The effect of adopting XBRL on credit ratings

The effect of adopting XBRL on credit ratings MSc in Accounting & Financial Management

Academic Year 2014-‐2015

Master Thesis Student Name: A.J.D. Vis

Student Number: 303151 Coach: Dr. S. Kramer, Department of Accounting & Control

Coreader: Dr. N. Dalla Via, Department of Accounting & Control

Date: 14/06/2015

2

Preface The copyright of the Master thesis rests with the author. The author is responsible for its

contents. RSM is only responsible for the educational coaching and cannot be held liable

for the content.

3

Abstract This study examines whether the use of eXtensible Business Reporting Language (XBRL)

influences credit ratings. XBRL use allows stakeholders to digitally import business

information into computer systems instead of digitalising paper-‐filed financial

statements. XBRL use, in theory, improves information efficiency: The costs of

processing information are reduced. Results of several studies analysing the benefits of

XBRL for a company and its stakeholders differed. Some reported a reduction in the

information gap when using XBRL; others reported none. Although the role of credit

rating agencies (CRAs) is to reduce the information gap between a company and its

external parties by providing credit ratings, previous research showed that CRAs are

reluctant to process huge amounts of data because of cost. Using XBRL provides CRAs

with cheaper data processing methods, resulting in more accurate credit ratings and

thus reduced split ratings, i.e., the difference in long-‐term issuer credit ratings provided

by the largest three CRAs. The Securities Exchange Commission (SEC) made XBRL use

mandatory for large accelerated filers in June 2009. Split ratings were analysed before

and after June 2009 using a regression model that included the moderator variables Size

and Leverage. Results showed XBRL use had no statistically significant influence on split

ratings, the moderator variables did not result in a significant influence of XBRL on split

ratings, and there was no statistical difference in split ratings before and after XBRL’s

introduction. This study contributes to the debate regarding mandatory XBRL use by

testing proponents’ arguments on the benefits of XBRL.

4

Table of Contents

1 INTRODUCTION ....................................................................................................................................... 5 1.1 INTRODUCTION TO THE RESEARCH QUESTION ................................................................................................... 5 1.2 PROBLEM STATEMENT AND THESIS DEVELOPMENT ......................................................................................... 5 1.3 EXPECTED CONTRIBUTION .................................................................................................................................... 7 1.4 RESEARCH METHODOLOGY .................................................................................................................................... 7 1.5 CHAPTER SUMMARY ............................................................................................................................................... 7

2 LITERATURE REVIEW ............................................................................................................................ 8 2.1 INTRODUCTION ........................................................................................................................................................ 8 2.2 INTRODUCTION TO XBRL ...................................................................................................................................... 8 2.3 INFORMATION EFFICIENCY ................................................................................................................................. 10 2.3.1 Previous research on improving information efficiency ............................................................. 11

2.4 CREDIT RATINGS ................................................................................................................................................... 12 2.5 MODERATORS ....................................................................................................................................................... 13 2.5.1 Company size .................................................................................................................................................. 14 2.5.2 Leverage ........................................................................................................................................................... 14

2.6 CHAPTER SUMMARY ............................................................................................................................................ 15 3 RESEARCH DESIGN AND DATA ......................................................................................................... 17 3.1 INTRODUCTION ..................................................................................................................................................... 17 3.2 METHODOLOGY .................................................................................................................................................... 17 3.3 MEASUREMENT OF VARIABLES .......................................................................................................................... 17 3.3.1 Moderator variables ................................................................................................................................... 18 3.3.2 Control variables .......................................................................................................................................... 18

3.4 SAMPLE SELECTION ............................................................................................................................................. 20 4 RESULTS ................................................................................................................................................... 22 4.1 DESCRIPTIVE STATISTICS .................................................................................................................................... 22 4.2 PRELIMINARY TESTS ............................................................................................................................................ 23 4.2.1 Normality ......................................................................................................................................................... 24 4.2.2 Multicollinearity ........................................................................................................................................... 25 4.2.3 Outliers .............................................................................................................................................................. 26 4.2.4 Homoscedasticity ......................................................................................................................................... 26

4.3 RESULTS OF THE STATISTICAL TESTS ............................................................................................................... 27 4.3.1 Results of the multivariate regression model .................................................................................. 27 4.3.2 Robustness check .......................................................................................................................................... 28 4.3.3 Testing H2 and H3 ....................................................................................................................................... 29

4.4 CHAPTER SUMMARY ............................................................................................................................................ 30 5 CONCLUSION ........................................................................................................................................... 31 5.1 CONCLUSION AND DISCUSSION ........................................................................................................................... 31 5.2 LIMITATIONS ......................................................................................................................................................... 33 5.3 RECOMMENDATIONS FOR FUTURE RESEARCH ................................................................................................ 34

6 REFERENCES ........................................................................................................................................... 35 APPENDIX: FIGURES .................................................................................................................................... 38

5

1 Introduction

1.1 Introduction to the research question Before the rise of the Internet, business reports were printed on paper and distributed

by mail. Historically, investors had greater difficulty obtaining publicly available

information than in modern times. Today, one can easily go to a company’s website and

download the annual report on his or her own computer, print it out, and make his or

her own analysis. Using a different way of communicating makes it easier to distribute

information to investors.

The same kind of revolution is currently happening. Companies are providing their

company reports by using a digital business language, named XBRL (eXtensible Business

Reporting Language). Pepsi’s CEO, Nooyi (2006), stated that XBRL “make(s) looking at

financial information easy in every sense: easy to access, easy to use, easy to compare

with other companies” (para. 6).

XBRL enables computers to process business reports without human interaction. It is no

longer necessary to manually input the data of published business reports (Richards,

Smith, & Saeedi, 2006). Credit rating agencies (CRAs), which compose business reports

in order to determine credit ratings, can benefit from XBRL. In developed countries,

CRAs rely more on publicly available information since there are regulations that

prohibit the use of insider information (D’Amato, 2014). The use of XBRL will save CRAs

considerable time-‐consuming work1 and make it cheaper for them to prepare credit

ratings. This research will investigate the relationship between a company’s usage of

XBRL and assigned credit ratings.

1.2 Problem statement and thesis development This research is based on the notion that XBRL leads to a more efficient market by

reducing the cost associated with processing financial statements (Cong, Hao, & Zou,

2014). The usage of XBRL does not lead to a greater quantity of information; instead, it

leads to information of higher quality by adding tags to information. This addition

makes it cheaper to perform analyses/obtain financial information, which leads to the

1 Non-‐XBRL data needs to be manually re-‐entered before it can be viewed in computer

systems.

6

increased interest of analysts and investors (Chiang & Venkatesh, 1988). All investors

should benefit from this enriched information, especially those investors who utilize

ratings from credit rating agencies (Hodge, Kennedy, & Maines, 2004).

Credit rating agencies have several methods to analyse financial statements. The usage

of XBRL will enable them to better categorize and process the same information, for less

cost, which will allow credit agencies to perform more in-‐depth analyses on companies.

A more thorough analysis of a company might result in a different credit rating since

disclosed information can more efficiently be analysed. Different CRAs can provide

different ratings for companies; this difference is called a split rating. Using XBRL will

increase the quality of these ratings and result in reduced split ratings. This concept will

be explained more in detail in the literature review.

The research question is in what way credit ratings will be affected by using XBRL. Two

moderators of this effect (firm size and leverage) will be researched as well. Larger

firms are more difficult to analyse, and the change in credit rating when using XBRL will

be stronger for large firms (Weber, 2003)2.

Furthermore, highly leveraged firms are more likely to voluntarily disclose more

information in order to reduce the costs of debt (Dumontier & Raffournier, 1998). Higher

leveraged firms are, therefore, assumed to have a smaller change in credit ratings when

adopting XBRL3. In order to research this theory, the following hypotheses have been

formulated with respect to the U.S. capital market:

Hypothesis 1 (H1): The adoption of XBRL has a reducing effect on split ratings.

Hypothesis 2 (H2): The effect of XBRL adoption on split ratings is stronger in larger

firms.

Hypothesis 3 (H3): The effect of XBRL adoption on split ratings is weaker for firms

that are more leveraged.

In this research, a split rating is the difference in credit ratings of the three major CRAs

(S&P, Fitch, and Moody’s).

2 This will be discussed in detail in Section 2.5.1, Company size. 3 This will be discussed in detail in Section 2.5.2, Leverage.

