synonyms. Loughran and McDonald (2011) have noted that the Harvard-IV-4 PsychoSocial Dictionary2 is a commonly used source of word classification, in part because its composition is beyond the control of the researcher and the possible impact of researcher subjectivity is significantly diminished. For example, words such as “capabilities, collective, cooperation, etc.” are associated with the word “collaborate,” and a relatively high frequency of their use in corporate documents suggests that the company has a collaboration-oriented culture. Words such as “achievement, performance, excellence, etc.” are associated with the word “compete,” and a relatively high frequency of their use in corporate documents suggests that the company has a competition-oriented culture. Words such as “boss, efficiency, caution, etc.” are considered synonyms for “control,” and a relatively high frequency of their use in corporate documents suggests that the company has a control-oriented culture. Words such as “dream, begin, elaborate, etc.” are associated with the word “create,” and a relatively high frequency of their use in corporate documents suggests that the company has a creation-oriented culture. We estimated the four corporate culture dimensions for each listed US firm between 1994 and 2011 by determining the number of times that our synonyms occur in each annual report, using percentages to measure cultural emphasis. For example, if the estimate for a “competition-oriented” dimension is equal to 5, the synonyms used to capture this culture dimension (reported in Table 1) represent 5% of the entire document. One possible difficulty of our approach is that listed companies may tend to write official documents to “cater” for investors’ expectations and, consequently, most Although Loughran and McDonald (2011) show that the Harvard IV Psycho-Social Dictionary misclassifies words that are not likely to be correlated with the variables under consideration (e.g., taxes and liabilities), we use this dictionary because the list of synonyms used to capture Cameron et al.’s (2006) four dimensions of corporate culture are directly correlated with the variables and do not suffer from the problems shown by Loughran and McDonald (2011). 2 13 official documents exhibit significant similarity. This phenomenon may prevent us from detecting any cultural differences in the cross-section. Nonetheless, we document in Table 3 that there is significant cross-section heterogeneity among companies along the four corporate culture dimensions proposed by Cameron et al. (2006). <<< INSERT TABLE 3 >>> 4.3 Measuring CEO turnover and firm performance Starting from the information contained in the Execucomp database, we define three different turnover variables. First, we define the CEOT variable, a dummy variable taking the value of 1 if the company has changed its CEO with respect to the previous year and 0 otherwise. When we have more than one top manager in charge of the company in the same financial year, we choose the person who has been the CEO for the longest period (e.g., if there is a change in 2008, but the entrant CEO is in charge only from October to December, i.e., only for 3 months, we consider the 2008 CEO to be the predecessor and register a CEO change in 2009). To compute the other two measures of CEO change, we consider the difference between the case of an internal and an external turnover, a distinction that has proven to be fundamental in several previous papers (e.g., Karaevli, 2007; Zhang and Rajagopalan, 2010). The selection of an outsider successor may be the signal of a stronger desire for change, implying that the link between poor performance and succession should be stronger than in the case of an internal turnover (e.g., Datta and Guthrie, 1994). We construct the variables linked to an external turnover (External CEO Turnover 1 and External CEO Turnover 2), based on two different definitions of an internal successor as follows: a) someone who has been among the top managers at 14 least for two years before the nomination or b) someone who joined the company at least 2 years before the nomination. Furthermore, we do not consider turnovers in which the predecessor was older than 63, as they most likely reflect a retirement situation and not shareholders’ decision to change the company’s top management. Regarding the measurement of firm performance, we use accounting-based performance measures because, as noted by Jenter and Kanaan (2012), these are shortterm profit measures that are better predictors of CEO turnover than stock market returns if the market incorporates future benefits after replacing an underperforming CEO. First, we focus on the Return on Assets (ROA), obtained by the ratio between the Earnings Before Interest, Tax and Depreciation and Amortization (EBITDA) on total assets. We focus on EBITDA, rather than net income, as EBITDA is a good means of comparing companies within and across industries. In our sample, we have companies in different industries, and the amount of fixed effect (which is subject to depreciation charges) may be subject to distorted accounting and financing effects on company earnings, which is not the case in EBIDTA. We also focus on Return on Sales (ROS) as robustness check. ROA may be decomposed into two parts: ROS (i.e., ROS=EBITDA/total sales) and asset turnover (i.e., total sales /total assets). ROS is a useful indicator of the firm’s profitability and operational efficiency in particular, as it provides an insight into how much profit is being produced per dollar of sales. As a robustness check, we also estimate the industry-adjusted ROA (IAROA), which is the difference between ROA and the average ROA for companies in the same industry group for every financial year (the industry group is identified on the basis of the SPINDEX code). 