1 Do CFOs Have Style? An Empirical Investigation of the Effect of Individual CFOs on Accounting Practices* WEILI GE, University of Washington DAWN MATSUMOTO, University of Washington JENNY LI ZHANG, University of British Columbia I believe in being disciplined but aggressive. Chris Liddell, CFO, Microsoft Corp present 1 1. Introduction What factors impact the accounting choices of a firm? Numerous prior studies in accounting have examined this question, focusing on various firm-level (e.g., Klein 2002) and marketlevel characteristics (e.g., Leuz, Nanda, and Wysocki 2003) that impact accounting outcomes (see Fields, Lys, and Vincent 2001). The purpose of this paper is to consider another potential influence on accounting choices: manager-specific factors. In particular, we focus on chief financial officers (CFOs) because the CFO typically oversees the firm s financial reporting process and therefore he she likely has the most direct impact of all the senior managers on the accounting related decisions of the firm, such as choosing accounting methods and making accounting adjustments (Mian 2001; Geiger and North 2006; Gore, Matsunaga, and Yeung 2008). We examine whether accounting choices are influenced by differences in CFOs individual characteristics that arise from numerous factors including their dispositions, personal situations and prior experiences. Certainly the opening quote of the paper would suggest such a possibility. 2 For expositional purposes, we label these differences CFO style and examine whether CFO style impacts accounting choices. Built on the premise of bounded rationality, research in judgment and decision making has long recognized that individual characteristics play a role in decision outcomes (Bonner * Accepted by Steven Salterio. We thank Brad Blaylock, Peter Demerjian, Mei Feng, Jane Kennedy, Kevin Koh, Sarah McVay, Shivaram Rajgopal, Lisa Sedor, Terry Shevlin, Ryan Wilson, the workshop participants at Columbia University, University of Colorado at Boulder, Duke University, University of Oregon, University of Pennsylvania, Shanghai University of Finance and Economics, and University of Washington, and the conference participants at the 2010 Contemporary Accounting Research Conference for their helpful comments. We thank Io-Ieong Chio, Shun Sik Chan, Jared Jennings, Jeffrey Kim, Jared Kuioka, Elizabeth Lijanto, Yan Limarta, Sally Nguyen, Eric Tohni, Catherine Wijaya, and Amanda Winn for their valuable research assistance. We also thank our editor, Steve Salterio, Susan Krische (discussant) and two anonymous referees for their valuable comments. Ge would like to thank the William R. Gregory Faculty Fellowship and Matsumoto the Emmett S. Harrington Professorship at the University of Washington for financial support. Zhang acknowledges the financial support from the Social Sciences and Humanities Research Council of Canada (SSHRC) and the KPMG Research Bureau in Financial Reporting at UBC. 1. Daisuke Wakabayashi, CFO Brings Philosophy of Change to Microsoft, Reuters News, It is interesting to note that under Chris Liddell s tenure at Microsoft, the company changed its policy of not capitalizing any software development costs to one of capitalizing a portion of these costs. In addition, the company considered taking on debt for the first time in the company s history (which it has since done). Both actions are consistent with Liddell having a more aggressive philosophy. Contemporary Accounting Research Vol. 28 No. 4 (Winter 2011) pp Ó CAAA doi: /j x
2 1142 Contemporary Accounting Research 2008). However, the possibility that these individual characteristics might manifest in corporate-level decision outcomes was less widely recognized until the development of upper echelons theory (Hambrick and Mason 1984; Hambrick 2007). Specifically, upper echelons theory suggests that top managers individual characteristics affect how they assess or interpret their situations and therefore impact their decisions. Numerous studies have since documented evidence consistent with this theory evidence of correlations between various manager-specific measures and corporate decisions. Some examples are the relation between chief executive officer (CEO) house size and firm performance (Liu and Yermack 2007), CEO overconfidence and corporate investment (Malmendier and Tate 2005), and superstar CEOs and firm performance (Malmendier and Tate 2009). Moreover, Bertrand and Schoar (2003) demonstrate the impact of individual managers on corporate decisions by documenting a common manager effect across different companies for which the manager works. Taken together, the findings in this line of research lend support to the important role of individual managers in certain corporate decisions and performance. However, the findings in the above-mentioned studies as well as those in concurrent research that document managers effect on voluntary disclosure strategies (Bamber, Jiang, and Wang 2010; Yang 2010) and tax avoidance strategies (Dyreng, Hanlon, and Maydew 2010) do not necessarily imply that a CFO s style will significantly influence a firm s accounting choices. First, compared to other corporate decisions, accounting choices face a different set of constraints, such as the requirements of generally accepted accounting principles (GAAP), external audits, and SEC regulations. Second, in contrast to prior and concurrent research, we focus on CFOs rather than CEOs or the top management team. Because the CFO is the top-level executive most directly involved in accounting choices, focusing on CFOs provides a more direct test of the link between manager style and the corporate-level decisions they influence. However, it is also possible that CEOs actually set the tone from the top, which would dominate CFOs style in accounting choices. 3 Thus, despite prior findings of a CEO effect on other types of firm decisions, it remains an empirical question whether CFO styles manifest themselves in firms accounting decisions and such findings would contribute to the literature on determinants of accounting choices and earnings quality (Fields et al. 2001; Dechow, Ge, and Schrand 2010). To provide evidence of CFOs effect on accounting choices, we investigate a number of accounting choices that we believe are likely subject to CFOs discretion. We categorize our accounting variables as follows: (1) accounting tools that CFOs can choose to achieve financial reporting goals, including discretionary accruals, operating leases, and expected rate of return for pension assets; (2) outcome-based measures that capture either properties of earnings that are likely the result of managerial intervention in the accounting system (e.g., meeting beating analysts expectations and earnings smoothing) or the likelihood of accounting misstatements (i.e., the F-Score based on Dechow, Ge, Larson, and Sloan 2011). We recognize that these two groups are not clearly delineated but we use these rough categories to facilitate our discussion. We expect CFOs to be able to exercise discretion over these various dimensions of accounting decisions; therefore, these dimensions provide perhaps the best opportunity to detect the influence of CFOs individual styles on accounting choices. 3. For example, using a sample of material accounting manipulation firms, Feng, Ge, Luo, and Shevlin (2011) provide evidence suggesting that CFOs succumb to pressure from CEOs and get involved in making accounting decisions that inflate earnings. In addition, Hunton, Hoitash, and Thibodeau (2010) document that perceived tone at the top is positively associated with earnings quality based on survey data of 123 public companies. Hunton et al. (2010) also show that CFO age, tenure, and compensation are not related to perceived tone at the top. However, this evidence does not imply that CFO characteristics do not matter to accounting decisions, because CFOs could influence accounting decisions without affecting tone at the top. In our supplemental analyses, we do control for the impact of CEOs on accounting choices to ensure that our documented CFO effects are incremental to CEO effects.
