How XBRL Affects the Cost of Equity Capital? Evidence from Emerging Market 1

Transcription

1 How XBRL Affects the Cost of Equity Capital? Evidence from Emerging Market 1 Songsheng Chen, School of Management and Economics, Beijing Institute of Technology,China chenss@bit.edu.cn Tel: Wenying Li, School of Management and Economics, Beijing Institute of Technology,China liwenyinginearnest@gmail.com Donglin Wu, School of Management and Economics, Beijing Institute of Technology,China @bit.edu.cn Acknowledgements: We thank professor Liyan Wang, Lina Wu from Peking University, and other participants of the 12th International Symposium on Empirical Accounting Research in China (ISEAR), held in Weihai City on Dec. 20th, 2013, for their helpful comments on earlier drafts of this paper. We are especially grateful to professor Prasad Padmanabhan from St. Mary s University, and Sophie Kong from Western Washington University for their constructive suggestions. Songsheng Chen acknowledges financial support from the National Natural Science Foundation of China(NSFC ). 1 Since data in our analysis are fully available in CSMAR and Wind Database, we encourage readers to collect these data directly from the two database as they need.

2 How XBRL Affects the Cost of Equity Capital: Evidence from an Emerging Market ABSTRACT This study explores the impact of XBRL on CEC from the perspective of the full financial information chain, and the research starts at the beginning of the information chain the intracompany environment represented by corporate governance (CG) and the dominant shareholder identity. Using a one-group pre- and post-test design, we find that XBRL reduces CEC; when there is a high level of CG, XBRL adoption leads to a greater reduction in CEC. State-owned identity has a nonlinear relationship with CG state-owned enterprises (SOEs) with a low CG level experience less CEC reduction than non-state-owned enterprises (NSOEs) after XBRL implementation, while the CEC of SOEs with high CG levels reduces more than that of NSOEs. Key words: XBRL, Cost of Equity Capital (CEC), Corporate Governance (CG), SOEs 1

3 I. INTRODUCTION Traditionally, the data items in PDF or HTML formats of financial reports cannot be extracted automatically or combined into computers and analytical software. Users have to collect, read, and justify the contents of documents in these formats manually, increasing cost, difficulty, and opportunities for error in data comparison. XBRL is an international financial reporting code used to tag financial and non-financial data to standardize the preparation, publishing, and exchange of financial information in the world (Boritz and No 2008). XBRL was developed in 1998 by Charles Hoffman, and in 2002, XBRL International, Inc. was established. XBRL technology not only integrates complex financial information, but also assists in the analysis of financial and non-financial data for investors, bondholders, or other users in the capital market (Kim et al. 2012; Apostolou and Nanopoulos 2009).. Our motivation for researching the effect of XBRL on the cost of equity capital (CEC) is twofold. Prior researchers find that high earnings quality can decrease information asymmetry (IA) (Greenstein and Sami 1994; Hagerman and Healy 1992; Healy et al. 1999; Heflin et al. 2005; Leuz and Verrecchia 2000; Welker 1995; Brown and Hillegeist 2007) and CEC (Amihud and Mendelson 1986; Kyle 1985; Diamond and Verrecchia 1991; Botosan and Plumlee 2002; Easley et al. 2002; Lambert et al. 2007). Low IA can also decrease CEC (Francis et al. 2004; Francis et al. 2008; Lambert et al. 2007; Florou and Pope 2012). Besides that, Research on the impact of XBRL implementation on IA and thus CEC has yielded inconsistent results. 2

4 Blankspoor et al. (2012) and Liu et al. (2014) find an increase in CEC after XBRL adoption, while Kim et al. (2012), Li et al. (2012), Yoon et al. (2011), Bai et al. (2012), and Chen and Li (2013) provide evidence that XBRL decreases CEC. Using Chinese data, we want to examine the effect of XBRL on CEC, both in the short- and long-term,period and to explain these conflicting conclusions. According to the promotion destination of XBRL implementation, we initially hypothesize that XBRL adoption decreases CEC. Prior research has assessed XBRL s impact on CEC primarily by looking at information processing costs (Li et al. 2012) or direct IA (Yoon et al. 2011),, both of which are external market responses. We begin inside the corporation and focus on how different internal characteristics have different impacts on CEC. Our research incorporates the full information supply chain: the interior of the corporation at the front, XBRL as the information disclosure engine in the middle, and CEC as investors market response at the end. There is sufficient evidence to prove that corporate governance (CG), one critical and inevitable factor when studying intracorporation issues, plays an important role in financial reporting quality (FRQ) and CEC (Love and Klapper 2004; Doidge et al. 2007). Since we hypothesize that the directional effect of CG on CEC is the same as that of XBRL on CEC, we further hypothesize that, with the collective and collaborative effect of the front and middle of the financial information chain, high CG would lead to a greater reduction in CEC after XBRL implementation. China has a unique business ownership structure, with many state-owned 3