5. Results 15 The test of the credibility of the CEO change threat essentially aims to verify whether the poor performance of a company is related to a future change in the CEO. We test this assumption using both a univariate and a multivariate approach. In the univariate approach, we measure CEO turnover frequencies by past performance quintiles, defined in terms of both ROA and IAROA. We distinguish internal and external turnovers, analyzing both the case of one-year and two-year time intervals. We also drop observations for which the predecessor was older than 63, assuming that in this case, the change is due to retirement and not to the decision to remove a bad performer. As shown in Table 5, there is a very strong relationship between past poor performance and subsequent CEO turnover. The turnover frequency in the lowest quintile (worst performers) is always greater than in the highest quintile (best performers), for both t-1 and t-2. When we disentangle turnovers in internal and external, we observe that the inverse relationship between past performance and change decisions is mostly explained by external turnover. The Q1-Q5 difference is statistically significant (at the confidence level) for both ROA and IAROA with reference to the one-year lag, whereas for the two-year interval, it is statistically significant (at the 5% confidence level) only when we consider the unadjusted performance. The results are the same irrespective of the chosen definition of external and internal successors and are consistent with the idea that when a company uses CEO turnover to remove a bad performer, it will most likely choose someone who is not linked to the previous CEO. <<< INSERT TABLE 5 >>> In the multivariate approach, we borrow the empirical strategy used by Jenter 16 and Kanaan (2012) to test the strong-form relative performance evaluation hypothesis (i.e., industry and market performance have no predictive power for the likelihood of forced CEO turnovers: see also Bertrand and Mullainathan (2001), Wolfers (2002), Garvey and Milbourn (2006), and Jenter and Kanaan (2012)). Following Jenter and Kanaan (2012), we use a two-stage regression approach to assess the sensitivity of CEO turnover to peer performance. In the first stage, we decompose firm performance into a systematic component caused by peer group performance and a firm-specific component that should, in part, reflect CEO ability. We use accounting-based performance measures (specifically, the ROA and ROS) because these are short-term profit measures that are better predictors of CEO turnover than stock market returns if the market incorporates future benefits for replacing an underperforming CEO. In the second stage, we predict the probability of CEO turnover using both the estimated peer group component and the estimated residual component of firm performance. As noted by Jenter and Kanaan (2012), this approach is an effective instrumental variable estimation, with peer group performance serving as an instrument for firm performance3: (First stage) ππ,π‘−1 = π½0 + π½1 πππππππππ’π,π‘−1 + π£π,π‘−1 (1) (Second stage) ππ(πΆπΈπ πβπππππ,π‘ ) = πΎ0 + πΎ1 πΜπ,π‘−1 + πΎ2 π£Μπ,π‘−1 + πΎ3 πΆππΏπ,π‘−1 + πΎ4 πΆππΏπ,π‘−1 ∗ π£Μπ,π‘−1 + πΎ5 πΆπππ,π‘−1 + πΎ6 πΆπππ,π‘−1 ∗ π£Μπ,π‘−1 + πΎ7 πΆπππ,π‘−1 + πΎ8 πΆπππ,π‘−1 ∗ π£Μπ,π‘−1 + πΎ9 πΆπ πΈπ,π‘−1 + πΎ10 πΆπ πΈπ,π‘−1 ∗ π£Μπ,π‘−1 + ππ,π‘−1 (2) where πΜπ,π‘−1 = π½Μ0 + π½Μ1 πππππππππ’π,π‘−1, πΜπ,π‘−1 is the estimated exogenous component of The tests treat peer performance as a plausibly exogenous instrument for the “luck” that has aided or hampered the CEO’s running of the firm. For peer group performance to be a valid instrument, it is necessary that (i) peer performance be exogenous and (ii) peer performance itself not have an effect on CEO dismissals. 3 17 firm performance common to the peer group and not attributable to CEO actions, and π£Μπ,π‘−1 is the estimated firm-specific performance component. Following previous studies (e.g., Jenter and Kanaan, 2012), we estimate a common peer performance beta for all firms in the first-stage regression. Table 4 summarizes all variables used in the empirical investigation. <<< INSERT TABLE 4 >>> Compared to previous studies (e.g., Jenter and Kanaan, 2012), we added our four measures of corporate culture (COL, COM, CON, and CRE) and their interaction with the estimated firm-specific performance component, as we posit that the corporate culture plays a moderating role in the relationship between firm performance and CEO turnover. Regarding the prediction of estimated coefficients, we expect that 1) πΎ1 = 0. We expect that the exogenous performance component will not affect CEO turnover, consistently with the prediction of strong-form relative performance evaluation; that is, 2) πΎ2 < 0. To be a credible corporate governance instruments, the probability of CEO turnover must be inversely related to firm-specific performance, so we expect to find a negative coefficient; that is, 3) πΎ3,5,7,9 ≠ 0. Consistently with our research hypotheses (Section 1), we expect that the probability of a CEO change is positively related to control- and competition-oriented corporate cultures (respectively, πΎ5 > 0, πΎ7 > 0) and negatively related to collaboration- and creation-oriented cultures (respectively, πΎ3 < 0, and πΎ9 < 0); and 4) πΎ4,6,8,10 ≠ 0. We expect corporate culture to be a mediator of the relationship between CEO turnover and firm-specific performance: specifically, we posit that companies with different cultures react differently to poor firm-specific performance and thus that the probability of observing a CEO turnover depends on the interaction of firm-specific performance and corporate culture. Consistently with our 18 research hypothesis (Section 1), we expect that the probability of a CEO change will be negatively related to the interactions between firm-specific performance and the control- and competition-oriented corporate cultures (respectively, πΎ6 < 0, πΎ8 < 0) and positively related to the interactions between the firm-specific performance and the collaboration- and creation-oriented cultures (respectively, πΎ4 > 0, and πΎ10 > 0). As outlined by Powers (2005), the interpretation of the interaction term coefficients in logit and probit models requires great care. Specifically, in logit and probit models of the type turnover=f(performance, firm type, firm type*performance, controls), Power (2005) shows that the significance of the interaction term coefficient may be misleading. One possibility is that the true difference in the relationship could be stronger than is indicated by the estimated coefficient, potentially generating Type 1 errors. Alternatively, the true difference in the relationship could be weaker than is indicated by the estimated coefficient, potentially generating Type 2 errors. To overcome this problem, we use the methodology developed by Norton et al. (2004) to compute correct marginal effects and their standard errors (as in Lel and Miller, 2010; Bushman et al., 2010; Loureiro, 2010). The two-stage estimation is repeated for the probability of a CEO change and the probability of an external turnover (with two possible definitions of an internal successor as follows: a) someone who has been among top managers for at least two years before the nomination or b) someone who joined the company at least 2 years before the nomination). In addition to this, the company performance is analyzed with respect to both ROA and ROS. As a robustness check, we also conduct a set of simple logit models in which the past performance of the company is not disentangled from