3 Do CFOs Have Style? 1143 We adopt a methodology similar to Bertrand and Schoar 2003 that allows managers styles to be idiosyncratic. We construct a sample of 359 CFOs who have occupied the CFO position in at least two companies and have worked for each company for at least two years. This sample allows us to disentangle CFO-specific effects from firm-specific and time-specific effects by including firm and time period specific indicator variables along with CFO indicator variables. 4 Our results suggest that CFO-specific factors are a statistically significant determinant of accounting choices. In particular, adding CFO fixed effects to a base model (that includes firm and year fixed effects along with time-varying control variables) increases the adjusted R 2 s by 2 percent on average. Although these increases may appear small, they are similar in magnitude to the increases reported in Bertrand and Schoar (2003) from including CEO fixed effects in models explaining firms investment and financing policies. We also document that the effect we attribute to CFOs is not a demonstration of CEO style, and changes in accounting choices of the CFO s subsequent firms do not occur prior to the CFO s arrival. Together these results are consistent with the active influence of CFOs on accounting choices. We next analyze two potential moderators of CFO fixed effects predicted by upper echelons theory: CFO discretion and CFO job demands. 5 We find evidence consistent with CFO style being reflected more in accounting choices when CFO discretion is higher and when CFO job demands are higher, consistent with the predictions offered by upper echelons theory. Finally, we investigate whether the variations in CFO style can be explained by CFOs individual observable characteristics. We explore the impact of three observable CFO characteristics that are potentially associated with CFOs risk attitudes and or overconfidence gender, age, and educational background (specifically, CPA qualification and or MBA degree). The results of this analysis are mixed. Overall we find limited evidence of the impact of these observable CFO characteristics on CFOs reporting choices, suggesting that these common and observable characteristics capture only a small portion of CFO styles. Our findings have important implications for understanding accounting decisions because they indicate that a portion of the variation in accounting choices could be attributable to CFO-level managerial characteristics. We expand the literature on the determinants of accounting choices by highlighting manager-specific factors as a new dimension of determinants worth considering for future work in the area. The fact that CFOs accounting choices are impacted by their individual styles potentially complicates firms ability to obtain optimal reporting choices via economic incentives alone, unless firms are aware of these style differences and select CFOs accordingly based on the firm s time-varying needs. 6 Moreover, our findings on the moderating effects of CFO discretion and CFO job demands on the relation between CFO style and accounting choices contributes to the 4. Note that in order for a CFO fixed effect to be significant in the presence of firm-fixed effects, the CFO would need to consistently select accounting policies that are above (or below) the mean on some dimension (e.g., aggressiveness) in all the firms the CFO works for. Thus, the fact that CFOs tend to work for firms with similar characteristics (e.g., firms that provide equity incentives) would not, in itself, result in a significant CFO fixed effect unless the CFO tends to choose policies that are unusually high (or low) relative to what normally occurs at the firm. 5. According to upper echelons theory, managerial discretion exists when constraints on decision making are absent and when there is some ambiguity about the optimal decision (Hambrick 2007). The theory also defines job demands as the degree to which a given executive experiences his or her job as difficult or challenging (Hambrick, Finkelstein, and Mooney 2005). 6. Note that we do not attempt to distinguish between the explanation that CFOs exert their style resulting in suboptimal reporting choices (i.e., their personal style results in choices that the firm s shareholders do not desire) versus the alternative explanation that CFOs with a certain style are selected by firms to meet the firms timevarying needs. Under both explanations, CFO style impacts firms accounting choices.
4 1144 Contemporary Accounting Research literature of upper echelons theory by providing evidence consistent with the predicted moderators of upper echelons theory. Our paper is related to two recent streams in the accounting literature. First, recent studies have investigated the association between earnings quality and various managerial characteristics such as CEO reputation (Francis, Huang, Rajgopal, and Zang 2008), superstar CEOs (Koh 2011), managerial ability (Demerjian, Lewis, Lev, and McVay 2009), executive overconfidence (Schrand and Zechman 2009), and CEOs with prior CFO experience (Matsunaga and Yeung 2008). Because CFO style could result from a wide range of individual characteristics that are both observable and unobservable, our paper complements and extends this stream of research by focusing not on any single aspect of managerial characteristics (e.g., reputation) but rather on the manager s overall effect. Second, three concurrent studies use a methodology similar to ours to investigate managerial fixed effects with respect to tax avoidance behavior (Dyreng et al. 2010), and voluntary disclosure strategies (Bamber et al. 2010; Yang 2010). Our paper complements and differs from these papers in a few ways. First, we examine the effect of managerial style on a set of accounting choices that have different constraints, such as GAAP requirements and external audits. In addition, we explore discretion and job demands as two moderators of CFO style, which helps shed light on the conditions or circumstances under which CFOs styles are more likely to affect their accounting decisions. Finally, we focus on the fixed effect of the manager with the most direct impact on the financial reporting process, the CFO. In contrast, these concurrent studies examine all top managers on ExecuComp, with the vast majority being CEOs presumably because of data availability. We find that the effect of CFO style on accounting choices goes beyond the impact of CEOs, suggesting that CFOs play an important role in shaping certain corporate decisions and are not necessarily just carrying out the desires of their CEOs. The remainder of the paper is organized as follows. Section 2 reviews prior research and develops our hypotheses. Section 3 discusses our classifications and measures of accounting choices. Section 4 describes our sample construction and research design. Sections 5 and 6 present empirical results. Section 7 concludes. 2. Literature review and hypothesis development How do managers make accounting decisions? The neoclassical view of the firm assumes that managers are homogenous or perfect substitutes in other words, faced with the same economic circumstances, including economic incentives, different managers would make the same rational choices. Under this view, CFOs might be recognized as the manager responsible for making the firm s accounting choices, but these choices would not be influenced by his her individual style. In contrast, upper echelons theory recognizes that individual-specific attributes of top management (e.g., their biases and dispositions) matter to corporate strategy outcomes, at least to some extent (Hambrick and Mason 1984; Hambrick 2007). The predictions from upper echelons theory are largely built on the literature of how individual characteristics affect judgment and decision making (JDM). 7 The JDM literature suggests three factors that influence individuals decisions: person, task, and environment (Bonner 2008). The category of person-related factors refers to how individual characteristics such as confidence, risk attitudes, and intrinsic motivation affect the cognitive processes that lead to decision making. These person-related factors are what we deem manager style. Based on 7. The psychology literature classifies cognitive characteristics into cognitive style, cognitive abilities, and cognitive strategies (Ho and Rodgers 1993; Kozhevnikov 2007). All of these three types could affect decision outcomes; therefore we do not distinguish among these three types.
