Geographic distances between corporate headquarters and their subsidiaries : how do they affect audit quality?

(1)

1 Amsterdam Business School

Geographic Distances between Corporate Headquarters and Their

Subsidiaries: How Do They Affect Audit Quality?

Name: Li, Ting

Student number: 11089644

Thesis supervisor: Dr. Sander Van Triest Date:19 June 2016

Word count: 16,379

MSc Accountancy & Control, Specialization: Accountancy Faculty of Economics and Business, University of Amsterdam

(2)

2 Statement of Originality

This document is written by student Li, Ting who declares to take full responsibility for the contents

of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

Inspired by prior studies on the association between auditors-clients’ geographic proximity and absolute discretionary accruals (Choi et al.2012 and Jensen et al. 2013), I explore whether there is the same relation between absolute discretionary accruals and headquarter-subsidiaries’ geographic proximity. After controlling auditor-client proximity, non-audit fees, Big 4, financial performance indicators etc., my regression model has been built up with a purpose of testing how headquarter-subsidiaries distances are linked to absolute discretionary accruals, known as an inverse proxy for audit quality in accordance with prior studies. Discretionary accruals of my observations are calculated by two model DeAngelo Model (DeAngelo 1986, Jones et al. 1991), and Modified Jones Model (Jones et al. 1991, Dechow et al.1995, etc.). After estimation of absolute discretionary accruals (ADA1 and ADA2), empirical results show that there is a positive and significant association between headquarter-subsidiaries’ average distance and absolute discretionary accruals; that is, geographic distance between headquarters and subsidiaries has a negative and significant relation with audit quality, meaning that audit quality will be negatively influenced by subsidiaries’ distance on the conditions that other thing being equal. The finding is consistent with prior studies that information advantages dissipates with geographic distance. Whereas three main limitations are involved in this study, including, the limited number of observations, the short observing years, and inaccuracy of discretionary accruals measurement and average distances’ measurement.

Key words: Geographic proximity; information asymmetry; monitoring quality; discretionary

accruals; earnings management; audit quality; geographic distances; headquarters and subsidiaries.

Acknowledgements

I appreciate so very much the helpful guidance, precious suggestions and immediate feedback from Dr. Sander Van Triest during the most essential period. I sincerely thank the useful and practical suggestions for collecting geographic information and empirical modeling from Dr. W.H.P (Wim) Janssen at the beginning period of my thesis. I thank my friends who offered me great helps in data collection, regression analysis and formality.

(4)

4

1 Introduction

The purpose of the study is to examine the association between headquarter-subsidiaries’ geographic distances and audit quality by means of developing an empirical model and conducting regression analysis that uses absolute discretionary accrual as a proxy for audit quality and that take account of the geographic distances between headquarters and their subsidiaries.

In the recent decades, the effect of geographic proximity has been widely tested and empirical studies show that geographic distances between entities do have relation with possibility of company violations, the magnitude of opportunistic earning management in financial statements, and audit quality, etc. Referring back to accounting and audit studies, three main studies that are based on public companies in developed and developing backgrounds (Choi et al. 2012, Jensen et al. 2013 and Liu Wenjun et al.2014) find that auditor-client’s geographic proximity has a positive and significant association with absolute discretionary accruals; their empirical results provide a practical evidence to the argument that the information asymmetry is related to geographic proximity. Additionally, an increasing number of studies in finance and investment have demonstrated that geographic proximity lowers the information asymmetry between economic agents by facilitating information transfer and improving professional monitoring effectiveness (DeFond et al. 2011; Kedia and Rajgopal 2011; Malloy 2005). Inspired by prior researches, I expect that a similar association between audit quality and geographic proximity within headquarters and subsidiaries exists for the reasons that: firstly, headquarters where auditors conduct most audit activities are information transfer centers so that the distances between headquarters and their subsidiaries have may cause information asymmetry; secondly, geographic distribution of subsidiaries do matter information sharing (Lahiri. N. 2010), I expect that geographic distribution of subsidiaries could also thus has impact on professional monitoring power.

In order to examine my hypothesis, I build up a regression model to test my hypothesis that use absolute discretionary accruals as a proxy for audit quality and use average distance between headquarters and subsidiaries as a measurement of geographic proximity. The magnitude of absolute abnormal accrual in audited financial statements is widely used as a proxy for audit quality (Eshelman and Peng 2014, Defond et al. 2014 etc.). Theoretically speaking, a higher level of abnormal discretionary accruals is a result of a lower professional monitoring power and a poorer ability of auditors to detect abnormal accruals in financial reports. Several measurements of discretionary accruals have been modeled and updated in the past decades. In my research, I used

(6)

6

DeAngelo (DeAngelo et al. 1985, Jones et al.1991) expectation model to calculate discretionary accruals because DeAngelo expectation model is suitable to observations in two years and relatively straightforward. Besides, I followe the widely used measurement of discretional accruals, Modified Jones Model (Dechow et al. 1995).