7

1.3 Expected contribution One of the claimed benefits of XBRL is more easily obtained and less costly available

financial information (Pinsker & Li, 2008). Furthermore, XBRL makes it easier to

compare different financial reporting methods (Weber, 2003). There is still considerable

research being conducted on the effects of XBRL. This research study will investigate

whether there is a correlation between the adoption of XBRL and credit ratings.

Several governments demand the use of XBRL, and those who support it argue that its

use should be mandatory (O'Kelly, 2007). The U.S. Securities and Exchange Commission

(SEC) made XBRL use compulsory for U.S. listed companies in 2009 (SEC, 2008).

Therefore, researching the effects of XBRL is relevant to this debate. Enough evidence

supporting H1 will be likely to encourage analysts and investors to argue for the

mandatory use of XBRL, thus improving the quality of credit ratings and decreasing

investors’ perceived investment risk.

This research will also increase the understanding of the practical use of XBRL from a

corporate point of view. Companies’ management might consider why they should

implement XBRL technology in their current data systems. This research will provide

them insight into the perception of external stakeholders of a company using XBRL over

a company that does not use XBRL.

1.4 Research methodology A multiple linear regression will be conducted by using a sample of U.S. listed companies

and their differences in ratings as they are provided by the largest three CRAs. The SEC

made XBRL use compulsory in 2009. This use was required for all publicly listed

companies with a minimum public float of $5 billion (SEC, 2008). The split ratings of

these companies will be compared before and after the mandatory use of XBRL. The data

will be collected by using CRSP, Bloomberg, and the Compustat database.

1.5 Chapter summary This chapter showed the background of this study regarding the effect of adopting XBRL

on credit ratings. Both the problem statement and thesis development were explained.

Furthermore, the scientific relevance of this research and the methodology used were

described.

8

2 Literature review

2.1 Introduction Since this study is designed to provide more insight into the relationship between the

usage of XBRL by companies and their assigned credit ratings, this chapter will discuss

the relevant literature in order to provide the reader with a clear understanding of the

concept of XBRL. Three hypotheses will be developed based on the literature review.

2.2 Introduction to XBRL XBRL, eXtensible Business Reporting Language, is an open standard for digital business

reporting. It is under license of the non-‐profit organization XBRL International. This

digital language adds tags to financial information. These tags enable computers to read

the accounting numbers and process them into reports. The benefit of using XBRL is that

every end user can compile his or her own reports based on his or her own needs. XBRL

does not add information to the reports; it only describes the presented information by

using tags and, therefore, adds value to the information presented (Efendi, Park & Smith,

2014; Hodge, Kennedy, & Maines, 2004).

This research is based on the theory that XBRL use allows users of financial information

to use that information more cost-‐efficiently. This theory has been called information

efficiency and will be explained in Section 2.3, Information efficiency (Elliott & Jacobson,

1994). In theory, the use of XBRL will lead to a better analysis of companies since

information is more easily available. This might reduce the information gap between

companies and their external stakeholders by improving information efficiency for

external stakeholders (Verrecchia, 1980), which is one of the objectives of XBRL

according of the SEC: “[XBRL]…has the potential to increase the speed, accuracy and

usability of financial disclosure and eventually reduce costs for investors” (SEC, 2008,

para. 1).

Figure 1 is an example of the process of converting a line of an annual report into XBRL.

The XML information in the image is called the XBRL information. XBRL can be seen as a

specific type of XML computer language.

9

Figure 1. How XBRL Works (Kapoor, 2012)

Figure 1 shows the annual report states that Share Capital equals 3,273.37. Share Capital

is part of the category Shareholders’ Funds. This information is coded into XBRL, and a

computer can easily read the XBRL-‐code. Upon request, a PDF file can be generated with

relevant financial information. This option is emphasized by the third stage of the image

that shows a line of a computer-‐generated PDF report with the numerical value of paid-‐

up share capital.

There are different benefits of using XBRL for both companies and their stakeholders.

Firms can benefit from using XBRL since both transparency and informational quality

improves after introducing XBRL. Companies’ internal costs for bookkeeping and

processing financial reports is reduced as well (Pinsker & Li, 2008). For external users of

financial statements, XBRL use will significantly reduce the errors from manually re-‐

coding information from business reports into analysts’ databases (Vasarhelyi, Yang, &

Liu, 2003). Furthermore, the SEC specifically mentioned that the adoption of XBRL

would result in cost-‐savings for external users (including the SEC itself) of a firm’s

financial statements4 (SEC, 2008). This is one of the main advantages of using XBRL, and

it will be the topic of the next sections. 4 This article http://raasconsulting.blogspot.nl/2011/01/why-‐did-‐sec-‐mandate-‐

xbrl.html comments on the theory that cost savings for the SEC itself was one of the

main drivers for demanding the use of XBRL by filing companies.

10

2.3 Information efficiency This section, and the following sub-‐section, will discuss the efficiency benefits of using

XBRL for a company’s stakeholders and introduce the concept of information efficiency.

Afterward, previous studies on improving information efficiency in relation to XBRL will

be discussed.

Historically, business reports were published on paper and, more recently, in digital

files, like PDF reports. The company decides the layout and provides the same report to

every stakeholder. Each stakeholder requires different kinds of information. For

example, an analyst has a different perspective than the local tax authority. Thus,

companies provide information in addition to their regular business reports. This kind

of information is usually converted into a format that can be used by that particular user

(SEC, 2009). The use of XBRL will make this process more convenient since companies

can generate these different reports more cheaply and quickly; this benefit has been

called information efficiency (Pinsker & Li, 2008). Information efficiency occurs for both

investors and analysts.

Secondly, stakeholders who generate their own reports can benefit from XBRL, as well,

by improving their methods of analysing information. According to Hodge, Kennedy, and

Maines (2004), investors benefit from this since they can more easily obtain and

integrate information. Analysing information is streamlined by using an XBRL-‐enabled

search program. This is a form of information efficiency. Their research was based on

investors without professional knowledge, and they found that those who use XBRL data

benefit from it. Notably, this effect is stronger for investors with lower professional

knowledge of analysing investments (Efendi, Park, & Smith, 2014).

Furthermore, definitions used for or methods of calculating financial statements are not

always similar (Richards, Smith, & Saeedi, 2006), which makes converting information

time consuming since numbers have to be analysed thoroughly before they can be

imported by analysts (Hodge, Kennedy, & Maines, 2004). Firms pay analysts who operate

on the sell side, and these analysts are more likely to perform more extensive analyses

(Groysberg, Healy, & Chapman, 2008), which differs from analysts who operate on the

buy side. Sell-‐side analysts decide the minimum information needed to perform their

11

analyses and convert only that kind of data. Analysts who operate on the buy side have

to trade off the costs and benefits of converting additional information in order to input

it into their computer systems. Since XBRL makes it cheaper to process information, it is

more likely that (both types of) analysts will import more data into their computer

systems and perform additional analyses. Thus, XBRL results in a higher level of

information efficiency (Efendi, Dong Park, & Subramaniam, 2010).

Additionally, different users might use different definitions and mistakes can be easily

made. XBRL use implies that a tag identifies every item in the financial statements. This

tag describes the meaning of the information, which makes it possible to identify items,

regardless of international interpretations or differences in definitions (Richards, Smith,

& Saeedi, 2006). It is even possible to combine both financial and non-‐financial

information (like disclosures) in an automatic analysis (Weber, 2003).

2.3.1 Previous research on improving information efficiency An information gap exists between companies and their stakeholders. Information

efficiency is the way new information is distributed to a firm’s stakeholders. A low

efficiency rate indicates a significant information gap between a company and its

stakeholders (Elliott & Jacobson, 1994).

Several researchers have studied the theory that XBRL use will improve information

efficiency. The Korean, Japanese, and American authorities forced certain groups of

companies listed in their national stock markets to use XBRL at once (Bai, Sakaue, &

Takeda, 2012). The reported results were not the same and led to different conclusions.

Empirical research in the Chinese capital market suggested the usage of XBRL leads to

reduced information efficiency (Chen & Li, 2013). Different conclusions were found by

Blankespoor, Miller, and White (2014). They studied U.S. stock market data for

companies that had switched to XBRL for reporting purposes. Their research found

evidence that the information playing field did not improve for the first year after XBRL

use was mandatory. Geiger, North, and Selby’s (2014) study supported this perspective

on the effect of using XBRL in order to improve information efficiency. They performed

research on companies in the United States that voluntarily used XBRL. They argued

that, based on their research, XBRL reduces the information gap between a company

and its stakeholders for large companies. A study of companies listed on the Korean

12

stock market showed that XBRL use reduces the information gap. This effect is stronger

for large companies than for medium or small companies (Yoon, Zo, & Ciganek, 2011).