5 Do CFOs Have Style? 1145 these fundamental theories, the premise that managerial characteristics could have an impact on corporate decisions would seem reasonable. However, the magnitude of the effect of individual style on corporate decisions is an empirical question. Studies in the JDM literature, both in management and accounting, find evidence consistent with cognitive characteristics impacting decision making outcomes (e.g., Henderson and Nutt 1980; Casey 1980; Bonner and Lewis 1990). However, because these studies are based on experiments, it is potentially problematic to generalize these findings to organizational-level outcomes. In addition, experimental studies often control certain factors in an attempt to identify the impact of other factors (Libby, Bloomfield, and Nelson 2002). Thus, in real-world situations, when person, task and environmental factors are allowed to vary simultaneously, it is possible that task and environmental factors play a larger role in determining decisions than person-related factors or style (Hogarth 1993; Libby et al. 2002). In other words, even if one accepts that managers have different styles that influence their accounting choices, it is still possible that the economic significance of these differences is too small to be empirically documented because economic circumstances or situational incentives have a greater influence over these choices. In addition, CEOs might set most of the tone from the top, which would potentially dominate CFOs style in accounting choices. There are, however, other studies in management (based largely on upper echelons theory) that provide evidence consistent with managers having an impact on corporatelevel outcomes such as firm performance, acquisitions, or financial leverage. The majority of these studies rely on observable manager demographic characteristics (e.g., age, education, functional background, and so on) as proxies for the unobservable cognitive characteristics of top executives (e.g., Zajac and Westphal 1996; Pegels, Song, and Yang 2000; Marcel 2009). The main disadvantage of this approach is that demographic characteristics are arguably limited or incomplete proxies of managers cognitive frames. In addition, many of these studies are conducted on a purely cross-sectional basis, resulting in issues of reverse causality (Hambrick 2007). 8 An alternative approach to investigating the effect of managers styles on corporatelevel decisions is the approach used by Bertrand and Schoar In this study, the authors do not specify a particular relation between an observable managerial characteristic and a specific corporate-level decision. Rather, they simply measure commonalties in corporate-level decisions across different firms for which the same manager works. There are several advantages to this approach. First, it allows managers styles to be idiosyncratic and not necessarily related to any observable managerial characteristic (e.g., age, functional background, and so on). Second, it allows the researcher to control for firmspecific effects, thereby controlling (at least partially) for the impact of environmental factors on corporate outcomes as well as helping to address the issue of reverse causality. 9 The downside of this approach is that it does not provide further insight into the determinants of managers style differences. Nevertheless, because it provides a potentially more powerful test (by measuring both observable and unobservable manager-specific effects), we adopt the approach used in Bertrand and Schoar For example, upper echelons theory would predict that a manager with an research and development (R&D) background would invest more heavily in R&D. However, it is equally plausible that firms that invest heavily in R&D hire executives with R&D backgrounds. Thus, it is not clear from this result whether the investment in R&D is necessarily due to a manager s style (Hambrick 2007). 9. Technically, including firm-level fixed effects only controls for static firm effects and not time-varying effects. Thus, while it helps address issues with reverse causality, it does not eliminate that possibility if firms environments change over time.
6 1146 Contemporary Accounting Research Our main hypothesis is that CFOs faced with the same institutional environment (e.g., the same firm environment) do not necessarily make the same accounting choices but rather make choices that are influenced by their own personal styles. Formally: HYPOTHESIS 1. CFOs styles impact their accounting choices. It is important to remember that we are not conjecturing that CFOs are managers responsible for firms accounting choices, because they almost certainly are. Our conjecture is that CFOs personal styles are related to their accounting choices that is, an idiosyncratic aspect of a particular CFO influences his or her choices beyond what the economic or situational factors facing the firm might generally lead the manager to do. Whether this hypothesis holds is, therefore, not a foregone conclusion, particularly given the importance of situational factors in decision making, as well as the potential influence of the CEO on setting the tone at the top. Our next two hypotheses explore the conditions under which CFOs styles are more likely to influence their decisions. Upper echelons theory predicts two moderating factors that would cause cross-sectional variations in manager-specific effects: managerial discretion and manager job demands (Hambrick 2007). One important assumption underlying upper echelons theory is that managers have a certain level of discretion (Hambrick and Finkelstein 1987). Managerial discretion exists when constraints on decision making are absent and when there is some ambiguity about the optimal decision. Therefore, the first important moderator that upper echelons theory predicts will influence managerial effects is the level of managerial discretion. Specifically, managerial characteristics will have a greater impact on corporate decisions when the level of discretion is higher. Accordingly, we predict: HYPOTHESIS 2. CFOs styles are more reflected in accounting choices when the level of CFOs discretion is high rather than low. The second and more recently introduced factor that upper echelons theory predicts will moderate the effect of managerial characteristics on organizational outcomes is executive job demands. Hambrick et al. (2005) define this construct as the degree to which a given executive experiences his or her job as difficult or challenging. Hambrick et al. (2005) suggest that when job demands are high, top managers are less able to comprehensively and correctly process all the information they need in order to make rational and optimal decisions, consistent with the underpinnings of bounded rationality. Thus managers are more likely to make decisions relying on their past experiences and dispositions under conditions of high job demands. Similar to this argument, Bonner (1994) suggests that, in the auditing setting, an increase in task complexity would lead to less optimal use of knowledge (see also Bonner 2008). Accordingly, we expect CFOs styles to manifest more in accounting choices when their job demands are high. Therefore, our third hypothesis is as follows. HYPOTHESIS 3. CFOs styles are more reflected in accounting choices when the level of CFOs job demands is high rather than low. 3. Accounting practices: Classifications and measures This section describes the accounting choices we examine in this paper. Because there are numerous ways in which a manager could influence accounting outcomes, we examine a wide range of accounting choice variables. However, in selecting variables to study it is important to consider whether the variable is more likely affected by managerial discretion
7 Do CFOs Have Style? 1147 or by the firm s fundamental earnings process. We attempt to select variables that are more likely affected by the former than the latter (e.g., discretionary accruals versus earnings persistence). In addition, a manager s influence on the financial reporting system can be measured or observed at various levels. That is, one can attempt to measure or observe the actual choices managers make to influence the financial reporting system or one can measure the outcome of these choices on the financial reporting system. We group our proxies for reporting choices along these lines. The first category of accounting choice variables we consider are accounting tools that CFOs can choose to achieve financial reporting goals. The second category are outcome-based measures that capture either properties of earnings that are likely the result of managerial intervention in the accounting system (e.g., meeting beating analysts expectations) or the likelihood of accounting misstatements. While the likelihood of accounting misstatements is not technically an outcome of managerial influence on the financial reporting system, measuring actual misstatements is unlikely to uncover a managerial fixed effect because managers who engage in fraud or misstatements are unlikely to obtain similar positions in a subsequent firm (Karpoff, Lee, and Martin 2008). However, the likelihood of misstatement captures an overall effect on the firms financial system that results from managers microlevel decisions and choices. We organize our discussion along these two categories. Accounting tools One of the most widely studied ways in which managers can influence the financial reporting system is by managing the accrual component of earnings. Discretionary accruals are commonly used in the literature to proxy for earnings management (Dechow et al. 2010). We therefore investigate whether the propensity to report income-increasing discretionary accruals is CFO specific (i.e., whether individual CFOs have a fixed effect on discretionary accruals). We measure discretionary accruals (DISC_ACC) based on the cross-sectional performance-matched modified Jones model (DeFond and Jiambalvo 1994; Kothari, Leone, and Wasley 2005). We adjust discretionary accruals by performance to control for the potential effect of CFO style on firm performance. 10 Table 1 provides more precise definitions of this variable and all other variables discussed below. Another way in which CFOs can influence the financial reporting system is by exercising their discretion in the choice of off balance sheet activities to improve reported earnings and or certain balance sheet ratios. For example, CFOs can arrange leases to obtain operating lease classification for financial reporting purposes instead of purchasing equipment through loans or leases that fall into the capital lease category. Operating lease classification allows a company to report lower expenses during the early stage of the lease life, as well as to report a lower leverage ratio. Therefore, we analyze whether the tendency to use more operating leases versus on balance sheet debt is CFO-specific. In order to measure the extent of operating lease activities (i.e., off balance sheet debt) relative to on balance sheet debt, we calculate the present value (PVOL) of the next five years minimum operating lease payments using a 10 percent discount rate (Ge 2007). We then divide 10. We focus on the level of discretionary accruals rather than the absolute value of discretionary accruals because we believe CFO style is more likely to lead to consistently aggressive (or conservative) accrual choices, a style which is more effectively captured by studying levels of discretionary accruals. However, it is arguably difficult for a manager to consistently make aggressive (or conservative) accrual choices over long periods of time, so it is possible managerial style will manifest in consistently large magnitudes of accruals (not necessarily of the same sign). This possibility would suggest using the absolute value of discretionary. We also examined the absolute value of discretionary accruals and find qualitatively similar results.