5 enterprises (SOEs). SOEs have different policies and greater resources than non-state-owned enterprises (NSOEs), which impacts investors perceptions and leads to different CECs. Researchers find mixed and confusing evidence regarding the relationship between state ownership and firm performance and, therefore, CEC (Sun et al. 2002; Ben-Nasr et al. 2012; Ng et al. 2009; Hess et al. 2010; Yu 2013). Since SOEs both have inefficient board mechanisms, which increase CEC, and an invisible government guarantee, which negatively affects CEC, we hypothesize that state ownership has a nonlinear connection with CEC and that the reduction in CEC after XBRL implementation is significantly different for SOEs and NSOEs. [Insert Figure 1 here] To test our hypotheses, we obtain firm-year observations from 2005 to from China s publicly listed A-share firms. We construct a one-group, pre- and postdesign to test whether XBRL affects CEC and design matched pairs to find out how XBRL affects CEC in the case of different corporate characteristics. We use Dhaliwal et al. s (2005) and Claus and Thomas s (2001) models (referred to subsequently by the creators initials, DH and CL models) to measure CEC, and design a system of CG. Our results show that, XBRL adoption decreases CEC by leading to better FRQ and decreased IA, and results in lower CEC for firms with high CG. State ownership has a U-shaped relation to CEC. In China, SOEs generally experience a greater reduction in CEC after XBRL adoption than NSOEs, but this result is primarily driven 1 To eliminate the influence of the global financial crisis in 2008, we dropped the 2008 data. For our three-year pre-and post-test design, the pre-test period is from 2005 to 2007, and the post-test period is from 2009 to

6 by high CG SOEs. Low CG SOEs actually have less of a reduction in CEC than matched NSOEs. The robustness of our results is demonstrated by: 2) using the capital asset pricing model (CAPM); a market-based method, to measure CEC; 1) testing whether XBRL improves FRQ, measured by timeliness, accrual quality, smoothness and discretionary accruals; 3) using stepwise regression for CG; 4) using SUR method to test H2 and H3; 5) controlling for financial crisis influence. Section II of our paper introduces XBRL adoption and develops hypotheses; Section III describes the data and empirical design; Section IV presents and discusses the empirical results; Section V reports the robustness test results; and our summarization and conclusions are in Section VI. II. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT 2.1 Why China? We choose China for our research because it is leading developing countries in adopting and deploying XBRL in financial reporting, by having a mandatory program for its use. All publicly listed firms in the Shanghai and Shenzhen stock exchange centers have been required to report financial statements in XBRL format and PDF files simultaneously since the beginning of As to the effect of XBRL on IA, recent studies have shown mixed results. Using US data from June 15, 2008 to June 14, 2010, Blankespoor et al. (2012) uses a one-year, pre- and post-design and find an increase in IA. Kim et al. (2012), on the other hand, saw a decrease in IA. Li et al. (2012) use a three-year pre- and post-design covering the period from April 4, 2005 to 2 For more information about XBRL in China, see China s XBRL on two websites: Shenzhen stock exchange center: and Shanghai stock exchange center: 5

7 April 30, 2012, including both voluntary and mandatory programs, and identify a decrease in IA. It seems that sample sizes and period lengths are possible reasons for the differing results. In the short-term period, XBRL appears to increase adverse selection but, in the long-term, decreases transaction costs. Adding to the uncertainty is the difference in results worldwide. Data from a mandatory program in Korea led Yoon et al. (2011) to conclude that, in the short-term, XBRL decreases IA. Bai et al. (2012) finds the same for Japan s mandatory program. In China, Liu et al. (2014) chooses as the pre-phase period and as the post-phase, and finds an increase in IA. Chen and Li (2012) look at the six months before and after XBRL implementation and argue that XBRL decreases IA by choosing October 2005 as the starting of XBRL s implementation.however, Since 2004, the Shanghai stock exchange center requires 50 firms to use the XBRL format for their 2003 annual report submissions; 118 firms voluntarily submit their annual reports in the XBRL format. Beginning in 2004, all publicly listed firms are required to submit XBRL format and PDF files, but the firms can choose the XBRL format submission time. It is until 2009 that all publicly listed firms simultaneously submitted their 2008 annual reports in both XBRL format and as PDF files. Between 2004 and 2007, China s XBRL program is similar to the US s voluntary program which often see untimely submissions and potentially small sample sizes; therefore, we choose 2008, the year that XBRL became simultaneously with traditional files and can be visited by investors simultaneously, as the starting point for our study, in order to eliminate self-selection bias. Testing the short- and long-term effects of XBRL on IA and CEC 6

8 is important, as China now has the third largest market capitalization in the world, and is sure to become increasingly important to global investors. China also has a unique intracompany environment. In China, many listed firms are SOEs, and their internal governance mechanisms, such as Boards of Directors and Supervisory Boards, are often inefficient, with undeveloped operations, providing a unique setting for SOE s research stream. All of these characteristics affect FRQ and IA. Therefore, the intracompany environment in China provides a representative perspective to assess how XBRL adoption affects CEC. 2.2 XBRL and CEC Earnings Quality, IA, and CEC Our study is based on information risk theory. Information risk mainly derives from governance risk caused by agent costs (Ashbaugh et al. 2004), as well as poor disclosure quality ((Leuz and Verrecchia 2004).A fundamental role of accounting information in financial markets is to serve as a basis for capital allocation (Bhattacharya et al. 2011) and the theory of information risk indicates that information risk premium affects CEC, a basic explanation of which is the IA theory (Admati 1985). Our empirical research derives from three theory streams. It has been well established that an increase in information quality leads to a decline in CEC (Lambert et al. 2007). Francis et al. (2004) examines the relationship between CEC and seven attributes of earnings quality and finds a negative relationship between earnings quality and CEC. Firms with good earnings quality have more expansive voluntary 7