specific and systematic components but measured in terms of industry-adjusted ROA. The other regressors are our four measures of corporate culture (COL, COM, CON, and 19 CRE) and their interaction with the firm performance, as we posit that corporate culture plays a moderating role in the relationship between firm performance and CEO turnover. Our predictions for the signs of the estimated coefficients are absolutely similar to those elaborated for the two-stage model. <<< INSERT TABLE 6 >>> First, we examine the results from the two-stage multivariate analysis of the relationship between past ROA and CEO change for both one-year and two-year lags (see Table 6). Consistent with prior papers, we find a strong negative relationship (statistically significant at the 1% confidence level) between the firm-specific ROA and the probability of a CEO turnover in all models without the introduction of cultural variables (1, 4, and 7) and in both time horizons considered (panel A, and B). Regarding the exogenous component of the ROA, the estimated coefficient is always positive but statistically significant (at least at the 5% confidence level) only for the two-year period (Table 6, panel B). This finding should represent a signal that a management change is more probable when the rest of the industry has been doing well, making bad performance more evident and demanding an intervention. The most interesting results are that we show that different corporate cultures have a different impact on the probability of changing a CEO. With respect to our four cultural dimensions, the results are highly consistent with our predictions. The sign of this coefficient is negative for the collaboration-oriented culture (consistently with H1), although not highly statistically significant, for both t-1 and t-2. The estimated coefficients for the competition-oriented corporate cultures are always positive and also statistically significant (at the 1% confidence level). As predicted by H3, we find 20 statistically significant evidence that the probability that shareholders fire a CEO increases as it increases the culture orientation toward competition. Contrary to our expectations (H7), the sign for the creation-oriented culture is positive, even if statistically significant only for the model predicting CEO change from firm performance in t-1. Estimated coefficients for the control-oriented culture are never statistically significant at the 10% level or less, so we do not find support for H5. Regarding the interactions measuring the moderating role of corporate culture, we find that the product of control-oriented culture and firm-specific performance always has a negative coefficient (significant at the 1% confidence level), as predicted by H6. This finding suggests that when firm-specific performance is positive, the probability of a CEO turnover is decreased by a control-oriented culture, whereas when firm-specific performance is negative, the probability of a CEO turnover is increased. Therefore, a control-oriented culture amplifies the magnitude of the inverse relationship between past performance and CEO turnover. We also find that the interaction with the creation-oriented culture always has a positive coefficient. However, this finding does not provide a statistically significant indication that this type of culture reduces the size of the inverse relationship between past performance and CEO turnover, as predicted by H8. The correct marginal effects calculated following Norton et al. (2004) show that the effect is generally positive (as predicted by H8), but we do not observe this positive link to be statistically significant at the 10% or less. As for the other cultural dimensions, collaborate and compete, the interactions are never statistically significant, rejecting H2 and H4. The results from the logit models predicting CEO change or external turnover, for both t-1 and t-2 (reported in the Table 6, in panel A and B, respectively) are always consistent with one another. 21 To test the robustness of our results, we perform two additional analyses. First, we repeat the analysis measuring firm performance focusing on ROS (rather than ROA). We find very similar results regarding the role of culture (Table 7). The only relevant difference is the sign of the coefficient for the exogenous component of the firm performance, which is negative and strongly significant. <<< INSERT TABLE 7 >>> Second, we run a one-stage logit model in which we consider the industryadjusted ROA without distinguishing between specific and exogenous components. As shown in Table 8, there is evidence of a strong inverse relationship between past performance and CEO change, confirming that turnover is a credible corporate governance instrument. The results also corroborate the hypothesis that corporate culture influences turnover decisions, signaling a positive relationship between a competition-oriented culture and CEO change. However, this simpler model does not allow us to highlight the moderating role of culture with interaction terms that are never statistically significant. <<< INSERT TABLE 8 >>> 7. Conclusions and discussion CEO change must be credible to be considered a valuable corporate governance instrument. Although there is convergent evidence that CEO change is credible, there are no studies assessing the reason for this link. In this paper, we show that corporate culture plays a moderating role in the relationships between CEO turnover and 22 performance pre-CEO change. Our paper is the first to provide empirical evidence using a large dataset (US listed companies from 1994 to 2010) that corporate culture influences the decision to fire the CEO. Specifically, we find a strong negative relationship between the firm-specific ROA (i.e., the ROA component not caused by peer group performance) and the probability of a CEO turnover. We also find that corporate culture is related to CEO change: the collaboration-oriented culture has a negative connection with the probability of a CEO change, whereas the other three dimensions of corporate culture (i.e., creation-, control- and competition-oriented corporate cultures) display a positive link with the probability of a CEO change. Overall, our results support that different cultures have a different impact on the probability of a CEO change. Once we consider what happens to the probability of a CEO change in the case of poor performance among companies with different cultures, we find strong, statistically significant evidence that the interaction between control-oriented culture and firm-specific performance has a negative link with the probability of a CEO change. CEO turnover is more credible in companies with a high control-oriented