Data Collection Conclusion, Limitations and Future Research

218 to estimate real earnings management. To eliminate bias, 918 samples that were in an industry-year group with less than 30 samples were excluded from the sample pool. Furthermore, to eliminate the effect of outlier bias, the top and bottom 1 of independent and dependent variables were winsorized. A total of 6,440 firm-year samples were used for the analyses. TABLE 3 shows the majority shareholder ownership of the companies. The majority shareholders‘ ownership of all companies was about γ0. It was about 17 in the lower- level group, and 43 in the higher-level group. When we compared the majority shareholders‘ ownership of companies according to earnings management brackets Eε1, EM2, and EM3, we found that it was almost the same for EM1 and EM3 but relatively larger in EM2. Insert TABLE 3 about here

V. EMPIRICAL RESULTS 5.1 Descriptive statistics

The descriptive statistics of the variables used in the analyses are presented in TABLE 4. We compared the lower ownership group with the higher ownership group for all samples and for each earnings management bracket EM1, EM2, and EM3. First, we compared the difference between earnings management brackets. Abnormal cash flow from operations ACFO, abnormal production costs APC, and abnormal discretionary expenses ADE were relatively small in EM1 and EM3, but large in EM2 compared to the full sample. This difference implies upward earnings management in EM1 and EM3, and downward earnings management in EM2. 219 Second, we compared all the variables between higher and lower majority shareholder ownership groups in each earnings management bracket. In the whole sample and the EM1 bracket, the REM of high majority shareholder ownership was lower. However, in the EM2 and EM3 brackets, there was no difference between high and low majority shareholder ownership. The results with ACFO, APC and ADE were almost identical to those obtained using REM. Insert TABLE 4 about here TABLE 5 shows the correlations between variables. OWNH had significantly negative relationships with REM and ACFO while it had negative but insignificant relationships with APC and ADE. This result is consistent with the idea that when the majority shareholders‘ ownership is higher, a company generally reduces real earnings management. Next, we use multivariate analyses to examine the effect of majority shareholder ownership on real earnings management according to earnings management incentive brackets. Insert TABLE 5 about here 5.2 Multivariate results 5.2.1 Test of hypothesis 1 TABLE 6 shows whether majority shareholder ownership is associated with real earnings management. A negative positive OWNH coefficient implies that a company avoids conduct real earnings management when majority shareholder ownership level is high. The result shows that OWNH was significantly negatively associated only with ACFO at the one percent level. However, we were not able to find evidence of monitoring by majority shareholder ownership with REM, APC, and ADE. These results generally 220 suggest that there is no systematic relationship between the ownership and real earnings management. Insert TABLE 6 about here SIZE, MTB, and NI were significantly negative in all models, which is consistent with the expectations discussed earlier.

5.2.2 Test of Hypothesis 2

When we tested hypothesis 1, we found no systematic relationship between the majority shareholders‘ ownership and real earnings management. However, majority shareholders have differential incentives in monitoring real earnings management if their sensitivity to real earnings management related to future performance is different as the prospect theory assumed. That is, monitoring by majority shareholders may only be effective in the upward earnings management bracket since it causes low operating performance in the future unlike downward real earnings management. Thus, to test hypothesis 2, we investigated the effects of the majority shareholders ‘ ownership on real earnings management according to earnings management incentives. TABLE 7 shows the results of the test of hypothesis 2, and confirms that the results can differ according to their incentives. Insert TABLE 7 about here Hypothesis 2a refers to EM1, which includes companies with upward earnings management incentive. All the coefficients of EM1 were significantly positive, and Eε1×OWNH was significantly negative correlated with REε. This result was consistent across the ACFO, APC, and ADE models. Positive sign of EM1 indicates that these