9 disclosures, which is associated with a lower CEC (Francis et al. 2008). Theory predicts that higher quality financial reporting can improve stock market liquidity and consequently reduce the cost of capital (Florou and Pope 2012). Second, the significantly positive relationship between IA and the CEC has been demonstrated in prior studies (Amihud and Mendelson 1986; Kyle 1985; Diamond and Verrecchia 1991; Botosan and Plumlee 2002; Easley et al. 2002). Easley and O Hara (2004) identify the proportion of information that is public versus private as a determinant of CEC. In their model, uninformed investors face an undiversifiable risk that arises from asymmetric information; an increase in the amount of private information increases the required rate of return. However, Lambert et al. (2007) dispute that it is IA per se that causes the CEC effect in pure competition settings, such as the one in Easley and O Hara (2004). Rather, reducing IA can affect CEC only when the reduction in asymmetry is accompanied by an increase in the average level of information precision. Third, prior research also indicates that an increased level of corporate disclosure reduces IA in the capital market (Greenstein and Sami 1994; Hagerman and Healy 1992; Healy et al. 1999; Leuz and Verrecchia 2000; Welker 1995). If investors differ in their ability to process earnings-related information, poor earnings quality can result in differentially informed investors and exacerbate IA (Diamond and Verrecchia 1991; Kim and Verrecchia 1994). Welker (1995) conducts the first empirical study to document an inverse association between disclosure quality and bid-ask spreads. Heflin et al. (2005) also finds that higher quality disclosures are associated with 8

10 greater liquidity. In a recent study, Brown and Hillegeist (2007) find an association between disclosure quality (based on AIMR scores) and the probability of informed trade (PIN) measure XBRL Impact Since XBRL provides investors with an effective way to prepare, publish, exchange, and analyze financial information (Boritz and No 2008; Debreceny and Robertsson 2001; Hodge et al. 2004; Stantial 2007), it has the potential to reduce IA by improving information disclosure quality and decreasing information processing costs (Yoon et al. 2011). Diamond and Verrecchia (1991). Diamond and Verrecchia (1991) report that, if the level of disclosure is increased, the level of IA is decreased, resulting in more demand from investors. Therefore, if XBRL adoption leads to an increase in the level of financial disclosure, IA should decrease, which could lead to a decrease in CEC and an increase in the firm's valuation. Li et al. (2012) finds that XBRL adoption results in a significant reduction in CEC by reducing investors information processing costs. We use China Data to investigate whether, in China, XBRL adoption reduces firms CEC from the perspective of improving FRQ. We expect that, after XBRL adoption, there will be a change in the two paths' strengths, and hypothesize the following: H1: XBRL adoption reduces CEC. 2.3 XBRL and CG Under the divergence between management and ownership rights, principals entrust professional agents to manage a firm s operations. An agency risk is generated 9

11 when management has an incentive to maximize their self-interest rather than firm value, creating IA problems between shareholders and managers, and IA creates a moral hazard problem. Agency costs also arise when investors cannot discern the true economic value of the firm, which is partially a function of the indistinguishable quality of management, i.e., IA leads to adverse selection problem. This lack of transparent financial information results in greater information risk being imposed on the shareholder. Without adequate controls, effective monitoring, and transparent financial information, rational investors will price-protect against expected agency costs, effectively raising the firm s CEC. CG encompasses a broad spectrum of mechanisms intended to mitigate agency problems by increasing the monitoring of managements actions, limiting managers opportunistic behavior, and reducing the information risk borne by shareholders (Ashbaugh et al. 2004). The study of this issue is motivated by the economic theory that greater disclosure lowers IA (e.g., Glosten and Milgrom 1985; Diamond and Verrecchia 1991; Healy et al. 1999) and estimation risk (e.g., Barry and Brown 1996; Lang and Lundholm 1996) and, therefore, CEC (Botosan 1997; Botosan and Plumlee 2002). Prior research (Coombes and Watson 2000; Doidge et al. 2007; Kermani 2008) suggests a negative relationship between other CG mechanisms and CEC, indicating that CG enhances firm value by reducing CEC. The results show that investors are willing to pay a premium for good CG and, in fact, already do so. Good CG can create greater investor protection, which increases investors willingness to provide 10

12 financing, which should be reflected in lower costs and greater availability of external financing (Love and Klapper 2004). Additionally, there is strong evidence that good non-disclosure CG leads to good earnings quality (Lin, Jerry W. et al. 2010; Pergola et al. 2009; Iyengar et al. 2010; Haw et al. 2011). Since XBRL creates a natural linkage between FRQ, external investors, and CEC, we use different levels of CG to determine how XBRL affects CEC. According to our analysis, XBRL and CG should have the same directional effect on both FRQ and CEC. With better financial information provided by high CG at the front of the financial information chain, the effect of XBRL on CEC, as an information disclosure engine in the middle of the information chain, will be greater. Therefore, our second hypothesis is as follows. H2: High CG leads to greater reduction in CEC after XBRL adoption. 2.4 XBRL and Identity of Dominant Shareholders Based on China s unique ownership structure, in state-owned firms, the government intervenes in firms operating decisions by imposing taxes, supervision, licensing requirements, approvals, etc., resulting in inefficiency in governance boards as well as checks and balance mechanisms. This impact is particularly evident in emerging markets, such as China (Sun et al. 2002; Ben-Nasr 2013). SOEs also serve as the government s tools for achieving political and social objectives. As a result, SOEs have less of an incentive than NSOEs to provide transparent financial reports to outside shareholders, leading to increased information risk. With higher agency cost and information risk, outside investors usually ask for a risk premium as a result of higher CEC from SOEs more often than from NSOEs (Huyghebaert and Wang 11