culture (i.e., when the firmspecific performance is negative, the probability of CEO turnover increases). We thus show that the control-oriented culture amplifies the magnitude of the inverse relationship between past performance and CEO turnover. We do not find statistically significant evidence that the other cultural dimensions have a moderating role. These results are consistent with different definitions of CEO change, different time periods (1 and 2 years before the CEO change) and performance indicators (ROA and ROS). Our results are robust to various specifications and obtained by focusing on firm performance that are related to the CEO responsibility (rather than sectors). Specifically, we disentangled firm performance by splitting it into two components 23 (systematic and firm specific), as in Jenter and Kanaan, (2012). We acknowledge some limitations of our analysis that suggest some interesting directions for future research. First, we have very limited information on both the predecessor and the new CEO, e.g., data on their origin, education, and previous experience. This type of data might help to better explain the moderating role of corporate culture. In addition to this limitation, we believe that cultural variables are likely crucial not only in determining the probability of dismissal for bad results but also in influencing opinions on the current CEO, irrespective of performance and in directing the choice for a successor to the candidate that best fits the corporate orientation. Further investigations should be devoted to enriching the database and exploring these issues. Second, it would be interesting to study the moderating role of culture not only on the relationship between CEO turnover and past firm performance (as in our paper) but also on the performance effects of succession to assess both CEO credibility and CEO effectiveness. Third, it is probable that a cultural approach is able to offer interesting results if applied to top managers other than the CEO, as the upper echelons perspective suggests. In conclusion, this study represents a first step toward opening a new perspective in the strategic literature, contributing to the studies on both the antecedents and the consequences of top management change. 24 References Antweiler W., Murray Z.F. (2004). Is all that talk just noise? The information content of internet stock message boards. The Journal of Finance, 59, pp. 1259-1293. Bernhardt D., Hughson E., Kutsoati E. (2006). The evolution of managerial expertise: How corporate culture can run amok. American Economic Review, 96, pp. 195-221. Bertrand, M., Mullainathan S. (2001). Are CEOs rewarded for luck? The ones without principles are. The Quarterly Journal of Economics, 116, pp. 901-32. Bushman R., Dai Z., Wang X. (2010). Risk and CEO turnover. Journal of Financial Economics, 96, pp. 381-398. Cameron K.S., De Graff J., Quinn R.E., Thakor A. (2006). Competing values leadership: Creating value in organisations. Cheltenham: Edward Elgar; 2006. Carretta A., Farina V., Fiordelisi F., Martelli D., Schwizer P. (2011). The impact of corporate governance press news on stock market returns. European Financial Management, 17, pp. 100-119. Datta D.K., Guthrie J.P. (1994). Executive succession: Organizational antecedents of CEO characteristics. Strategic Management Journal, 15, pp. 569-577. Deal T., Kennedy A.A. (1982). Corporate culture: Rites and rituals of organizational life. Reading, MA: Addison- Wesley. Easterwood J.C., Δ°nce Ö. Ε., Raheja C.B. (2012). The evolution of boards and CEOs following performance declines. Journal of Corporate Finance, 18, pp. 727-744. Fang H. (2001). Social culture and economic performance. American Economic Review, 91, pp. 924-979. Garvey, G.T., Milbourn T.T. (2006). Asymmetric benchmarking in compensation: Executives are rewarded for good luck but not penalized for bad. Journal of Financial Economics, 82, pp. 197-226. Guiso L., Sapienza P., Zingales L. (2006). Does culture affect economic outcomes? Journal of Economic Perspectives, 20, pp. 23-48. Guiso L., Sapienza P., Zingales L. (2009). Cultural biases in economic exchange. The Quarterly Journal of Economics, 124, pp. 1095-131. Hoberg, G., Phillips, G. (2010). Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis. Review of Financial Studies, 23, pp. 37733811. Hoberg, G., Hanley, K. (2010). The Information Content of IPO Prospectuses, Review of Financial Studies, 23, 2821-64. 25 Jenter D., Kanaan F. (2013). CEO turnover and relative performance evaluation. Journal of Finance (Forthcoming) Karaevli A. (2007). Performance consequences of new CEO ‘outsiderness’: Moderating effects of pre and post-succession contexts. Strategic Management Journal, 28, pp. 681-706. Lel U, Miller D.P. (2008). International cross-listing, firm performance, and top management turnover: A test of the bonding hypothesis. The Journal of Finance, 63, pp. 1897-1937. Levinson S. (2003). Language and mind: Let's get the issues straight!. In Gentner D., Goldin-Meadow S., (Eds), Language in mind, MIT Press, Cambridge, U.K., pp. 25-45. Loughran T., McDonald B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. The Journal of Finance, 66, pp. 35-65. Loureiro G. (2010). The reputation of underwriters: A test of the bonding hypothesis. Journal of Corporate Finance, 16, pp. 516-532. Norton, E., Wang H, Ai C. (2004) Computing interaction effects and standard errors in logit and probit models. Stata Journal, 4, pp. 103-116. Peters T. J., Waterman, R.H.Jr. (1982). In search of excellence. New York: Harper & Row. Pitcher P., Chreim S., Kisfalvi V. (2000). CEO succession research: Methodological bridges over troubled waters. Strategic Management Journal, 21, pp. 625-648. Powers, E. (2005). Interpreting logit regressions with interaction terms: An application to the management turnover literature. Journal of Corporate Finance, 11, pp. 504-522. Sackmann A.S. (2010). Culture and performance. In (Eds.) Ashkanasy M.N., Peterson M.F., Wilderom C.P.M. The Handbook of Organizational Culture and Climate, SAGE publications, pp. 188-224. Schein, E.H. (1992). Organizational culture and leadership, 2d ed. San Francisco: Jossey-Bass. Stone P.J., Dunphy D.C., Smith M.S., Ogilvie D.M. (1966). The General Inquirer: a computer approach to content analysis. MIT studies in comparative politics, MIT Press, Cambridge, U.S. Tabellini, G. 2008. The scope of cooperation: Values and incentives. The Quarterly Journal of Economics, 123, pp. 905-950. Tetlock P.C. (2007). Giving content to investor sentiment: The role of media in the stock market. The Journal of Finance, 62, pp. 1139-1168. Tetlock P.C., Saar-Tsechansky M., Mackassy S. (2008). More than words: Quantifying language to measure firms’ fundamentals. The Journal of Finance, 63, pp. 1437-1467. 