Earnings Management around Founder CEO Re-appointments and Successions in Family Firms

This paper studies reappointment of a chief executive officer (CEO) and succession events in listed family firms with an incumbent family CEO. We explore whether family firms with a founder CEO are more likely to engage in earnings management preevent than other family firms. We find evidence of preevent upward earnings management for firms that reappoint their founder CEO but no for other family firms. These findings suggest that the costs and benefits from earnings management change around founder CEO reappointments in family firms. Investors, auditors, policymakers and regulators should be aware of the temptation of founder CEOs to inflate earnings preceding their reappointment.


Introduction
While the literature on earnings management is extensive and mature, Burgstahler and Chuk (2017, p. 741) nevertheless identify "[a] potentially important area for future research [which] is exploration of changes over time in costs and benefits of earnings management to meet benchmarks". We study earnings management in listed family firms with an incumbent family CEO around the re-appointment or the replacement of the family CEO. We argue that this event is a natural breaking point where the costs and benefits of earnings management to meet benchmarks are likely to be significantly different as the family is at a crossroads facing two choices. 1 First, the family can turn its firm into a firm managed by a professional non-family CEO, likely combined with gradual divestment by the familyor to the very least the emergence of an arm's length relationship between the family and the firm. Second, the family can maintain the status of a family firm, i.e., a firm managed and monitored by successive generations of the family.
Based on the above, we argue that the costs and benefits of family firms to engage in earnings management change around founder-CEO re-appointments. In particular, we maintain that founder CEOs who opt for re-appointment, given their greater emotional attachment to their firm than later-generation family CEOs, engage in earnings management to report good performance in the year preceding their re-appointment. This will ensure that minority shareholders are less likely to oppose the family's ongoing control and management of the firm. While we corroborate prior evidence that overall family firms practice less earnings management compared to non-1 Prior studies in the field identify other instances of natural breaking points where the controlling shareholders have incentives to manage earnings. Hou et al. (2015), for example, find that Chinese firms with a controlling shareholder entering into performance commitment contracts manage earnings upward to achieve the pre-specified performance target when actual performance falls short. This suggests that, supporting the more general argument put forward by Burgstahler and Chuk (2017), firms with controlling shareholders have incentives to use earnings management to achieve benchmarks and these incentives are only temporal. family firms (e.g., Martin et al. 2016), we also find evidence of upward earnings management but only in the year preceding the re-appointment of the founder CEO. Martin et al. (2016) hypothesize that family firms are less likely to use earnings management given the socio-emotional wealth the family holds in its firm. They argue that, the family is keen on preserving the firm's reputation and hence does not engage in earnings management, which might lead to reputational damage. Although maintaining the firm's reputation is essential to any shareholder of a firm, protecting the firm's reputation is even more important for a family shareholder. Indeed, the family's association and identification with its firm is likely to be stronger than is the case for other, more short-term shareholders. Further, that reputation is more important for founder family firms than for later generation family firms given the founder's greater socio-emotional wealth in the firm. Martin et al. find evidence in support of their two hypotheses.
Following Healy and Wahlen (1999), we define earnings management as changes made to reported earnings by insiders to mislead certain stakeholders or to affect contractual outcomes.
There are two types of earnings management. The first type is accrual-based earnings management, which consists of insiders manipulating reported earnings via discretionary accruals (see e.g., Jones 1991;Ball and Shivakumar 2005). The second type is real earnings management, which consists of manipulating reported earnings via changing the timing and scale of various real transactions such as research and development (R&D), production, investment and financing (see e.g., Roychowdhury 2006). Our paper makes the following four contributions to the extant literature. First, it shows that the costs and benefits from earnings management change around founder CEO re-appointments in family firms (Burgstahler and Chuk 2017). We argue that the founder CEO coming up for reappointment represents a natural breaking point where the incentives of founder family firms to Electronic copy available at: https://ssrn.com/abstract=3433367 manage their earnings have changed. Our results support the earlier evidence by Martin et al. (2016) that family firms are less likely to manage their earnings. However, we also find evidence supporting our argument that founder family firms behave differently from other family and non-family firms in the year preceding the founder CEO re-appointment. Importantly, whereas Martin et al. (2016) find that overall founder family firms are less likely to engage in earnings management, we show that the exact opposite pattern applies to founder family firms whose founder CEO is up for re-appointment. 2 Our results suggest that re-appointed founder CEOs engage in upward accrual-based earnings management prior to their re-appointment. They also engage in real earnings management, as they shift earnings to the year preceding their reappointment by foregoing expenditures, such as R&D and advertising. Second, and more generally, the paper adds to the literature on earnings management around CEO turnover, a literature, which as yet disagrees as to the effects of outgoing and incoming CEOs on earnings management. We find that founder CEOs in family firms manage earnings upward in the year before coming up for re-appointment whereas other family CEOs do not practice upward earnings management. Third, to date, both the theoretical and empirical literature disagrees as to the effects of family control and ownership on earnings management. Theory predicts that there are two competing effects of concentrated ownership and control  and family ownership and control more specifically  on earnings management: the entrenchment effect and the alignment effect. According to the entrenchment effect, large shareholderssuch as familiesextract private benefits of control from their firm. Manipulating reported earnings via earnings management is one way of extracting such benefits. Hence, earnings management should be greater and earnings quality, also called earnings informativeness, should be lower in family firms. Conversely, according to the alignment effect, families reduce agency costs via their monitoring of the firm's management, which benefits all the shareholders. Families may also be keen on protecting their reputation, especially if they intend to transfer the control over the firm to future generations of the family. Our results show support for the entrenchment effect as founder CEOs who are up for re-appointment engage in accrual-based and real earnings management. Finally, studying three very different corporate governance systems, i.e., France, Germany and the UK, rather than just a single country, increases the generalizability of our results. Given the greater levels of investor protection in the UK (Djankov et al. 2008), 3 one would expect there to be less earnings management around CEO re-appointments and successions than in France and Germany. However, we find no significant differences in terms of earnings management across the three countries. This is contrary to Leuz et al. (2003) who find that earnings management is more prevalent in countries with weak investor protection. This is not surprising given that, contrary to Leuz et al. (2003), the three countries in our sample are well-developed economies, with large stock markets and a high proportion of firms crosslisted on foreign stock exchanges associated with high investor protection.
We proceed by reviewing the three strands of literature that are directly related to earnings management around CEO re-appointments and successions in family firms. The first strand of literature is on earnings management in firms with concentrated ownership and control, including family firms, whereas the second strand is on earnings management around CEO turnover in listed firms. The final strand of the literature is on the socio-emotional wealth of families in their firm. To the best of our knowledge, there is as yet no study specifically investigating earnings management around CEO re-appointments and successions in family firms. The literature review is followed by Section 3, which explains the sample selection and methodology. The next section discusses the results from the empirical analysis. Section 5 contains a number of robustness and additional tests. Finally, Section 6 concludes.

