Influence of CEO's personal characteristics on short-term M&A performance in Russia

In accordance with another assumption of OLS method, the error term has a constant variance. In order to conduct the test for homoskedasticity, Breusch-Pagan-Godfrey test is used. This test is applied for large samples. The squares of residuals are standardized by dividing by the mean square of residuals (sum of regression squares divided by N), resulting in generalized residuals. These residuals are then regressed to all dependent variables, which are suspected of heteroskedasticity. The statistically significant value means that the null hypothesis is rejected and the assumption of homoskedasticity is not fulfilled.

2.6.4 Testing for Autocorrelation

At this stage it is necessary to conduct an autocorrelation test. It is assumed that the remnants of the regression are not related to each other over time. Autocorrelation testing is performed using the Durbin-Watson test. According to Brooks (2008), in order to confirm the absence of autocorrelation, it is necessary to obtain the Durbin-Watson statistic value close to 2. The Durbin-Watson test is calculated using the SPSS statistical package.

2.6.5 Testing for Normality

In order to check the normality of the distribution the graphic method is used. For this purpose, outliers are determined by plotting the "box plot" in the statistical package in SPSS. Financial and economic data is likely to have no normal distributed data. In order to smooth the difference in data, Brooks (2008) recommends to either convert variables using the natural logarithm to smooth outliers or use dummy variables to exclude outliers. Theoretically, outliers can be eliminated but this is not recommended because valuable observations can be lost. In addition, the exclusion of variables artificially improves the model.

2.7 Robustness test

In order to modify findings of the regression model, some variables and conditions are slightly changed. As stated in Kilian and Schindler (2014) research, the dependent variable CAR can be changed in order to get more accurate results. It is concluded that the smaller event window - the more exact results of CARs. As the process of the choosing the event-window for this research is based on a big number of researches, CARs are calculated in advance for two event windows. The CARs that is calculated for 11 event windows are used for modifying the findings and to check whether it influences the results of the final regression model.

As for Age variable, the study of Yim (2013) suggests to divide CEO age groups into categories. To do this a dummy variable AgeScale is created. Since this study is conducted for companies in the Russian market, to determine the age groups it was customary to take data on the distribution of the population employed in the country's economy by age groups, created by the sites of the Federal Statistics Service. Thus, age groups are encoded as 15-29 years old - "young" = (0); 30-49 - "middle" = (1); 50 - 72 - "old" = (2).

Another change is creating two sub-samples, corresponding to the time period from 2012 to 2014 and from 2015 to 2019. This modification is made in order to assess the influence of currency crisis of 2014. Despite the fact that the length of the time periods is unequal, the number of observations in each sub-sample is greater than 100 and the samples are almost equal. All variables that involve the time periods in the calculation process are changed in accordance with the time frame of the sub-sample. For calculation of the regression of each sub-sample the same methodology is applied as described above.

2.8 Validity and Limitations

The sample of data consists of 251 friendly M&A deals, conducted by Russian acquirers from 2012 to 2019. All of them are friendly, as hostile attitude may influence the overall results of the transactions and it would have been examined separately. The period of 2012-2019 years is chosen to compare the situation before and after the financial crisis of 2014. The data source for company's share price is the financial analytical service, called Refinitiv, by Thomas Reuters Eikon. Considering personal characteristics of the CEO, data is gathered from multiple online resources, including official websites of companies, their press-centers and archives, online media resources, magazines, etc. Full table with the collected data can be seen in appendices (see Appendix 1).

Process of data filtering is multi-staged. The primary downloaded data (more than 1200 results) firstly filtered by the criteria of type of the company: private companies are excluded to the absence of information, individual entrepreneurs and bank are also excluded from the sample (banks have different organization of financial statements). During the process of collecting data some companies have incomplete information about CEO's characteristics and thus are also removed from the list. That is why the primary sample reduces to 251 deals.

The crisis of 2014, caused by the imposed sanctions by the United States and European countries, negatively affected Russian economy. According to the 2015 bulletin of the socio-economic crisis in Russia from the analytical center under the government of the Russian Federation, the Russian economy in 2014 was influenced by a number of negative factors, including reduced consumption, increased inflationary threats and low investment activity. All these factors were caused by the imposed sanctions against Russia, the fall in oil prices and the devaluation of the ruble that arose from this. Industries such as construction, manufacturing, retail, and manufacturing have also suffered. Nevertheless, a stressful situation can become a driver for an intensive modernization and updating of systems operating in the Russian market, thereby contributing to subsequent growth. For example, agro-complexes and companies, mining minerals were able to find benefits in this situation, as the country was intensively introducing a policy of import substitution.

