How do factors related to corporate governance affect earnings management

The author Based on an overview of previous studies, the thesis listed these following hypotheses as follows: H1: Size of BOD has a negative correlation with earnings management H2: The number of non-executive board members has an inverse correlation with earnings management H3: The concurrent role of CEO and chairman in BOD has a positive correlation with earnings management H4: Percentage of directors with expertise in financial accounting in the board of directors has a negative correlation with earnings management. H5: The number of female board members has a negative correlation with earnings management Control variables - Business size (SIZE) - Return of Equity (ROE) - Financial leverage (LEV) - Cash flow(CF) - Revenue growth (GROWTH) - Independent audit (BIG4) Dependent variables Earnings management (EM) Characteristics of corporate governance - Size of BOD (BOARD) - Number of non-executive board members (NED) - The concurrent role of CEO and chairman in BOD (DUAL) - Percentage of directors with expertise in financial accounting in the board of directors (FAD) - Number of female board members (BSR) - Average age of the board members (AGE) - Ownership concentration (CO) - Management's ownership percentage (MO) - State’s ownership percentage (SO) - Foreign investors' ownership percentage (FO) 8 H6: Average age of the board members has an inverse correlation with earnings management. H7: Ownership concentration has a positive correlation with earnings management H8: Management's ownership percentage has a negative correlation with earnings management. H9: State's ownership percentage has a negative correlation with earnings management. H10: Foreign investors' ownership percentage has a negative correlation with earnings management H11: Business size has a positive correlation with earnings management H12: ROE has an inverse correlation with earnings management H13: Financial leverage has a negative correlation with earnings management H14: Operating cash flow (OCF) has a positive correlation with earnings management H15: Revenue growth has a positive correlation with earnings management H16: Independent auditor has a negative correlation with earnings management 3.2. Methodology 3.2.1. Development of regression The regression is developed with the dependent variables being the absolute value of discretionary accruals (EM) representing the firm's earnings management (Paul Hribar and D. Craig Nichols, 2007). To test hypotheses from H1 to H16, the impacts of corporate governance on earnings management and the control of the impacts of other relevant variables are calculated by using multivariate regression as follows: EM = αo + α1BOARD + α2NED + α3DUAL + α 4FAD + α 5BSR + α 6AGE + α 7CO + α 8MO + α 9SO + α 10FO + α 11SIZE + α 12ROE + α 13LEV + α 14CF + 15GROWTH + α 16BIG4 Table 3.1: Measurement of variables in the research model No Variables Acronyms Measurement References I Dependent variables 1 Earnings management EM The absolute value of discretionary accruals Kasznik (1999) II Independent variables 2 Size of BOD BOARD Number of board members Xie et al (2003), Rahman and Ali (2006) 3 Number of non-executive board members NED Number of board members not participating in the administration and management Baysinger and Hoskisson (1990), Forker (1992) 4 The concurrent role of CEO and chairman in BOD DUAL The dummy variable, which equals 1 if the CEO is also the chairman, equals 0 if the CEO is not the chairman Jensen (1993), Chtourou et al. (2008) 5 Percentage of directors with expertise in financial accounting in the board of directors FAD Number of directors with expertise in finance and accounting/Total number of board members Abbott et al (2004), Bédard et al (2004) 6 Number of female board members BSR Number of female board members Pearce and Zahra (1992), Srinidhi et