Earnings management in firms with concentrated ownership and control
As stated in the introduction, theory suggests two competing effects of concentrated ownership and control  and family ownership and control more specifically  on earnings management: the entrenchment effect and the alignment effect. Whereas the entrenchment effect results in more earnings management in family firms, the alignment effect results in less. Broadly speaking, the literature is supportive of the alignment effect while evidence for the entrenchment effect tends to be limited to firms with dual-class stock. 4 The study by Leuz et al. (2003) is an exception to the above pattern as it finds wider support for the entrenchment effect. Their study attempts to explain differences in earnings management across 31 countries. It finds that earnings management is more prevalent in countries with relatively concentrated ownership, weak investor protection and less developed stock markets. Francis et al. (2005) as well as Martin et al. (2016) study the effect of dual-class stock on earnings management. Both studies argue that the entrenchment effect is likely to be stronger the greater the deviation of cash flow rights from control rights. Francis et al. (2005) find evidence supporting their hypothesis as earnings management is greater in dual-class firms than in firms with one type of stock. Similarly, Martin et al. (2016) find that earnings management is more likely in family firms with dual-class stock.
Apart from Leuz et al. (2003) and the specific case of (family) firms with dual-class stock, most other studies find support for the alignment effect. Wang (2006) and Ali et al. (2007) study earnings management in S&P 500 companies. Both studies find that family firms have lower abnormal accruals. A more recent study by Martin et al. (2016) also investigates accrual-based earnings management in US family and non-family firms in the S&P 500 index. They find that family firms are less likely to manage their earnings than non-family firms. They also find that the negative effect of family firms on earnings management is more pronounced in founder family firms. Chen et al. (2008) study S&P 1500 firms. They find that family firms are less likely to report earnings forecasts and use conference calls with shareholders. They interpret this as evidence in favor of the alignment effect and the lower asymmetry of information between managers and shareholders in family firms. Chen et al. also report that family firms are more likely to provide earnings warnings, which suggests that families are concerned about protecting their reputation.
Finally, they also find that family ownership is better at explaining differences in the disclosure of earnings forecasts across firms than institutional ownership and other types of insider ownership (Ajinkya et al. 2005;Karamanou and Vafeas 2005). Achleitner et al. (2014) investigate earnings management in German family and non-family firms. They study not only accrual-based earnings management, the focus of much of the empirical literature, but also real earnings management. They hypothesize that family firms, given the socio-emotional wealth the family attach to their firm, are less likely to engage in real earnings management, which is not purely paper based and therefore more likely to harm future firm value, than non-family firms. Achleitner et al. find support for their hypothesis for a sample of German family firms as such firms are less likely to engage in real earnings management. This is in direct contrast to their findings for non-family firms, which are likely to engage in both types of earnings management, if they manage their earnings.
Electronic copy available at: https://ssrn.com/abstract=3433367 Murphy and Zimmerman (1993) propose three hypotheses as to why the outgoing and incoming CEO may engage in earnings management around CEO changes. First, outgoing CEOs approaching retirement may be tempted to boost reported earnings in their final years to increase their performance linked bonuses. This is the short-term horizon hypothesis. Second, CEOs of poorly performing firms, who are in danger of being dismissed, may manage earnings upwards to mask the firm's declining performance. 5 This is the cover-up hypothesis. Third, incoming CEOs may 'take a big bath' by making major write-offs to boost performance in subsequent years and to show investors that they perform better than the outgoing CEO. This is the big bath hypothesis. 6 Murphy and Zimmerman (1993) do not find any support for the short-term horizon hypothesis. However, they find evidence in support of the cover-up hypothesis and some support for the big bath hypothesis. DeAngelo (1988) studies the effect of proxy contests on earnings management. Importantly, she finds that the incumbent managers use upward earnings management to gain support from the shareholders. This is further evidence in favor of the cover-up hypothesis. In addition, dissenting shareholders who are elected to the board tend to make massive write-offs, blaming the previous management for bad decisions, followed by earnings increases the following year. This supports the big bath hypothesis. 5 This argument is further supported by the finding reported by Kaplan and Minton (2012) that firm performance is a strong determinant of internal (board initiated) CEO turnover in the US in terms of forced as well as voluntary CEO turnover. 6 Brickley et al. (1999) advance a fourth reason for earnings management around CEO changes (see also Reitenga and Tearney 2003). They argue that retiring CEOs may engage in upward earnings management to secure nonexecutive positions on their firm's board and/or other firms' boards post-retirement. We refer to this hypothesis as the labor market hypothesis. In untabulated results, we find evidence in support of this hypothesis as departing CEOs who also chair the board pre-event practice upward earnings management to maintain their chair position of the board.

Earnings management around CEO turnover
More generally, Pourciau (1993) looks at earnings management around all non-routine, i.e., forced, changes of top executives in US firms during 1985-1988. 7 She finds that, contrary to previous studies, the outgoing executives use negative accruals and write-offs in their last year.
Further, the incoming executives make large write-offs and record large special items during the year of the executive change whereas they record positive accruals in the following year. Wells (2002) confirms Porciau's (1993) results for Australia: incoming CEOs engage in downward earnings management in the year of the CEO change and this negative earnings management is strongest for non-routine CEO changes. Hence, both Pourciau (1993) and Wells (2002) provide further support for the big bath hypothesis.
Further, Davidson et al. (2004) compare announcements of duality creating successions, whereby the incumbent CEO is replaced by a new CEO who also chairs the board, to announcements of non-duality creating successions for the case of US firms. They find that there is a greater likelihood of earnings management following the former rather than the latter. The authors also find that earnings management is more likely to occur if past firm performance has been poor. Davidson et al.'s results not only provide further support for the cover-up hypothesis, but they also suggest a greater likelihood of cover-up of bad performance when private benefits of control are large as evidenced by duality.
Finally, Hazarika et al. (2012) suggest that forced CEO changes are more likely following earnings management and that it is the amount rather than the direction of the earnings management that increases the likelihood of a forced CEO change. They find that this pattern holds even after adjusting for financial performance. The authors interpret the results as evidence that boards of directors punish CEOs engaging in aggressive earnings management given its costs (i.e., reduced earnings quality). 8