26 Warner J.B., Watts R.L., Wruck K.H. (1988). Stock prices and top management changes. Journal of Financial Economics, 20, pp. 461-492. Wiersema M.F., Zhang Y. (2011). CEO dismissal: The role of investment analysts. Strategic Management Journal, 32, pp. 1161-1182. Wilkins A.L., Ouchi W.G. (1983). Efficient cultures: Exploring the relationship between culture and organizational performance. Administrative Science Quarterly, 28, pp. 468-481 Wolfers J. (2002). Are voters rational? Evidence from gubernatorial elections, Working Paper, Graduate School of Business, Stanford University, March. Zajac A.J. (1990) CEO selection, succession, compensation and firm performance: A theoretical integration and empirical analysis. Strategic Managament Journal, 11, pp. 217-230. Zhang Y., Rajagopalan N. (2010). Once an outsider, always an outsider? CEO origin, strategic change, and firm performance. Strategic Management Journal, 31, pp. 334346. 27 Table 1: The corporate culture dimensions investigated This table illustrates Cameron et al.’s (2006) corporate culture dimensions investigated in our study. Panel A defines the four culture dimensions, and Panel B reports the bag of words used in the text analysis to capture each culture dimension. The word banks were obtained in two steps: first, we selected synonyms suggested by Cameron et al. (2006) to identify each culture dimension. Second, all words selected in the first step were cross-referenced with the Harvard-IV-4 Psycho Social Dictionary to identify other synonyms. Words ending with an “*” mean that we accept all suffixes for them. In this way, we are able to count as many words as possible with close meaning without reporting all of them. Panel A) Culture dimensions definition Control (CON) Control-oriented cultures pursue improvements in efficiency by implementing better processes (e.g., by the extensive use of processes, systems and technology) and quality enhancements (such as statistical process control and other quality control processes). These banks usually make great use of standardized procedures and emphasize rule-reinforcement and uniformity. Compete (COM) Competition-oriented cultures are aggressive and forceful in the pursuit of competitiveness by focusing on organizational effectiveness, fast response, and customer focus. These banks usually set customers as the highest priority and are defined as the ultimate objective of being in business (e.g., success is judged on the basis of indicators such as market share, revenues, meeting budget targets, and profitability growth). Collaborate (COL) Collaboration-oriented cultures address building human competencies, developing people, and solidifying an organizational culture by following a consensual and cooperative process rule to build cohesion through consensus and satisfaction through involvement (e.g., clarifying and reinforcing organizational values, norms, and expectations; developing employees and cross-functional work groups; implementing programs to enhance employee retention; and fostering teamwork and decentralized decision making). These banks usually succeed because they hire, develop, and retain their human resource base. Create (CRE) Creation-oriented cultures engage innovation in the products and services the organization produces by effectively focusing on entrepreneurship, vision and constant change (e.g., allowing for freedom of thought and action among employees so that rule breaking and reaching beyond barriers are common characteristics of the organization’s culture). These banks usually aim to include innovative product-line extensions, radical new process breakthroughs, innovations in distribution and logistics that redefine entire industries and developing new technologies. Source: Adapted from Cameron et al. (2006) Panel B) Bag of words Control (CON) capab*, collectiv*, commit*, competenc*, conflict*, consens*, control*, coordin*, cultur*, decentr*, employ*, empower*, engag*, expectat*, facilitator*, hir*, interpers*, involv*, life*, long-term*, loyal*, mentor*, monit*, mutual*, norm*, parent*, partic*, procedur*, productiv*, retain*, reten*, skill*, social*,tension*, value* Compete (COM) achiev*, acqui*, aggress*, agreem*, attack*, budget*, challeng*, charg*, client*, compet*, customer*, deliver*, direct*, driv*, excellen*, expand*, fast*, goal*, growth*, hard*, invest*, market*, mov*, outsourc*, performanc*, position*, pressur*, profit*, rapid*, reputation, result*, revenue*, satisf*, scan*, succes* signal*, speed*, strong, superior, target*, win* Collaborate (COL) boss*, burocr*, cautio*, cohes*, certain*, chief*, collab*, conservat*, cooperat*, detail*, document*, efficien*, error*, fail*, help*, human*, inform*, logic*, method*, outcom*, partner*, people*, predictab*, relation*, qualit*, regular*, solv*, share*, standard*, team*, teamwork*, train*, uniform*, work group* Create (CRE) adapt*, begin*, chang*, creat*, discontin*, dream*, elabor*, entrepre*, envis*, experim*, fantas*, freedom*, futur*, idea*, init*, innovat*, intellec*, learn*, new*, origin*, pioneer*, predict*, radic*, risk*, start*, thought*, trend*, unafra*, ventur*, vision* 28 Table 2: Sample descriptive statistics This table reports the main characteristics of the sample selected. Our sample included all listed companies in the US between 1994 and 2011 for which both financial and culture information was available. Data have been obtained from Compustat and Execucomp. We consider as financial all firms operating in the following industry groups: Asset Management and Custody banks; Consumer finance; Diversified banks; Diversified REITs; Industrial REITs; Insurance brokers; Investment banking and brokerage; Life and health insurance; Mortgage REITs; Multiline insurance; Office REITs; Other diversified financial services; Property and casualty insurance; Real investment trust; Regional banks; Reinsurance; Residential REITs; Retail REITs; Specialized finance; Specialized REITs; Thrifts and mortgage finance. Companies without an assigned industry group were excluded from the sample. Overall Year Financial Non-financial No. obs No. CEO changes No. obs No. CEO changes No. obs No. CEO changes 1994 772 69 98 6 674 63 1995 695 63 78 4 617 59 1996 767 78 98 7 669 71 1997 963 84 112 7 851 77 1998 1003 107 130 10 873 97 1999 945 95 128 5 817 90 2000 930 129 125 13 805 116 2001 915 126 126 20 789 106 2002 1080 99 150 10 930 89 2003 1511 166 234 27 1277 139 2004 1517 138 232 15 1285 123 2005 1310 151 220 18 1090 133 2006 1579 149 271 19 1308 130 2007 1528 146 283 28 1245 118 2008 1604 160 275 25 1329 135 2009 1574 146 271 28 1303 118 2010 586 52 62 7 524 45 2011 174 14 6 2 168 12 Total 19453 