This result was confirmed by later research (Kim, Lim, & No, 2012).

2.4 Credit ratings Credit rating agencies (CRAs), like Moody’s, Fitch, and S&P, provide third-‐party opinions

about the solvency of debt instruments to the public. Historically, investors paid for

these credit ratings, but this tradition has shifted. Companies who issue debt generally

need to pay for this kind of service, and these fees are a major part of a CRA’s revenues.

Companies need these credit ratings in order to attract investors and are forced to

cooperate with the issuer paid CRAs (Forster, 2008; Funcke, 2015).

CRAs provide ratings based on both publicly available information and information that

is only available to market insiders. D’Amato (2014) argued that CRAs mostly use

publicly available information in more developed countries and more insider

information in less developed countries. This theory is supported by the argument that

developed countries have stricter regulations that prohibit the spread of insider

information. The exact method of calculating credit ratings has not been disclosed by

CRAs, but this has changed since the Dodd–Frank Act (2010) required CRAs to provided

more information on their rating processes. This change was a result of the ongoing

debate as to the trustworthiness and impact of CRAs. For example, the day that Lehman

Brothers went bankrupt, the company was still rated as investment grade. However, the

exact details of the rating processes are still not made public (Funcke, 2015).

CRAs can be seen as information processing agencies that reduce the information gap

between investors and companies and thus improve information efficiency (Boot &

Milbourn, 2002). Their aim is to reduce the information gap between companies and

their (potential) investors by making information available in the form of trading advice

and credit ratings.

13

However, CRAs are reluctant to process huge amounts of data since this practice is

costly (Millon & Thakor, 1985)5. Using XBRL will enable CRAs to better categorize and

process the same information but at less cost, which will allow CRAs to perform more in-‐

depth analyses on companies. A more thorough analysis of a company might result in a

revised credit rating. Split ratings are the difference between the ratings as they are

provided by different CRAs. This research will investigate the relationship between

these two variables: 1) The adoption of XBRL by a company and 2) the difference in

credit ratings provided by CRAs on the same company6. The independent variable is the

usage of XBRL, and this influences the dependent variable, the split ratings, which leads

to the development of the following hypothesis:

Hypothesis 1 (H1): The adoption of XBRL has a reducing effect on split ratings.

The adoption of XBRL will reduce split ratings because, as Blankespoor (2012)

demonstrated in her dissertation, that reduction in the cost of processing information

leads to increased levels of voluntarily disclosure by firms. I anticipate that this

increased level of voluntarily disclosure will induce more accurate estimations of credit

ratings. As discussed in the literature review, the usage of XBRL will improve

information efficiency. More efficient and precise ratings provided by different CRAs

(i.e., reduced split ratings) will be the result of this process.

2.5 Moderators The previous sections have shown that XBRL use will improve the information efficiency

for information processors like CRAs. As previously explained, information-‐processing

companies have to determine what information is relevant for them to convert into their

analysing tools. They always need to find a trade-‐off between the costs and benefits of

processing additional information. Therefore, improved information efficiency will

result in more processed data and analyses performed, and in turn, more analyses

5 Although processing data has sped up since 1985, the total amount of data has

expanded as well, which makes this research still relevant (Rubini, 2000). 6 Moderators will be discussed in Section 2.5, Moderators, and control variables in

Section 3.3.2, Control variables.

14

performed can result in reduced split ratings. This research will measure to what extent

such a relationship exists.

However, there might be factors that will influence this relationship; these moderators

will be researched as well. Based on the literature, two moderators were selected:

Company size and leverage. These moderators will be explained in the following sub-‐

sections.

2.5.1 Company size The change in split ratings should depend on the company size. The absolute amount of

information not used for analysis purposes for larger firms is greater than that of

smaller firms. This amount of information not used is a result of CRAs who

predetermine (based on the trade-‐off between their costs and benefits) what

information seems to be relevant for them to convert for analyses. Thus, the possibility

that the credit rating changes depends on the number of additional analyses. Since more

additional analyses can be performed for larger companies, it is more likely that the

change in split ratings will be stronger for large firms.

Furthermore, larger companies operate in more business reporting jurisdictions, which

results in different methods of reporting (Premuroso & Bhattacharya, 2008). The

improvement of information efficiency for larger companies due to XBRL use will be

greater since the usage of XBRL will increase efficiency when comparing different

business reporting methods (Weber, 2003). These two factors will result in a potentially

significant reduction in split ratings for larger companies than for smaller firms when

using XBRL, which leads to the second hypothesis:

Hypothesis 2 (H2): The effect of XBRL adoption on split ratings is stronger in larger firms.

Implementing the variable Size in the regression model will test this hypothesis. Firm

size will be measured by using total assets.

2.5.2 Leverage A second important variable is a firm’s level of leverage. This variable is based on the

efficient market theory. The efficient market theory states that information is reflected

in stock prices (Fama, 1970). Both voluntarily disclosed information and hidden

15

information is returned in those prices. The level of reflection can be different and

depends on the degree of market efficiency.

This theory also applies to the market’s pricing of corporate bonds. Leveraged firms

need to disclose information to debt holders. Disclosing information directly influences

prices. Jensen and Mechling (1976) stated that firms that disclose more information

reduce the monitoring costs for creditors, which will be reflected in costs charged on

loans. Firms that disclose more information have, therefore, less costs of debt (Elliott &

Jacobson, 1994). These less costs of debt is one of the main benefits for firms to use the

services of CRAs (Sufi, 2009).

Less costs of debt provide firms the possibility to attract more debt. Higher leveraged

firms are expected to have voluntarily disclosed more information in order to reduce

costs of debt (Dumontier & Raffournier, 1998; Wallace & Naser, 1995). CRAs are

expected to obtain fewer new insights into these highly leveraged companies when they

start using XBRL. Leverage is, therefore, negatively correlated to a reduction in split

ratings, which leads to the third hypothesis:

Hypothesis 3 (H3): The effect of XBRL adoption on split ratings is weaker for firms that are

more leveraged.

Implementing the variable Leverage (debt as a percentage of equity) into the regression

model will test this hypothesis7.

2.6 Chapter summary This chapter provided an overview of the current literature in the XBRL field with

respect to information efficiency. The theoretical purpose of XBRL is clear: Improving

information efficiency. In practice, several studies were conducted to analyse the

benefits of XBRL on the information gap between a company and its stakeholders.

Results differed; some studies reported a reduction in the information gap when using

XBRL, while others report none.

7 This will be explained in Section 3.3.1, Moderator variables.

16

The role of credit rating agencies (CRAs) is to reduce the information gap between a

company and its external parties by providing credit ratings. Previous research showed

that CRAs are reluctant to process huge amounts of data, as this is costly. Using XBRL

will provide CRAs cheaper methods to process data, which will result in more accurate

credit ratings and thus reduced split ratings. Split ratings are the difference in ratings

provided by the largest three CRAs. This idea was formulated into the first hypothesis

(H1): The adoption of XBRL has a reducing effect on split ratings.

The expected reduction in split ratings will be larger for larger firms since the use of

XBRL will make it less costly to perform analyses. Larger firms have more potential data

to analyse and operate in more countries, which results in different methods of

reporting. Since more additional analyses can be performed for larger companies, it is

more likely that the reduction in split ratings will be stronger for larger firms. This idea

was formulated into the second hypothesis (H2): The effect of XBRL adoption on split

ratings is stronger in larger firms.

Research showed that firms that are more leveraged tend to voluntarily disclose more

information in order to reduce costs of debt. Voluntarily disclosing more information

will reduce the potential benefits of using XBRL on calculating credit ratings and the

reducing effect on split ratings will, therefore, be less for more leveraged firms. The idea

was formulated in the third hypothesis (H3): The effect of XBRL adoption on split ratings

is weaker for companies that are more leveraged.

Several statistical tests were performed to test these three formulated hypotheses and

will be explained in the next chapter.

17

3 Research design and data

3.1 Introduction This chapter will explain the statistical tests used to gain insight into the relationship

between XBRL use and the difference in assigned credit ratings, as well as how the data

was collected.