The devaluation of the ruble led to a decrease in imports, the replacement of money for durable goods and also complicated the work of banks under threat of crisis. The fall in oil prices, as one of the fundamental sectors of Russian economy, led to a reduction in investment programs in the sector and also caused a drop in M&A market, as described in the literature review section. However, the wave-like development of the M&A market indicates that after a sharp economic crisis there is a rising wave - the right fiscal policy and supporting measures by the governing authorities help the market to stabilize and increase capacity again.

The results of data collecting show that gender diversity among Russian CEOs is almost zero - all CEOs are male. There were few companies (2-3) with female CEO but they were excluded from the list due to the absence of the information. The situation is quite predictable as even global pool of CEOs consists of only 9% of female CEO representatives, according to Athanasopoulou et al. (2017). They also mention that there are some factors, influencing such dynamics. For example, social context: females are still raised with certain gender patterns, limiting women from accessing the CEO role within the organization (Fitzsimmons et al., 2014). Athanasopoulou et al. (2017) also highlight balance between career and role of wife, lack of sponsorship, over mentoring, prejudgment, and female psychological features as factors, restricting female leadership within the organization.

According to the Deloitte Research on Women CEOs in Russia (2020), only 20% of Russian CEOs are women, but the women CEOs' share depends on the size of the company and its sector in which it operates: 35% women CEOs are in social sector and only 8% in mining and energy sector as well as in the government higher positions. Furthermore, the research shows that Russian female-led businesses, operating on the micro level, show better financial performance, while the picture is different with the large companies (6.5% of female CEOs are among 200 largest Russian companies).

Размещено на http://www.allbest.ru/

Figure 7 Share of women CEOs by sector and company size Source: Deloitte Research on Women CEOs in Russia (2020)

Beltran (2019) says that female presence in leadership roles positively affects the overall firm's performance. Marcelo and Brooks (2016) conclude that female participation in the executive boards and leading positions is beneficial to the organization's performance as women are better in understanding customers' needs and preferences and are less risky taking. It is important to say that not all studies suggest low risk taking as a positive point, as sometimes risk is a driver for increase of the organization's performance. The modern world is changing, and gender stereotypes are being violated due to the increasing awareness of the feminism movement, which promotes the equal rights of women not only in society but also in their careers.

Fitzsimmons et al. (2014) explain male predominance in the position of CEO by psychological factor: people are prone to choose people, whose mind and behavior are more or less similar to their and, as mostly the board consists of male members, they choose male CEOs as more understandable and similar ones. Vinkenburg et al. (2011) mention that, comparing to women who display better transformational behavior as leaders, male CEOs show higher score of inspirational motivation, which is valued by people during the CEO selection.

During the process of gathering data such limitation as absence of information about the sum of the deal has arisen. This possible variable defines the size of the deal, which in turn shows the significance of importance of this specific transaction for the specific company, whether it creates value for the firm. Previously conducted studies give this information but because the research sample mostly consists from international M&As.

Additionally, there are such limitations as the fact that several companies use USD currency in their financial statements and thus there is need to convert the indexes into the ruble equivalent, using the exchange rate of previous years.

3. Empirical results

This chapter presents descriptive statistics of the sample and describes the variables and the tested data on the validity of the OLS method application on the sample. After that, the results of the regressions are presented and the statistical significance of the results is described.

3.1 Descriptive statistics

First of all, the essential step is to characterize the data collected, based on indicators such as the distribution of the number of transactions by year and industry. Looking at the table below, it can be concluded that the number of transactions in the sample is almost evenly distributed by years. However, the largest number of transactions is in 2013. According to Ivanov and Peredunova (2017), that year dates as the year of the wave downturn in the global M&A market, and companies are less likely to complete transactions during this period. Therefore, it can be concluded that, despite the decline in the wave, Russian public companies completed a significant number of transactions in the pre-crisis period. This notice is important for further division of the sample by sub-samples.

Table 3

Distribution of the observations of the sample by years

Source: The authors' calculations

To support the analysis of the Russian market that is done in the literature review, the table below shows that the largest number of M&As of the sample is conducted by takeover companies that operate in the oil and gas, chemicals and metal and mining industries, as well as telecommunications services. The majority of the transactions included in the sample are made by companies such as Rosneft Oil Co (27), SSA Sistema PJSFC (26), Gazprom PAO (18), Rostelekom PAO (14), Polymetal International PLC (14), and Mobile TeleSystems PJSC (13).

