**4. Methodology and Data**

To test our hypotheses, we used two different methodological approaches, which were also applied in the previous literature. First, similar to Brunello et al. (2003) and Hu and Leung (2012), among others, we applied a balanced panel data analysis. The (forced) CEO turnover dummy variable is equal to 1 if the firm had at least one (forced) CEO replacement during the year, and 0 otherwise. Second, we considered each CEO departure separately, and we applied an unbalanced panel data analysis (as a robustness check procedure), due to the fact that there are companies in our sample that registered more than one CEO turnover during the same year.

First, we identified and classified the reasons for CEO turnover using the methodology also applied by Warner et al. (1988) and Dherment-Ferere and Renneboog (2002), with necessary adjustments for the Romanian case, described below. Additionally, we considered the reasons for CEO turnover motivated by political manipulations and the battle for power as in Cannella and Shen (2001). We identified 12 reasons for CEO turnover (including both forced and voluntary CEO departures): retirement, health problems, political reasons, promotion, persona non-grata, change of ownership, common succession, pressure from shareholders, separation of the positions of CEO and Chairman of the Board, CEO resignation, other reasons, and no public reasons. A more detailed presentation for these reasons for our sample of Romanian listed companies can be found in the Appendix A.

Second, we divided these turnovers into two classes: forced and voluntary. However, identification of forced departures can sometimes be very difficult when press releases are the main source of documentation, because they rarely describe them as such. To isolate most likely forced departures, firstly we followed Fee and Hadlock's (2004) methodology for classifying turnover as forced. If an article in the business press indicates that the CEO was banished, forced, or equivalent, then turnover was defined as forced. Secondly, similar to Conyon and Florou (2002), we considered the turnover as forced if the CEO was replaced for reasons such as political misunderstandings, conflicts with the Board of Directors, weak performance, personality differences, and scandals. Thirdly, we followed Dahya et al.'s (2002) methodology, considering dismissal and resignation as forced turnover. Thus, we considered the following reasons as forced turnovers: persona non-grata, change of ownership, pressure from shareholders, political reasons (except for the CEO of Oil Terminal who was elected in the Romanian Parliament, due to incompatibility reasons), CEO resignation (due to conflicts with the Board of Directors), as well as the cases in which no reason was given for CEO replacement, in case of weak pre-succession performances.

The sample contained 58 CEO turnover events that occurred in 36 companies. In the balanced panel data approach, they represent 53 company year CEO turnover observations out of the 426 observations collected between 2005 and 2010 for the whole sample of 71 companies. Hence, CEO turnover occurred in 12.47% of the cases and forced CEO turnover in 6.81% of the observations.

In the balanced panel analysis, we considered both accounting-based and market-based returns, computed at the end of the year previous to that when the CEO turnover occurred. In subsidiary, we also took into account in our models the annual returns of the year when the CEO was changed. We presumed that effects of the dismissed CEO's activity might also occur after the CEO was changed, due to his/her strategic position in the company. In the unbalanced panel analysis, the performance indicators were determined for 12 months before the date of the succession announcement and, because we did not exclude observations for CEOs who served less than a year, we computed the return for the period when they held the position and annualized the result.

To compute industry annual return, we used all of the Romanian companies listed on the BSE and on the over-the-counter market (former RASDAQ) that had the same first two industry code digits. It has been proven that this classification offers a higher homogeneity of returns than statistical clustering or three- or four-digit classification (Chan et al. 2007; Clarke 1989). To minimize the possibility of biases in our results due to outliers, the performance variables, the price/book value, and the leverage ratios were winsorized at 1% and 99%. We did not include the same regression variables that had a correlation coefficient equal to or greater than 0.4. Following the literature and the

hypotheses presented in Section 2, we applied a binary logit model to estimate the forced CEO turnover equation. Logit regressions are commonly used in management turnover studies in order to show that the likelihood of turnover is sensitive to changes in firm performance (see Powers 2005 for an extensive literature in this field). Consistent with previous literature, we included as independent variables several accounting-based and market-based performance indicators, various corporate governance indicators, variables related to the financial features of the firm, and industry dummies. Additionally, we included new variables for foreign CEOs and female CEOs, for the financial crisis that started in 2008, as well as an interactional variable for government ownership and political changes.

The use of composite variables in logit regressions was proved to be problematic by Powers (2005) due to the non-linearity of the model employed. However, in our analysis, the composite variable acts as a stand-alone variable, as its components were not used separately in the regression. Consequently, it can be interpreted as such, the sign of its coefficient being statistically significant.