8 1148 Contemporary Accounting Research TABLE 1 Variable definitions Variable Definition Financial reporting measures DISC_ACC Residuals from the following pooled regression based on two-digit SIC code: TAi;t ASSETi;t 1 ¼ a0 þ a1 1 ASSETi;t 1 þ a2 DSALESi;t DARi;t ASSETi;t 1 þ a3 PPEi;t ASSETi;t 1 þ a4 NIi;t ASSETi;t 1 Where for firm i year t, TA i,t is total accruals, which equal Net Income minus Cash Flow from Operations (data18-data308); ASSETi,t)1 is lagged Total Assets (data6); DSALESi,t is the change in Sales (data12); DARi,t is the change in Accounts Receivables (data2); and PPEi,t is Net Property, Plant, and Equipment (data8). NI is Net Income (data18). OPLEASE Operating lease deflated by the sum of long term debt and operating lease, where operating lease is defined as the present value of the next five years minimum rent commitment under operating leases, discounted at 10 percent: ððdata96=1:1 þ data164=1:1^2 þ data165=1:1^3 þ data166=1:1^4 þ data167=1:1^5þ=ðdata34 þ data9 þðdata96=1:1 þdata164=1:1^2 þ data165=1:1^3 þ data166=1:1^4 þ data167=1:1^5þþ PENSION_RET The expected rate of return for pension assets (data336). FSCORE The scaled predicted probability from plugging time variant firm characteristics into the following logit model, which uses estimated coefficients from Dechow et al. 2011: Manipulationt ¼ 7:184 þ 0:702 RSSTaccrualst þ 3:035 Change in receivablest þ 2:678 Change in inventory t þ 0:105 Change in cash salest 1:124 Change in earnings t þ 0:839 Actual issuancet 0:199 Abnormal change in employees t þ 0:615 Existance of operating leases t þ ei;t (The table is continued on the next page.)
9 Do CFOs Have Style? 1149 TABLE 1 (Continued) Variable Definition In the above model, RSST accruals are the change in non-cash net operating assets; Change in receivables is DAccounts Receivables (data 2) Average total assets; Change in inventory is DInventory (data 3) Average total assets; Change in cash sales is percentage change in cash sales [Sales(data 12) ) DAccounts Receivables (data 2)]; Change in earnings is [Earnings t (data 18) Average total assets t ] ) [Earnings t)1 Average total assets t)1 ]; Actual issuance is an indicator variable which is one if the firm has issued new debt or equity during the time period; Abnormal change in employees is percentage change in the number of employees (data 29) ) percentage change in assets (data 6); Existence of operating leases is an indicator variable coded 1 if future operating lease obligations are greater than zero. FSCORE is the predicted probability from the above model, scaled by the unconditional probability of having accounting manipulations. An FSCORE of 1 indicates that the firm has the same probability of manipulation as the unconditional expectation. An FSCORE less (more) than one indicate a lower (higher) probability of manipulation than the unconditional expectation. SMBE The percentage of quarters the CFO meet or beat analyst forecasts by three cents per share or less (EPS ) Meanest $0.03) each year, where Meanest is the last consensus forecast for the quarter. EARN_SMOOTH The variance of the residuals for each CFO-firm from the following pooled regression (Myers et al. 2007): DNIi;t ¼ a1leveragei;t þ a2growthi;t þ a3debt ISSi;t þ a4equity ISSi;t þ a5asset TURNi;t þ a6sizei;t þ ei;t Where for firm i quarter t, DNI i,t is the current quarter net income taking fourth differences to remove the effects of seasonality, then deflated by total assets ([data69 ) lag4(data69)] data44); LEVERAGEi,t is total liabilities divided by assets (data54 (data60 + data172)); GROWTHi,t is the percentage change in sales (data2 lag(data2) ) 1); DEBT_ISS i,t is percentage change in long-term debt (data51 lag(data51) ) 1); EQUITY_ISS i,t is percentage change in common shares outstanding (data61 lag(data61) ) 1); ASSET_TURN i,t is quarterly asset turnover ratio (data2 data44); SIZEi,t is calculated as the natural log of sales (log(data2)). We measure EARN_SMOOTH for a CFO as this variance taking out the firm effect (VARIANCECFO ) VARIANCEfirm). For easier interpretation, we define EARN_SMOOTH as (VARIANCECFO ) VARIANCEfirm) *()1000). (The table is continued on the next page.)
10 1150 Contemporary Accounting Research TABLE 1 (Continued) Variable Definition Control variables ROA Return on assets ratio, defined as income before extraordinary items over lagged total assets: (data18 lag(data6)) for annual data. SIZE Log transformation of sales, defined as log(data12) for annual data. BTM Book to market ratio, defined as book value of equity over market value of equity: (data60 (data25*data199)) for annual data. LEVERAGE Leverage ratio, defined as long-term debt plus debt in current liabilities over long-term debt plus debt in current liabilities plus the book value of common equity: ((data9 + data34) (data9 + data34 + data60)) for annual data. GROWTH Sales growth, defined as percentage change in total sales: (data12 lag(data12) ) 1) for annual data. CFF Cash flow from financing activities deflated by total assets: (data313 data6) for annual data. Moderating factors Auditor Expertise Following Krishnan 2003, we first calculate auditor portfolio shares for each Big N auditor, industry (2-digit SIC code), and year using the COMPUSTAT population ( ). Auditor portfolio share is the sum of sales revenue for clients of an auditor in a specific industry, divided by the sum of sales revenue of all clients of the auditor. Then we calculate the average portfolio shares over the sample period for each auditor in each industry. We consider an auditor an industry expert (Auditor Expertise = 1) if the industry is among the auditor s top ten in terms of average portfolio share. Complexity The sum of the number of operating segments and geographic segments as reported in COMPUSTAT. CFO characteristics WOMEN An indicator variable for female CFOs. AGE CFO s age. CPA An indicator variable that is one for a Certified Public Accountant and zero otherwise. MBA An indicator variable that is equal to one if the CFO has a master s degree in business administration and zero otherwise.