Geographic proximity is an essential element that allows auditors to acquire audit evidence and other soft information beneficial to detect discretionary accruals (Choi et al 2012). Bearing in mind of effect of geographic proximity, I manually collect distance information of 112 observations out of 1000 total samples; I searched for subsidiaries names in SEC 10-K forms, Exhibit 21 and find out addresses of subsidiaries in Opencorporate website. Finally, calculated distances between headquarters and subsidiaries via a distance measurement tool in Google Maps. After data collection, I conduct regression analysis in order to examine my hypothesis along with robustness tests for empirical model.

Empirical results from regression analysis demonstrate that there is a positive and significant relation between average distances of subsidiaries and the absolute value of discretionary accruals (both ADA1 estimated by DeAngelo model and ADA2 calculated by Modified Jones Model), suggesting that audit quality is negatively associated with headquarter-subsidiaries’ geographic distances. My research gains special insights into the studies of geographic economics research and information asymmetry in more sophisticated angle; my study expands prior studies, which tested audit-client proximity and audit quality, and offers an empirical evidence to the hypothesis that the same association between discretionary accruals and geographic proximity within companies’ headquarters and subsidiaries exists. Furthermore, my hand-collected subsidiaries information could be used for further studies; those data are not available in any currently used database. Practically, the finding, the positive and significant association between the magnitude of discretionary accruals in audited financial statements and headquarters-subsidiaries proximity, could remind accounting firms to pay more attention to financial information or audit evidence from distant subsidiaries. It is meaningful for regulators and lawmakers to enact specific regulations for companies that distant subsidiaries with a purpose of constraining the level of discretionary accruals.

The rest of the paper is arranged as follow: I conduct a literature review and hypothesis development in the second section, pointing out the current research on audit quality and discretionary accruals development in recent research papers, stating proximity and geographic theory and then detailing the hypothesis development. In the third part, I build up the research question, research methodologies, regression models and data sorting criteria and sorting records.

(7)

7

Following that, the fourth section consists of descriptive statistics and Pearson correlations and the fifth part gave an empirical results interpretation in details and robustness tests for empirical model. Conclusions, future research directions, contributions and limitations of my study are arranged at the last part of my research paper.

(8)

8

2 Literature Review & Hypothesis Development

2.1 Audit Quality and Discretionary Accruals

Many extant literatures have examined the association between accruals quality and audit quality, and most of those empirical studies agree on that high quality audit deters corporate managements from reporting discretionary accruals in financial reporting (Persakis and Latridi et al. 2016, Choi et al. 2012, etc.). Accrual quality is the magnitude of absolute discretionary accruals in audited financial statements (Choi et al.2012, Dechow et al. 1995, etc.). Audit quality is the likelihood that auditors detect material misstatements and omissions in financial statements; as a result, auditors provide reasonable but not absolute assurance to the fair and unbiased presentation of the financial statements on which investors depend when making investment decisions (DeAnglo et al. 1980).

In former studies, there are three main proxies for audit quality: magnitude of abnormal accruals (Choi et al. 2012), earning distribution tests and auditors’ issued modified reports (Defond et al. 2002). The first two methods are based on the theory that good audit quality enables less earning management to exit in financial statement. Modified statement method depends on the idea that indifferent and independent auditors have a bigger probability to come to going concerning opinions. Collectively, magnitude of discretionary accrual is widely used in studies of earnings management and audit quality (Choi et al. 2012 and Dechow et al. 2010, etc.). Prior studies document that corporate management bears incentives to manage their financial performance by using accruals in order to reach their desired outcomes (DeFond and Park 1997, Jones et al. 1991; Healy 1985; Becker et al.1998). Accruals are earned revenue and incurred expenses, which would have influences on income statement and balance sheets in corporate financial reporting. Total accrual consists of normal accrual and abnormal accrual; normal accrual is indicative of adjustment of companies’ fundamental performance where discretionary accrual (abnormal) accrual is the result of manipulation of accounting rules and opportunistic earnings management according to corporate goals and management incentives (Choi et al.2012, Dechow and Dichev 2002 et al.). The magnitude of discretionary accrual is used as a distortion of good reporting quality and normal accrual is regarded as a proper earnings adjustment. Since discretionary accruals reflect undetected opportunistic earnings managements in audited financial statements, they should be related to the possibility of reporting a qualified audit reports and regarded as a signal of low audit quality (Bartov et al. 2000). Real earning management (normal accruals) is also used as a proxy for audit quality,

(9)

9

empirical findings show that higher audit quality is positively and significantly associated with higher magnitude of real earnings management (Chi et al. 2011). Compared to normal accruals, discretionary accruals measurement is a more popular and mature method that scholars use to test the association between opportunistic earnings management reflected in financial statements and professional monitoring efficiency (Kedia and Rajgopal et al. 2011, Choi et al. 2012, Jensen et al. 2013, etc.).

