Endogeneity and Self-selection Bias

Since firms will choose their level of disclosure based upon consideration of the costs and benefits associated with comprehensive disclosure, CP_i will be endogenous, and as such, the OLS regressions would tend to suffer from self-selection bias. As noted in the political cost hypothesis of Wagenhofer (1990), if the comparative benefits of comprehensive disclosure are greater than the costs, a firm will self-select its preferred choice to signal a higher level of transparency (Christensen and Feltham, 2000; Suijs, 2005). Leuz and Verrecchia (2000) also adopt the self-selection model to examine the effects of increased disclosure levels.

For taking into account the potential self-selection bias, the process of deciding the choice of voluntary disclosure has to be specified. It is supposed that there is an unobservable utility of interest UCP concerning the disclosure of compensation paid to directors and executives. The utility of choice can be specified as following.

, 1 1 , 1; , 0 0 , 0

i CP i CP i CP i CP i CP i CP

U ₌ =Z γ ₌ +η ₌ U ₌ =Z γ ₌ +η ₌ (5)

where Ui,CP=1 is the utility concerning the comprehensive disclosure of compensation for firm i (CP = 1); whilst Ui,CP=0 is the one concerning non-comprehensive disclosure (CP = 0); Z is a vector of determinants of whether a firm will voluntarily disclose comprehensive information on compensation for directors and executives; γ is a vector of parameters to be estimated; and ηi,CP is the error term.

If the relative benefits of comprehensive disclosure are larger than its costs, the firm will choose to signal high quality of transparency by providing comprehensive disclosure of compensation (Christensen and Feltham, 2000; Hyytinen and Pajarinen, 2005). This study follows this proposition and suggest that if the utility concerning the comprehensive disclosure is larger than the one concerning the non-comprehensive disclosure, Ui,CP=1 > U i,CP=0, the firm will voluntarily disclose comprehensive

information on director and executive compensation in the annual financial report. The difference between the two utility functions can be specified as the bivariate choice.

(

^{, 1 , 0}

) (

^{* * 0}

) (

^{* 0}

)

i i CP i CP i i i

CP =I U ₌ >U ₌ =I Z γ +η > =I CP >

% % (6)

where ηi* = ηi,CD=1 – ηi,CD=0; and γ*= γCP=1 – γCP=0.

Following Heckman (1979), the error term, ηi* is assumed to follow a symmetric distribution, determined as standard normal assumption. CPi* is an unobserved continuous latent variable of interest that can be measured by another observable discrete variable CPi and I(•) is an identity mapping. That is, if a firm evaluate the disclosure decisions and find that the benefits of comprehensive disclosure is larger than its costs (UCP=1>Ui,CP=0; CPi* > 0), then it will voluntarily disclose comprehensive information on compensation paid to directors and executives (CP = 1).

In response to the potential problems of endogeneity and self-selection bias, the two-stage Heckman model is adopted. Following Heckman (1979), the probit model could be adopted to control the problem of endogeneity and self-selection effect. It is also able to determine the expectation of index function as a probability through the mapping from the transformation function, denoted by F.

( )

ⁱ

^{( )}

two-stage self-selection model, the parameters are estimated using the Heckit procedure (Heckman, 1979). In the first stage, the inverse Mills ratio that captures the self-selection effect is obtained from the probit model in the selection (disclosure) equation.

( )

where γ is a vector of the parameters to be estimated by regressing CP_i on a vector of the instrumental variables (Z). The vector Z includes the control variables in Model (1) and the instruments for comprehensive disclosure. α is the parameter to be estimated.

The conditional expected market value of the firm with or without voluntary disclosure of comprehensive information on compensation paid to directors and executives are further specified as following.

( ) ( )

where λi,CP=1(•) is the inverse Mills ratio used to adjust the self-selection effect caused by the choice of comprehensive disclosure, λi CP, ₌1

(

Z γi

) (

=ϕ Z γi

) (

Φ Z γi

)

and

( ) ( ) ( )

, 0 1

i CP i i i

λ ₌ Z γ = −^⎡⎣ ϕ Z γ ^{⎤ ⎡}⎦ ⎣ − Φ Z γ ^⎤⎦

% . φ(•) is the standard normal distribution and Φ(•) is its corresponding cumulative distribution function.

The first-stage disclosure equation can be estimated by the probit model:

*

where CP_i^* is the unobservable net benefit of comprehensive disclosure; CP_i is equal to 1 only when the net benefits of comprehensive disclosure (CP_i^*) are greater than zero, thereby inducing the firm to voluntarily provide comprehensive disclosure of compensation.