11 Do CFOs Have Style? 1151 PVOL by the sum of PVOL and on balance sheet long-term debt and term this variable OPLEASE. Finally, accounting for pension obligations and plan assets in defined benefit plans provides another opportunity for managers to influence the financial reporting system. Specifically, CFOs have substantial flexibility in deciding the assumptions that affect reported pension expense. Prior research has shown that CFOs can assume higher expected returns on the plan assets to reduce reported pension expense (Comprix and Muller 2006; Picconi 2006). We therefore investigate whether the tendency to make higher assumptions of the expected rate of return for pension assets (PENSION_RET) is CFO specific. 11 Outcome-based measures We also examine measures that focus on outcomes of the financial reporting system. Our first two measures are based on properties of reported earnings and are intended to capture the outcome of two particular earnings management strategies that have been widely discussed in the literature. The first measure is an earnings smoothing measure. Earnings smoothing involves both downward and upward earnings management to hide the true variance of economic performance and can be accomplished by using either real transaction management or accrual management. It is well established in the literature that managers have strong incentives to show a smooth string of earnings rather than volatile earnings (Graham, Harvey, and Rajgopal 2005). However, managers may vary in the extent to which they believe smoothing earnings is appropriate or beneficial as well as in their knowledge of actions that result in smooth earnings; therefore, CFO-specific factors might impact the degree of earnings smoothing by a firm. Our earnings smoothing measure is based on the approach used in Myers, Myers, and Skinner Specifically, we use quarterly earnings data and take fourth differences to remove the effects of seasonality. We then compute the variance of the residuals obtained from a regression of the changes in quarterly earnings on a set of control variables, including leverage, sales growth, debt issuance, equity issuance, asset turnover, and size (EARN_SMOOTH). The second outcome-based measure we investigate focuses on the earnings management strategy related to meeting earnings targets, particularly analysts forecasts. DeGeorge, Patel, and Zeckhauser (1999) document a kink in earnings around analyst forecasts and interpret this as evidence of earnings management to meet analysts expectations. Based on these findings, beating analysts expectations (particularly by a small amount) has been commonly used as a proxy for accounting discretion (e.g., McVay 2006; Hribar, Jenkins, and Johnson 2006). However, while prior studies have documented various firm-specific factors that are associated with the propensity of firms to meet or beat analysts forecasts (e.g., Matsumoto 2002), it is also possible that managers themselves have predispositions toward avoiding negative earnings surprises. Therefore, we predict a CFO fixed effect with respect to this behavior. We compute the percent of firm quarters in a year in which the firm reports earnings that meet or beat analysts forecasts by a small amount (less than or equal to three cents) (SMBE) and consider these events as indicative of upward earnings management to achieve this target. 12 We recognize that meeting or beating analysts expectations is possibly the result of downward biased earnings guidance (Matsumoto 2002) as much as it is a result of earnings-related accounting choices, and results should be interpreted accordingly. Finally, we also consider an outcome-based measure that captures the likelihood of accounting misstatements. As discussed previously, it is unlikely that there is a CFO fixed 11. Following Comprix and Muller 2006, we focus on the expected rate of return as the key pension assumption because it is generally viewed as the assumption most subject to managerial discretion. 12. Our results are similar if we define small meet beat to be 1 or 2 cents.
12 1152 Contemporary Accounting Research effect with respect to the actual incidence of fraud or misstatements, but it is possible that CFOs vary in their predisposition toward engaging in behaviors that increase the likelihood of misstatements. Dechow et al. (2011) develop a measure, which they call the F-Score, by modeling the factors associated with actual accounting misstatements, and have shown their models to have reasonable predictive abilities using out-of-sample tests. The F-Score (FSCORE) uses both accruals and off balance sheet activities (e.g., operating leases) to measure the overall likelihood of accounting misstatements. Specifically, FSCORE is a scaled logistic probability for each firm-year based on a model of the determinants of accounting manipulations (see Table 1 for further detail). Thus, FSCORE reflects a combination of managerial actions that could be considered aggressive financial reporting. While a managers style may not impact all eight factors that are used to estimate FSCORE, we hypothesize that managers style will manifest itself in the composite measure of financial reporting aggressiveness that is captured by FSCORE. If some CFOs have a tendency to engage in more aggressive reporting than other CFOs, we expect CFO fixed effects to explain a portion of the overall reporting aggressiveness proxied by FSCORE. 4. Sample construction and research design Sample construction Following the methodology of Bertrand and Schoar 2003 we construct a CFO-firm matched panel data set tracking the same CFOs across different firms over time as well as including data for the same firm under different CFOs. This data set allows us to include both CFO and firm fixed effects, thereby enabling us to disentangle the impact of the CFO from the underlying economic factors that are specific to the firm. To construct this sample we combine three databases: ExecuComp (1992 to 2006), Management Change Database (2002 to 2004), and AuditAnalytics (2002 to 2006). We first use ExecuComp to track the names of the CFOs in 1,500 publicly traded U.S. firms. 13 We next obtain additional CFO data from Management Change Database and AuditAnalytics. The Management Change Database is a new database that gathers executive changes from press releases. AuditAnalytics provides information on the top executive changes from firms 8- Ks. 14 Both databases provide information on CFO name, the name of the company the CFO is leaving (Firm 1), the name of the company the CFO is going to (Firm 2), and the dates of the CFO change. However, we also need to know the number of years the CFO worked for each company in order to conduct our analyses. We identify the exact years that a CFO worked with each firm by searching for CFOs biographies using Google or firms SEC filings. CFOs whose tenure cannot be identified are deleted from our sample. We then combine the above data sets and limit our sample to CFOs that can be traced to at least one other firm that is, that have worked as a CFO for at least two companies. 15 In addition, because it likely takes some time for a CFO to impose his her style on 13. We use the variable titlean in ExecuComp to identify the CFO of the firm. The following key words are chosen: Chief Financial Officer, CFO, Vice President in Finance, VP Finance, and the like. Note that ExecuComp collects data from proxy statements; therefore, it only includes CFOs who are in the top five paid executives and have compensation higher than $100,000. This constraint does not apply to the Management Change Database, because the source is company press releases, nor to the Audit Analytics database, whose source is 8-Ks. 14. Most of the manager changes in these data sets are in 2005 and 2006, because firms were required to file 8-Ks for chief executive changes starting in There are very few observations from This research design enables us to control for unobservable firm characteristic but, as a result, our sample leaves out (1) CFOs that do not move to another public company, and (2) CFOs that do not move companies at all during the sample period, either because they are unable or unwilling to move firms or are promoted to higher positions in the same firm. Thus, we cannot generalize our findings to CFOs who have not occupied CFO positions in at least two companies. However, we do not believe our sample selection biases our inference in any systematic way.