3.2 Methodology This research was performed by analysing a dataset. The American stock market was

selected because of the mandatory use of XBRL. The SEC made XBRL use compulsory in

2009, requiring its use for all publicly listed companies with a minimum public float of

$5 billion (SEC, 2008). The difference in split ratings for these companies were

compared with companies who did not have to file by using XBRL.

The appropriate statistical test for testing the hypotheses (H1, H2, and H3) is a multiple

regression analysis. This analysis made it possible to measure the difference in split

ratings for two time periods (before and after the mandatory use of XBRL). First, the

data and data sources will be discussed. Afterward, the regression model will be

presented and will be followed by an overview of the selected sample.

3.3 Measurement of variables Several variables were used in this research. The dependent variable, Difference, was

the difference between the credit rating provide by the largest three CRAs. These three

CRAs (S&P, Moody’s, and Fitch) provide similar long-‐term company ratings that can be

converted into numbers. Only companies that were rated by at least two of the three

CRAs were used. For companies with three ratings provided, the largest difference in the

split rating was used. The assigned credit ratings conversion table and the

corresponding points are shown in Table 1 on the next page.

18

Table 1. Credit rating conversion

The explanatory (independent) variable was XBRL and refers to the mandatory use of

XBRL. XBRL was a categorical variable with the value of 0 or 1. The value for XBRL was 1

when the companies were required to file reports using XBRL and 0 when they did not

have to file by using XBRL. Two moderator variables, Size and Leverage, were measured

in the model as well.

3.3.1 Moderator variables Size Firm size was expected to be positively correlated to the increase of information

efficiency. This expectation is based on the literature review, Section 2.5.1, Company

size. The firm size was measured as the total assets of a company in millions of euros.

This measure (Size) is based on previous research (Yoon, Zo, & Ciganek, 2011).

Leverage

The leverage of a firm was expected to be negatively correlated to the increase of

information efficiency, which was explained in Section 2.5.2, Leverage. The degree of

leverage was measured as the book value of total debt as a percentage of total equity.

3.3.2 Control variables As explained in the literature review, previous studies into the effects of XBRL on

information efficiency showed that several aspects are highly important (Yoon, Zo &

Ciganek, 2011; Bini, Giunta & Dainelli). These aspects have resulted in two control

variables: Turnover and Profitability. The variances in the performed tests will be

explained by using these control variables.

SP Mooy Fitch Points SP Moody Fitch Points AAA Aaa AAA 20 BB Ba2 BB 9 AA+ Aa1 AA+ 19 BB-‐ Ba3 BB-‐ 8 AA Aa2 AA 18 B+ B1 B+ 7 AA-‐ Aa3 AA-‐ 17 B B2 B 6 A+ A1 A+ 16 B-‐ B3 B-‐ 5 A A2 A 15 CCC+ Caa1 CCC+ 4 A-‐ A3 A-‐ 14 CCC Caa2 CCC 3

BBB+ Baa1 BBB+ 13 CC-‐ Caa3 CC-‐ 2 BBB Baa2 BBB 12 C CaC C 1 BBB-‐ Baa3 BBB-‐ 11 C C C 0 BB+ Ba1 BB+ 10

19

Turnover A high turnover rate is an indicator of information efficiency, according to Copeland and

Galai (1983). The turnover rate was calculated by dividing the average daily trading

volume by the total number of outstanding shares. The average daily trading volume

was calculated by dividing the total trade volume for a given fiscal quarter by 90 days;

the total number of outstanding shares were taken from the end of the corresponding

fiscal quarter.

Profitability

Research has shown that the more profitable a firm, the higher the number of

voluntarily disclosures (Singhvi & Desai, 1971), which makes Profitability an important

control variable for this research. Profitability is negatively correlated to a reduction in

credit rating and is measured as the ROA ratio (Net income/total assets) since this

relates profit to the size of a company.

Profitable

The variable, Profitable, is a binary representation of Profitability. This variable is 0 for

companies that took a loss and 1 for companies that made a profit. This variable was

added since the profitability of the companies in the collected sample varies greatly, so it

might add explanatory value to the regression model.

The second and third hypotheses (H2 and H3) addressed whether Size and Leverage are

moderator variables by creating new variables, XBRL*SIZE and XBRL*LEVERAGE, which

were calculated as the product of XBRL and Size and Leverage, respectively.

This together will result in the following regression model:

𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒! =

𝛽! + 𝛽!𝑋𝐵𝑅𝐿! + 𝛽!𝑆𝑖𝑧𝑒! + 𝛽!𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒! + 𝛽!𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟! + 𝛽!𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦! +

𝛽!𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑙𝑒! + 𝛽!𝑋𝐵𝑅𝐿 ∗ 𝑆𝑖𝑧𝑒!+ 𝛽!𝑋𝐵𝑅𝐿 ∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 + 𝜀!

Where,

i = firm

t = period: pre-‐XBRL, post-‐XBRL period 1 or post-‐XBRL period 2

20

Three periods were used in this research. The first time period was the pre-‐XBRL

period. This period was compared to two post-‐XBRL periods.

3.4 Sample selection Since the use of XBRL was mandatory for companies with a public float of over $5

billion, companies with a public float of over $5 billion by the beginning of 2009 were

selected, resulting in a data set of approximately 500 companies (SEC, 2008). This

method is based on previous research (Yoon, Zo, & Ciganek, 2011).

Credit ratings

The publicly listed companies had to file their reports within 40 to 45 days after the end

of the corresponding fiscal quarter (SEC, 2015). The use of XBRL was mandatory in the

US from the first fiscal quarter, ending after the 15th of June 2009 (SEC, 2008), which is

by the end June, July, or August. Reports had to be filed within these 40 to 45 days, but

they might have been released earlier. Credit ratings before the 15th of June 2009 were

certainly based on non-‐XBRL filings and, as a result, credit ratings of the 15th of June

2009 were determined to be those of the pre-‐XBRL period.

Credit ratings published after the 15th of June to July 2009 could be based on pre-‐XBRL

(fiscal Q1 2009) or post-‐XBRL (fiscal Q2 2009) filings. Therefore, it was necessary to

exclude the months June and July from the time period to ensure all data was in the

post-‐XBRL period. The first XBRL filings were filed by August 15th for those fiscal

quarters ending in June and by October 15th for fiscal quarters ending in August. It is

good practice to consider credit rating changes within one month as being linked to the

same event. This consideration resulted in a post-‐XBRL period 1 sample selection of

credit ratings for the 15th of November 2009. The post-‐XBRL period 2 sample selection

was one fiscal quarter later, and thus by the 15th of February 2010.

Company fundamentals

Companies can use different fiscal book years. The pre-‐XBRL data for the variables Size,

Leverage, Turnover, and Profitability were retrieved for the last fiscal quarter ending

before the 15th of June 2009. The post-‐XBRL period 1 data for the variables Size,

Leverage, Turnover, and Profitability was retrieved for the first fiscal quarter ending

after the 15th of June 2009. The post-‐XBRL period 2 data for the variables Size, Leverage,

21

Turnover, and Profitability was retrieved for the second fiscal quarter ending after the

15th of June 2009.

A total of 433 companies were identified as required to file using XBRL in 2009. A total

of 104 companies that did not have at least two credit ratings per each credit rating

selection moment (August 15th, November 15th, February 15th) were excluded. The same

applied to 75 companies that had missing information for the variables Size, Leverage,

Turnover, or Profitability for the pre-‐XBRL or post-‐XBRL period. A total of 36

companies had missing information for both split ratings and the variables Size,

Leverage, Turnover, or Profitability.

From the remaining companies, 13 firms participated in the SEC Voluntary Filing

Program (SEC, 2011) and filed at least one quarterly report using XBRL in a 12 month

period before June 2009. These 13 firms were excluded from the dataset, which resulted

in a sample of 277 companies.

22

4 Results This chapter will describe the statistical tests performed on the selected sample. First,

descriptive statistics will be discussed. This discussion will be followed by several

preliminary tests in order to prepare for a multivariate regression.

4.1 Descriptive statistics The collected data was analysed using Stata. The data was validated and no missing

values were present in the dataset. The sample consisted of 277 companies with pre-‐

XBRL and two moments of post-‐XBRL observations. These three time periods will be

referred to as the pre-‐XBRL period, the post-‐XBRL period 1, and the post-‐XBRL period 2

groups.