Founder CEOs, socio-emotional wealth and earnings management
Socio-emotional wealth refers to the pursuit by family firms of objectives other than purely financial objectives (see e.g., Cruz et al. 2011;Gómez-Mejía et al. 2007;Gómez-Mejía et al. 2011;Martin et al. 2016). Gómez-Mejía et al. (2007) argue that families are willing to take risk to preserve their socio-emotional wealth. If this socio-emotional wealth in the firm is considerable, families may be willing to engage in significant risk to preserve this wealth.
Importantly, Martin et al. (2016, p. 457) argue that the emotional attachment to the firm is much stronger for founders than for their descendants as "they have toiled to launch and develop the firm". 9 Combining these two arguments, we argue that founder CEOs who are up for reappointment and intend on remaining as the CEO are more likely to use earnings management to improve net earnings in the year(s) preceding the re-appointment. The reasons are as follows.
First, reporting low earnings may increase pressure from minority shareholders for the founder CEO to step down and appoint a successor, ideally a professional, non-family manager who may improve the firm's performance. Second, founder CEOs intend on being re-appointed are likely to have in mind major projects enhancing the future success of the firm, which require their continuation in their position as CEO. In other words, their ongoing emotional and economic investment in the firm is such that having to step down as CEO would result in a major loss to 8 We adjust for the potential endogeneity of earnings management in Section 5.1. 9 A somewhat different level of attachment of the founder CEO and subsequent generations is identified by Fan et al. (2012). The authors argue that founder CEOs create specialized assets that are highly valuable to the firm, but that cannot be easily transferred across individuals or organizational frontiers. Such specialized assets include implicit contracts with other family members, whereby the latter provide important services to the firm without the need for formal contracts. Although the sons and daughters are the best successors to inherit the founder's reputation and networks, they often fail to preserve the entirety of the specialized assets after the succession. They also have less emotional attachment to the firm and, as a result, a firm under the control of an heir is likely to shift to arm'slength contracts that rely less on personal networks. their investment. Hence, they are keen on reducing that probability. This leads us to our main hypothesis: HYPOTHESIS: Founder CEOs who are up for re-appointment are more likely to use upward earnings management to ensure support for their re-appointment.

Sample selection and data sources
Appendix 1 provides details on the sample selection process. We start with the full population of domestic firms (including active, dead and suspended firms) listed in France, Germany and the UK from 2001 to 2016, which comprises 2,679 French firms, 2,352 German firms and 7,747 UK firms. We then apply a series of filters. First, all financial firms with a Datastream Industry Classification Benchmark (ICB) code 8000 are excluded, because they follow different reporting conventions, which make comparisons with non-financial firms difficult. They are also subject to additional governance and reporting requirements. Second, following prior studies (e.g., Achleitner et al. 2014), we exclude firms that have only their preference shares listed, i.e., held by outsiders. As the holders of the preference shares have no voting rights but have the right to a fixed guaranteed dividend − typically expressed as a percentage of the nominal share value − they are less likely to be concerned and influenced by reported earnings. 10 Third, given that all the measures of earnings management used in this paper necessitate data on total assets, we require that each firm has such data for at least one year during the 2001-2016 period. The remaining sample includes 1,384 French firms, 1,269 German firms and 3,568 UK firms. Fourth, given the aim of the study, we distinguish between family and non-family firms. Following Ansari et al. (2014), a family firm is defined as a firm with a family CEO as well as with a family owning at least 25% of the votes and remaining the largest shareholder for at least half of the period of the study. 11 The latter three selection criteria result in 187 French, 120 German and 88 UK family firms. The remaining firms are classified as non-family firms. 12 Fifth, given the focus of this paper, we retain only those family and non-family firms in the sample with at least one event, i.e., one change in the CEO or at least one re-appointment of the incumbent CEO during The corporate governance characteristics for the sample firms and the biographies of the incumbent CEO as well as the successor CEO are collected from the company reports, Reuters, Thomson One Banker, company websites and country-specific company guides. 14 LexisNexis, various newspaper archives, Forbes and Capital IQ are used to identify the event date. Financial information is sourced from Datastream and Osiris. 11 A CEO is considered to be a family CEO if the CEO is the founder or a descendant of the founder, the CEO's surname is identical to the firm's name, or he/she shares his/her surname with another member of the firm's board of directors. 12 Any firm with a family holding at least 25% of the votes even just for a single year during our period of study is excluded from the sub-sample of non-family firms. 13 Comparing the identity of the CEO stated in the annual reports from one year to another, we know that there was a change in the CEO in a particular year, but we are unable to identify the exact date of the succession. 14 In addition, we use Hoppenstedt Aktienführer for Germany, and Companies Handbooks for the UK.

Types of events
We distinguish between four types of events in the family firms: (a) founder re-appointments, (b) other re-appointments, i.e., re-appointments of non-founder family CEOs, (c) appointments of new family CEOs, and (d) appointments of non-family CEOs. The latter two being actual CEO changes or successions whereas the former two maintain the status quo. The events in nonfamily firms are all actual CEO changesi.e., there are no re-appointments in the sub-sample of non-family firms. A re-appointment is defined as the appointment of the incumbent family CEO for another term. 15 If the incumbent family CEO is not re-appointed, then there are two alternative succession options: appointing a new family CEO or appointing a non-family CEO. 16 Following Choi et al. (2014), we define year 0 as the event year, i.e., the first full fiscal year during which the incumbent CEO (assuming the incumbent was not re-appointed) is no longer in office. We require the new CEO to be in office during the first quarter of that year. If this is not the case, the next year is then treated as year 0. By implication, year -1 is the last fiscal year when the incumbent CEO is in office throughout the entire year.