13 Do CFOs Have Style? 1153 the financial reporting practices of a company, we require CFOs to have occupied the CFO position at a company for at least two years. 16 In order to separate the firm fixed effects from the CFO fixed effects, it is also necessary to have the firms in our sample appear under more than one CFO. Thus, for those firms appearing under only one CFO, we add data for three years prior to the starting year of the CFO at the firm and three years following the final year of the CFO at the firm. We call these years filler years. 17 We require that data be available for at least two of these six filler years in order for the firm to remain in the sample. The CFOs employed by the firm during these filler years are unidentified and not part of the CFO fixed effect estimation. We provide an example of this process in Figure 1. Bennett Nussbaum worked as a CFO for Kinko from 1997 to 2000 and then at Burger King from 2001 to However, our sample did not initially include firm-years for these two firms under a different CFO. Therefore, we add six filler years for both companies in order to disentangle the Kinko effect and Burger King effect from the Nussbaum effect. Panel A of Table 2 reports the results of our sample selection procedure. We identify a total of 691 CFOs who have worked for at least two firms as CFOs. Of these CFOs, 183 are excluded because their tenure at one of their two firms was less than two years. Next, we remove observations with missing data on our control variables as well as reporting variables, resulting in the loss of 71 CFOs from our sample. 18 The data requirements for the control and reporting variables also results in the loss of either Firm 1 or Firm 2 data for a given CFO and or the loss of filler year data. Thus, to ensure that the firm fixed effects can be separated from the CFO fixed effects based on the final sample, we further remove CFOs with only available data in one firm, as well as CFOs who have worked for firms who do not have at least two filler years of data. The resulting CFO-firm matched sample contains 2,565 firm-years, 705 firms, and 359 individual CFOs (i.e., excluding the unidentified CFOs associated with the filler years). Panel B of Table 2 presents the frequency of CFO-firm pairs based on the number of years the CFO worked with a given firm. 19 For about 62 percent of our CFO-firm pairs, the CFO stayed in a firm for at least three years. The average tenure of stay of a CFO in our sample is 3.35 years, indicating that CFOs are given a reasonable time to have an influence on a firm s financial reporting outcomes. Panel C of Table 2 tabulates the distribution of the sample firms based on the number of distinct CFOs in our sample. Of the 705 firms identified in our sample, 60 have at least two distinct CFOs in the sample. As discussed previously, for the 645 firms that have only 16. If we restrict our sample to CFOs with tenure equal to or longer than three years at both firms, our sample size would drop to 148 CFOs. The F-tests of all five accounting choice variables yield a significance level less than 5 percent. The earnings smoothness variable is significant at a 10 percent level. 17. Note that the inclusion of filler years is necessary because two of our data sources Audit Analytics and Management Change database are not panel data sets (like ExecuComp). As a result, our sample construction is somewhat different and more complicated than in similar studies (Dyreng et al. 2010; Bamber et al. 2010; Yang 2010). We do not include all the firm-year observations (as filler years) for the firms in our sample because, for firm-years that are not covered by the ExecuComp database, we need to hand -collect CEO information in order to estimate CFO effects beyond CEO effects. 18. We use COMPUSTAT, Center for Research in Security Prices, and I B E S data to construct our annual and quarterly financial reporting variables. Variables used are described in Table In panel B of Table 2, multiplying the number of years a CFO is in each firm by the number of CFO-firm pairs and summing the resulting amounts result in 2,565 firm-year observations, as shown in Table 2, panel A. Note that our final sample includes CFOs with only one year at a given firm (as shown in panel B) even though our initial requirement was that the CFOs have at least a two-year tenure at each firm. This is because one of the two years might have missing data for our analysis. If we delete these CFOs with only one year s data at a given firm, the F-tests of all five accounting choice variables yield a significance level less than one percent. The earnings smoothness variable is significant at a five percent level.
14 1154 Contemporary Accounting Research Figure 1 Example of sample construction for F-tests (Tables 4 and 5) CFO: Bennett Nussbaum Filler Year Filler Year Filler Year Kinko Kinko Kinko Kinko Filler Year Filler Year Filler Year Filler Year Filler Year Filler Year Burger King Burger King Burger King Filler Year Filler Year Filler Year Notes: This figure demonstrates the process used to construct our sample. We first combined data from ExecuComp, Management Change Database, and Audit Analytics. We then identified 359 CFOs who have occupied the CFO position in at least two companies based on this combined data. In the above example, CFO Bennett Nussbaum worked for Kinko from 1997 to 2000 and for Burger King from 2001 to We then added filler years (three years prior to the starting year of the CFO at a firm and three years following the final year of the CFO at the firm) when the firm was under a different CFO so that we can disentangle the CFO fixed effect from the firm fixed effect. We require at least two filler years for each firm. We also require all the firmyear observations to have data on all of our control variables and at least one accounting choice variable. one CFO in the CFO-matched sample, we add filler years (a minimum of two filler years and a maximum of six filler years per firm) when the firm is under a different CFO so that we can disentangle the CFO fixed effect from the firm fixed effect. In our final sample, we have 2,476 filler year observations and 2,565 non filler year observations. Panel D of Table 2 focuses on the distribution of CFOs according to how many times they have changed their jobs. All of the 359 CFOs in our sample have assumed the CFO position in at least two companies, and 42 of them have changed their jobs more than once. Descriptive statistics Panel A of Table 3 presents means and standard deviations for our variables of interest.we report summary statistics for the CFO-firm matched sample as well as the descriptive statistics for the COMPUSTAT universe between 1993 and It appears that firms in our CFO-firm matched sample are larger than the COMPUSTAT average in terms of total assets and market value. 21 This is not surprising for two reasons. First, one of our data sources is ExecuComp, which covers relatively large firms; second, we limit our sample to firms whose CFO moves to another firm. This procedure would lead us to larger firms because executives from larger firms are more likely to move between public firms. CFOs from smaller firms might move to a private firm or to a divisional CFO position in a large firm. The average firm in our sample also has a lower book-to-market ratio, more operating leases, somewhat higher pension assumptions, and lower F-scores. They also tend to have a higher likelihood of meeting or beating analyst forecasts by a small amount. While we include variables such as firm size, growth, leverage, and book-to-market ratios as controls in our analyses, we acknowledge that our conclusions might not be generalizable to CFOs who have only worked for other firms in the COMPUSTAT universe. 20. We compare our sample to the COMPUSTAT sample from 1993 to 2006 because less than 10 percent of our observations occur prior to However, firms in our sample are smaller than firms in Bamber et al (e.g., mean market value in our sample is $2,848 million and is $11,066 million in Bamber et al.). This size difference is likely due to the fact that our sample of CFOs is not limited to managers covered in ExecuComp, but also includes managers covered in Management Change Database and Audit Analytics.
15 Do CFOs Have Style? 1155 TABLE 2 Sample selection and sample description Panel A: Sample selection Number of firm-years Number of distinct firms Number of distinct CFOs CFO-firm matched sample 5,170 1, for CFOs that worked for at least two firms as CFOs, from AuditAnalytics, Management Change database, and ExecuComp Less: CFOs who worked (958) (381) (183) in a firm for less than two years Less: firm-years that have (1,060) (138) (71) missing data for our control variables or have missing data for all of our accounting choice variables Less: CFOs that have only (587) (161) (78) available data for one firm and CFOs that have worked for firms without at least 2 filler years CFO-firm matched sample 2, Panel B: Frequency of CFOs based on the number of years at each firm N of years in each firm N of CFO-firm pairs Percentage (%) Total Panel C: Frequency of firms based on the number of different CFOs N of different CFOs Freq of firms Percentage (%) N of CFO-firm pairs Total (The table is continued on the next page.)