Our sample included all firms listed, during the 2005–2010 period, on the first and second tiers of the BSE, except for the BSE itself (a listed company which became public in June 2010). We studied public companies because, for small and/or unlisted Romanian companies, it would be very difficult to collect these data, and the corporate governance mechanisms are expected to not be properly implemented. The sample cannot be extended due to the small dimension of the Romanian capital market. The 36 companies that experienced CEO turnovers represent one half of all listed firms on the first and second tiers of the BSE in 2010, with average sales of 904 million RON, an average number of employees of 1728, and an average stock market capitalization of 985 million RON. The analyzed time span includes a period of exceptional development of the Romanian capital market (2005–2008), but also one of economic recession (2009–2010). Besides, this period is particularly interesting due to the ongoing corporate governance reforms and the adoption of the Romanian Code of Corporate Governance in 2009.

Most of the data regarding the CEO's identity, CEO ownership, board independence, the reasons for CEO turnover, the succession type (voluntary/forced), and the exact date when the CEO was changed were hand-collected through corporate websites. Additionally, different Romanian business newspapers (Ziarul Financiar, Capital, Săptămâna Financiară, and Business Magazin) and news agencies (Mediafax and Daily Business) were accessed. Data for corporate governance variables and the accounting and financial data were from quarterly and annual company reports provided by the BSE. The explanatory variables are described in Table 1.


**Table 1.** Description of explanatory variables used in the model.


**Table 1.** *Cont*.

Notes: Furthermore, we constructed two variables for the second largest shareholder ownership, as a percentage of common shares, and respectively as a dummy variable equal to 1, if the second shareholder's ownership equals or exceeds 10% of the total voting rights and 0 otherwise. We found no empirical evidence of a correlation between these variables and the CEO turnover occurrence.

Some descriptive statistics for the variables used in the model are presented in Table 2. These data emphasize that the average values for the firm's performance ratios are low, reflecting a negative influence of the global financial crisis starting in late 2008. The same trend is sustained by the price to book median value of 0.95, suggesting that, for more than half of the companies in the sample, the market value of the company is lower than the book value of shareholders' equity. This finding may also be related to the low liquidity and low degree of informational efficiency of the Romanian capital market.

Table 2 shows that 11% of the listed companies have foreign CEOs. We found evidence that the foreign origin of a manager is associated with the fact that the company is part of a foreign group. Almost 12% of the managers in our sample were women. This is not a high figure, but compared to other statistics, we can expect a relative significance of the manager's gender on the likelihood of CEO turnover occurrence. For instance, Kato and Long (2006) emphasize that in 2002, the proportion of female CEOs was only 4% in China.

The average first shareholder's ownership was around 47%, with a median value slightly above 50%. This result is important to characterize the corporate governance context of the Romanian market. It is presumably associated with important agency problems, especially between controlling and

minority shareholders, and with high opportunities of extracting private benefits from control on a capital market with high control premiums.<sup>3</sup>

**Table 2.** Summary statistics on CEO turnover for the Romanian listed companies during the 2005–2010 period.


Source: Authors' calculation based on data provided by Bucharest Stock Exchange and the annual reports of the companies in the sample.

The average percentage of votes held by all institutional investors, each of them holding more than 5% of the votes, was 26.07%, but in half of the cases, it was below 11%. The average ownership of institutional investors was around half of the average institutional ownership for US firms in the work by Helwege et al. (2012). The standard deviations for all the variables regarding institutional ownership recorded high values, showing significant differences between the listed companies in our sample. Only 29% of the companies had institutional investors who own more than 50% of the equity. This may lower the incentives of institutional investors to be actively involved in controlling

<sup>3</sup> Dragotă et al. (2013) documented an average value of control premium around 115% and a median one of 25% for 173 tender offers carried through the BSE and RASDAQ for the period 2000–2011.

the companies efficiently. Due to the fact that, in the existing literature, it was shown that the role and involvement of institutional investors in the decision to change CEOs (forced) is even greater as they hold more shares (Kang et al. 2018), the results for this independent variable in the case of Romania are somewhat uncertain.

The average board size is around five members, with a maximum value of eleven members. These values are somewhat similar to the size in other one-tier board systems, such as in the US (Yermack 1996; Denis et al. 1997; Denis and Sarin 1999), Belgium (Renneboog 2000), Italy (Brunello et al. 2003), and China (Kato and Long 2006). We can bring into the discussion the arguments of Yermack (1996) for smaller and more efficient boards that monitor the executive managers more effectively. Thus, smaller board size can emphasize the negative relation between firm performance and CEO turnover.