13 2012).Since XBRL can decrease CEC, we assume that SOEs could benefit more from XBRL use and, to some extent, decrease their CEC more, than NSOEs. On the other hand, however, political connections can also provide SOEs with priority access to rare natural and intellectual resources, beneficial tax policies, and can lower their financial costs (Ng et al. 2009; Hess et al. 2010; Yu 2013)). Under this invisible government guarantee, outside investors may not absolutely require a higher risk premium and thus a higher CEC for SOEs. Therefore, state ownership would have a nonlinear association with CEC. Since XBRL can decrease CEC, as we hypothesize, it is unclear whether SOEs experience a greater reduction in CEC than NSOEs after XBRL implementation, and we hypothesize the following: H3: State ownership has a nonlinear relation with CEC and there is a significant difference in CEC reduction between SOEs and NSOEs after XBRL implementation. III. SAMPLE SELECTION AND RESEARCH DESIGN 3.1 Sample Selection Our sample included firms from a broad cross-section of different industries, all of which traded on the Shanghai and Shenzhen A-share stock markets during the period of 2005 through Since XBRL adoption in China began at the end of 2008, we are able to compare the firms CECs for the three-year period before and after XBRL adoption. We eliminate financial firms from the sample because they face regulations that other firms do not. We also eliminate firms with missing codes in the China Securities Markets and Accounting Research Database (CSMAR) and the WIND database. Moreover, high-frequency stock market transaction data are 12

14 collected from the WIND database. All of the continuous firm-year variables included in our tests are required to have non-missing values, and are Winsorized at the 1st and 99th percentiles. To test our hypothesis that XBRL reduces CEC, we use a one-group pre- and post-test design. The major advantage of this is that it allows for adopters, in the pre-adoption phase, to serve as their own control group to evaluate post-adoption effects (Brazel and Dang 2007). Firms are selected that have data available for at least one year before 2008, during the time span of t-1 to t-3, and at least one year after year 2008, during the time span of t+1 to t+3. See Table 1 for sample selection procedures (Panel A) and sample sizes (Panel B). 3 [Insert table 1 here] 3.2 CEC Analysis To test H1, we examine the effect of XBRL on CEC via a pre- and post-test design, dropping 2008 data to control for the impact of the financial crisis. Therefore, the primary independent variable is a dummy variable, XBRL, which is set to 1 for the years after XBRL adoption (2009 to 2011) and 0 for the years before XBRL adoption (2005 to 2007). Following previous literature (Claus and Thomas 2001; Dhaliwal et al. 2005), we rely on two estimation models to construct measures of CEC. We also report results of individual measures separately in our analysis. Our model estimates CEC as follows: 3 The number of observations for each year is different because not all firms selected have data available from 2005 to 2011, the same condition as the different variables shown in Table 1. 13

15 Our first measure of CEC is based on this model (Dhaliwal et al. 2005) (called the DH model in the later literature). (1) (2) (3) (4) Where: Our second measure of CEC is based on this model (Claus and Thomas 2001) (called the CL model in the later literature). (5) (6) g = the yield on 5-year treasury bonds minus 3% For the control variables, we use those that have been shown to affect CEC. We control size (SIZE) and market-to-book ratio (MB), which Fama and French (1992) argue are the two key determinants of expected stock returns. Leverage (LEV) is included, as it is positively associated with expected returns. We add market beta (BETA) to control for a firm s systematic risk and include the long-term growth rate 14

16 (LNGROW) and analyst forecast dispersion (LNDISP), following Li et al. (2012). We include year indications to address potential time series variation in CEC and to make it possible to control for fixed year effects. Industry effects are also controlled. Our formal regression model is as follows: (7) 3.3 CG Impact In China, CG is taken into account by researchers later than in western countries. We design and divide CG into five subgroups. First, we choose ownership concentration, as do Dahlquist and Robertsson (2001), and management structure, as used by Bushee et al. (2010). We add SOE/NSOE and reclassify these as Ownership Structure and Board Governance. Quality of disclosure is also important to consider because it is the board of directors that manages information disclosure in annual reports, which is related to CG (Haniffa et al. 2002; Abdelrazik et al. 2013). Corporate social responsibility is another measure of CG, which reflects the quality of CG. An important advantage of governance structures and systems for corporate social responsibility research is their formal nature, which makes the nature of the interactions between firms and their stakeholders much easier to identify (Graaf and Stoelhorst 2013). On average, better-governed corporations tend to pursue a more socially responsible agenda through increased corporate social responsibility practices (Ntim and Soobaroyen 2013). Therefore, disclosure and ethics, as measures of 15

17 corporate social responsibility, are included in our CG system. Finally, managerial behavior is included because it also plays an important role in CG. (See the system of CG in Appendix B). Using PCA, we then investigate the principle components and calculate the firm-year CG scores. We conduct a multivariate regression as follows: (8) where Gov means the CG scores, and we expect a negative and significant sign of the interactive variable XBRL*Gov. To test the different effects of XBRL on CEC under different levels of CG (i.e., H2), high and low CG groups are divided by two quartile methods. One pair is divided by median and another pair is chosen from the lower quartile (25%) and upper quartile (75%) of CG scores. 3.4 Identity of Dominant Shareholders Impact To test H3, we use a matched pair design to determine if the identity of dominant shareholders (i.e., SOEs and NSOEs) affects XBRL s impact on CEC. We also conduct regression analysis to test the relationship between state ownership and CEC. SOEs are divided into two groups by CG score (i.e., high CG and low CG), and the groups are used to determine whether and how CG level influences differences in the effect of XBRL adoption on CEC between SOEs and NSOEs. 16