16 1156 Contemporary Accounting Research TABLE 2 (Continued) Panel D: Frequency of CFOs based on the number of changes N of changes Freq of CFOs Percentage (%) N of CFO-firm pairs Total Notes: Panel A presents our sample selection process. Panel B presents the frequency of the CFOs for the CFO-firm matched sample, based on how many years they worked for each firm. Panel C presents the frequency of the firms for the CFO-firm matched sample, based on how many different CFOs have worked with each firm. For the 645 firms which only have one CFO in the CFO-firm matched sample, we add filler years to disentangle the CFO effect from the firm effect. Panel D presents the frequency of CFOs for the CFO-firm matched sample, based on how many times they change their jobs. Research design For each financial reporting variable, with the exception of EARN_SMOOTH, we regress the variable of interest (DISC_ACC, OPLEASE, PENSION_RET, FSCORE, and SMBE) on a set of CFO indicator variables as well as a set of firm indicator variables, year indicator variables, and time-varying control variables 22 : FINANCIAL REPORTING it ¼ a 0 þ a 1 ROA it þ a 2 SIZE it þ a 3 BTM it þ a 4 GROWTH it þ a 5 LEVERAGE it þ a 6 CFF it þ FIRM i þ YEAR t þ CFO j þ e it ð1þ: In each case, we perform an F-test for the joint significance of the CFO indicator variables to test for a CFO fixed effect. While year and firm fixed effects control for year- and firm-specific factors associated with accounting choices, we also control for time-varying control variables that have been shown to be associated with accounting choices. We control for performance using return on assets (ROA) because prior research shows a relation between firm performance and discretionary accruals. 23 In addition, the probability of engaging in accounting fraud likely increases when firm performance deteriorates (Dechow et al. 2011). We control for firm size (SIZE) because larger firms may face greater political costs (Watts and Zimmerman 1986). Growth firms may face greater capital market pressure to over-report their earnings (Lee, Li, and Heng 2006); therefore, we include two controls for growth, the book to market ratio (BTM) and sales growth (GROWTH). We also control for a firm s leverage ratio (LEVERAGE), because prior research suggests that managers face incentives to influence the financial reporting system because of debt 22. Note that SMBE is the percent of quarters in a year in which a firm meets or beats analyst expectations by a small amount. Alternatively, we could have defined SMBE as a dummy variable indicating firm-quarters in which a firm meets or beats expectations and run our analysis at the quarterly level using a logistic regression specification. However, when we attempted this specification the model would not converge. 23. As mentioned earlier, we also directly control for performance when measuring discretionary accruals (see Table 1 for variable definitions).
17 Do CFOs Have Style? 1157 TABLE 3 Descriptive statistics Compustat data Our sample COMPUSTAT Mean Median Mean Median t-statistics (sample vs. COMPUSTAT) TOTAL ASSETS 2, , MARKET VALUE 2, , RETURN ON ASSETS ) ) )4.04 SIZE BTM )12.49 LEVERAGE GROWTH CFF )7.07 DISC_ACC )0.012 )0.010 )0.007 )0.009 )4.24 OPLEASE PENSION_RET FSCORE )3.51 CFO-firm matched sample I B E S I B E S data Mean Median Mean Median t-statistics of t-tests SMBE Notes: Our sample refers to the set of firm-year observations for firms that have at least one CFO observed in multiple firms. This sample includes observations for these firms in the years in which they have other CFOs that we do not observe in multiple firms. COMPUSTAT is a comparison sample of all listed firms on COMPUSTAT over the period 1993 to I B E S is a comparison sample of all listed firms on I B E S over the period 1993 to The maximum number of observations is 5,041 and 54,941 for CFO-firm matched sample and COMPUSTAT, respectively. For SMBE, the number of observations is 3,853 and 71,413 for CFO-firm matched sample and COMPUSTAT, respectively. Not all variables are available for each firm-year observation. TOTAL ASSETS is the total asset (data6); MARKET VALUE is the market value of equity (data199*data25). All other variables are described in Table 1. Each of the continuous variables are winsorized at 1 percent and 99 percent to mitigate outliers. covenants (DeFond and Jiambalvo 1994). Finally, external financing needs provide an incentive to influence the financial reporting system (Teoh, Welch, and Wong 1998). We include cash flow from financing (CFF) as a control variable to capture this effect Prior research has also shown that CFOs respond to compensation incentives when making accounting decisions (Jiang, Petroni, and Wang 2010). To the extent compensation structures tend to be fairly sticky at the firm level, our firm fixed effects would control for compensation structure. Consistent with this notion, Graham, Li, and Qiu (2009) document that time-invariant firm effects explain a significant portion of the variation in executive compensation. However, they also find that manager fixed effects explain a significant portion of the variation in compensation. Thus, it is possible that CFOs choose firms that offer a certain type of compensation structure. In this case, a CFO s fixed effect might be associated with his her compensation structure, but the association is a manifestation of the managers style impacting both their choice of compensation structures as well as their accounting choices. Therefore, we do not view incentive compensation as an alternative explanation, per se.
18 1158 Contemporary Accounting Research To calculate EARN_SMOOTH requires time-series data; thus, the variable is computed at the CFO-firm pair level and we have only two observations per CFO for the majority of our CFOs. This fact makes estimating a CFO fixed effect in the traditional manner infeasible. Instead, we choose an alternative approach to examine the commonality across the different firms in which a CFO works. We regress EARN_SMOOTH measured at the firm the CFO moves to (FIRM 2) onearn_smooth measured at the firm the CFO moves from (FIRM 1). Specifically, we run the following regression: EARN SMOOTH FIRM2 ¼ a 0 þ a 1 EARN SMOOTH FIRM1 þ Controls þ e ð2þ: We expect a 1 to be positive if a CFO has a style with respect to earnings smoothness; that is, a CFO who prefers to smooth earnings (and is capable of it) will likely smooth earnings at his subsequent employer. One potential problem with this approach is that a 1 could be positive due to the common factors between firm 1 and firm 2. To address this potential concern, we calculate EARN_SMOOTH for the firm (over the CFO s tenure as well as the filler years) and subtract the firm-level EARN_SMOOTH from the CFO-firmlevel measure of EARN_SMOOTH. Thus, EARN_SMOOTH in equation 2 measures the smoothness of the firm s earnings during the CFO s tenure relative to the average smoothness of the firm s earnings, and any relation between EARN_SMOOTH at a CFO s first and second firm should not be affected by static firm effects. In addition, we include firmlevel control variables that are used in the first stage (see Table 1 for a list of variables in the first stage) in regression (2) to further control for potential firm-level determinants of smoothness. 5. Empirical results The effect of CFOs on accounting choices (Hypothesis 1) Table 4 presents the results of our analyses of the effect of CFOs on firms accounting choices. Panel A of Table 4 reports the regression results for equation 1. For each measure, the first row reports the adjusted R-square from a base regression excluding the CFO indicator variables. The second row reports the F-statistics, the associated p-value from tests of the joint significance of the CFO fixed effects, and the adjusted R-square when the CFO indicator variables are added into the regression (i.e., equation (1)). 25 The first variable in Table 4, panel A is discretionary accruals (DISC_ACC). The adjusted R 2 in the base regression (regressing DISC_ACC on firm and year fixed effects only) is 26 percent. When we include CFO fixed effect, the adjusted R 2 increases by 2 percent to 28 percent. The F-test also yields a significance level less than 1 percent, which allows us to reject the null hypothesis of no CFO fixed effect on discretionary accruals. The next two variables measure the extent of aggressive reporting via off balance sheet activities, such as increasing operating leases or changing pension accounting assumptions. The adjusted R 2 s for the base regressions are high 79 and 93 percent for operating leases and pension assumptions, respectively likely due to the fact that firmspecific factors explain a significant portion of the cross-sectional variation in the use of operating leases and pension rate assumptions (and the base regression includes firm fixed effects). However, adding CFO fixed effects still provides an additional 1 to 2 percent 25. We do not report the results for our control variables due to space limitation. Overall, the results for our control variables are largely consistent with those from prior research. Specifically, most accounting choice variables (i.e., three or more out of five accounting variables) are significantly positively associated with ROA, SIZE, LEVERAGE, and CFF.