Measurement of earnings management
As stated in the literature review, most empirical studies focus on accrual-based earnings management. We also investigate whether there is evidence of real earnings management. Our measure of accrual-based earnings management is based on Ball and Shivakumar's (2005) and 15 'Another term' is specific to the firm and may be equivalent to the period recommended in the country's respective governance code or shorter. Appendix 3 provides a description of the legal and institutional framework pertaining to re-appointments in France, Germany and the UK. It also discusses the criteria met by all the reappointments included in the sample. This description suggests that the re-appointments are genuine re-elections of the incumbent CEO rather than just a simple rubberstamping of an extension of their term. 16 A new family CEO consists of a family member of the incumbent CEO who succeeds him/her. A family member may include the spouse, child, sibling, cousin or in-law. The family relations of the family CEOs are confirmed through the IPO prospectuses. A non-family CEO is a person not related (by blood/marriage or other ties) to the incumbent family CEO who succeeds him/her. Wang's (2006) modifications of Dechow and Dichev (2002). This model is the non-linear version of the traditional, linear model by Jones (1991). The main advantage of the non-linear model is the greater explanatory power obtained from allowing for a non-linear accruals process (Ball and Shivakumar 2005;Wang 2006). 17 The approach adopted to measure accruals is similar to that used in extant literature. Accruals are determined by estimating the following equation: where: ACCt is total accruals at year t, scaled by total assets at t-1; CFt is cash flow from operations at t scaled by total assets at t-1; DCFt equals one if the cash flow from operations at t is negative, and zero otherwise; DCFt * CFt is the proxy for economic losses; and t is an error term.
Equation (1) includes a proxy for economic losses as it is likely that such losses are recognized in a timelier manner, as unrealized charges against income, whereas economic gains are more likely to be recognized only once realized as charges against cash flow (see Ball and Shivakumar 2005). Importantly, equation (1)  17 Accounting income is likely to be non-linear as it is generated by two different processes. The first one is a moving average of current and past economic gains and the second one is a more transitory, i.e., less smoothed, incorporation of economic losses (Ball and Shivakumar 2005).
More precisely, as in Ball and Shivakumar (2005), equation (1) is estimated by country, industry and year, requiring a minimum of 30 observations. For this purpose we use the Industry Classification Benchmark (ICB) from Datastream. The country-, industry-and year-specific parameter estimates are then used to obtain the normal discretionary accruals for each firm-year observation. The abnormal accruals, proxying for accrual-based earnings management, are then the residuals obtained from that equation.
To measure real earnings management, we follow Roychowdhury (2006). We employ three different measures of real earnings management. The first measure is based on the abnormal cash flow from operations. Managers are able to inflate current earnings by reducing prices, thereby shifting sales from the next fiscal year to the current one. Such strategies come at the cost of reduced future profitability. Hence, negative discretionary cash flow from operations in year 0 is evidence of upward real earnings management in the pre-succession year. Normal levels of cash flow from operations are estimated with the help of the following equation: where: CFt is cash flow from operations at t; TAt1 is total assets at the end of t1; St is sales at t; St is change in sales from t1 to t; and t is an error term.
Equation (2) is estimated for each country, industry and year, with a minimum of 15 observations as in Roychowdhury (2006). The estimated parameter values are then used to determine the normal level of cash flow from operations for each firm-year observation.
The second measure is abnormal production costs, which is defined as the sum of the cost of goods sold and the change in inventory. Another way of inflating current earnings is by overproducing goods. This will reduce the cost of goods sold per unit as the fixed costs will be allocated across a larger number of units produced. The reduction in costs of goods sold will increase the reported profit margins. However, as the reduction in costs of goods sold does not result in greater sales the overproduction results in greater production and inventory costs.
Hence, cash flow from operations is lower than normal. Therefore, positive abnormal production costs are another measure of upward real earnings management. Normal production costs are estimated using the following equation: A procedure similar to that used to calculate the abnormal cash flow from operations is then used to determine the abnormal production costs. 18 The third and final measure of real earnings management is abnormal discretionary expenses.
Managers can also inflate current earnings by reducing non-operating expenses such as R&D and advertising. This boost in current earnings, via the increase in current cash flow, may however come at the cost of lower future cash flow. Hence, negative discretionary expenses in year 0 would be a sign of upward real earnings management in the pre-succession year. Normal discretionary expenses are estimated using the following equation: where: DISCEXPt is discretionary expenses at t calculated as the R&D plus selling, general and administrative expenses; and t is an error term.
The procedure to determine the discretionary expenses is similar to that used for the case of the previous two measures of real earnings management. 19 We then estimate the following ordinary least squares (OLS) regression model to test our main hypothesis. where: EARNINGS MANAGEMENTt is one of the four earnings management measures at t, i.e., accrual-based earnings management, abnormal cash flow from operations, abnormal production costs or discretionary expenses; EVENT TYPEi is one of the dummy variables indicating the event type(s), such as founder re-appointment, our main variable of interest; FOUNDER CEOi,-1 equals one if the CEO is the founder of the firm, and zero otherwise; lnTAt is the natural logarithm of total assets at t; ROAt is the return on assets at t; LEVERAGEt is debt over total assets at t; BOOK-TO-MARKETt is the book value of equity divided by the market value of equity at t; LOSSt equals one if net income is negative, and zero otherwise; BIG FOURt is one if the auditor belongs to one of the top four audit firms in year t, and zero otherwise; and t is an error term.
In the regressions estimated on the entire sample of family and non-family firms, we also include FAMILY FIRM DUMMYi, which equals one if the firm is a family firm, and zero otherwise.
To account for the entrenchment effect on earnings management the following six measures of private benefits of control are used: family wedge (which would adjust for the presence of dualclass shares, including the proportion of non-voting shares), board independence, CEO-chair duality, duality destroying event, founder CEO, and departing founder CEO on board post-event.
We expect that the greater the private benefits of control, the greater is the earnings management in family firms around the founder CEO re-appointments and successions.

Descriptive statistics
Panel A of Table 1 reports the distribution of the firms across the three countries. While the majority of family firms are from France, the vast majority of non-family firms are from the UK.
These percentages reflect the relatively greater concentration of control, including family control in France, and the relatively greater percentage of widely held firms in the UK. Panel B shows the distribution across the three countries of the 793 events. The total number of events in the 614 firms is 793. Out of the 614 firms, 467 firms are involved in one event, 120 are involved in two events during the period of study, 23 firms in three events, three firms in four events and one firm in five events (not tabulated). The most frequent type of event in the family firms is founder re-appointments, followed by the replacement of the family CEO by a non-family CEO.
Across the three countries, the only slight exception to this rule is France where the second most important type of event is the re-appointment of the non-founder family CEO. 22 [Insert Table 1 about here] Four auditor compared to slightly less than the majority for the family firms.
[Insert Table 3 about here]

Accrual-based earnings management
In what follows, we estimate a number of OLS regressions to test the validity of our main hypothesis that founder CEOs who get re-appointed are more likely to manage their firm's earnings in the year preceding their re-appointment. Before proceeding with this test, we run regressions in Table 4 comparing family firms with non-family firms. The regressions, which are based on equation (1), are estimated separately for each of the five years centered on year 0, i.e., the first full fiscal year during which the incumbent CEO (assuming the incumbent was not re-appointed) is no longer in office.
The results show evidence of earnings management in non-family firms in the years following the appointment of the new CEO as the constant is significant at the 5% level or better in years 0, 1 and 2. With one exception, the regressions also suggest that family firms are less likely to manage earnings when compared to non-family firms. Indeed, the family firm dummy is negative and significant in all of the five years, except for year -1. The fact that the dummy is not significant in year -1 suggests that family firms behave differently in the year preceding the re-appointment or the succession of the family CEO.
It is also noteworthy that the founder CEO dummy is significant at the 10% level in year -1 while not being significant in the other years. To sum up, similar to extant research (e.g. Martin et al. 2016), we find that family firms are less likely to use earnings management than non-family firms.
[Insert Table 4 about here] Next, Table 5 distinguishes between founder re-appointments and other events in family firms while the base case is still non-family firms as in Table 4. The table suggests that firms with a founder CEO who is re-appointed manage their earnings upwards in year -1 while the remaining family firms do not. Indeed, the dummy variable for events other than the CEO founder reappointment in family firms is not significantly different from zero in year -1 while it is negative and significant in years -2, 0 and 1. Taken together, these results suggest again that family firms are less likely to manage earnings when compared to non-family firms. However, in year -1 family firms are no different from non-family firms, except for family firms that re-appoint their CEO founder. Again, the latter are more likely to manage their earnings upwards in year -1. This trend is clearly depicted in Figure 1 which shows upward accrual-based earnings management by the re-appointed founder CEO in year -1. Hence, these regression results and diagrammatical evidence provide support for our main hypothesis that founder CEOs who are up for reappointment engage in upward earnings to ensure their re-appointment. Next, similar to Table   4, Table 5 provides evidence that non-family firms manage earnings in the years following the appointment of the new CEO. Finally, our measures of private benefits of control do not affect accrual-based earnings management. Once more, this is in line with the results from Table 4.
[Insert Table 5 and Figure 1 about here] Table 6 focuses on the family firms only. It also distinguishes between founder re-appointments, other re-appointments (i.e., re-appointments of non-founder family CEOs) and the appointments of new family CEOs. The base case is appointments of non-family CEOs (in family firms). We now find strong evidence (at the 1% level of significance) of earnings management in both years -2 and -1 for firms re-appointing their founder CEO. This provides further support for our main hypothesis. Similarly, there is also evidence (albeit at lower significance levels) of upward earnings management in years -2 and -1 for firms appointing a new family CEO. In contrast, firms appointing a non-family CEO are less likely to manage their earnings in year -1 as reflected by the negative and significant constant.
We also find that the departing CEOs in duality destroying events inflate earnings upwards preevent. It is possible that the departing CEOs, who also act as the chair of the board of directors pre-event, inflate earnings upwards in order to ensure a seat on the board of directors and maintain oversight over the successor CEOs post-event. These incentives are even greater when the departing family CEO is succeeded by a non-family CEO as this is the first time the firm is managed by a non-family member. Untabulated results show that 94% of the duality destroying events result in the departing CEO remaining on the board post-event (in 88% of the cases the CEO assumes the chair position post-event) and, in the vast majority of the events, the departing CEOs who act as the chair pre-event are replaced by non-family CEOs. 27 Further regression analysis (not tabulated) shows that the significant results reported for the duality destroying event dummy in years -2 and -1 are driven by the departing CEOs, who also act as chairs preevent, and are replaced by non-family CEOs. Hence, these departing CEOs, who also act as the chair of the board in year -1, engage in upward earnings management pre-event to ensure oversight of the succeeding non-family CEO.