al (2011) 7 Average age of the board members AGE Average age of the board members Bantel and Jackson (1989), He, W.F. and Liu, Q.L. (2010) 9 No Variables Acronyms Measurement References 8 Ownership concentration CO Total number of shares held by shareholders owning 5% of share or more in the company Shleifer and Vishny (1997), Gabrielsen et al. (2002) 9 Management ownership MO Management's ownership percentage Cheng & Warfield (2005), 10 State ownership SO State's ownership percentage Koh (2003), Hsu and Koh, (2005) 11 Foreign investors' ownership percentage FO Foreign investors' ownership percentage Firth et al (2002), Ho, Wu and Xu (2010) III Control variables 12 Business size SIZE Log10 of Total Assets of the company Kim, Y., Liu, C., & Rhee, S. G. (2003); Naz, Bhatti, Ghafoor and Khan (2011) 13 ROE ROE Return On Equity Kiel & Nicholson (2003), Carter et al. (2003) 14 Financial leverage LEV Total Debt To Total Assets Ratio Agrawal and Knoeber, (1996), Jelinek, (2007), Jiang et al. (2008) 15 Cash flow CF Operating cash flow to total assets in the previous year ratio Dechow et al (1995), Peasnell et al (2005) 16 Revenue growth GROWTH Ratio of revenue year t - revenue t-1/Revenue year t-1 Chan et al (2001), Lee et al. (2006) 17 Independent audit BIG4 The dummy variable, equals 1 if the firm is audited by Big4 (KPMG, EY, PWC, Deloitte) and zero if the firm is not audited by Big4. Becker, (1998), Francis, (2009); Brown (2014) Source: The author 3.2.2. Research data The collected data was provided by Vietstock. The collected data was designed in the form of tabular data and entered into the statistical data processing software Stata 12. The sample consisted of499 companies listed on Vietnam stock market in the period of 10 years from 2009 to 2018. Baseline data included 4990 observations, after eliminating ineligible observations due to lack of information, the remaining sample of 3013 observations was in stage 1 and stage 2. 3.2.3. Research methods During the data analysis, the author used Stata 12 software to run two-stage regression. The stages were carried out as follows: Stage 1: The author regressed the sample 1 containing 3013 observations to measure discretionary accruals (DA) value representing by the firm's earnings management through non-discretionary accruals according to the model by Kasznik (1999). Specifically, the calculation was performed as follows: - Calculation of total accruals (TA): According to Collins and Hribar (2000), total accruals were calculated as follows: TAit = NIit - CFOit (1) In which: TAit: Total accruals in year t of the company i 10 NIit: Operating profit after tax before changes in working capital in year t of Company I CFOit: Operating cash flow in year t of Company I - Calculation of non-discretionary accruals (NDA) According to Kasznik (1999), the NDA variable regression model according to TA is presented as follows:
  = β x  + βx    + βx   + βx   + ε (*) In which: TAit: Total accruals in year t of the company i DAit: Discretionary accruals in year t of the company i NDAit: Non-discretionary Discretionary accruals in year t of the company i Ait-1: Total assets of company i in year t -1 DtREVit: Changes in the company's revenue in year t compared to year t-1 DtRECit: Changes in the company's receivables in year t compared to year t-1 PPEit: Tangible fixed asset costs of company i in year t DtCFOit: Changes in OCF of company i in year t compared to year t-1 β , β, β, β: The coefficients indicate the degree of impacts of independent variables on dependent variables : Error in the model Using data including 3013 observations of 499 enterprises from 2009 to 2018, the study used Stata 12 software to perform regression according to the Pooled OLS, FEM, REM, GLS model to find the most suitable model. The equation (1) after regression produced β ’, β’, β’, β’ being the estimates of the regression coefficients β , β, β, β in the most suitable model. The author replaced β ’, β’, β’, β’ in equation (2) to calculate the variable NDAit. NDAit = β ’x  + β’x !!  + β’x   + β’x !  (2) - Calculation of discretionary accruals (DA) After calculating NDAit/Ait-1, the author used equation (2) to calculate discretionary accruals DAit DAit = TAit/Ait-1 – NDAit (3) - Calculation of the dependent variable EM of the regression in stage 2: In the research model, the author used the variable EM as a representative of earnings management calculated by the absolute value of the variable DA according to equation (4): EMit = Abs (DAit) (4) Stage 2: After reviewing the data, the study still used the data of 499 non-financial companies listed on Vietnam stock market from 2009 to 2018 with unbalanced panel data of 3013 observations. The stage-2 regression proposed by the author: EM = αo + α1BOARD + α2NED + α3DUAL + α 4FAD + α 5BSR + α 6AGE + α 7CO + α 8MO + α 9SO + α 10FO + α 11ROE + α 12SIZE + α 13LEV + α 14CF + 15GROWTH + α 16BIG4 (**) Using the data set including 3013 observations, the author separated the data set into: negative DA group with 1826 observations and positive DA group with 1187 observations. To clarify the impacts of corporate governance on earnings management, the author also studied the model (**) phase 2 with 10 independent variables and 6 control variables that affect the 11 dependent variable EM (calculated by taking the absolute value of DA) in two cases: Split the data set into negative DA and positive DA and aggregate all DA data. - Firstly, the attendant conducted descriptive statistics for all the variables in the model. - Secondly, the author carried out correlation analysis of variables in the stage-2 model according to two cases including the split of negative and positive DA and the aggregation of DA to examine the relationship between the independent variables and the dependent variables. - Thirdly, the author performed regression according to the Pooled OLS model. In experimental studies, separate effects are one of the most frequent phenomena (Baltagi et al., 2005). Therefore, the resolution of unobserved factors will be carried out with two models, the FEM-Fixed effects models and the REM-Random effects models. Next, the attendant continued to regress the GLS model to resolve the model defects (Arellano & Bond, 1991). Although the GLS model had solved the variance changes and autocorrelation, however, Greene (2005) proposed that in the models related to earnings management, endogenous problems can arise between the variables. To solve this issue, the author used Dynamic GMM according to the study of Arellano and Bond (1991). Hoang Thi Mai Khanh and Nguyen Vinh Khuong (2018) used lag being the degree of earnings management as an endogenous variable in the current earnings management model. Therefore, the author also used the delay variable EM_L1 as an endogenous variable in the dynamic regression model to analyze the impacts of corporate governance on earnings management. Hansen testing results on the overidentification and Abond testing(AR (2)) - testing of 2-order series correlation to know whether the results in the dynamic GMM are significant or not. - Finally, after selecting the most suitable model, the author compared the results of the split of discretionary accruals (DA) into negative and positive DA with the aggregation of negative and positive DA to discover errors in results. Also this was to show how different the overall results in the aggregation are compared to the split of negative and positive DA. CHAPTER 4 THE CURRENT SITUATION OF IMPACTS OF CORPORATE GOVERNANCE ON EARNINGS MANAGEMENT OF COMPANIES LISTED ON VIETNAM STOCK MARKET 4.1. Descriptive statistics of corporate governance and earnings management of companies listed on Vietnam stock market 4.1.1. Descriptive statistics of corporate governance Quản Corporate governance is a concept