Table 4

Distribution of the observations of the sample by industries

Source: The authors' calculations

Then, it is necessary to analyze the obtained values of the dependent variable. In this study, the dependent variable is the CAR values calculated for two event windows. Thus, for 251 calculated values of CAR21Days varies from -31.29% to 23.85%. At the same time, the median CAR value for the 21 days event window around the day of the announcement of the transaction is -0.005%, and its average value is 0.14%. Based on this, it can be concluded that in the majority of cases companies are unable to create value for the company in the result of M&A deals. For CAR values calculated for the 11 days event window, the minimum and maximum values are -19.97% and 16.77% respectively, and the median value is at the level of 0.34%, which also indicates the poor performance of transactions included in the sample of this research. The distribution of the dependent variables is checked on the normality. The graphs below demonstrate the results of normal distribution check that is conducted by SPSS. There are several outliers that are not excluded advisedly for the analysis, in order to not lose valuable observations.

Размещено на http://www.allbest.ru/

Figure 8 Tasting of the CAR variables on the normality of the distribution Source: The authors' calculations

Considering the variables that describe the personal characteristics of CEOs, the following conclusions can be made. The age of the sample varies from 31 to 72 years old. The manager's experience in the company until the transaction is completed is from 0 to 23 years. In this case, the median value of age is almost 48 years. Thus, based on the age distribution of the economically employed population of the country according to Rosstat data, it can be concluded that in this sample the majority of managers involved in the implementation of the M&A transactions can be assigned to the middle age segment.

As for independent variables, four dummy variables are included in the regression model both for the full sample and for the sub-samples. It is noteworthy that 67.5% of CEOs included in the full sample have experience in performing M&A. More than 50% had experience in a related industry before becoming a CEO in a company that had a deal. As for work experience in the government, a little more than a half of the observations has no ties with the state. Finally, the distribution of the observed data by the degree in management or business among the directors is almost even. However, more than a half of the considered CEOs do not have business education. The table with the frequencies of variables is presented below.

Table 5

Distribution of the observations of the dummy variables Source: The authors' calculations

Considering the values of control variables, we can conclude that most of the companies included in the sample can be characterized as `large' according to the distribution by revenue. About 5% of the companies are micro-enterprises, and only 5 observations are representatives of small and medium-sized businesses. It is found that the ROA of companies are distributed from -RUB 60.55 mln to RUB35.40 mln. As for values of financial leverage, they vary greatly from 0.57 to 2313.23. In order to smooth outliers when constructing a linear regression, this variable is transformed using the construction of natural logarithms. At the same time, values of the companies' assets logarithmic data is taken in advance in order to avoid large emissions. Thus, the mean value of the logarithmic assets is 12.63. As for the tangibility, the median value is concentrated at 0.3, while the maximum and minimum values ??are 0.88 and 0 respectively. This suggests that the sample contains companies with different volumes of tangible assets. As a rule, companies operating in the producing or mining industries own more technical capacities than companies in the telecommunications industry, in which intangible assets make up the majority. Finally, the date includes the information about companies' ages. The age of the companies begins from 18 years old. As for the maximum value, the older company is 154 years old. The mean value is about 40 years. Despite the fact, that the data has emissions, they are not excluded, as it threatened with loss of valuable information.

3.2 Cumulative abnormal returns

According to the methodology of the event analysis described above, for each day of the announcement of the transaction, abnormal returns were calculated for each day of the event window at 21 days and 11 days. The obtained abnormal returns were averaged for 251 observations and the obtained average abnormal returns were checked for normal distribution and evaluated for significance using Student's t-test. Statistically significant CAR values ??are those that correspond to statistically significant abnormal returns. Thus, for a sample containing 251 observations statistically significant is a value that is greater than tcrit = 1.65097 at a significance level of 0.01. Therefore, for the investigated event window at 21 days, the CAR values ??corresponding to `-9', `0' and `6'days of the event window were statistically significant. As for the 11-day event window, only the advising value `0' of the day is statistically significant and can be used for further calculations. Tables with the obtained values ??of AARs, CARs and calculated day values ??of the test statistics are presented in the Appendix 2. For further use of CARs as a dependent variable for constructing a regression, the CARs values ??calculated for the `0' day of the transaction announcement are used.