Real earnings management
Tables 7, 8 and 9 are the equivalent of Table 6, but focus on real earnings management, i.e., abnormal cash flow from operations, abnormal production costs and abnormal discretionary expenses, respectively. 28 Similar to Table 6, these tables compare events in family firms while distinguishing between founder re-appointments, other re-appointments and new family CEO.
Hence, the base case is appointments of non-family CEOs (in family firms). As a reminder, negative abnormal cash flow, positive abnormal production costs and negative discretionary expenses are evidence of earnings-increasing real earnings management. Table 7 suggests that firms re-appointing their CEOs (founder CEOs as well as non-founder CEOs) shift earnings from year 0 to year -1 as reflected by the negative and significant dummies on CEO founder re-appointments and other re-appointments in year 0. In contrast, Table 8 does not suggest the presence of abnormal production costs in the immediate aftermath of CEO reappointments or other events in family firms. However, there is evidence in Table 9 of positive rather than the expected negative abnormal discretionary expenses in year 0 for firms reappointing their founder CEO and non-family CEOs (the base case). The positive discretionary expenses suggest that founders intend on investing in projects, such as R&D or advertising, hold back from making these investments until their re-appointment. This pattern is further supported by the fact that the dummy variable indicating whether the departing founder CEO remains on the board post-event is positive and significant at the 10% in Table 9. This suggests that founder CEOs who stay on the board invest in projects that they have refrained from considering preevent. 29 [Insert Tables 7, 8 and 9 about here] Pulling together the evidence from all of the regression tables (i.e., Tables 4 to 9), there is evidence that founder CEOs who are re-appointed manage earnings upwards in the year preceding their re-appointment. Hence, there is strong and consistent support for our main hypothesis.

Endogeneity concerns
This paper studies earnings management around CEO re-appointments and successions. This focus implicitly assumes that the actual CEO changes happen irrespective of whether there is earnings management or not, and irrespective of the type and/or direction of earnings management. However, it might be the case that the CEO changes are not exogenous.
Importantly, Hazarika et al. (2012) suggest that forced CEO changes are more likely following earnings management and that it is the amount rather than the direction of the earnings management that increases the likelihood of a forced CEO change. They find that this pattern holds even after adjusting for financial performance. They interpret this as evidence that boards 29 We also find evidence that the incoming CEOs in duality destroying events inflate earnings post-event via abnormal discretionary accruals. As stated in footnote 24, there are 35 CEOs associated with duality destroying events. Twenty-two out of the 35 are replaced by a non-family CEO and the remaining 13 are replaced by a new family CEO. Given the small number of cases, any further analysis is not representative.
of directors punish CEOs engaging in aggressive earnings management given its costs (i.e., reduced earnings quality). Hence, it is important to adjust for the potential endogeneity of earnings management. To address this issue we run logit regressions using the events in family firms and the forced succession dummy variable as the dependent variable, the absolute value of our measures of earnings management in year -1 or year -2, and country, industry and year dummies as the independent variables. The regressions (not tabulated) show that all our measures of absolute earnings management in year -1 as well as year -2 are insignificant, suggesting that our results are not driven by reverse causality.
Also, the motives of the controlling family to retain control of the firm are idiosyncratic, unobservable, and may be correlated with the decision to manage earnings. To alleviate these further endogeneity concerns, we utilize propensity score matching (PSM) to match events in family firms with those in non-family firms (Rosenbaum and Rubin 1983). To achieve this, we first run a logit regression using the family firm dummy variable as the dependent variable, while including our measures of private benefits of control as well as the control variables used in the regressions reported in Table 5 on the right-hand side. 30 Again, the family firm dummy variable takes a value of one for the family firms, and zero otherwise. All independent variables are measured in the year preceding the event (i.e., year -1). The logit regression is estimated using all the events in the family and non-family firms (i.e., 307 and 486 events respectively). Second, a propensity score is then generated for all the events. Our aim is to seek a match for events in family firms with those in non-family firms in the same industry (using the Industry Classification Benchmark from Datastream). Hence, next we generate an amended propensity score to account for the industry-specific characteristics. 31 Using the latter score we then match 30 We do not include the family wedge dummy as this is a family firm characteristic only. 31 The propensity scores obtained from the previous step are transformed via the following formula: Amended Propensity Score = Industry Code * 10 + Propensity Score. No matching by year is required as we use relative years in the analysis.
the events based on the nearest neighbor approach with a maximum caliper distance of 0.1 in order to minimize the systematic differences between the matches. Our measures of earnings management in year -1 are used as outcomes in the matching process. This approach allows us to match the events in terms of their propensity score while preserving the industrial closeness of the family and non-family firms. Given the use of earnings management in year -1 as the outcome in the matching process, we end up with four different matched samplesone for each measure of earnings management. The pre-matching logit and the four post-matching logits are reported in Table 9.
Following the above procedure (see Table 10), we successfully match 134 of the 307 events in family firms for the accrual-based earnings management, 142 events for the abnormal cash flow from operation, 145 for abnormal production costs and 98 events for the abnormal discretionary accruals. The pre-matching logit has a high pseudo-R 2 with a value of 0.272 and most of the independent variables have statistically significant coefficients. However, as expected, the postmatching logits on the matched sample reported in the last four columns of Table 10 have little explanatory power as their pseudo-R 2 is close to zero and all the coefficients on the independent variables are insignificant. We also use mean and median comparisons to test for differences between the events in family and matched non-family firms for each of the explanatory variables included in the propensity score logit regression. Untabulated results show that there are no statistically significant differences between the two samples. Hence, we are confident that our matching is of good quality.
Next, we re-run the OLS regressions reported in Table 5 on the matched sample of events in family and non-family firms. The regression results reported in Table 9 provide further, strong evidence for our main result that re-appointed founders engage in upward accrual-based earnings management in the year preceding the event. The regression coefficient is positive and significant at the 5% level for year -1 and insignificant in all the remaining years. Untabulated results using propensity score matching also show that, in line with our main results related to real earnings management, re-appointed founders shift earnings from year 0 to year -1 in order to portrait better performance in the pre-event year. Hence, we find strong support for our findings using propensity score matching.  Finally, only six out of the 307 events in family firms relate to the death of the CEO and in all six events is the family CEO replaced by another family member. The results are upheld when we exclude these events from the sample.