18 IV. EMPIRICAL RESULTS 4.1 CEC Analysis We investigate whether XBRL has an impact on CEC, as measured by the DH model (Dhaliwal et al. 2005, as described in Section 3.2), the CL model (Claus and Thomas 2001, as described in Section 3.2), and the average value of the DH and CL models. Panel A, Table 2 shows that the descriptive statistics do not indicate the existence of any extreme values, rdh(diff=-0.023, t=-4.9), rcl(diff=-0.047, t=-1.897), and the average of rdh and rcl(diff=-0.037, t=-9.635), all have significant differences before and after XBRL adoption. Significant differences are also observed for all of the control variables (Size, Mb, Lev, Beta, Lngrow, Lndisp) before and after XBRL adoption. Panel B, Table 2 indicates that the correlation matrix does not show any apparent multicollinearity problems and that CEC appears to be negatively correlated with the XBRL indicator. [Insert table 2 here] To test whether the value of CEC significantly decreases or increases after XBRL adoption, we conduct the multivariate regression analysis discussed in the previous section. Table 3, Panel A presents the regression results, reporting coefficient estimates, and t-values based on robust standard errors. In all three of the models, there is a significant reduction in CEC after firms initial XBRL filings (coeff.=-0.06, t=-6.53 for the DH measure; coeff.=-0.03, t=-2.06 for the CL measure; coeff.=-0.02, t=-2.18 for the average result). These results provide support for H1, that XBRL adoption decreases CEC. From an economic perspective, the estimated coefficients 17

19 translate into a range of basis point reductions, from 172 to 600 basis points in CEC, after firms initiated XBRL filings. With regard to control variables, we find that firms with large size (take mean(re) for example, Coeff.= , t=-2.12) and high market-to-book value (take mean(re) for example, Coeff.= , t=-2.45) have a lower CEC, while high leverage firms (take mean(re) for example, Coeff.=0.0039, t=3.59) tend to have a higher CEC. High growth firms (take rcl for example, Coeff.=-0.017, t=-3.53) have a lower CEC. [Insert table 3 here] When analyzing the one-year, pre- and post-test design, as shown in Table 3, Panel B, we find that XBRL decreases CEC; the results are the same in both the two-year and three-year pre- and post-test designs. In China, XBRL helps to reduce CEC in the short-term, as in the Japan and Korea markets, and in the long-term, as seen in the US market. The reason for the difference between our results and Liu et al. s (2014) and their similarity to Chen and Li s (2013), all using data from Chinese firms, may be the use of different starting points for XBRL implementation. As mentioned in Section II, it seems more appropriate to choose 2008 as the XBRL mandatory starting point when testing XBRL s effect on capital markets. We also find no evidence that XBRL increases IA in China in the short-term period as it does in the US market. One potential influencing factor may be the cultural differences between Eastern Asia and the US. Hofstede et al. (1997) states that cultural differences across nations or regions will lead people to have different 18

20 judgments and actions, and Liu et al. (2014) further finds that the effect of XBRL on investors forecast accuracy is affected by national culture. However, there are other potential explanations for XBRL s differing effects on CEC, such as differences in investors acquisitions and characteristics in different capital markets. More study is required to determine whether cultural differences are the cause of the differing effects of XBRL on CEC in the short- and long-term periods. 4.2 CG Analysis To test H2, following prior research (Dahlquist and Robertsson 2001; Bushee et al. 2010; Haniffa and Cooke 2002; Graaf and Stoelhorst 2013; Ntim and Soobaroyen 2013), we build a corporate governance system with the five CG subgroups and conduct a principal component analysis (PCA) to analyze the elements of CG (See Appendix B). The first four subgroups, Ownership Structure, Board Governance, Managerial Behavior, and Disclosure, account for 95.3 percent of CG. Table 4 shows the CG scores for the 7-year period from 2005 to 2011, where we see CG scores increasing significantly each year. Two elements of CG, Ownership Structure and Board Governance, have a greater impact on CG scores than Managerial Behavior and Disclosure. [Insert table 4 here] Using the CG scores described above, we perform a multivariate regression analysis to test the effect of CG on the impact that XBRL has on CEC (Table 5). For the full sample, a significant and negative correlation between the interactive effects of XBRL*governance and CEC (Coeff.=-0.025, t=-1.68) is observed, as expected. 19

21 Specifically, for firms with better CG, XBRL adoption resulted in a 68 basis-point reduction in CEC (Coeff.=-0.068, t=-3.03). These results suggest that the effect of XBRL on CEC is greater in firms with high CG, with a better FRQ, and substantiate H2. The results of an analysis of the control variables are consistent with the aforementioned regression regarding XBRL s effect on CEC. [Insert table 5 here] We also divide CG into two equal parts by median quartile and find that high CG helps XBRL to reduce more CEC; this finding is robust and immune from the quartile division method. 4.3 Identity of Dominant Shareholders Impact We examine the effect of state ownership on CEC and the regression results in the first column of Table 6, Panel A show a U-shaped correlation between state ownership and CEC, as we hypothesize in H3. Initially, state ownership has a negative relationship with CEC; however, beyond a certain level (about percent), state ownership is positively associated with CEC. Although state ownership is not high enough, SOEs can also take advantage of political connections to obtain more valuable resources than NSOEs, contributing to better firm performance (Sun et al. 2002) and lower CEC. When state ownership is above a certain level, these controlling shareholders have more power to intervene and weaken the corporate monitoring mechanism for their own self-interest and decision making, which negatively impacts the incentive of small outside investors (Huyghebaert and Wang 2012), leading to a higher CEC. 20