consisting of many different factors. However, within the scope of the thesis and collected data, the author focused on factors of corporate governance as shown in Table 4.1 as follows: 12 Table 4.1: Descriptive statistics of characteristics of corporate governance Variables Observations Mean Standard deviation Minimum value Data split Data aggregation Data split Data aggregation Data split Data aggregation Data split Data aggregation Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA BOARD 1826 1187 3013 5.581051 5.476832 5.539993 1.206944 1.126903 1.176975 3 3 NED 1826 1187 3013 3.661555 3.499579 3.597743 1.330947 1.273928 1.310959 0 0 DUAL 1826 1187 3013 0.2617744 0.336984 0.2914039 0.4397209 0.4728786 0.4544846 0 0 FAD 1826 1187 3013 0.4697769 0.4731711 0.4711141 0.3203491 0.3214502 0.3207343 0 0 BSR 1826 1187 3013 0.8165389 0.7413648 0.7869233 0.9581776 0.8824688 0.9296654 0 0 AGE 1826 1187 3013 48.88907 48.03223 48.55151 4.908722 5.143014 5.018985 24 31 CO 1826 1187 3013 0.5242506 0.4689683 0.5024716 0.2205713 0.2132022 0.2193327 0 0 MO 1826 1187 3013 0.0992685 0.1307657 0.1116771 0.1510853 0.1567483 0.1540862 0 0 SO 1826 1187 3013 0.2501655 0.1806241 0.222769 0.2653965 0.2429362 0.2589812 0 0 FO 1826 1187 3013 0.1259313 0.0926652 0.1128258 0.149 0.1256367 0.1411786 0 0 Source: The author 13 4.1.2. Descriptive statistics of earnings management In stage 1, the author conducted the regression by Kasznik (1999) (*) to measure earnings management and to calculate the value of the variable EM as the dependent variable representing earnings management in the model in stage 2 (**). The accurate estimate of earnings management in the first stage will help the attendant evaluate the impacts of corporate governance on earnings management in a more comprehensive direction. Therefore, the author regressed the equation (*) according to three models including Pooled OLS, FEM, REM. The regression results are detailed in the Appendix. To correct the variance change and autocorrelation, the author used the GLS regression model with the option to resolve the defects and obtained the results demonstrated in Table 4.3. Table 4.3: Regression results of generalized least squares (GLS) Variables Statistical significance 1/At-1 0.00892*** [3.07] (DtREVt-DtRECt)/At-1 0.0667*** [21.12] PPEt/At-1 0.0885*** [24.88] DtCFOt/At-1 -0.513*** [-72.92] Degree of significance: * p < 0.1, ** p < 0.05, *** p < 0.01 Source: The author Testing results showed that the degree of impacts of all four independent variables on the dependent variable is statistically significant. The model chosen in the first stage is the GLS model as the cointegration and heteroskedasticity covariance can be solved. Finally, the thesis achieved a model to calculate non-discretionary accruals (NDA t) as follows: NDA t = 0,00892*1/At-1 + 0,0667*(DtREVt-DtRECt)/At-1 + 0,0885*PPEt/At-1 - 0,513* DtCFOt/At-1 Based on that, the author calculated discretionary accruals (DA t) and take the absolute value of discretionary accruals to measure the value (EM t) that represents earnings management according to the formula: EM t =Abs(DA t) = Abs(TA t/At-1 – NDA t) Results in Table 4.4 showed the current situatiopn of earnings management of listed companies on the Vietnamese stock market from 2009 to 2018. 14 Table 4.4: Descriptive statistics of earnings management of companies Data Observations Mean Standard deviation Minimum value Maximum value Data split Negative DA 1286 -0.1172128 0.1075023 -1.26398 -0.0000206 Positive DA 1187 0.1027302 0.1309019 0.0000573 2.186293 Data aggregation Aggregated DA 3013 -0.0305641 0.1590696 -1.26398 2.186293 Source: The author Discretionary accruals is a variable that represents earnings management of companies Data in Table 4.4a illustrated that out of 499 listed companies on the Vietnam stock market that were sampled, there were 418 firms with earnings management (accounting for 85.48%), which shows that the majority of enterprises tend to adjust their earnings to achieve their goals. Firms in the sample managed earnings by both decreasing and increasing, and the degree of earnings management is not equal between enterprises and between years within the same enterprise. In the downward trend of management, the degree of earnings management changed between enterprises in the range from 0.0000206 to 1.264; in the upward trend of management, the degree of earnings management of businesses ranged from 0.0000573 to 2.186. 