Incumbent and successor CEO characteristics
We further verify the robustness of the results by including in the regressions reported in Tables 6 to 9 the age and tenure of the incumbent CEO and the age of the successor CEO (not tabulated). 32 These variables are overall insignificant in all the regressions, whether using accrual-based or real earnings management as the dependent variable. Importantly, our main results are upheld.

Number of times a founder CEO is re-appointed
It may be the case that founder CEOs who are re-appointed more than once are more likely to engage in earnings management. In other words, such CEOs may be re-appointed because of successfully manipulating earnings. To address this possibility, we generate a dummy variable, which equals one if the firm re-appoints its founder CEO more than once, and zero otherwise.
We then include this dummy variable in the regressions in Tables 6 to 9. Twenty-six out of the 241 family firms re-appoint their founder CEO twice during the period of study, five firms reappoint their founder CEO three times and another two firms re-appoint their founder CEO four times. Our main results are upheld after controlling for the number of times a founder CEO is re-appointed.

Market reaction and earnings management
Our results suggest that firms that re-appoint their founder CEO manage earnings upwards in year -1. However, is the stock market fooled by this earnings manipulation? To answer this question, we compute cumulative abnormal returns (CARs) around the announcement of these events. The CARs are based on daily data for the market model, where day 0 is the day of the event announcement. 33 The parameters of the market model are estimated from day -270 to day -20. The STOXX Europe600 index is the proxy for the market portfolio. 32 We also attempted to collect data on the incumbent CEO's gender and education. Of the 307 events in family firms, only six involve a female CEO. The data on education (e.g., university degree) proved to be difficult to obtain and we were able to obtain this information for only 66 events out of the 307 events. 33 We use LexisNexis, the Forbes database and newspaper archives to identify the announcement date of each founder CEO re-appointment. Wherever possible, the date is confirmed using more than one source.
Untabulated results suggest that there is no significant market reaction, as measured by CAR [-1,1] and CAR [-3,3], to the announcement of founder CEO re-appointments as well as other events in family firms. The mean and median differences in abnormal returns across the two types of events (i.e., founder CEO re-appointments and other events in family firms) are also insignificant. However, further untabulated results show that there is a positive market reaction to the announcements of founder CEO re-appointments, but only when the incumbent CEO engages in positive accrual-based earnings management in year -1. This suggests that founder CEOs who are up for re-appointment successfully manipulate accrual-based earnings upwards, triggering a positive short-term market reaction to the announcement of their reappointment.

Does it matter if the CFO is part of the family?
Prior evidence shows that chief financial officers (CFOs) have strong incentives to engage in earnings management. 34 It is possible that such incentives are even stronger if the CFO is connected to the controlling family, which will further enhance the ability of the founder CEO to manipulate earnings upwards prior to the re-appointment. Hence, we explore whether it matters whether the CFO in year -1 is a member of the controlling family or has ties with it. 35 Untabulated results suggest that 108 out of the 307 events in family firms have a CFO in year -1. Only ten out of the 108 CFOs are part of the family and 29 CFOs are not related to the family but have ties with it; the remaining CFOs are neither related to the family nor have any other ties to it. The majority of the events that take place in firms with a CFO related to the family or having ties with the family (34 out of 39 events) result in the re-appointment of the founder CEO 34 Jiang et al. (2010), for example, show that the magnitude of accruals and the likelihood of beating analyst forecasts are more sensitive to CFO equity incentives than to those of the CEO. Further, Baker et al. (2018) conclude that the power of the CEO relative to the CFO is an important factor in the both the type and magnitude of earnings management. 35 The CFO is considered to have ties with the family if he/she has been a director/executive in the same firm for more than nine years or he/she serves on other boards with the family director(s).
(20 events), the re-appointment of the non-founder CEO (9 events) and the appointment of a new family CEO (5 events). A dummy variable, which takes the value of one if the CFO is part of the family or has ties to it, and zero otherwise, is included in all the regressions from Tables 6 and 9. Untabulated results suggest that this dummy variable is insignificant in all, except in two regressions. Importantly, our main results are upheld.

Further tests
It is plausible that firms, which re-appoint their founder CEO, are younger and also have greater risk. To control for these two firm characteristics, we re-run the regressions reported in Tables 6 to 9 using firm age and the standard deviation of earnings per share based on years -4 to -1.
Untabulated regression results suggest that firm age and the standard deviation of earnings per share in years -4 to -1 are insignificant in most of the regressions and our main results are upheld.
Finally, given the importance of market returns in the CEO succession literature (e.g., Kaplan and Minton 2012), we replace the return on assets with the annual stock performance. Stock performance is measured as the annual change in the total return index of the stock divided by the total return index of the stock in the previous year. Untabulated results show that the latter is insignificant in all the regressions reported in Tables 6 to 9 and our main results are upheld.

Conclusion
This paper studies accrual-based earnings management as well as real earnings management around the re-appointment and the replacement of the incumbent family CEO in family firms from France, Germany and the UK. Our definition of a family firm is a firm with a family as its largest shareholder and holding at least 25% of the votes. Given the aim of the study, we only retain family firms whose incumbent CEO is a member of the controlling family.
We argue that the costs and benefits of family firms to engage in earnings management change around the re-appointment of the founder CEO in family firms. More specifically, we hypothesize that founder CEOs who opt for re-appointment have greater incentives, compared to later-generation family firms, to engage in earnings management around their re-appointment. This is the case for the following reasons. First, founder CEOs who opt for re-appointment will engage in upward earnings management in order to report good performance in the year preceding their re-appointment. This will ensure that minority shareholders are less likely to oppose the family's ongoing control and management of the firm. Second, founder CEOs intend on being re-appointed are likely to have in mind major projects enhancing the future success of the firm, which require the continuation of their position as CEO. In other words, their ongoing emotional and economic investment in the firm is such that having to step down as CEO would result in a major loss to their investment. In line with this argument, we find evidence of positive earnings management in family firms around the re-appointment of the founder CEO. To the best of our knowledge, this is the first study on earnings management by founder CEOs around their re-appointment.
We find strong and consistent support for our hypothesis. In contrast to other family CEOs as well as CEOs in non-family firms, founder CEOs practice accrual-based earnings management in the year preceding their re-appointment. We also find some evidence of real earnings management by re-appointed founder CEOs suggesting that founders shift earnings from year 0 to year -1 and also refrain from investing in projects, such as R&D or advertising, until after their re-appointment.
Importantly, we do not find evidence that overall family firms are more likely to practice earnings management than non-family firms, confirming extant literature. Our evidence provides support for the statement by Burgstahler and Chuk (2017, p. 741) that there might be "changes over time in costs and benefits of earnings management to meet benchmarks". This suggests that CEO re-appointments and successions in family firms trigger such changes.
Our paper generates important policy implications: Investors in family firms, as well as internal and external auditors and policy makers should be aware of the temptation of founder CEOs to engage in upward earnings management preceding their re-appointment.