The instruments comprising of proxies of board independence encompass two dimensions of internal governance mechanisms: ownership structure and the board of directors. The criterion of ownership includes the proportion of shares owned by domestic trust funds, outside funds, managers, controlling directors, family funds, family unlisted corporations and the critical controlling shareholding. The criterion for the board of directors includes the proportion of directors appointed by outside listed corporations and family unlisted corporations, the proportion of supervisors appointed by outside listed corporations and outside funds, and the size of the board.

Using the instruments and their corresponding parameters, it can further calculate the term ^λi=⎡_⎣^φ

(

^{Z γ}_%^{i 13}

) (

^{Φ Z γ}_%^{i 13}

)

^⎤_⎦^{⋅CPi+ −⎡}_⎣

(

^φ

(

^{Z γ}_%^{i 13}

) )

^Φ

⁽

^{Z γ}_%^{i 13}

⁾

^⎤_⎦^{⋅ −}

⁽

^{1 CPi}

⁾

% % % %

to correct for the

self-selection bias, where φ(·) and Ф(·) are the respective density function and cumulative distribution function for the standard normal. The second stage involves the OLS estimation in the valuation model which includes the addition of the λi

regressor and the fitted values of comprehensive disclosure (CP^ˆ_i).

,

where σε,η is the error term covariance, with εi and ηi, satisfying certain assumptions.

These assumptions are: (i) that the two error terms follow bivariate normal distribution with unity normalized standard deviations, σε2 and ση2, and correlation ρε,η, i.e., εi , ηi ~ N (0, 0, σε2 , 1, ρε,η); and (ii) that εi|ηi=σε,ηηi+νi, where νi ~ N(0,1) and E(εi|ηi)=σε,ηηi.

As noted in many of the prior studies, the instruments determining comprehensive disclosure may be attributable to the level of board independence (Core et al., 1999;

Core, 2001; Ryan and Wiggins, 2004; Makhija and Patton, 2004; Gordon, 2005;

Laksmana, 2008). Khanna et al. (2004) find that Taiwan has a relatively lower

governance index, essentially as a result of higher family control, pyramidal structures and lower transparency. The internal governance mechanisms of ownership structure and board composition are therefore adopted as instruments in the present study.

Since Bushee and Noe (2000) and Barako et al. (2006) argue that outside shareholders require higher disclosure standards, this study takes the percentage of shares owned by domestic trust funds (%DTF) and the percentage of shares owned by outside funds that are not controlled by major shareholders (%OF) as additional instruments. Major shareholders are defined as investors owning a significant proportion (10 per cent) of the shares. For newly-listed companies, major shareholders are defined as investors who are ranked in the top ten, in terms of total holding stake, or those with more than a 5 per cent holding stake in the company.

The share interests of managers are maximized by their greater willingness to provide higher disclosure transparency, thereby further increasing the market value of the firm (Core, 2001; Nagar et al., 2003; Makhija and Patton, 2004). However, it is noted in Nagar et al. (2003) that managers focusing on personal interests may be reluctant to disclose private information, particularly in those circumstances where there is a lack of any real incentives. The proportion of shares owned by executives (%MNG) is therefore taken as an instrumental variable.

Concentrated ownership could lead to lower transparency and fraudulent activities, particularly in firms with poor board independence. Thus, several studies suggest that disclosure may reduce the information advantages of insiders (Holland, 1998; Huddart et al., 1999; Hossain et al., 2005) and Bannister and Newman (2006). This study therefore includes two additional instruments, controlling directors (%CD), which is measured as the percentage of shares owned by major shareholders on the board, and critical control (%CC), which is the percentage of shares the major directors need to

hold to achieve absolute voting power or control rights (Cubbin and Leech, 1983).

Another property of the Taiwanese data is the prevalence of family control, which is common to many Asian and European capital markets. Setia-Atmaja et al. (2007) find that the board independence is lower in family firms, whilst Chen et al. (2008) suggest that firms with higher family ownership tend to disregard voluntary disclosure practices. Therefore, the percentage of shares that are owned by foundations controlled by family directors (%FF) and the percentage of shares that are owned by unlisted companies controlled by family directors (%FU) are both included as instruments.

Byrd et al. (1998) argue that factors encouraging compensation committees to provide voluntary disclosure are the major concerns of external stakeholders. Soffer (1998) also suggests that such concerns provide a mechanism which can effectively disclose poor compensation practices. Furthermore, both Ho and Wang (2001) and Hossain et al. (2005) find positive associations between voluntary disclosure and the proportion of independent directors on the board. Therefore, the proportion of directors and supervisory representatives appointed by outside listed corporations, (OCD and OCS), and the percentage of supervisory representatives of outside funds (OFS), none of which are controlled by major shareholders, are included as instruments.