3.3 Regression Model

In order to use the OLS method to assess the influence of personal characteristics of CEOs on a variable expressing abnormal returns of companies' stock prices, several assumptions were tested for the model. Thus, an assessment of the multicollinearity of variables was carried out by constructing the Pearson correlation matrix. Multicollinearity was revealed for FirmSize and LN (Assets) variables, this is due to the fact that both variables explain the size of the company. Since the FirmSize variable was created as a dummy variable, a decision is made to exclude this one. As for the correlation of other variables, there is a correlation between the age variable and other independent variables. However, this correlation value does not exceed the threshold value defined in the methodology section. A negative correlation is observed between the independent variables Business education and CEOIndRelation, the correlation value is -0.207, which also does not correspond to the threshold value. The table that represents Pearson correlation matrix is presenting in the Appendix 3. Taking this into account, the multivariate regression models are following:

(14) CAR21Dayit = б + в1 Age_it+ в2 CEOIndustryrelation_it + в3 CEO Governmentworkexperience_it + в4 CEOM&AExperience_it + в5 Businesseducation_it + в6 ROA_it + в7 Financial leverage_it + в8 CompAge_it + в9 Tangibility_it + в10 LN_Assets_it + е_it

(15) CAR11Dayit = б + в1 Age_it+ в2 CEOIndustryrelation_it + в3 CEO Governmentworkexperience_it + в4 CEOM&AExperience_it + в5 Businesseducation_it + в6 ROA_it + в7 Financial leverage_it + в8 CompAge_it + в9 Tangibility_it + в10 LN_Assets_it + е_it

3.3.1 Multivariate regressions

The table below presents the results of four fundamental regressions. The model (1) describes the dependence of CAR values ??calculated for 21 days around the day of the announcement of the transaction and includes control variables. In the model (2), the CAR values ??calculated for 11 days around the event window were used as the dependent variable, and the model includes control variables. For regressions (3) and (4), the same dependent variables were used as they were for the first two models but excluding control variables. Autocorrelation testing was performed for each model using the Durbin-Watson test. As a result, the values ??2.049, 1.968, 2.031, and 2.686 respectively were obtained, that exclude autocorrelation. Moreover, the Breusch-Pagan-Godfrey test showed that the presence of heteroscedasticity was not indicated. Testing the dependent variables for heterogeneity showed that p-values are less than 0.05, it means that for CARs the null hypothesis of equal population variances is rejected, and the standard Student t-test can be used to test the significance of the values of variables in the models.

Regarding the results of the regressions, it was found that for full sample the coefficient of the variable Business Education is statistically significant at a significance level of 0.1. The coefficient has a positive value, which indicates a positive correlation of the abnormal stock returns of the company and the CEOs having degree in business and management. With the exclusion of control variables, this variable also has a statistical significance at the level p = 0.01, but for the values ??of the dependent variable calculated for 21 days around the event. In this case, the coefficient of the independent variable also has a positive value and confirms a positive correlation. Thus, the Hypothesis 5 that a CEO with a business education is more likely to complete M&A transactions that create value for the enterprise can be accepted.

For a variable describing the CEO's experience in the related industry, a statistically significant value appears in model (4), excluding control variables, where the dependent variable is presented in the form of CARs calculated for 11 days around the event. The observed value has a positive coefficient, which indicates a linear correlation between the cumulative stock returns of the company for the 11 days event window and the presence of experience in the related industry of the CEO. Hence, the Hypothesis 2 can be accepted, and CEO's working experience in the related industry has a positive influence on M&A outcome.

Considering the obtained coefficients of other independent coefficients, it can be concluded that the Age variable for all 4 regressions has a negative value, which indicates that the age of CEOs has an inverse correlation with the CARs values. In other words, the younger the CEO, the more successful the outcome of transactions. However, it is important to say that these values ??cannot be accepted because they are not statistically significant. Positive variables were obtained for the dummy variables CEOM&AExp and CEOGovExp in regressions (1), (2) and (3). This suggests that there is a linear relationship between the CARs variables calculated for the two event windows, and the variables that describe the CEO's previous experience in M&As and government experience. These coefficients ??cannot be accepted for subjecting or rejecting hypotheses involving these variables since the coefficient of significance of these variables is less than the critical value.

Despite the fact that the models were tested for the validity of using the OLS method, the regression results are not statistically significant at the 1% significance level based on the Fisher test, except Model (2). However, the obtained Adjusted R Square values ??indicate that models (1), (2), (3), (4) describe less than 1% of the data in the sample and, hypotheses cannot be accepted based on the results of the models.

Table 6

The results of the regression models (1) - (4) Source: The authors' calculations