22 We also set out to determine whether state-ownership has a greater impact on the reduction of CEC after XBRL adoption. For this analysis, we construct a matched-pair design of SOEs and NSOEs, by controlling for size and industry factors. The second and third columns in Table 6, Panel A show that there is a significant difference in the effect of XBRL on CEC between SOEs and NSOEs, supporting H3. (The untabulated results of t-tests to check for differences in CEC between SOEs and NSOEs are: rdh= for SOEs, rdh= for NSOEs, t= ). XBRL adoption results in a significantly greater basis point reduction (121 basis points) in CEC for SOEs than for NSOEs (108 basis point reduction), suggesting that the effect of XBRL on CEC is greater for SOEs than for NSOEs. Does that mean SOEs necessarily have higher CG than NSOEs? Since CG is negatively associated with CEC, there is also a nonlinear relation between state ownership and CG. According to the U-shape, we divide SOEs into two parts by the quartile method. SOEs above the upper and below the lower quartiles usually have low CG than the ones in intervals between 25 percent and 75 percent. We choose SOEs at a low CG level to create matched pairs with NSOEs and carry out a process similar to that shown in Table 6, Panel B. The results show that SOEs with low CG experience less reduction in CEC than matched NSOEs, after XBRL adoption. Within SOEs, the ones with high CG experience more of a reduction in CEC than matched NSOEs. We conclude that SOEs with high CG experience a greater reduction in CEC than NSOEs, and that NSOEs benefit more after XBRL adoption than SOEs when the 21

23 SOE has low CG. Not all SOEs will have a higher CG level than NSOEs, but CG is a critical factor for the effect of XBRL on CEC within both SOEs and NSOEs. This indicates that improving the corporate control mechanism contributes substantially and addressed during SOEs stock reform to maximize firm value. Notably, there are mixed results regarding the relationship between state ownership and firm performance and, therefore, CEC. Ben-Nasr et al. (2013) finds a positive relationship between state ownership and CEC, while Sun et al. (2002) sees the relationship between state ownership and firm performance represented by a concave curve. Ng et al. (2009) and Yu (2013) find a convex curve between state ownership and firm performance, while our results show a convex curve between state ownership and CEC. A possible explanation for the differences may be the sample sizes and sample selection period. In 2005, China carried out split share structure reform, which means that there will be substantial differences in SOEs before and after Most of the studies mentioned above choose samples either before 2005 or with 2005 data included. Results about state ownership s relation to CEC may change along with the reform process and time span. Our sample period from 2005 to 2011 is post-reform, in order to eliminate changes resulting from reform and to maintain the credibility of our results. Future studies can explore these issues and testify regarding these potential hypotheses. [Insert table 6 here] V. ROBUSTNESS TESTS We hypothesize that XBRL adoption improves information quality, and greater 22

24 disclosure enhances stock market liquidity and reduces the estimation risk associated with investors assessments, leading to a decrease in CEC. To test the robustness of our results, we choose financial reporting proxies to substantiate the finding that XBRL increases FRQ. Second, we change the measure of CEC to see if XBRL s effect on CEC changes correspondingly. If the results remain the same, our results related to H1-H3 are robust. We use another measure of CG stepwise regression to check the robustness of our H2 results. Finally, we control for the impact of the financial crisis during our sample period. 5.1 Does XBRL Really Improve FRQ in China? As mentioned in Section II, investors can benefit more when analyzing financial information by using XBRL, since XBRL standardizes financial disclosure and eliminates costly manual processes (Yoon et al. 2011; Kim et al. 2012). As a result, financial information will become more understandable, more accurate, and available more rapidly (SEC 2009), which translates to improved FRQ. In our robustness testing, we verify whether XBRL improves FRQ in China Measurement of FRQ Following previous literature (Francis et al. 2004), we construct four measures of FRQ, proxied by accrual quality, smoothness, discretionary accruals and timeliness 4, 4 The reason we did not choose other methods of Francis (2004) is that studies of accounting conservatism in China have shown controversial results. Early studies showed little evidence of conservatism in East Asian countries (Ball et al. 2003), and similar results were shown for China (Ball et al. 2000). More recent studies going beyond 2000 also showed controversial findings. Du and Du (2010) studied the relationship between accounting standards, fair value, and accounting conservatism based on Chinese data from 1998 to They found that, although the level of conservative accounting varies to some extent during different phases of fair value s application, Chinese listed firms showed conditional accounting conservatism in 2006 to Zhao (2004) also found that Chinese firms adopted significantly higher levels of accounting conservatism since Chen and Huang (2006) tested the time-series properties of accounting conservatism in China and found the level of 23

25 to see if XBRL improves FRQ and confirm that our result is robust. Smoothness is calculated by the ratio of firm j s standard deviation of net income divided by lagged total assets, to its standard deviation of cash flows from operations divided by lagged total assets, Smoothness NI ) / ( CFO ) t ( t t. Large values of Smoothness indicate less earnings smoothness(francis et al. 2004). Following (Francis et al. 2004), our measure of timeliness is based on the explanatory power of the following equation: (9) where Equation (9) is the estimate on a firm- and year-specific basis, using rolling five-year windows. We use as the dependent variable and use the opposite value of the adjusted R 2 as the value of timeliness; as a result, larger values of timeliness imply less timeliness. accounting conservatism increased after 1998, and Chinese firms showed significant evidence of accounting conservatism after The above studies examined the existence of accounting conservatism using measurements suggested by Basu (1997). Xia and Zhu (2009) studied the same topic using measurement suggested in Ball and Shivakumar (2005) using samples from 1999 to Their results showed that the negative relationship between total accruals and operating cash flows is more evident for firms with negative operating cash flow. However, the predicted mitigated negative relationship by the model of the relationship between accruals and cash flow is mainly caused by non-operating accruals. For these reasons, persistence, predictability, and relevance are used less often than the other four proxies. 24