However, a higher proportion of family directors may also be detrimental to board independence (Chen et al., 2008). The percentage of directors appointed by unlisted corporations under family director control (FCD) is therefore taken as an additional instrument.

Since the positive association between board size and firm performance can be strengthened by the greater monitoring power arising from larger board size (Dalton et al., 1999; Certo et al., 2001), firms with such monitoring power will tend to voluntarily provide comprehensive disclosure. However, several studies suggest that as boards

become smaller, they may also become more capable of holding frank discussions and engaging in more effective monitoring (Jensen, 1993; Vafeas, 2000; Gordon et al., 2002). As a result, smaller board size can enhance the informative nature of disclosure.

Board size (BSIZE), measured as the ratio of the total number of directors to the natural log of total assets, is therefore taken as an additional instrument.

I further examine whether comprehensive disclosure leads to a higher market value after controlling for potential problems of endogeneity and self-selection bias.

Table 8 reveals that λi is significant, indicating that the choice of comprehensive disclosure is not random, and that self-selection bias is prevalent in our setting. In particular, various monitoring mechanisms, including the pressure provided by fewer insiders, more outsiders, smaller board size, higher managerial shareholdings, lower family control and diversified ownership, are all found to lead to higher levels of transparency. The evidence suggests that under a scenario within which firms are given broad discretion with regard to their provision of greater levels of transparency, firms with higher levels of board independence will voluntarily provide comprehensive disclosure of compensation, thereby leading to higher market value.

Although the evidence shows that the instruments of board independence can reduce the potential problem of endogeneity and self-selection bias, those instruments themselves could also be faced with another endogeneity problem. Since firms with better governance mechanisms may voluntarily provide comprehensive disclosure of compensation, whilst they are also more likely to provide higher independence of the board. Such phenomenon leads to a potential problem that board independence taken as exogenous instrument and comprehensive disclosure of compensation are simultaneously determined by other factors, such as other criteria of internal governance mechanisms or the overall governance practices. Another candidate of the

exogenous instrument is the level of competitiveness in their industries. Since firms could face non-discernible pressures from manager labor market, further changing the transparency level of compensation information that is valuable to identify the firm’s future strategies, the executives’ capability, and competitiveness in its industry.

Table 8 The Effects of Comprehensive Disclosure of Compensation with Heckman Self-selection Model

This table presents the estimated coefficients from the following regression models.

Disclosure Model:

Coefficient z-statistic Coefficient t-statistic

CP – – 0.963 8.29***

%FF – – – –

a In the first-stage disclosure equation (Model I), the dependent variable is comprehensive disclosure (CP). If the firm voluntarily provides comprehensive disclosure not only on the level of compensation paid to its executives, but also on the level of compensation paid to its directors, then the compensation disclosure variable, CP, takes the value of 1, otherwise 0. The instrumental variables (Z) include the proportion of shares owned by domestic trust funds (%DTF), the proportion of shares owned by outside funds (%OF), the proportion of shares owned by managers (%MNG), the proportion of shares owned by controlling directors (%CD), the critical controlling shareholding (%CC), the proportion of shares owned by family funds (%FF), the proportion of shares owned by family unlisted corporations (%FU), the proportion of directors appointed by outside listed corporations (OCD), the proportion of supervisors appointed by outside listed corporations (OCS), the proportion of supervisors appointed by outside funds (OFS), the proportion of directors appointed by family unlisted corporations (FCD) and the size of the board (BSIZE). The control variables come under two categories of firm characteristics and profitability. Firm characteristics include firm size (SIZE) measured by the natural log of total assets; debt ratio (DEBT) measured by the ratio of total liabilities to total assets; proprietary ratio (INVST) measured by the ratio of total investment to total assets; and free cash flow (FCF) measured by the ratio of operating cash flow to total assets. The profitability includes return on assets (ROA) and earnings per share (EPS).

b In the second-stage valuation equation (Model II), the dependent variable is the Tobin’s Q (Q), which is defined as the sum of the firm’s market capitalization and the book value of debt, divided by the book value of total assets. the coefficient on λ examines the effect of self-selection bias; ρε,η is the correlation of the error terms in the disclosure and valuation equation; σε is the standard deviation of the error term in the second-stage equation.

c *** indicates significance at the 1% level; ** indicates significance at the 5% level; and * indicates significance at the 10% level.

在文檔中 薪酬資訊揭露與公司價值 (頁 43-53)