1972 2899 251 16554 1721 Source of data: Compustat and Execucomp 29 Table 3: Corporate culture dimension estimates Panel A reports the descriptive statics for the four culture dimensions proposed by Cameron et al. (2006) investigated in our study. Data have been obtained from 10-K reports in the SEC’s Edgar database. Industry Overall Time Financial Non-Financial Collaborate Mean 1.386 1.620 1.346 1.315 1.264 1.332 1.504 1.461 (COL) St. Deviation 0.440 0.509 0.413 0.629 0.427 0.395 0.495 0.349 0 0 0 0.43 0 0 0 0.83 Min 1994 1998 2002 2006 2011 Max 8.60 5.36 8.60 4.580 4.290 3.220 4.320 4.360 Compete Mean 4.844 4.521 4.901 4.887 4.904 4.914 5.089 5.076 (COM) St. Deviation 1.184 1.103 1.189 1.989 1.045 1.133 1.700 0.885 Min 0 0 0 1.22 0 0 0 3.44 Max 25.72 18.04 25.72 19.82 9.57 8.85 15.80 9.76 Control Mean 1.410 1.342 1.422 1.460 1.328 1.324 1.589 1.451 (CON) St. Deviation 0.509 0.363 0.530 0.629 0.329 0.926 0.524 0.303 Min 0 0 0 0.58 0 0 0 0.77 Max 33.33 5.76 33.33 6.500 3.740 28.570 5.620 3.100 Create Mean 1.081 1.262 1.049 0.877 0.932 1.082 1.266 1.214 (CRE) St. Deviation 0.387 0.381 0.379 0.429 0.343 0.372 0.416 0.267 Min 0 0 0 0 0.16 0 0 0.69 Max 20.00 5.16 20.00 3.03 3.71 7.14 5.48 2.3 30 Table 4: Variables description This table defines the variables used in the paper. Data have been obtained by Compustat. Variable Symbol Definition and calculation method CEO Turnover CEOT CEOT is a dummy variable taking the value 1 if the company has changed its CEO with respect to the previous year and 0 otherwise External CEO Turnover 1 ECT1 EXT1 is a dummy variable taking the value 1 if the company has changed its CEO with respect to the previous year with an external successor (i.e., someone who has not been among the top managers at least for two years before the nomination) and 0 otherwise External CEO Turnover 2 ECT2 EXT2 is a dummy variable taking the value 1 if the company has changed its CEO with respect to the previous year with an external successor (i.e., someone who did not join the company at least 2 years before the nomination) and 0 otherwise Control-Oriented Culture CONi,t CONi,t is the control-oriented corporate culture estimate of the company i at the time t obtained using text analysis. Competition-Oriented Culture COMi,t COMi,t is the competition-oriented corporate culture estimate of the bank i at the time t obtained using text analysis. Collaboration-Oriented Culture COLi,t) COLi,t is the collaboration-oriented corporate culture estimate of the bank i at the time t obtained using text analysis. Creation-Oriented Culture CREi,t CREi,t is the creation-oriented corporate culture estimate of the bank i at the time t obtained using text analysis. Return on Assets ROA ROA is obtained by the ratio between Earnings Before Interest, Tax and Depreciation and Amortization (EBITDA) and Total Assets. Industry-adjusted ROA IAROA IAROA is the difference between ROA and the average ROA for companies in the same industry group for every financial year (the industry group is identified on the basis of the SPINDEX code). ROA firm-specific performance component π£Μπ,π‘−1 π£Μπ,π‘−1 is the estimated firm-specific performance component. We follow Jenter and Kanaan (2012) to disentangle the peer group component and the estimated residual component of ROA. ROA exogenous component πΜπ,π‘−1 πΜπ,π‘−1 is the estimated exogenous component of firm performance common to the peer group and not attributable to CEO actions. We follow Jenter and Kanaan (2012) to disentangle the peer group component and the estimated residual component of ROA. Return on Sales ROS ROS is obtained by the ratio between Earnings Before Interest, Tax and Depreciation and Amortization (EBITDA) and Total Sales. 31 Table 5: CEO change credibility: the univariate analysis This table presents CEO turnover frequency by performance quintile in US companies over the period 1994-2011. Performance is measured in terms of ROA and industry-adjusted ROA (i.e., for every financial year, the difference between the single-company ROA and the average ROA for companies in the same industry group, considering the SPINDEX code in the Execucomp database). Quintiles are calculated with respect to two different periods: one year and two years before the turnover. Companies without an assigned industry group are not considered. Internal and external turnover are classified on the basis of two different definitions of an insider successor: 1) someone who has been among top managers for at least two years before the nomination (Panel A) or 2) someone who joined the company at least 2 years before the nomination (Panel B). CEO turnovers in which the predecessor was older than 63 were not considered. ***,**,* indicate statistical significance at the 1%, 5% and 10% levels, respectively, for the t-test of differences in turnover likelihood between the top and bottom performance quintiles. Data have been obtained from Compustat and Execucomp Panel A Q1 Q2 Q3 Q4 Performance quintiles based on ROA in t-1 P (Ceo Turnover in t) 9.9% 8.8% 7.4% 7.4% P (External Turnover in t) 5.4% 4.0% 3.2% 2.6% P (Internal Turnover in t) 4.6% 4.8% 4.2% 4.8% Performance quintiles based on industry-adjusted ROA in t-1 P (Ceo Turnover in t) 9.1% 8.0% 7.1% 6.9% P (External Turnover in t) 4.5% 3.3% 2.8% 2.9% P (Internal Turnover in t) 4.5% 4.7% 4.3% 4.1% Performance quintiles based on ROA in t-2 P (Ceo Turnover in t) 9.4% 9.1% 8.3% 8.1% P (External Turnover in t) 4.7% 4.1% 3.5% 2.7% P (Internal Turnover in t) 4.8% 5.0% 4.8% 5.4% Performance quintiles based on industry-adjusted ROA in t-2 P (Ceo Turnover in t) P (External Turnover in t) P (Internal Turnover in t) 8.7% 3.9% 4.8% 8.6% 4.0% 4.6% 8.6% 3.2% 5.4% Q5 Q1 vs Q5 6.7% 2.5% 4.2% *** *** 7.5% 2.9% 4.6% 8.2% 3.3% 4.9% 8.1% 2.9% 5.2% 8.5% 3.5% 5.0% Q1 Q2 Q3 Q4 Performance quintiles based on ROA in t-1 Q5 *** ** Panel B Q1 vs Q5 P (Ceo Turnover in t) 9.9% 8.8% 7.4% 7.4% P (External Turnover in t) 5.2% 4.1% 3.2% 2.6% P (Internal Turnover in t) 4.7% 4.8% 4.2% 4.8% Performance quintiles based on industry-adjusted ROA in t-1 6.7% 2.4% 4.3% P (Ceo Turnover in t) P (External Turnover in t) P (Internal Turnover in t) 7.5% 2.8% 4.6% *** P (Ceo Turnover in t) 9.4% 9.1% 8.3% 8.1% P (External Turnover in t) 4.6% 4.1% 3.5% 2.6% P (Internal Turnover in t) 4.8% 5.0% 4.8% 5.5% Performance quintiles based on industry-adjusted ROA in t-2 8.2% 3.3% 4.9% ** P (Ceo Turnover in t) P (External Turnover in t) P (Internal Turnover in t) 8.5% 3.5% 5.0% 9.1% 8.0% 7.1% 6.9% 4.5% 3.3% 2.7% 2.8% 4.6% 4.7% 4.4% 4.1% Performance quintiles based on ROA in t-2 8.7% 3.7% 5.0% 8.6% 4.1% 4.5% 8.6% 3.2% 5.4% 8.1% 2.9% 5.2% *** *** 32 Table 6: CEO change credibility: The two-stage multivariate analysis This table reports