4.1.3. Descriptive statistics of control variables In addition to the characteristics of corporate governance, the author also included control variables in the research model to more comprehensively evaluate the impacts of corporate governance on earnings management. Descriptive statistics of control variables is demonstrated in Table 4.7 as follows: 15 Table 4.7: Descriptive statistics of control variables Variable s Observations Mean Standard deviation Minimum value Maximum value Data split Data aggregation Data split Data aggregation Data split Data aggregation Data split Data aggregation Data split Data aggregation Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA Negative DA Positive DA Aggregated DA SIZE 1826 1187 3013 11.89799 11.95865 11.92189 0.6527297 0.5830846 0.6268199 10.26529 10.46512 10.26529 14.32999 14.45935 14.45935 ROE 1826 1187 3013 0.1452977 0.0861736 0.1220052 0.1232605 0.128586 4 0.1286512 - 0.5712161 - 0.759308 4 -0.7593084 0.7834526 0.9821288 0.9821288 LEV 1826 1187 3013 0.4689134 0.5720845 0.5095587 0.2118081 0.198247 3 0.212604 0.0019807 0.007524 0.0019807 0.9650876 0.964009 7 0.9650876 CF 1826 1187 3013 0.0244657 -0.014416 0.0091478 0.1050754 0.0930501 0.1022747 - 0.5998674 - 0.642043 6 -0.6420436 1.408307 0.8807944 1.408307 GROW TH 1826 1187 3013 0.1619891 0.1908709 0.1733674 0.5861763 0.792028 6 0.6748331 - 0.9894291 -0.930426 -0.9894291 9.482057 9.876404 9.876404 BIG4 1826 1187 3013 0.3417306 0.2982308 0.3245934 0.4744197 0.457674 2 0.4683005 0 0 0 1 1 1 Source: The author 4.2. The current situation of the impacts of corporate governance on earnings management of companies listed on Vietnam stock market 4.2.1. Correlation analysis of research variables Before regression, the author conducted a correlation analysis to detect the multicollinearity. The results showed that the correlation coefficients between the independent variables in the regression model in both data split of negative and positive accruals and data aggregation of negative and positive accruals. The positive discretionary accruals were all less than 0.8, which meant that the model is less likely to have multi-collinearity . 4.2.2. Regression of the impacts of corporate governance on earnings management of listed companies on Vietnam stock market To clarify the impacts of corporate governance on earnings management, the author also studied model (**) stage 2 with ten independent variables and six control variables that affect the dependent variable EM ( calculated by taking the absolute value of DA) in the data split of negative and positive DA. Then, the author carried out the regression of data aggregation of negative and positive DA into one model to see how the aggregation affects the results of experimental analysis compared to the above-mentioned data split. EM = αo + α1BOARD + α2DUAL + α 3FAD + α 4BSR + α 5AGE + α 6CO + α 7MO + α 8SO + α 9FO + α 10ROE + α 11LEV + α 12PM + α 13FTR + α 14CF + 15GROWTH (**) 4.2.2.1. Data split of negative and positive discretionary accruals - Firstly, the author conducted the regression according to the Pooled OLS - Secondly, the author implemented the regression through FEM and REM - Thirdly, the author performed regression according to the GLS with the option of resolving model defects. 