26 Our measure of accrual quality is based on Dechow and Dichev s (2002) model relating current accruals to lagged, current, and future cash flows from operations: (10) We estimate this equation using rolling five-year windows. These results yield five firm- and year- specific residuals, accrual quality metric, of firm j s estimated residuals (, t=t-4,,t, which form the basis for the, equal to the standard deviation ). Large values of correspond to poor accrual quality. As for control variables, we follow prior literature that has been shown to affect financial reporting quality. Dechow and Divhev (2002) find that accrual quality is inversely associated with firm size, and positively related to cash flow variability, sales variability, operating cycle, and incidence of losses. Firm size is the log of total 25

27 assets (Size) and the cash flow variability is calculated as the standard deviation of the firm s rolling five-year cash flows from operations, scaled by total assets. Sales variability is defined as the standard deviation of the firm s rolling five-year sales revenues, scaled by total assets and operating cycle is the log of the sum of the firm s days accounts receivable and days inventory. Finally, incidence of negative earnings realizations is measured as the firm s proportion of losses over the prior five years. We examine the relationship between XBRL and FRQ using four proxies accrual quality, timeliness, smoothness, and discretionary accruals and the multivariate regression is as follows: (11) Panel A, Table 7 presents descriptive statistics of key variables of FRQ. The values of accrual quality, smoothness, and absolute discretionary accruals (ADA) are significantly smaller after XBRL adoption, which means that the XBRL can improve earnings quality. Standard deviations of accrual quality, smoothness, and ADA are very small. Panel B, Table 7 shows Pearson correlations between the key variables of earnings quality and accrual Quality. Accrual quality appears to be negatively correlated with XBRL adoption. It is also negatively correlated with cash flow variability, sales variability, operating cycle, and incidence of negative earnings realizations, but positively correlated with firm size. We also perform a multivariate regression analysis to determine whether 26

28 XBRL actually increases FRQ after controlling for influential factors (Table 8). Using four variables accrual quality, smoothness, absolute discretionary accruals, and timeliness as proxies for FRQ, we find a significant and negative correlation between XBRL and CEC. (Coeff.=-0.008, t= for the accrual quality measure; Coeff.=-1.227, t=-2.17 for the smoothness measure; Coeff.=-0.007, t=-2.13 for the discretionary accruals; Coeff.=-0.006, t= for the timeliness measure). With regard to the control variables, we find that large firms have a lower value for FRQ., which denotes cash flow variability, whereas they tend to have a higher value for accrual quality. Sales variability, operating cycle, and the incidence of negative earnings realizations are also positively associated with FRQ. [Insert table 7 here] [Insert table 8 here] 5.2 Alternative Measure of CEC Following Da et al. s (2012) approach, we choose the CAPM to calculate the expected return on equity: E ( R) R [ E( R ) R ] (21) f m f where indicates systematic risk, measured by using five-year monthly return data, which is tradable capitalization-weighted prior to return measurement in month t; R f is risk-free rate, measured as one-year deposit interest rate 5. As to E(R m ), the expected market return is defined as yearly return of stock indexes Shanghai composite 5 Rf in this study is weighted by the days from one adjustment date to the next adjustment date divided by 365 days, as announced by The People's Bank of China. During our period of study from March 17, 2005 to July 7, 2011, there were a total of 20 adjustments. 27

29 A-share index (code:000002) and Shenzhen composite A-share index (code:399107), Pt E( R ) 1 m which is measured by the following equation: Pt 1 6 Panel A, Table 9 presents descriptive statistics of key variables of the CAPM model. Consistent with previous results, using the CAPM model, CEC is significantly different before and after XBRL adoption. Panel B, Table 9 documents that Pearson correlations between key variables do not show any apparent multicollinearity problems, and the Re (CAPM) appears to be negatively correlated with XBRL adoption. It is negatively correlated with firm size, market-to-book ratio, and leverage, and positively correlated with the market beta, analyst forecast dispersion, and long-term growth. In general, R squares in Table 10 and Table 12 have high values, compared with the ones in Tables 3, 5 and 6, indicating that market-based measures seem to outperform accounting-based measures for calculating CEC. The multivariate regression results (Table 10) indicate a significant and negative correlation between XBRL and CEC (Coeff.=-1.656, t=-47.95) when the CAPM model is used to measure CEC. The results of the control variables are consistent with previous empirical results, confirming that the results related to H1 are robust. We also assess how CG impacts the effect of XBRL on CEC, and how XBRL s effect on CEC differs for SOEs and NSOEs (Tables 11 and 12). Not all of the results of the control variables are found to be consistent with previous results when using rdh, rcl Mean of rdh, and rcl to measure CEC. There is a significant and 6 The reason we chose the composite A-share index rather than the component A-share index is that our sample included most of the firms listed in the Shanghai and Shenzhen stock exchange centers, and the composite A-share index can better represent our sample s price exchange. Meanwhile, in China, there are two stock exchange centers and the two composite A-share indexes are calculated separately, so we choose different composite A-share indexes for listed firms from the corresponding stock exchange center. 28