The descriptive statistics for all three groups are shown in Tables 2, 3 and 4. The most

significant difference between the minimum and maximum values for Size were

inspected and were determined to be logical8. The data was corrected for unusual

values, a total of four companies with a negative leverage as a result of a reported

negative equity9. This correction reduced the sample size to 273. The maximum value

for the variable Leverage differed for the pre-‐XBRL and post-‐XBRL periods. Inspection of

the data showed that this was caused by just a few companies and was corrected for, as

seen in Section 4.2.3, Outliers.

Table 2. Pre-‐XBRL Period Variable Mean Std. Dev. Min Max

Difference 1.16 1.17 0 7 Size 90,057 282,325 2,525 2,789,352 Leverage 4.01 6.83 0.14 61.45 Turnover 0.01 0.01 0.00 0.09 Profitability 0.01 0.02 -‐0.19 0.13 Profitable 0.84 0.37 0 1

Table 3. Post-‐XBRL Period 1 Variable Mean Std. Dev. Min Max Difference 1.14 1.16 0 7 Size 89,172 266,507 2,613 2,429,488 Leverage 3.50 5.30 0.14 49.45

8 Size was measured in millions of euros. Firms in the dataset with a size greater than one trillion euros were banks. 9 For example, total equity of Ford Motor Company was negative by the end of 2009.

23

Turnover 0.01 0.01 0.00 0.09 Profitability 0.01 0.01 -‐0.03 0.07 Profitable 0.88 0.32 0 1

Table 4. Post-‐XBRL Period 2 Variable Mean Std. Dev. Min Max

Difference 1.11 1.18 0 7 Size 90,425 267,013 2,666 2,427,932 Leverage 3.33 4.93 0.14 46.98 Turnover 0.01 0.01 0.00 0.06 Profitability 0.01 0.02 -‐0.18 0.06 Profitable 0.90 0.30 0 1

The paired t-‐test results are shown in Tables 5 and 6. These results show that there was

no statistical difference for the variable differences for both periods in relation to the

pre-‐XBRL period. The same applied to Size. The p-‐value of the paired t-‐test for the

variables Leverage, Turnover, and Profitability was less than 0.05 therefore, the

difference was statistically significant.

Table 5. Paired t-‐test Post-‐XBRL Period 1

Table 6. Paired t-‐test Post-‐XBRL Period 2 Variable T-‐value p-‐value

Variable T-‐value p-‐value

Difference -‐1.51 0.13

Difference -‐1.39 0.17 Size -‐0.63 0.53

Size 0.24 0.81

Leverage -‐3.59 0.00

Leverage -‐3.79 0.00 Turnover -‐8.05 0.00

Turnover -‐11.54 0.00

Profitability 2.12 0.03

Profitability 2.36 0.02 Profitable 1.91 0.06

Profitable 2.75 0.01

4.2 Preliminary tests The first hypothesis assumes that there is a relationship between the use of XBRL and

credit ratings: The adoption of XBRL has a reducing effect on split ratings. The paired t-‐

test showed that there was no statistically significant difference between the means of

the difference in split ratings of these groups. Thus H1 is rejected and the null

hypothesis (H0), that there is no statistically significant difference for the variable

Difference in the pre-‐XBRL and post-‐XBRL periods, is accepted.

However, this research continued by performing a regression analysis. Before this test

could be conducted, the dataset was checked on normality, significant outliers,

24

multicollinearity and homoscedasticity by performing several preliminary tests. The

preliminary tests ensured that the various conditions of each statistical test held.

4.2.1 Normality A result of empirical data is that the dataset is usually not normally distributed. The

dependent variable, Difference, was visually and numerically checked for all periods on

normality. The normality of a variable is theoretically bell-‐shaped with most values in

the middle. Less frequent scores are reported on the sides. The variable, Difference, did

not seem to be normally distributed, which was confirmed when examining the

frequency histograms shown in Figures 2 through 4 below. This distribution was a

result of the coding process; Difference was described as the absolute value of the

largest difference among the credit ratings, creating the variable’s absolute results in

this positively skewed distribution.

Figures 2-‐4. Frequency of Difference of respectively pre-‐XBRL, post-‐XBRL period 1 and post-‐XBRL period 2

Normality can be checked numerically as well by assessing the skewness and kurtosis

values of variables, which was accomplished by using the SKTEST command in Stata.

This command tests the dataset for normality by testing against the null hypothesis that

there is normality. The p-‐values for the skewness and kurtosis values of Difference are

seen in Table 7. The p-‐value was below 0.05 for all groups, which rejects the null

hypothesis that there is normality.

Table 7. Skewness and Kurtosis test P-‐values Skewness Kurtosis Joint Pre-‐XBRL 0.00 0.00 0.00 Post-‐XBRL period 1 0.00 0.00 0.00 Post-‐XBRL period 2 0.00 0.00 0.00

Since the dependent variable was not normally distributed, several transformations

were used to normalize it: Square root, quartile, inverse, and logarithmic (Bowerman,

O'Connell, & Murphree, 2009). These transformations were applied, and the logarithm

020

4060

8010

0Fr

eque

ncy

0 2 4 6 8DIFFERENCE PRE

020

4060

8010

0Fr

eque

ncy

0 2 4 6 8DIFFERENCE POST

020

4060

8010

0Fr

eque

ncy

0 2 4 6 8DIFFERENCE POST2

25

transformation resulted in the most normal distribution. Thus, the variable Difference

became the logarithm transformation of Difference.

4.2.2 Multicollinearity A regression analysis was performed to determine the separate influences of the

independent variables on the difference in split ratings. The independent variables

should not be strongly correlated to each other. Multicollinearity occurs when two or

more independent variables correlate with each other. The data was checked on

multicollinearity by showing the Pearson correlation coefficients and the Variance

Inflator Factor (VIF) and tolerance (1/VIF) (O'Brien, 2007).

The Pearson correlation coefficients are shown in Tables 8 to 10. In all three groups

(pre-‐XBRL, post-‐XBRL Period 1, and post-‐XBRL period 2), no variables strongly

correlated to each other. The strongest correlations were, in all three groups, between

Leverage and Size. However, this correlation was still considered moderate. The

correlation between Profitability and Profitable is obvious since the variable Profitable

is a binary variable and is based on the variable Profitability.

Table 8. Pre-‐XBRL Group correlation coefficients Variable Difference Size Leverage Turnover Profitability

Difference Size 0.12 Leverage 0.14* 0.61** Turnover 0.11 0.18* 0.11 Profitability -‐0.08 -‐0.08 -‐0.08 -‐0.21** Profitable -‐0.06 -‐0.01 -‐0.07 -‐0.24** 0.54**

Table 9. Post-‐XBRL Period 1 correlation coefficients Variable Difference Size Leverage Turnover Profitability

Difference Size 0.11 Leverage 0.16* 0.65** Turnover 0.15* 0.14* 0.07 Profitability -‐0.15* -‐0.18* -‐0.25** -‐0.33** Profitable -‐0.06 0.05 0.05 -‐0.31** 0.55**

Table 10. Post-‐XBRL Period 2 correlation coefficients Variable Difference Size Leverage Turnover Profitability

Difference Size 0.11

26

Leverage 0.15* 0.67** Turnover 0.12 0.01 -‐0.04 Profitability -‐0.17* -‐0.15* -‐0.23** -‐0.22** Profitable -‐0.14* -‐0.09 -‐0.07 -‐0.32** 0.52** *p < 0.01 **p < 0.01

The VIF and tolerance are shown in Tables 11 and 12 on page 28. No variable had a VIF

value greater than 10 and the tolerance values (1/VIF) were below 1 as well (O'Brien,

2007). These results imply that these variables can be seen as linear combinations of

other independent variables. Therefore, there was no multicollinearity.

4.2.3 Outliers Another important preliminary test was to check if there were significant outliers. As

described in the sample selection and descriptive statistics, the sample was already

corrected for erroneous data entry. However, some highly leveraged data points, which

would influence the results, might still exist. Since they would be correct data points,

they could not simply be excluded from the dataset. Robust regression corrects for

highly leveraged data points (Rousseeuw & Leroy, 1987). This correction was

accomplished by performing the regression twice, a regular regression and a robust

regression.