the results for the two-stage logit regression of CEO turnover on firm performance reported in model 2. The first stage is an OLS regression of individual ROA on contemporaneous industry ROA. Predicted values and errors from the first stage are used to disentangle firm performance in specific and systematic components. The second-stage logit models predict CEO turnover or External Turnover, classified on the basis of two different definitions of an insider successor: a) someone who has been among top managers for at least two years before the nomination or b) someone who joined the company at least 2 years before the nomination. CEO turnovers in which the predecessor was older than 63 are not considered. πΜπ,π‘−1 is the estimated exogenous component of firm performance common to the peer group and not attributable to CEO actions, and π£Μπ,π‘−1 is the estimated firm-specific performance component. COL, COM, CON and CRE are the four corporate culture dimensions proposed by Cameron et al. (2006). Panel A reports the results for the second-stage logit regression of CEO turnover on firm performance in t-1. Panel B reports the results for the second-stage logit regression of CEO turnover on firm performance in t-2. The mean interaction effects and the corresponding level of significance are estimated following Norton et al. (2004). The graphs displaying the interaction effects and the corresponding z-statistics are reported in Appendix A.1, Figure 1. ***, **, * indicate statistical significance at the 1%, 5% and 10% levels, respectively. Source: Authors’ elaboration of Execucomp data. Panel A Ceo Turnover π£Μπ,π‘−1 πΜπ,π‘−1 -0.1807*** (0.0348) 0.0565* (0.0327) COL COM CON CRE π£Μπ,π‘−1 * COL π£Μπ,π‘−1 * COM π£Μπ,π‘−1 * CON π£Μπ,π‘−1 * CRE Constant -2.4947*** (0.0331) -0.1864*** (0.0359) 0.0547 (0.0346) -0.0722* (0.0373) 0.1082*** (0.0343) -0.0033 (0.0403) 0.0694** (0.0351) -0.0065 (0.0372) 0.0025 (0.0294) -0.0927** (0.0469) 0.043 (0.0269) -2.499*** (0.0335) External Turnover a -0.1879*** (0.0362) 0.042 (0.0343) -0.051 (0.0365) 0.0956*** (0.0346) 0.0179 (0.0396) 0.0843** (0.0348) -0.0051 (0.0367) 0.0025 (0.0295) -0.0932** (0.0454) 0.0409* (0.0233) -2.5896*** (0.3273) -0.3601*** (0.0496) 0.0681 (0.0476) -3.4579*** (0.0519) -0.3797*** (0.0495) 0.0475 (0.0514) -0.0937 (0.0586) 0.1878*** (0.0472) 0.0288 (0.057) 0.0454 (0.0526) -0.0244 (0.0525) 0.0415 (0.0412) -0.1264** (0.0642) 0.1026*** (0.0363) -3.4812*** (0.0529) External Turnover b -0.3831*** (0.0498) 0.0447 (0.0523) -0.1013* (0.0607) 0.2051*** (0.0501) 0.0407 (0.0577) 0.0235 (0.0566) -0.0256 (0.0538) 0.0383 (0.045) -0.1335** (0.0656) 0.1041*** (0.0368) -2.9948*** (0.3814) -0.3543*** (0.05) 0.072 (0.0481) -3.4684*** (0.0522) -0.374*** (0.0499) 0.051 (0.0521) -0.089 (0.0588) 0.1945*** (0.0474) 0.0248 (0.0579) 0.0352 (0.0533) -0.0284 (0.0527) 0.043 (0.0419) -0.1224* (0.0651) 0.1011*** (0.0371) -3.4925*** (0.0531) -0.3773*** (0.0503) 0.0493 (0.0532) -0.0962 (0.0608) 0.2127*** (0.0502) 0.0354 (0.0586) 0.0118 (0.0575) -0.0291 (0.054) 0.0401 (0.0456) -0.1296* (0.0666) 0.1027*** (0.0375) -2.9943*** (0.3812) Year dummies No No Yes No No Yes No No Yes Observations 13,186 13,186 13,186 13,186 13,186 13,186 13,186 13,186 13,186 -0.00001 (0.002) -0.0006 (0.002) -0.0043** (0.0021) 0.003 (0.0021) 0.00004 (0.002) -0.0009 (0.0021) -0.0047** (0.0023) 0.0033 (1.4335) Mean interaction effects (on the basis of the model with year dummies) π£Μπ,π‘−1 * COL π£Μπ,π‘−1 * COM π£Μπ,π‘−1 * CON π£Μπ,π‘−1 * CRE 0.0003 (0.0026) -0.001 (0.0027) -0.0066** (0.0027) 0.0023 (0.0027) 0.0002 (0.0026) -0.0009 (0.0026) -0.0069** (0.0028) 0.002 (0.0026) 0.0002 (0.002) -0.0006 (0.002) -0.0046** (0.0021) 0.003 (0.0021) 0.0002 (0.002) -0.0009 (0.0021) -0.005** (0.0024) 0.0033 (0.0023) 33 Panel B Ceo Turnover π£Μπ,π‘−2 πΜπ,π‘−2 -0.1231*** (0.0363) 0.088*** (0.0342) -2.4263*** (0.0343) -0.1212*** (0.037) 0.079** (0.0362) -0.0135 (0.0359) 0.0942*** (0.0348) -0.0517 (0.0406) 0.0101 (0.0367) -0.0139 (0.0368) 0.0016 (0.031) -0.0955** (0.0436) 0.0125 (0.0329) -2.8134*** (0.3877) No 11,489 No 11,489 Yes 11,489 COL COM CON CRE π£Μπ,π‘−2 * COL π£Μπ,π‘−2 * COM π£Μπ,π‘−2 * CON π£Μπ,π‘−2 * CRE Constant Year dummies Observations External Turnover a -0.1248*** (0.0368) 0.0822** (0.036) -0.028 (0.0366) 0.1001*** (0.0347) -0.0457 (0.0406) -0.0156 (0.0383) -0.0168 (0.0377) 0.0005 (0.0319) -0.1009** (0.045) 0.018 (0.0398) -2.4287*** (0.0347) -0.293*** (0.0562) 0.1557*** (0.0498) External Turnover b -3.3756*** (0.0533) -0.2843*** (0.0562) 0.1633*** (0.0526) -0.0028 (0.0522) 0.1395*** (0.0484) -0.0487 (0.0589) 0.0404 (0.0471) 0.0051 (0.0532) -0.0222 (0.0407) -0.0732 (0.0691) 0.0886** (0.0446) -3.382*** (0.0537) -0.2816*** (0.0564) 0.1793*** (0.0534) -0.0093 (0.0534) 0.1542*** (0.0496) -0.0776 (0.0616) 0.0311 (0.0511) 0.0023 (0.0546) -0.0207 (0.0408) -0.0735 (0.07) 0.0904* (0.0472) -3.3881*** (0.5065) -0.2949*** (0.0565) 0.1732*** (0.0507) -3.3834*** (0.0536) No 11,489 No 11,489 Yes 11,489 No 11,489 -0.2867*** (0.0565) 0.1762*** (0.0535) -0.0063 (0.0524) 0.1302*** (0.0487) -0.0471 (0.0589) 0.023 (0.0483) 0.0146 (0.0534) -0.0158 (0.0409) -0.0662 (0.0688) 0.0879* (0.0468) -3.3884*** (0.0539) No 11,489 -0.2834*** (0.0567) 0.1936*** (0.0544) -0.0133 (0.0536) 0.1468*** (0.05) -0.0764 (0.0615) 0.0134 (0.0532) 0.0125 (0.055) -0.0145 (0.0409) -0.066 (0.0697) 0.0905* (0.0503) -3.3809*** (0.5065) Yes 11,489 Mean interaction effects (on the basis of the model with year dummies) π£Μπ,π‘−2 * COL π£Μπ,π‘−2 * COM π£Μπ,π‘−2 * CON π£Μπ,π‘−2 * CRE -0.001 (0.0027) -0.0007 (0.0028) -0.0072** (0.0028) 0.0015 (0.0027) -0.0009 (0.0027) -0.0006 (0.0027) -0.0067** (0.0028) 0.0009 (0.0027) 0.0002 (0.0018) -0.002 (0.0019) -0.0021 (0.0018) 0.0027 (0.0019) 0.0002 (0.0018) -0.002 (0.0019) -0.0019 (0.0019) 0.0029 (1.3928) 0.0006 (0.0018) -0.0017 (0.0019) -0.0019 (0.0018) 0.0029 (0.0019) 0.0005 (0.0018) -0.0016 (0.0019) -0.0019 (0.0019) 0.003 (0.002) 34 Table 7: CEO change credibility: The two-stage analysis focusing on Return on Sales This table reports the results for the two-stage logit regression of CEO turnover on firm performance in t-1 reported in model 2. We decompose the ROA into two components, Return on Sales (ROS=EBITDA/Total Sales) and Asset Turnover (AT=Total Sales/Total Assets), and we report the results for ROS. The first stage is an OLS regression of individual ROS on contemporaneous industry ROS. Predicted values and errors from the first stage are used to disentangle firm performance in specific and systematic components. The second-stage logit models predict CEO turnover or External Turnover, classified on the basis of two different definitions of an insider successor: a) someone who has been among top managers for at least two years before the nomination or b) someone