Notes:
Panel A reports the distribution across countries of the 241 family firms and 373 non-family firms. Panel B report the distribution of the events across the three countries and firm type. Out of the 614 firms, 467 firms are involved in one event, 120 in two events during the period of study, 23 firms in three events, three firms in four events and one firm in five events.

Notes:
This table reports the distribution across industries of the 241 family and the 373 non-family firms. The reported numbers are based on the first event for the firm during the period of study and are measured in the year before the event (i.e., year -1). 120 out of the 614 firms are involved in two events during the period of study, 23 firms in three events, three firms in four events and one firm in five events. The significance of the differences in percentages is assessed using a test for differences in proportions.

Notes:
Panel A of this table reports summary statistics for the 241 family firms and Panel B for the 373 non-family firms. The reported numbers in both panels are based on the first event for the firm during the period of the study and are measured in the year before the event (i.e., year -1). ROA, leverage, book-to-market and the standard deviation of earnings per share pre-event (based on the years -4 to -1) are winsorized at the 2nd and the 98th percentiles. 120 out of the 614 firms are involved in two events during the period of study, 23 firms in three events, three firms in four events and one firm in five events. The significance levels for the differences in means and medians across the two panels are reported in Panel A. The differences in means are assessed using a t-test whereas the differences in medians are assessed using a z-test (Wilcoxon rank-sum test).

Notes:
This table reports the OLS regressions for accrual-based earnings management on family firm dummy, the measures of private benefits of control as well as a set of control variables measured in year t. The family firm dummy variable takes a value of one for family firms, and zero otherwise. The six measures of private benefits of control are: the family wedge dummy, board independence, duality, duality destroying event, founder CEO and departing founder CEO remains on board post-event. The control variables, except for the loss, Big Four, country and year dummies, are winsorized at the 2nd and 98th percentiles. The regressions are based on the 793 events. Due to missing values, the actual number of observations in all the regressions is smaller than 793. Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively.

Notes:
This table reports the OLS regressions for accrual-based earnings management on event type dummies in family firms, the measures of private benefits of control as well as a set of control variables measured in year t. The founder reappointment dummy takes a value of one if the founder CEO is reappointed, and zero otherwise. The other events in family firm dummy takes a value of one for all the remaining events in family firm, and zero otherwise. All the events in non-family firms are actual changes in CEOs. The five measures of private benefits of control are: the family wedge dummy, board independence, duality, duality destroying event and departing founder CEO on board post-event. The control variables, except for the loss, Big Four, country and year dummies, are winsorized at the 2nd and the 98th percentiles. The regressions are based on the 793 events. Due to missing values, the actual number of observations is smaller than 793. Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively. This table focuses on the 307 events in family firms only. It reports the OLS regressions for accrual-based earnings management on the event type dummies, the measures of private benefits of control as well as a set of control variables measured in year t. The regressions compare founder re-appointments, other re-appointments, new family CEO appointment with non-family CEO appointment. The founder re-appointment dummy takes a value of one if the existing founder CEO is reappointed, and zero otherwise. Other re-appointment dummy takes a value of one if the non-founder CEO is reappointed, and zero otherwise. The new family CEO is set to one if a new family CEO is appointed, and zero otherwise. The five measures of private benefits of control are: the family wedge dummy, board independence, duality, duality destroying event and departing founder CEO on board post-event. The control variables, except for the loss dummy, Big Four, country and year dummies, are winsorized at the 2nd and the 98th percentiles.
Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively.  This table reports the OLS regressions for the abnormal production costs on the event type dummies in family firms, the measures of private benefits of control as well as a set of control variables measured in year t. The regressions compare founder re-appointments, other re-appointments, new family CEO appointment with non-family CEO appointment. The founder re-appointment dummy takes a value of one if the existing founder CEO is reappointed, and zero otherwise. Other re-appointment dummy takes a value of one if the non-founder CEO is reappointed, and zero otherwise. The new family CEO is set to one if a new family CEO is appointed, and zero otherwise. The five measures of private benefits of control are: the family wedge dummy, board independence, duality, duality destroying event and departing founder CEO on board post-event. The control variables, except for the loss dummy, Big Four, country and year dummies, are winsorized at the 2nd and the 98th percentiles. Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively.  These regressions report the estimation results of the logit underlying the propensity score matching for the pre-matching sample and the post-matching sample when using accrual-based earnings management and real earnings management in year -1 as an outcome in the matching process. The dependent variable is a family firm dummy which takes a value of one for family firms and zero otherwise. The control variables, except for the loss dummy and Big Four are winsorized at the 2nd and the 98th percentiles. Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively. This table reports the replicated OLS regressions for accrual-based earnings management reported in Table 5 using propensity score matching. The founder re-appointment dummy takes a value of one if the founder CEO is reappointed, and zero otherwise. The other events in family firm dummy takes a value of one for all the remaining events in family firm, and zero otherwise. All the events in non-family firms are actual changes in CEOs. The five measures of private benefits of control are: the family wedge dummy, board independence, duality, duality destroying event and departing founder CEO on board post-event. The control variables, except for the loss, Big Four, and year dummies, are winsorized at the 2nd and the 98th percentiles. Year 0 is the event year, i.e., the first full fiscal year during which the incumbent CEO is no longer in office. The t-values presented in parentheses are heteroscedasticity consistent and the standard errors are clustered by firm. *, ** and *** stand for statistical significance at the 10%, 5% and 1% level, respectively.

FIGURE 1
The dynamics of the mean accrual and real earnings management around the re-appointment of a founder CEO in family firms

Notes:
This figure plots the mean accrual and real earnings management for the re-appointments of the founder CEO. Year 0 is the event year, i.e., the first full fiscal year following the re-appointment of the founder CEO.