4.2.4 Homoscedasticity Another preliminary test investigated whether the variables were homoscedastic.

Variables are homoscedastic if the residuals have similar variances. Homoscedasticity is

the opposite of heteroscedasticity and can be tested mathematically. A mathematical

method of testing was performed by using the Breusch-‐Pagan test. The Breusch-‐Pagan

test investigates the dependency of the residuals’ variances on the independent

variables (Breusch & Pagan, 1979). The tests were performed against H0 that there is

constant variance. The test for the post-‐XBRL period 1 group resulted in a χ2 score of

0.01 with a corresponding p-‐value of 0.92. Secondly, the result for the post-‐XBRL period

2 group was a χ2 score of 0.01 with a corresponding p-‐value of 0.92. Therefore, there

was no evidence for significant heteroscedasticity for these groups.

27

4.3 Results of the statistical tests After the preliminary tests were performed, the three different hypotheses were tested.

The first hypothesis was rejected by performing a paired t-‐test. All three hypotheses

were then tested against the multivariate regression model.

4.3.1 Results of the multivariate regression model A multivariate regression is described as “a technique that allows additional factors to

enter the analysis separately so that the effect of each can be estimated. It is valuable for

quantifying the impact of various simultaneous influences upon a single dependent

variable” (Sykes, 2000, p. 8). Before such a regression can be performed, the preliminary

tests should be used to verify that the data meets certain assumptions. These

preliminary tests were performed as stated in the previous sections. The regression

formula was used to test for the influence of the independent (XBRL), moderator (Size

and Leverage), and control (Turnover and Profitability) variables on the dependent

variable (Difference).




i = firm

t = period: pre-‐XBRL, post-‐XBRL period 1, or post-‐XBRL period 2

Regarding period 1, the regression model statistically significantly predicted

Difference (F = 3.66, p < 0.005). Notably, the overall fit of the model was extremely low

(adj. R2 = 0.0284), which implies that the regression model explained 2.84% of the

differences in the variable Difference. The variables XBRL, Size, Profitability, and

Profitable were not statistically significant to the prediction. The variables Leverage and

Turnover were statistically significant (p < 0.05) with beta coefficients of 0.005 and

2.897, respectively. Regression coefficients and standard errors can be found in Table 11

on the next page.

Similar results were found for period 2: The regression model statistically significantly

predicted Difference (F = 3.73, p < 0.005). The overall fit of the model was slightly higher

than for period 1 but still low (adj. R2 = 0.292). The variables XBRL, Size, Turnover,

28

Profitability, and Profitable were not statistically significantly to the prediction. The

variable Leverage was statistically significant (p < 0.05) with a beta coefficient of almost

zero (0.004). Regression coefficients and standard errors can be found in Table 12.

Table 11. Post-‐XBRL Period 1 VIF Tolerance Beta t-‐value p-‐value Model statistics

XBRL 1.01 0.99 0.004 0.23 0.82 Dependent variable Size 1.68 0.59 0.000 0.17 0.86 Difference Leverage 1.65 0.61 0.005 2.34 0.02 N = 546 Turnover 1.13 0.88 2.897 2.29 0.02 Adjusted R 2 =0.0284 Profitability 1.45 0.69 -‐0.650 -‐1.17 0.24 F = 3.66 (p-‐value = 0.0014) Profitable 1.46 0.68 0.003 0.08 0.94 Mean VIF 1.40

Table 12. Post-‐XBRL Period 2 VIF Tolerance Beta t-‐value p-‐value Model statistics

XBRL 1.05 0.95 0.001 0.14 0.89 Dependent variable Size 1.66 0.60 0.000 0.47 0.64 Difference Leverage 1.67 0.60 0.004 2.10 0.04 N = 546

Turnover 1.15 0.87 2.560 1.82 0.07 Adjusted R2 = 0.0292 Profitability 1.43 0.70 -‐0.683 -‐1.35 0.18 F = 3.73 (p-‐value = 0.0012) Profitable 1.46 0.69 -‐0.018 -‐0.54 0.59 Mean VIF 1.40

4.3.2 Robustness check The regression was performed again with the command robust to control for points of

high leverage (significant outliers). This model predicted Difference for period 1 as still

significant (F = 6.11, p < 0.005). The overall fit of the model was low with an R2 of

0.0391. XBRL, Size, Profitability, and Profitable were not statistically significant to the

regression model, whereas the variables Leverage and Turnover were statistically

significant. Results can be seen in Tables 17 and 18, in the appendix.

Similar findings are found for period 2. The robust model still predicted the value of

Leverage as significant (F = 6.07, p < 0.005). The overall fit of the model was still low

(R2 = 0.0398). XBRL, Size, Profitability, and Profitable were not statistically significant to

the regression model. The variables Leverage and Turnover were statistically significant

to the regression model. The difference in the non-‐robust model is that Turnover was

29

statistically significant; the p-‐value was 0.05. The overall fit of the model was higher

with the robust model (R2 was higher).

4.3.3 Testing H2 and H3 The second (H2) and third (H3) hypotheses were tested by implementing XBRL*Size

and XBRL*Leverage into the regression model. The output is shown in Tables 13 and 14

(for XBRL*Size for H2) and Tables 15 and 16 (for XBRL*Leverage for H3).

Table 13. Testing H2: Post-‐XBRL Period 1 VIF Tolerance Beta t-‐value p-‐value Model statistics

XBRL 1.12 0.89 0.004 0.19 0.85 Dependent variable Size 2.66 0.38 0.000 0.10 0.92 Difference Leverage 1.66 0.60 0.005 2.34 0.02 N= 546 Turnover 1.13 0.88 2.899 2.29 0.02 Adjusted R 2= 0.0266

Profitability 1.46 0.69 -‐0.649 -‐1.16 0.25 F = 3.13 (p-‐value = 0.003)

Profitable 1.46 0.68 0.003 0.08 0.94 XBRL*Size 2.01 0.50 0.000 0.07 0.94 Mean VIF 1.64


XBRL 1.15 0.87 0.001 0.05 0.96 Dependent variable Size 2.72 0.37 0.000 0.19 0.85 Difference Leverage 1.68 0.60 0.004 2.11 0.04 N = 546 Turnover 1.16 0.86 2.605 1.84 0.07 Adjusted R2 = 0.0275


Profitable 1.46 0.69 -‐0.018 -‐0.53 0.60 XBRL*Size 2.04 0.49 0.000 0.28 0.78 Mean VIF 1.66

The regression model with XBRL*Size was still statistically significantly predicting the

dependant variable, Difference. However, the overall fit of the model was still extremely

low with an adjusted R2 of less than 0.0275 for both period 1 and 2. The p-‐values for the

variable XBRL*Size in period 1 and 2 were, respectively, 0.94 and 0.78. Therefore,

XBRL*Size did not add a significant explanation to the regression model. Similar findings

were discovered for the variable XBRL*Leverage in both period 1 (p = 0.57) and period

2 (p = 0.62), which is shown in Tables 15 and 16.

30


XBRL 1.41 0.71 -‐0.002 -‐0.11 0.91 Dependent variable Size 1.70 0.59 0.000 0.12 0.91 Difference Leverage 2.13 0.47 0.004 1.79 0.07 N = 546 Turnover 1.13 0.88 2.913 2.31 0.02 Adjusted R2 = 0.0272


Profitable 1.47 0.68 0.001 0.03 0.98 XBRL*Leverage 2.00 0.50 0.002 0.57 0.57 Mean VIF 1.62


XBRL 1.44 1.44 0.69 -‐0.002 -‐0.20 Dependent variable Size 1.7 1.70 0.59 0.000 0.37 Difference Leverage 2.03 2.03 0.49 0.004 1.63 N = 546 Turnover 1.16 1.16 0.86 2.644 1.87 Adjusted R2 = 0.0281

Profitability 1.44 1.44 0.69 -‐0.655 -‐1.29 F = 3.25 (p-‐value = 0.0022)

Profitable 1.46 1.46 0.69 -‐0.018 -‐0.55 XBRL*Leverage 1.94 1.94 0.52 0.001 0.62 Mean VIF 1.60

4.4 Chapter summary The selected data sample was analysed, and the results were provided in this chapter.

After describing the dataset with descriptive statistics, preliminary tests regarding

normality, outliers, multicollinearity, and heteroscedasticity were performed. The

regression model was tested by using a multivariate regression analysis. The data was

corrected for significant outliers by using a robust model. The variables XBRL, Size,

Profitability, and Profitable were not statistically significant to the regression model in

period 1. The variables Turnover and Leverage were significantly to the regression with

p-‐values below 0.05. Regarding period 2, the variables XBRL, Size, Turnover,

Profitability, and Profitable were not statistically significant to the prediction. However,

the variable Leverage was statistically significant (p < 0.05). The robust regression

showed both Turnover and Leverage to be statistically significant to the regression

model for both period 1 and period 2. The variables XBRL*Size and XBRL*Leverage were

not statistically significantly to the regression model.