16 - Fourthly, the author conducted regression according to the dynamic GMM to solve the endogenous problem that exists in the research model of the impacts of corporate governance on earnings management The regression of dynamic GMM model in the data split of negative and positive discretionary accruals are presented in Table 4.8 below and detailed in the Appendix (Appendix 31, 32). Table 4.8: Results of the dynamic GMM in related to data split of separation of negative and positive discretionary accruals Variables Significance Negative DA Positive DA EM(NEG) EM(POS) EM_L1 0.0642** 0.0573* [2.11] [1.96] BOARD -0.00412 -0.00487 [-0.94] [-0.77] NED -0.00159 0.00347 [-0.42] [0.65] DUAL 0.0124 0.00156 [1.19] [0.12] FAD 0.016 0.00837 [1.55] [0.45] BSR -0.00414 0.00235 [-1.02] [0.36] AGE 0.00144** 0.000561 [2.30] [0.54] CO 0.0432** 0.0457 [2.03] [1.45] MO -0.0604** -0.0576 [-2.39] [-1.39] SO -0.0251 -0.0676** [-1.27] [-2.32] FO -0.0446* -0.0966** [-1.81] [-2.29] SIZE 0.00429 0.00638 [1.23] [1.28] ROE 0.248*** 0.126** [5.02] [2.16] LEV -0.0713*** -0.0156 [-3.68] [-0.54] CF 0.162 -0.13 [1.29] [-1.10] GROWTH -0.0125 0.0279 [-0.53] [1.15] BIG4 -0.00636 -0.0265*** [-0.82] [-2.68] Source: The author 17 4.2.2.2. Data aggregation of negative and positive discretionary accruals Next, the author performed the sequence of regression steps of the model (**) for the data aggregation of negative and positive discretionary accruals. Finally, the most suitable model being selected was the dynamic GMM as it can solve the endogenous problem existed among variables that the previous models such as Pooled OLS, FEM, REM, GLS cannot. Table 4.9: Results of the dynamic GMM in related to data aggregation of separation of negative and positive discretionary accruals Variable Significance EM_L1 0.0913*** [3.12] BOARD -0.00446 [-1.15] NED 0.00187 [0.55] DUAL 0.00748 [0.86] FAD 0.0132 [1.44] BSR -0.00219 [-0.60] AGE 0.000606 [1.02] CO 0.0460** [2.44] MO -0.0289 [-1.10] SO -0.0405** [-2.24] FO -0.0704*** [-2.94] SIZE 0.00674** [2.21] ROE 0.220*** [5.06] LEV -0.0518*** [-3.00] CF 0.0721 [0.78] GROWTH -0.0108 [-0.62] BIG4 -0.0117* [-1.76] Significance: * p < 0.1, ** p < 0.05, *** p < 0.01 Source: The author 18 4.2.3. The comparison results of regression of the data split of negative and positive discretionary accruals Table 4.10: The comparison results of regression of the data split of negative and positive discretionary accruals Variable Data split Data aggregation Negative DA Positive DA Aggregated DA EM(NEG) EM(POS) EM EM_L1 0.0642** 0.0573* 0.0913*** [2.11] [1.96] [3.12] BOARD -0.00412 -0.00487 -0.00446 [-0.94] [-0.77] [-1.15] NED -0.00159 0.00347 0.00187 [-0.42] [0.65] [0.55] DUAL 0.0124 0.00156 0.00748 [1.19] [0.12] [0.86] FAD 0.016 0.00837 0.0132 [1.55] [0.45] [1.44] BSR -0.00414 0.00235 -0.00219 [-1.02] [0.36] [-0.60] AGE 0.00144** 0.000561 0.000606 [2.30] [0.54] [1.02] CO 0.0432** 0.0457 0.0460** [2.03] [1.45] [2.44] MO -0.0604** -0.0576 -0.0289 [-2.39] [-1.39] [-1.10] SO -0.0251 -0.0676** -0.0405** [-1.27] [-2.32] [-2.24] FO -0.0446* -0.0966** -0.0704*** [-1.81] [-2.29] [-2.94] SIZE 0.00429 0.00638 0.00674** [1.23] [1.28] [2.21] ROE 0.248*** 0.126** 0.220*** [5.02] [2.16] [5.06] LEV -0.0713*** -0.0156 -0.0518*** [-3.68] [-0.54] [-3.00] CF 0.162 -0.13 0.0721 [1.29] [-1.10] [0.78] GROWTH -0.0125 0.0279 -0.0108 [-0.53] [1.15] [-0.62] BIG4 -0.00636 -0.0265*** -0.0117* [-0.82] [-2.68] [-1.76] Significance: * p < 0.1, ** p < 0.05, *** p < 0.01 Source: The author Testing results of hypotheses are presented in Table 4.11 as follows: 19 It can be seen from the testing results of hypotheses that the impacts of independent variables in the characteristics of corporate governance and control variables on earnings management with the same model provided different the results in two cases including data split and data aggregation of negative and positive discretionary accruals. Testing results of hypotheses Hypotheses Contents Testing results Data split Data aggregatio