Notes:
This table provides details about the sample selection. We start with the full population of firms listed in France, Germany and the UK from 2001 to 2016, which comprises 2,679 French firms, 2,352 German firms and 7,774 UK firms. We then exclude the financial firms (with an ICB code of 8000), firms with their preference shares listed only and missing total assets for all 16 years of the study. This sample is then used to estimate the measures of earnings management discussed in Section 3.2. We distinguish between family and non-family firms. A family firm is defined as a firm with an incumbent family CEO as well as with a family owning at least 25% of the votes and remaining the largest shareholder for at least half of the period of the study. We retain only those family and non-family firms with at least one change in the CEO or at least one re-appointment of the incumbent CEO during 2001 and 2016. Finally, after dropping the firms without a clear-cut succession date (i.e., those firms whose annual reports suggest there was a change of the CEO in a particular year, but we are unable to identify the exact succession date), we end up with a final sample of 621 firms of which 248 are family firms and the remaining 373 firms are non-family firms. Deviation of the actual accruals from their predicted value estimated using the corresponding country-industry-year regressions. = 0 + 1 + 2 −1 + 3 +1 + 4 + 5 * +

Cash flow from operation (CF)
Earnings before exceptional and extra-ordinary items less accruals. Datastream Deviation of the actual cash flow from operations from the predicted value estimated using the corresponding country-industry-year regressions.

Discretionary expenses (DISCEXPt)
R&D expenses plus selling, general and administrative expenses. R&D expenses are assumed to be zero, if information for this item is missing and the information on selling, general and administrative is available.

Types of events
Family firm dummy A dummy variable, which equals one if the event took place in a family firm, and zero otherwise. A family firm is defined as a firm with a family CEO succession as well as with a family owning at least 25% of the votes and remaining the largest shareholder for at least half of the period of the study.

Annual reports, IPO prospectuses, Capital IQ
Non-family firm dummy A dummy variable, which equals one if the event took place in a nonfamily firm, and zero otherwise.

Annual reports, Capital IQ
Founder re-appointment in family firm A dummy variable, which equals one if the founder CEO is reappointed for a further period of time, and zero otherwise.

Annual reports, LexisNexis
Other re-appointment in family firm A dummy variable, which equals one for all the remaining events in a family firm except the re-appointment of a founder CEO, and zero otherwise.

New family CEO in family firm
A dummy variable, which equals one if the incumbent family CEO in a family firm is replaced by another family member, and zero Annual reports, IPO prospectuses otherwise. A family member may include the spouse, child, sibling, cousin or in-laws.

Non-family CEO in family firm
A dummy variable, which equals one if the incumbent family CEO in a family firm is succeeded by a person not related (by blood/marriage or other ties) to the latter, and zero otherwise.

Measures of family power
Founder CEO A dummy variable, which equals one if the founder (the person that founded the firm) is the incumbent CEO, and zero otherwise.

IPO prospectuses, Osiris, annual reports
Departing founder CEO remains on board postevent A dummy variable, which equals one if the departing founder CEO is still on the board of directors in year 0, and zero otherwise. For German firms and French firms with two boards, the departing founder CEO moves to the supervisory board.

IPO prospectuses, Osiris, annual reports
Family wedge Excess of family control over family ownership. Family control is measured as votes held by the family shareholders plus any additional votes resulting from pyramidal ownership (measured by the weakest link in the chain of control) as a percentage of votes outstanding in year -1. Family ownership is calculated as the number of shares of all classes held by the family as a percentage of total shares outstanding in year -1. The numerator also includes shares held by family representatives.

Board independence
The proportion of non-executive directors on the board as reported in the annual report for year -1.

Annual reports, IPO prospectuses
Duality A dummy variable, which equals one if the CEO also acts as the chair of the board in year -1, and zero otherwise. German company law (paragraph 105, AktG) prevents duality and hence this variable is always zero for Germany.

Annual reports
Duality destroying event A dummy variable, which equals one if there is CEO-chair duality in year -1 but there is no CEO-chair duality in year 0, and zero otherwise.

Control variables
Total assets Total assets of the firm. Datastream

Return on assets
Earnings before interest, tax, depreciation and amortization divided by total assets.

Leverage
Total debt divided by total assets. Datastream

Book-to-market
Book value of equity divided by the market value of equity. Datastream

Loss
A dummy variable, which equals one if net income before extraordinary items is negative, and zero otherwise.

Own calculations
Big Four A dummy variable, which equals one if the auditor belongs to one of the top audit firms, and zero otherwise.

Thomson One Banker
Firm age Number of years the firm has been incorporated. Own calculation, Datastream Standard deviation of EPS pre-event Standard deviation of earnings per share between year -4 and year -1. Own calculation, Datastream Notes: This table presents the definitions of the variables used in the paper.
Appendix 3 Legal and institutional framework for France, Germany and the UK used to identify the CEO re-appointments

Country
Legal and institutional framework Criteria satisfied to be included in the sample

France
The French Commercial Code (Code de Commerce) stipulates that the maximum term of a director in the office is six years. It also states that directors are eligible for re-election unless otherwise specified in the Memorandum and Articles of Association (Article L225-18). Exceptions are permitted only when they are made in accordance with the conditions specified in Article L225-24 of the Code.
All the French re-appointments (or re-elections) of the CEO included in the sample satisfy the legal requirements stated in the French Commercial Code. They are also approved through the formal procedures, i.e. subject to majority vote at an annual general meeting (AGM).

Germany
The German Stock Corporation Act (AktG) states that members of the management board, including the CEO, are generally appointed for five years. Section 84(3) of AktG also indicates that directors can be removed by the supervisory board but only for a "good reason" (ein wichtiger Grund). This, for example, includes vote of no confidence by the shareholders. Re-appointments of directors are allowed but they require a formal proposal from the supervisory board (Section 84(1) of AktG). The reappointment process is similar to the first appointment to the board if the event takes place after the end of the five-year period.
All the German re-appointments in the sample followed the regular appointment process through the supervisory board. In other words, these are genuine re-elections (Wiederwahl) rather than rubberstamping extensions (Verlängerung) as indicated in the respective annual reports of the firms.

UK
The Companies Act of 2006 in the UK states that all the directors, including the CEOs, are subject to reappointment by the shareholders every three years and in some cases every year depending on the conditions included in the Article of Association. Directors can be re-appointed either at the AGM via shareholders' vote or via approvals of the board based on directors' appraisal. The shareholders are privy to the detail of the board's appraisal of individual directors but the board is required to inform the shareholders why it believes an individual director should be re-elected, particularly in the case of the CEO (see the UK Corporate Governance Code 2010-2012).
Similar to France and Germany, for the UK, we only consider those re-appointments which go through the formal voting process at an AGM and are explicitly stated as such in the respective annual report.
The re-appointment events in all three countries are also confirmed via reliable news sources and also the AGM agenda presented in the annual report of the firms.

Notes:
The information included in this appendix describes the legal and institutional framework pertaining to re-appointments in France, Germany and the UK and it describes the procedure used to select the re-appointments included in the sample. The above description suggests that the re-appointments are genuine re-elections of the incumbent CEO rather than just a simple rubberstamping of an extension of their term.

about ECGI
The European Corporate Governance Institute has been established to improve corporate governance through fostering independent scientific research and related activities.
The ECGI will produce and disseminate high quality research while remaining close to the concerns and interests of corporate, financial and public policy makers. It will draw on the expertise of scholars from numerous countries and bring together a critical mass of expertise and interest to bear on this important subject.
The views expressed in this working paper are those of the authors, not those of the ECGI or its members.