31

5 Conclusion This chapter will summarize and discuss the results of the data analysis, as shown in the

previous chapter. The implications for both theory and practice will be addressed.

Furthermore, the limitations of this research will be examined. To conclude,

recommendations for future research will be made.

5.1 Conclusion and discussion The aim of this research was to examine whether the difference in companies’ assigned

credit ratings by credit rating agencies (CRAs) were influenced by the introduction of

XBRL. The introduction of XBRL would enable external parties to analyse a company

more cheaply and quickly since they were no longer supposed to determine what data

was relevant in their computer systems. XBRL would allow them to import all available

data at a low cost, thus improving information efficiency (Pinsker & Li, 2008). The SEC

made the use of XBRL mandatory in June 2009 (SEC, 2008), and the split ratings before

and after this period were analysed. Split ratings are the difference in ratings provided

by the largest CRAs.

The findings in the literature resulted in three hypotheses: (H1) The adoption of XBRL

has a reducing effect on split ratings, (H2) The effect of XBRL adoption on split ratings is

stronger in larger firms, and (H3) the effect of XBRL adoption on split ratings is weaker for

firms that are more leveraged. The size and leverage of a company were expected to be

moderating variables for the difference in credit ratings as a result of using XBRL.

The findings in literature have been transformed into a regression model with the

following equation:




Where,

i = firm

t = period: pre-‐XBRL, post-‐XBRL period 1, or post-‐XBRL period 2

With respect to the first hypothesis, two tests were performed: A paired t-‐test and a

32

regression analysis. The paired t-‐test showed that there was not enough evidence to

state that XBRL use reduced split ratings in periods 1 or 2. The regression showed

similar results. Hypothesis 1 was thus rejected, and the null hypothesis (H0) was

accepted at a 95% confidence level. There was significant evidence that the dependent

variable (Difference) was not influenced by XBRL. These tests were conducted for two

time periods: Directly after the mandatory use of XBRL and one quarter later. The

results were similar for both time periods.

The variables Size, Turnover, Profitability, and Profitable were not as significant in this

model as presumed. They did not add explanatory value to the regression model. These

variables were based on previous research on the effect of XBRL on the relative spread

of shares (Yoon, Zo, & Ciganek, 2011). Therefore, the difference in credit ratings is not

controlled by these variables.

The coefficient of the control variables Leverage and Turnover was statistically

significant; the p-‐value was below 0.05. This result implies that these variables did add

explanatory value to the model. The variable Leverage was positively correlated to

Difference with a beta almost equal to zero (0.004), which is extremely weak. The beta

coefficient of Turnover was 2.90 for period 1 and 2.56 for period 2.

The robust regression model had a low R-‐square of 0.0391 for period 1 and 0.0398 for

period 2. These results imply that the provided model explained the changes in the

difference in split ratings for less than 4%. Therefore, the statistically significantly

explanatory variables, Leverage and Turnover, which were based on the literature

review, together minimally explain the difference in split ratings.

Size

Theoretically, the improvement of information efficiency for larger companies will be

greater from the usage of XBRL since the usage of XBRL will make it more efficient to

compare different business reporting methods (Weber, 2003). The size of a firm was,

therefore, expected to be positively correlated to reduction in split ratings.

This theory was formulated into the second hypothesis, The effect of XBRL adoption on

split ratings is stronger in larger firms, and was tested by adding the variable XBRL*Size

to the regression model for both period 1 and period 2. This variable was not

33

statistically significant to the regression model. At a 95% confidence level, there was not

enough evidence to support this hypothesis, and H2 was, therefore, rejected.

Leverage

Higher leveraged firms were expected have voluntarily disclosed more information in

order to reduce costs of debt (Dumontier & Raffournier, 1998; Wallace & Naser, 1995).

CRAs were expected to obtain fewer new insights into these highly leveraged companies

when they started using XBRL. Leverage was, therefore, expected to be negatively

correlated to a reduction in split ratings.

This idea was formulated into the third hypothesis: The effect of XBRL adoption on split

ratings is weaker for firms that are more leveraged. This hypothesis was tested by adding

the variable XBRL*Leverage to the regression model for both period 1 and period 2. This

variable was not statistically significant to the regression model. At a 95% confidence

level, there was not enough evidence to support this hypothesis, and H3 was, therefore,

rejected.

To conclude, this research investigated the U.S. stock market and enough evidence was

found to state that the difference in credit ratings is not influenced by the use of XBRL.

This study helps in the debate regarding making XBRL use mandatory. Those in favor of

using XBRL focus on the benefits of implementing XBRL for both companies and

stakeholders (Pinsker & Li, 2008), and it is important to perform research to test these

arguments.

5.2 Limitations One of the limitations for this research is the fact that CRAs do not easily change their

credit ratings. For example, Moody’s only takes a rating action “when it is unlikely to be

reversed within a relatively short period of time” (Cantor, 2001). Changes in credit

ratings involve a process in which the company itself can provide its opinion on changes

in credit ratings. CRAs might be conservative with certain insights since they “attempt

to avoid unnecessary rating volatility by ignoring changes in a client’s business or

financial risk profiles that occur as part of the regular business cycle and are likely to be

reversed shortly after” (Funcke, 2015, p. 20). The effect of adopting XBRL might,

therefore, occur on a longer timeframe.

34

Furthermore, this research was based on companies that were listed to the U.S. stock

market. Credit ratings for equity of listed companies are common in the US and are less

standard in other parts of the world. Therefore, it is not possible to generalize these

findings on a global level.

5.3 Recommendations for future research The recommendations for future research are linked with the limitations of the study

performed. The overall fit of the presented model was low; other control variables than

those used should be researched. It is recommend to include market-‐related factors in

future research investigating the role of the financial crisis. The difference in credit

ratings might have been influenced by the financial crisis. In addition to market factors,

different time periods might be addressed, as well, in order to strength this model for

other parts of the economic cycle.

Furthermore, this research was based on companies listed in the U.S. stock market. As

discussed in the previous section, the extensive use of credit rating agencies in the US

differs from other parts of the world. It is recommend to perform similar research in

other countries.

35

6 References Bai, Z., Sakaue, M., & Takeda, F. (2012, December). The Impact of XBRL Adoption on the

Information Environment in Japan. IPRC Discussion Paper Series No.05 . Tokyo, Japan. Bini, L., Giunta, F., & Dainelli, F. (2010). Signalling Theory and Voluntary Disclosure to the

Financial Market-‐Evidence from the Profitability Indicators Published in the Annual Report. Available at SSRN 1930177.

Blankespoor, E. (2012). The impact of investor information processing costs on firm disclosure choice: Evidence from the XBRL mandate. Michigan: Doctoral dissertation, The University of Michigan.

Blankespoor, E., Miller, B., & White, H. (2014). Initial evidence on the market impact of the XBRL mandate. Review of Accounting Studies , 19 (4), 1468-‐1503.

Boot, A., & Milbourn, T. (2002). Credit Ratings as Coordination Mechanisms. Amsterdam: Faculty of Economics and Econometrics University of Amsterdam.

Bowerman, B., O'Connell, R., & Murphree, E. (2009). Business Statistics in Practice. McGraw-‐Hill International Edition (5).

Breusch, T., & Pagan, A. (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrica , 47 (5), 1287-‐1294.

Cantor, R. (2001). Moody’s investors service response to the consultative paper issued by the Basel Committee on Banking Supervision and its implications for the rating agency industry. Journal of Banking and Finance , 175.

Chen, H., & Li, F. (2013). Analysis the Impact of XBRL in China’s Capital Market Using Methods of Empirical Research. Research Journal of Applied Sciences, Engineering and Technology , 1521-‐1527.

Chiang, R., & Venkatesh, P. (1988). Insider holdings and perceptions of information asymmetry: a note. Journal of Finance , 43 (4), 1041-‐1048.

Cong, Y., Hao, J., & Zou, L. (2014). The Impact of XBRL Reporting on Market Efficiency. JOURNAL OF INFORMATION SYSTEMS , 28 (2), 181–207.

Copeland, T., & Galai, D. (1983). Information Effects on the Bid-‐Ask Spread. The Journal of Finance , 38 (5), 1457-‐1469.

