Self-selection Bias
Our analyses have been so far performed on the assumption that auditor designation is exogenous. However, as shown in Table 2, high leverage, concentrated ownership structure, and extensive intra-group loans are major factors leading a firm to have its auditor designated by the FSC. If firms influence the decision of being designated as well as discretionary accruals and the error term of the model for being designated is correlated with the error term of Eq. (4) due to unobservable omitted variables, an OLS regression of discretionary accruals on a dummy for auditor designation in Eq. (4) would suffer from self-selection bias (Heckman 1979). A simple OLS estimation of Eq. (4) that ignores this selection bias leads to inconsistent coefficients.
To correct for this potential self-selection bias problem, we add the following probit regression to our basic regression model (4) to form a two-stage treatment effects model (Barnow et al. 1981; Maddala 1983; Leuz and Verrecchia 2000). We posit the following probit model to explain the FSC’s decision on auditor designation:
it it
it
it it
it it it
u mmies IndustryDu GROUP
WEDGE
OWN LOAN
LEV ROA
DESIG
+ +
+ +
+ +
+ +
=
−
−
−
−
−
−
1 6
1 5
1 4
1 3
1 2 1 1
0
*
β β
β β
β β
β (5)
where:
DESIG*it = the firm’s unobservable extent of being auditor-designated.
If DESIG* it is positive, the firm’s auditor is designated,
27
thereby a dummy variable DESIGit taking the value of 1;
otherwise, DESIGit=0;
ROAit−1 = return on assets for firm i and year t - 1;
LEVit−1 = the ratio of total debts to total assets for firm i and year t - 1;
LOANit−1 = the sum of loans, including loan guarantees, to directors, large shareholders, affiliated companies, and other related parties/lagged total assets for firm i and year t - 1;
OWNit−1 = the largest shareholder’s and his/her family member’s ownership for firm i and year t - 1;
−1
WEDGEit = the difference between control rights and cash flow rights held by the largest shareholder for firm I and year t - 1;
−1
GROUPit = A dummy variable that takes the value of 1 if a firm is affiliated with a business group and zero otherwise;
IndustryDummies = Dummy variables controlling for industry differences; and uit = error term.
We model the decision of auditor designation as a function of return on assets (ROA), leverage (LEV), the extent of intra-group loans (LOAN), concentrated ownership (OWN), the difference between control rights and ownership rights (WEDGE), and a firm’s affiliation with a business group (GROUP). These variables are all measured as of year t- 1 since the value of these variables in year t - 1 affects the decision of auditor designation in year t. We include return on assets (ROA) because, as shown in Table 3, earnings performance measured by ROA is quite different between the designated and non- designated samples. We also include LEV, LOAN, and OWN since these variables are related to the major reasons for auditor designation in Table 2.
In addition, we include two more variables that reflect the features of corporate governance systems of Korean firms. Corporate Korea is well characterized by chaebols
28
(large business groups or conglomerates) that play a dominant role in many aspects of the Korean economy. As of 1997, the largest 30 chaebols in Korea accounted for 62.54%
and 72.63% of total assets and gross sales, respectively, of all firms listed on the Korea Stock Exchange (Bae et al. 2002). Like other business groups in emerging markets, chaebols can be viewed as a collection of business enterprises in a wide range of industries that are typically controlled by members of a founding family. Using a large sample of Korean firms, Joh (2003) finds that during the 1993-1997 period, group- affiliated firms show significantly lower profitability compared to independent firms.
Bae et al. (2002) provide evidence that group-affiliated firms in Korea make takeover decisions that do not necessarily contribute to maximizing the market value of an acquiring firm, a finding consistent with the “tunneling” view of Johnson et al. (2000b).
The aforementioned studies suggest that firms affiliated with diversified business groups suffer from greater agency problems than independent firms. To the extent that the agency problems reduce earnings performance, one can predict that firms affiliated with business groups are more likely to be auditor designated. Thus, we include a dummy variable (GROUP) indicating chaebol membership in Eq. (5). GROUP takes the value of 1 when a firm belongs to the business groups identified by the Korea Fair Trade Commission (KFTC).11
Previous research suggests that as the divergence of management control rights and ownership rights becomes larger, potential agency problems associated with the separation of control from ownership become exacerbated and thus corporate earnings performance becomes poorer. Firms with higher divergence are more likely to engage in
11 The KFTC legally defines a business group as “a group of companies, more than 30% of whose shares are owned by some individuals or by companies controlled by those individuals or those that are practically
29
opportunistic earnings management to mask poor earnings performance, and thus they are more likely to have their auditors designated by the FSC. We measure WEDGE as the simple difference between control rights and cash flow rights in the hands of the largest shareholder. The control rights include the direct ownership (cash flow rights) of the largest shareholder and ownership stakes held by his/her family members and affiliated firms. We calculate OWN and WEDGE from the KIS database, which provides the names and shareholdings of the largest shareholder and his/her family members. The inclusion of ownership-related variables in Eq. (5) reduces our sample size to 2,843, of which 504 firm-year observations are auditor-designated.
In the first stage, we estimate a probit model in Eq. (5) to obtain the inverse Mills ratio. In the second stage, we include the inverse Mills ratio as an additional control variable to correct for potential self-selection bias in Eq. (4). That is,
it it
it
it it
it it
it it
v mmies IndustryDu Mills
BIG
TAC L LEV
SIZE OCF
DESIG DAC
+ +
+ +
+ +
+ +
+
=
7 6
5 4
3 2
1 0
6
1 β
β
β β
β β
β
β (6)
where Millsit denotes the inverse Mills ratio for firm i and year t, vit is an error term, and other variables are as defined earlier.
Table 8 reports the estimation results of Eqs. (5) and (6). Panel A of the table reports coefficients of the probit model, along with z-statistics. The likelihood ratio statistics indicates that Eq. (5) has significant explanatory power and the classification rate suggests that 77.1% of the choices were predicted correctly by Eq. (5). The coefficients on ROA, LEV, and OWN are highly significant with expected signs. The coefficients on LOAN, WEDGE and GROUP are insignificant.
controlled by them despite lower ownership control”.
30
Panel B of Table 8 presents the regression results for Eq. (6). The results show that the auditor designation dummy remains highly significant with an expected negative sign. Other control variables become more significant and preserve expected signs.
Further, the significance of the inverse Mills ratio (Mills) indicates that it is important to adjust for self-selection bias. Overall, the results in Table 8 suggest that auditor designation effectively limits the ability of managers to boost reported earnings through opportunistic earnings management.