30 negative correlation between XBRL and CEC, as shown in Table 11, and firms with high CG experience a greater reduction in CEC after XBRL adoption (Coeff.= , t=-19.81) than those with low CG (Coeff.= , t=-17.88). Table 12 also shows that XBRL has a significantly negative effect on CEC, and that SOEs have more of a reduction in CEC after XBRL adoption (Coeff.= , t=-19.48) than NSOEs (Coeff.= , t=-19.92). Table 10 and Table 12 provide the robustness results indicating that our conclusions regarding H1-H3 are sound. [Insert table 9 here] [Insert table 10 here] [Insert table 11 here] [Insert table 12 here] 5.3 Alternative Measure of CG: Stepwise Regression Given that PCA is mainly dependent on statistics theory, to verify our findings for H2, we employ stepwise regression to find the CG variables with the strongest relationship to CEC. The three variables most related to CEC are Sindex (the percentage of shares held by the top 10 shareholders, except the top 1 shareholders), Bosunum (the total number of board directors and supervisory board members), and Lnsalary (the logarithm of managerial salary) (Table 13). Compared with the CG scores in Table 4, these three variables are respectively representative of three parts of the CG system, specifically, Ownership Structure (Sindex), Board Governance (Bosunum), and Managerial Behavior (Lnsalary). Further, the regression results suggest that CG has a significant impact on CEC, and that XBRL still has a 29

31 significant and negative effect on CEC (Coeff.=-0.047, t=-3.87) after controlling for CG. [Insert table 13 here] 5.4 SUR Method The SUR method is a formal test to determine if there are significant differences between the coefficients of different regressions (Bhattacharya 2001; Blankespoor et al. 2012). Using the SUR method, we test to see if SOEs experience significantly more reduction in CEC than NSOEs after XBRL implementation, as documented in Section IV. Untabulated results show the same conclusions (For SOEs, Chi-squared statistics and P-value are and 0.000, Coeff. Of XBRL = , z=-6.17; For NSOEs, Chi-squared statistics and P-value are and 0.000, Coeff. Of XBRL = , z=-4.77). The robustness test of the findings related to CG using SUR yields similarly stable results. 5.5 Controlling for the Financial Crisis Since China s mandatory XBRL program started in 2008, the control group method for eliminating the financial crisis effect does not seem appropriate for our study. At the same time, since the 2008 financial crisis increases IA and information risk (Bai et al. 2012), there can be an increase in CEC that is counteracting the XBRL effect. When firms in the financial industry are removed from the 2008 data, XBRL decrease CEC; after removing the 2008 data entirely, XBRL decrease CEC even more 30

32 (three-year pre-and post-test design without 2008 data included: Coeff.=-0.06, t=-6.53; with 2008 data included: Coeff.= , t=-8.26 ). VI. CONCLUSIONS AND IMPLICATIONS This study investigates whether XBRL adoption decreases CEC and explains how XBRL affects CEC from a new, intracompany perspective. Using the DH model, the CL model, the mean model of rdh and rcl, and the CAPM model as measures of CEC, we find that, in general, XBRL adoption reduces CEC, both in the short- and long-term period. We also find that corporations internal environments play an important role in the effect of XBRL on CEC. In China, state ownership has a U-shaped nonlinear correlation with CEC, and the effect of XBRL on CEC in SOEs significantly differs from that in NSOEs. Specifically, high levels of CG with high FRQ reduce the impact of XBRL on CEC. Although SOEs with low CG have a lower reductions in CEC than matched NSOEs after XBRL adoption, high CG SOEs cause SOEs overall to experience a greater reduction in CEC than NSOEs. Therefore, in China, CG level plays an important role in information technology efficiency and capital cost for both SOEs and NSOEs. This study makes several contributions to the extant literature. First, we provide substantial evidence of XBRL s effect on capital markets and XBRL adoption in practice. Our results are consistent with prior research (Yoon et al. 2011, Li et al. 2012), that the experiences of advocates of XBRL, such as the SEC and CSRC (China Securities Regulatory Commission), should be of interest to other emerging markets 31

33 or other countries intending to adopt XBRL. Second, while prior research focuses only on exterior market response, we provide a new perspective for testing the effect of XBRL on CEC by starting with the internal environment of companies, and establishing a linkage between IT efficiency and intracompany issues such as CG. When testing and assessing IT efficiency, prior studies should account for the influential factors of specific corporate characteristics. Third, by using the unique Chinese capital setting, where most firms are state-owned, we enrich the research stream of SOEs. In China, SOEs seem not always bad apples, and our research suggests that it depends, to some extent, on CG, the result of which has immediate benefits for foreign investors who are investing in SOEs in markets like China, and international regulation collaborators. There are also some limitations to our research. Because all publicly listed firms are required to provide both PDF files and XBRL format files starting in 2008, for lack of group that did not adopt XBRL during the sample period, we cannot use the control group method to verify our results, although we employ alternative methods to support our results robustness. At the same time, we choose a one-year pre-and post-test design to examine the short-term effect of XBRL on CEC, without further using quarterly data to test shorter-term effect. Our study may provide some new ideas for further research. Firstly, future studies can use the information technology suggested by our study, to test the underlying mechanism between FRQ and the allocation efficiency of capital, with both investment efficiency and CEC included. Studies can also adopt the same 32