D’Amato, S. (2014). An analysis of the credit rating agencies. International Journal of Critical Accounting , 6 (3), 258-‐283.

Dodd-‐Frank. (2010). Dodd-‐Frank Wall Street Reform and Consumer Protection Act. Washington, DC: Government Printing Office.

Dumontier, P., & Raffournier, B. (1998). Why Firms Comply Voluntarily with IAS: an Empirical Analysis with Swiss Data. Journal of International Financial Management and Accounting , 9 (3).

Efendi, J., Dong Park, J., & Subramaniam , C. (2010, September 5). Do XBRL Reports Have Incremental Information Content? – An Empirical Analysis. Retrieved March 24, 2015 from http://ssrn.com/abstract=1671723

Efendi, J., Park, J., & Smith, L. (2014). Do XBRL filings enhance informational efficiency? Early evidence from post-‐earnings announcement drift. Journal of Business Research , 67, 1099-‐1105.

Elliott, R., & Jacobson, P. (1994). Costs and benefits of business information disclosure. Accounting Horizons , 8 (4), 1-‐80.

Fama, E. (1970). The Journal of Finance. EFFICIENT CAPITAL MARKETS: A REVIEW OF THEORY AND EMPIRICAL WORK , 25 (2), 383-‐417.

Forster, J. (2008). Essays on the Economics of Credit Rating Agencies and Banking. Ludwig-‐Maximilians-‐Universitat Munchen.

36

Funcke, N. (2015). Credit Ratings and the Auditor’s Going-‐Concern Opinion. Maastricht: Maastricht University.

Geiger, M., North, D., & Selby, D. (2014). Releasing Information in XBRL: Does It Improve Information Asymmetry for Early U.S. Adopters? Academy of Accounting and Financial Studies Journal , 18 (4), 66-‐83.

Groysberg, B., Healy, P., & Chapman, C. (2008). Buy-‐Side vs. Sell-‐Side Analysts’ Earnings Forecasts. Financial Analysts Journal , 64 (4), 25-‐39.

Hodge, F., Kennedy, J., & Maines, L. (2004). Does search-‐facilitating technology improve the transparency of financial reporting? Account Rev , 79 (3), 687-‐703.

Jensen, M., & Mechling, W. (1976). Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics , 3 (4), 305-‐360.

Kadakal, O. (2013). XBRL adoption and information asymmetry. Rotterdam: Rotterdam School of Management.

Kapoor, R. (2012, October 19). XBRL. Retrieved Februari 14, 2015 from http://www.slideshare.net/ManuVrs/webxbrl

Kim, J., Lim, J., & No, W. (2012). The Effect of First Wave Mandatory XBRL Reporting across the Financial Information Environment. Journal of Information Systems , 26 (1), 127-‐153.

Leuz, C., & Verrecchia, R. (2000). The Economic Consequences of Increased Disclosure. Journal of Accounting Research , 38 (Supplement 2000), 91-‐124.

Mann, H., & Whitney, D. (1947). On a test of whether one of two-‐random variables is stochastically larger than the other. The Annals of Mathematical Statistics , 18 (1), 50-‐60.

Millon, M., & Thakor, A. (1985). Moral Hazard and Information Sharing: A Model of Financial Information Gathering Agencies. The Journal of Finance , 40 (5), 1403-‐1422.

Nooyi, I. (2006, October 4). CEO Pepsi. (S. Johnson, Interviewer) http://ww2.cfo.com/risk-‐compliance/2006/10/the-‐good-‐and-‐bad-‐about-‐xbrls-‐future/.

O'Brien, R. (2007). A Caution Regarding Rules of Thumb for Variance Inflation Factors. Quality & Quantity 41.5 , 673-‐690.

O'Kelly, C. (2007). XBRL Europe driving a business reporting renaissance. Accounting Ireland: Financial Reporting , 39 (6), 16-‐18.

Pinsker, R., & Li, S. (2008). Costs and benefits of XBRL adoption: early evidence. Communications of the ACM , 51 (3), 47-‐50.

Premuroso, R., & Bhattacharya, S. (2008). Do early and voluntary filers of financial information in XBRL format signal superior corporate governance and operating performance? International Journal of Accounting Information Systems , 9 (1), 1-‐20.

Richards, J., Smith, B., & Saeedi, A. (2006). An Introduction to XBRL. Retrieved Februari 11, 2015 from http://ssrn.com/abstract=1007570

(1987). In P. Rousseeuw, & A. Leroy, Related Statistical Techniques. Hoboken, NJ, USA: John Wiley & Sons, Inc.

Rubini, H. (2000). De-‐Regulated Credit Bureaus: Could They Raise Credit Rationing? Pennsylvania State University.

SEC. (2008, December 18). SEC Approves Interactive Data for financial Reporting by Public Companies and Mutual Funds. Retrieved Januari 25, 2015 from http://www.sec.gov/news/press/2008/2008-‐300.htm

SEC. (2015). SEC Form 10-‐Q. Retrieved Februari 18, 2015 from http://www.sec.gov/answers/form10q.htm

SEC. (2011, September 19). SEC Interactive Data Filings and RSS Feeds. Retrieved June 04, 2015 from http://www.sec.gov/spotlight/xbrl/filings-‐and-‐feeds.shtml

SEC. (2009). SEC Interactive Data to Improve Financial Reporting. Retrieved 2 19, 2015 from http://www.sec.gov/rules/final/2009/33-‐9002.pdf

37

Singhvi, S., & Desai, H. (1971). An empirical analysis of the quality of corporate financialdisclosure. The Accounting Review (46), 129-‐138.

Sufi, A. (2009). The real effects of debt certification: Evidence from the introduction of bank loan ratings. Review of Financial Studies , 22 (4), 1659–1691.

Sykes, A. (2000). An Introduction to Regression Analysis, In E. Posner (Ed.), Chicago Lectures in Law and Economics.

Vasarhelyi, M., Yang, D., & Liu, C. (2003). A note on the using of accounting databases. Industrial Management and Data Systems , 103 (3), 204-‐210.

Verrecchia, R. (1980). Consensus Beliefs, Information Acquisition, and Market Information Efficiency. The American Economic Review , 70 (5), 874-‐884.

Wallace, R., & Naser, K. (1995). Firm-‐specific determinants of the comprehensiveness of man-‐datory disclosure in the corporate annual reports of firms listed on the stock exchangeof Hong Kong. Journal of Accounting and Public Policy (14), 311-‐368.

Wang, J. (1993). A Model of Intertemporal Asset Prices Under Asymmetric Information. The Review of Economic Studies , 60 (2), 249-‐282.

Weber, R. (2003). XML, XBRL, and the Future of Business and Business Reporting. In rust and data assurances in capital markets: The role of technology solutions (pp. 3-‐6). PricewaterhouseCoopers LLP.

Williams, R. (2009). Using heterogenous choice models to compare logit and probit coefficients across groups. Sociological Methods & Research , 37, 531-‐559.

Yoon, H., Zo, H., & Ciganek, A. (2011). Does XBRL adoption reduce information asymmetry? Journal of Business Research (64), 157-‐163.

38

Appendix: Figures

Table 17. Robust regression Post-‐XBRL Period 1 VIF Tolerance Beta t-‐value p-‐value Model statistics

XBRL 1.01 0.99 0.004 0.23 0.82 Dependent variable: Size 1.68 0.59 0.000 0.17 0.86 Difference Leverage 1.65 0.61 0.005 2.34 0.02 N=546 Turnover 1.13 0.88 2.897 2.29 0.02 R2=0.0391 Profitability 1.45 0.69 -‐0.650 -‐1.17 0.24 F=6.11 (p-‐value=0.0000) Profitable 1.46 0.68 0.003 0.08 0.94 Mean VIF 1.40

Table 18. Robust regression Post-‐XBRL Period 2 VIF Tolerance Beta t-‐value p-‐value Model statistics

XBRL 1.05 0.95 0.001 0.14 0.89 Dependent variable: Size 1.66 0.60 0.000 0.69 0.49 Difference Leverage 1.67 0.60 0.004 2.25 0.03 N=546 Turnover 1.15 0.87 2.560 1.98 0.05 R2=0.0398 Profitability 1.43 0.70 -‐0.683 -‐1.40 0.16 F=6.07 (p-‐value=0.0000) Profitable 1.46 0.69 -‐0.018 -‐0.53 0.60 Mean VIF 1.40

The effect of adopting XBRL on credit ratings

Economy & Finance