who joined the company at least 2 years before the nomination. CEO turnovers in which the predecessor was older than 63 are not considered. πΜπ,π‘−1 is the estimated exogenous component of firm performance common to the peer group and not attributable to CEO actions, and π£Μπ,π‘−1 is the estimated firm-specific performance component. COL, COM, CON and CRE are the four corporate culture dimensions proposed by Cameron et al. (2006). The results for logit regression of CEO turnover on firm performance in t-2 are available on request from the authors. Mean interaction effects and the corresponding level of significance are estimated following Norton et al. (2004). The graphs displaying the interaction effects and the corresponding z-statistics are reported in Appendix A.1, Figure 2. ***, **, * indicate statistical significance at the 1%, 5% and 10% levels, respectively. Source: Authors’ elaboration of Execucomp data. . Ceo Turnover π£Μπ,π‘−1 πΜπ,π‘−1 -0.1371*** (0.0326) -0.1014*** (0.0331) -2.5016*** (0.0331) -0.1588*** (0.0328) -0.0988*** (0.0336) -0.0832** (0.0374) 0.1037*** (0.035) -0.0209 (0.0403) 0.0703** (0.0339) 0.0118 (0.0319) 0.0008 (0.0264) -0.0903* (0.0493) 0.0367 (0.0232) -2.5096*** (0.0335) -0.1589*** (0.0332) -0.115*** (0.0344) -0.0574 (0.0364) 0.0863** (0.0353) -0.0028 (0.0397) 0.0928*** (0.0335) 0.0118 (0.0312) -0.0004 (0.0269) -0.0887* (0.0484) 0.0402* (0.0213) -2.6262*** (0.327) No No Yes COL COM CON CRE π£Μπ,π‘−1 * COL π£Μπ,π‘−1 * COM π£Μπ,π‘−1 * CON π£Μπ,π‘−1 * CRE Constant Year dummies External Turnover a Observations 13,186 13,186 13,186 Mean interaction effects (on the basis of the model with year dummies) π£Μπ,π‘−1 * COL π£Μπ,π‘−1 * COM π£Μπ,π‘−1 * CON π£Μπ,π‘−1 * CRE 0.0016 (0.0027) -0.0009 (0.0027) -0.0063** (0.0028) 0.002 (0.0028) 0.0014 (0.0026) -0.0008 (0.0027) -0.0063** (0.0028) 0.002 (0.0027) -0.2629*** (0.0609) -0.2087*** (0.049) -3.4629*** (0.0519) -0.3116*** (0.0455) -0.207*** (0.0505) -0.1054* (0.0582) 0.1796*** (0.0474) -0.0132 (0.0595) 0.0525 (0.0535) 0.0105 (0.0487) 0.0404 (0.0342) -0.1702** (0.0678) 0.0489 (0.0397) -3.4904*** (0.0528) External Turnover b -0.317*** (0.0452) -0.2141*** (0.0521) -0.1081* (0.0599) 0.191*** (0.0506) -0.0043 (0.061) 0.0415 (0.0574) 0.0122 (0.0491) 0.0381 (0.0348) -0.1686** (0.0681) 0.046 (0.0412) -3.4904*** (0.0528) -0.2592*** (0.0609) -0.2038*** (0.0491) -3.4734*** (0.0521) -0.3082*** (0.0457) -0.2015*** (0.0508) -0.100* (0.0583) 0.1865*** (0.0478) -0.0165 (0.0603) 0.0427 (0.0546) 0.0093 (0.0486) 0.0399 (0.034) -0.1677** (0.0682) 0.0542 (0.042) -3.5017*** (0.0531) -0.3136*** (0.0455) -0.2076*** (0.0524) -0.1026* (0.06) 0.199*** (0.0508) -0.0089 (0.0617) 0.0297 (0.0588) 0.0108 (0.0491) 0.0378 (0.0345) -0.1664** (0.0685) 0.0513 (0.0433) -3.0591*** (0.383) No No Yes No No Yes 13,186 13,186 13,186 13,186 13,186 13,186 0.0013 (0.002) -0.0002 (0.0021) -0.0055** (0.0022) 0.0011 (0.002) 0.0014 (0.0021) -0.0004 (0.0021) -0.0055** (0.0025) 0.0011 (0.0021) 0.0012 (0.002) -0.0003 (0.002) -0.0053** (0.0022) 0.0014 (0.002) 0.0013 (0.0021) -0.0005 (0.0021) -0.0053** (0.0025) 0.0014 (0.0021) 35 Table 8: CEO change credibility: A robustness check using a one-stage approach This table reports the results for the one-stage logit regression of CEO turnover on firm performance measured as industry-adjusted ROA (IAROA). The logit models predict CEO turnover or External Turnover, classified on the basis of two different definitions of an insider successor: a) someone who has been among top managers for at least two years before the nomination or b) someone who joined the company at least 2 years before the nomination. CEO turnovers in which the predecessor was older than 63 are not considered. πΌπ΄π ππ΄ is the industry-adjusted performance (i.e., for every financial year, the difference between the single company ROA and the average ROA for companies in the same industry group, considering the SPINDEX code in the Execucomp database) . COL, COM, CON and CRE are the four corporate culture dimensions proposed by Cameron (1993). We report the results for logit regression of CEO turnover on firm performance in t-1. The results for logit regression of CEO turnover on firm performance in t-2 are available on request from the authors. Mean interaction effects and the corresponding level of significance are estimated following Norton et al. (2004). The graphs displaying the interaction effects and the corresponding z-statistics are not reported in the Appendix because the interactions are never statistically significant at the 10% confidence level or less. These graphs are available from the authors upon request. ***, **, * indicate statistical significance at the 1%, 5% and 10% levels, respectively. Source: Authors’ elaboration of Execucomp data. Ceo Turnover πΌπ΄π ππ΄π,π‘−1 -0.0682** (0.0315) -2.4864*** (0.0327) -0.0964*** (0.0356) -0.0545 (0.036) 0.1097*** (0.0344) 0.0125 (0.0388) 0.068** (0.0326) -0.0166 (0.0331) 0.0063 (0.0302) -0.0305 (0.0327) 0.0326 (0.022) -2.6032*** (0.3273) No No Yes COL COM CON CRE πΌπ΄π ππ΄π,π‘−1 * COL πΌπ΄π ππ΄π,π‘−1 * COM πΌπ΄π ππ΄π,π‘−1* CON πΌπ΄π ππ΄π,π‘−1* CRE Constant Year dummies Observations External Turnover a -0.0869*** (0.0327) -0.0791** (0.037) 0.1237*** (0.0341) -0.0096 (0.0394) 0.0478 (0.0332) -0.02 (0.0336) 0.0093 (0.0299) -0.0259 (0.0329) 0.0296 (0.0244) -2.4952*** (0.0336) 13,186 13,186 13,186 -0.1602*** (0.0472) -0.1602*** (0.0472) No 13,186 -0.1978*** (0.0469) -0.0881 (0.0567) 0.1986*** (0.0461) 0.0177 (0.0561) 0.0103 (0.0535) 0.0091 (0.0493) 0.0311 (0.0442) -0.0598 (0.047) 0.0756* (0.045) -0.1978*** (0.0469) No External Turnover b -0.2133*** (0.0531) -0.0925 (0.0581) 0.2128*** (0.0492) 0.0347 (0.0566) -0.0021 (0.0569) 0.0048 (0.0509) 0.0329 (0.0472) -0.0674 (0.0487) 0.0782* (0.0449) -0.2133*** (0.0531) Yes 13,186 13,186 0.0008 (0.0017) -0.00012 (0.0025) -0.002 (0.0017) 0.0023 (0.0017) 0.0007 (0.0018) -0.00022 (0.0018) -0.0024 (0.0019) 0.0025 (0.0018) -0.155*** (0.0479) -3.4266*** (0.0501) -0.1939*** (0.0475) -0.0838 (0.0568) 0.2051*** (0.0465) 0.0147 (0.0569) -0.0002 (0.0543) 0.003 (0.0497) 0.0321 (0.0445) -0.0574 (0.0479) 0.0721 (0.0465) -3.4497*** (0.0515) -0.2073*** (0.0537) -0.0879 (0.0583) 0.2201*** (0.0495) 0.0302 (0.0573) -0.0141 (0.058) -0.0015 (0.0512) 0.0339 (0.0474) -0.0641 (0.0494) 0.0743 (0.0465) -3.0083*** (0.3828) No No Yes 13,186 13,186 13,186 Mean interaction effects (on the basis of the model with year dummies) π£Μπ,π‘−1 * COL π£Μπ,π‘−1 * COM π£Μπ,π‘−1 * CON π£Μπ,π‘−1 * CRE -0.001 (0.0026) 0.00002 (0.0025) -0.0018 (0.0025) 0.0019 (0.0025) -0.0009 (0.0026) -0.00018 (0.0025) -0.0022 (0.0026) 0.0019 (0.0025) 0.0006 (0.0016) -0.0001 (0.0017) -0.0019 (0.0017) 0.0023 (0.0017) 0.0005 (0.0018) -0.0002 (0.0018) -0.0022 (0.0018) 0.0024 (0.0018) 36