19 Do CFOs Have Style? 1159 TABLE 4 CFO effects on accounting choices Panel A: Earning aggressiveness F-test on fixed effects for CFOs N Adj. R 2 (%) Aggressiveness DISC_ACC DISC_ACC 1.33 (<.001, 357) OPLEASE OPLEASE 2.05 (<.001, 358) PENSION_RET PENSION_RET 3.27 (<.001, 180) FSCORE FSCORE 1.46 (<.001, 356) SMBE SMBE 1.26 (.002, 349) 4, , , , , , , , , , Panel B: Earnings smoothness Dependent variable: EARN_SMOOTH FIRM 2 Predicted sign Coefficient estimate One-tailed p-value Adj. R 2 (%) EARN_SMOOTH FIRM Panel C: Distribution of CFO fixed effects N Mean Median Lower quartile Upper quartile Variable (1) (2) (3) (4) (5) DISC_ACC 357 ) ) OPLEASE ) PENSION_RET ) FSCORE 356 )0.047 )0.026 ) SMBE 349 )0.021 )0.013 ) Notes: This table reports the test results for CFO fixed effects on accounting choices. Sample is the CFO-firm matched panel data set as described in Table 1. Reported in panel A are the results from fixed effects panel regressions. For each dependent variable (as reported in column 1), the fixed effects included are row 1: firm and year fixed effects; row2: firm, year, and CFO fixed effects. Reported are the F- tests for the joint significance of the CFO fixed effects (column 2). For each F-test we report the value of the F-statistic and, in parentheses, the p-value and number of constraints. Also reported are the number of observations (column 3) and adjusted R 2 s (column 4) for each regression. Reported in panel B are the results of a second stage regression. There are a total of 367 observations used in panel B. The first stage is a panel regression using the firm-quarter level data, in which the absolute value of change in net income (the current quarter net income taking fourth differences to remove the effects of seasonality and then scaled by total assets) is regressed on the control variables. We calculate EARN_SMOOTH for the firm (over the CFOs tenure as well as the filler years) and subtract the firm-level EARN_SMOOTH from the CFO-firm-level measure of EARN_SMOOTH. The second stage is a panel regression at the CFO-firm level. EARN_SMOOTH for the CFO s second firm is regressed on EARN_SMOOTH for his first firm and control variables. The fixed effects used in panel C are obtained from the regressions reported in panel A. Column 1 reports the number of estimated CFO fixed effects. Columns 2 and 3 report the mean and median fixed effect for each accounting choice variable. Columns 4 and 5 report the CFO fixed effects at the lower quartile and the upper quartile of the distribution respectively. All variables are described in Table 1.
20 1160 Contemporary Accounting Research increase in the adjusted R 2 s. F-tests also reject the null hypothesis that there is no joint CFO effect in these off balance sheet activities. For our measure of the likelihood of accounting manipulations, FSCORE, the adjusted R 2 increases by 3 percent from the base regressions. The F-test rejects the null hypothesis of no joint CFO effect on the likelihood of accounting manipulations. Finally, with respect to our first earnings outcome-based measure, the F-test on SMBE allows us to reject the null hypothesis that jointly CFOs do not have effects on the likelihood of avoiding negative earnings surprises by a small margin. Adding CFO fixed effects increases the adjusted R 2 by 2 percent. The other earnings outcome-based measure we investigate is earnings smoothing. Using the methodology described in section 3 (equation (2)), we estimate whether the degree of earnings smoothing at the first firm that a CFO works determines the level of earnings smoothing at the next firm he she works. As reported in panel B of Table 4, the coefficient on EARN_SMOOTH FIRM1 is significantly positive at a p-value less than 5 percent, consistent with the hypothesis that individual CFOs vary in the degree to which they adopt an earnings smoothing strategy. The estimated coefficient is also economically significant. For example, a one percent increase in EARN_SMOOTH for a CFO at his first job is associated with percent increase in EARN_SMOOTH at his second job. To assess the economic significance of the CFO effects, we examine the distribution of the CFO fixed effects estimated in Table 4, panel A. 26 Panel C of Table 4 reports mean, median, lower quartile, and upper quartile values of the estimated CFO fixed effects. Overall, the differences between a CFO at the lower quartile and a CFO at the upper quartile of CFO fixed effects appear to be quite large. For example, the interquartile (IQ) range for DISC_ACC is 0.052, indicating that a manager at the third quartile has discretionary accruals that are higher than a manager at the first quartile by an amount equal to 5.2 percent of total assets. Similarly, the IQ range for OPLEASE indicates that a manager at the third quartile has operating leases that are higher than a manager at the first quartile in the amount of 12.6 percent of total debt, and also has a percent higher expected rate of return for pension assets and a 23.5 percent higher number of quarters that meet or beat analysts forecasts by a small margin. The IQ range for FSCORE is less easily interpreted because FSCORE is a composite measure of the likelihood of accounting misstatements; however, we note that the IQ range is 0.308, which is roughly 30 percent of the mean FSCORE of 1.08 (an FSCORE less (more) than one indicates a lower (higher) probability of manipulation than the unconditional expectation). To further assess the magnitude of these differences, we also computed the IQ ranges of the firm fixed effects estimated in Table 4, panel A. These ranges are for DIS- C_ACC, for OPLEASE, for PENSION_RET, for FSCORE, and for SMBE (not tabulated). Thus, firm-specific effects exhibit greater variations than manager-specific effects, which is perhaps not surprising given that the economic circumstances facing particular firms are likely to significantly affect their accounting choices (leading to commonalities in accounting choices across time for a given firm). Still, the magnitude of the variation in CFO fixed effects is comparable to that of firm fixed effects the IQ ranges for the CFO fixed effects are between 31 percent to 62 percent of the IQ ranges of the firm fixed effects for the same variables. Thus, the style of an individual manager can impact a firm s accounting choices with magnitudes that are often close to half the impact 26. The economic magnitude of the CFO fixed effects can be seen in the variation of the individual intercepts because, to the extent managers do not exhibit a consistent style, the magnitude of the manager-specific dummy variables will be reduced (i.e., the managers accounting choices will not be consistently high or low) and there will be less variation in the individual dummy variables across managers.