Various accrual measurement models have been built up and continuously updated to date as a result of an increasing significant role of accrual in financial and accounting research. Several models usually used are as follow: Deangelo (1986) model used total accrual in the past year as normal accrual benchmark under the assumption that average change of normal accrual is approximately close to 0 (Jones et al. 1991, Deangelo 1986). Jones model (1991) is based on a theory that accruals are results of revenue growth and that depreciation is an adjustment of fixed assets. Modified Jones model was brought into the light when considering growth in credit sales and should be effective in the setting of revenue manipulation (Dechow et al. 1995). Besides, Kothari (2005) updated modified Jones model (Dechow et al. 1995) and matched companies’ performance (return of assets) to observation in the same year and same industry; the method is used when companies’ performance is of first priority in research. Following on, other widely used model is Dechow and Dechec (2002); in this method, cash flow of past year current year and future year are taken into consideration in accrual measurement, focusing on accrual in short periods (Dechow 2002, 2005). 2.2 Geographic Proximity and Economic Studies

Existence of geographic proximity impacts on information transferring and sharing has been an ongoing debate since the past years. Morgan (2004) argue that globalization and digitalization end the power of geographic proximity. With the rapid development of information technology and enhanced logistical engagements, the importance of distance in global economic setting is no longer significant (Castells, 1996; Cairncross, 1997). Many researchers have conducted empirical studies to give supports to this debate. To directly answer this debate, Olivier (2012) contribute a new result that geographic proximity is playing an important role still today in information and knowledge sharing. What’s more, recent finance economics studies offer more empirical evidences that geographic proximity between business entities does matter to the behavior and performance of business participants, including regulators, lawmakers, corporate management, investors and other

(10)

10

stakeholders. Then, the effect of geographic proximity is not linear; in other words, information travels within a distance range, leading to information asymmetry (Kedia and Rajgopal 2013. Information asymmetry increases with the geographic distance (Jensen et al. 2011). That is, geographic proximity is beneficial for an entity to gather sensitive and ambiguous information. To be more specific, a shorter geographic distance gives one more valuable and reliable accesses to the latest information of the other entities. Also, the geographic proximity reduces information acquisition costs and provides channels to private information specific to certain entities (Coval and Moskowitz 1999, 2001).

To answer practical questions, the role of geographic proximity has been gradually explored in geographic economics, finance, and accounting studies. Coval and Moskowitz (1991) found that US fund managers are more likely to invest their money in the companies that their headquarters are near to their office locations. By taking advantage of geographic proximity, financial analysts are more likely to conclude higher projection accuracy in investing decision-making (Malloy et al. 2005). When investors change their residence locations, their investment portfolio will be reconstructed and increase their investment to companies that are closer to their new locations (Andriy et al. 2009). Furthermore, local investors can gain higher return on their equity investment activities (Baik 2010; Coval & Moskowitz et al 2011).

The role of geographic proximity in professional monitoring, the function that is aiming to monitor corporate performance presentation instead of earning a monetary return from investment activities, has been examined in several recent studies. Several scholars start to expect whether the association between geographic proximity and effectiveness of professional monitoring exists. Kedia and Ragopal (2011) used “differently informed criminal hypothesis” and “Constrained Cop” hypothesis to investigate whether the Securities and Exchange Commission’s (SEC) enforcement preferences influence a firm’s likelihood of committing a violation (Kedia & Rajgopal et al. 2011). They suggest that a positive association between high quality of financial statements and geographic proximity between companies and SEC offices exists. Their empirical contributions are in accordance with the former findings as mentioned above that information asymmetry is potentially embedded in geographic proximity.

After Enron scandal and bankruptcy of the accounting firm Arthur Andersen, financial regulators, lawmakers, researchers, and investors have transferred their focus to engagement-specific factors that influencing audit quality. Considering the quality of professional monitoring efficiency, audit quality is widely discussed. Choi (2012) conduct an empirical study to examine a linkage

(11)

11

between auditor–clients’ geographic proximity and the magnitude of absolute discretionary accruals, which are used as a proxy for audit quality. Their findings provide a practical evidence that there is a positive association between proximity and abnormal accruals and that informational advantages are related to the geographic proximity, suggesting that information advantages help auditors more effectively detect opportunistic earnings management in financial reporting. The same empirical finding is contributed from other two studies based on developed country - the United States of America (Jensen et al. 2013) and developing country – China (Liu Wenjun 2014). Jensen et al. (2013) used accruals quality as a proxy for audit quality and controlled two elements: monitoring costs and client’s firm’s selection of the auditors. They collect data from Audit Analytics database and find that information advantage dissipates with the geographic distance, indicating that the geographic proximity between auditors and headquarters leads to better audit quality, lower level of absolute discretionary accruals. Liu Wenjun (2014) collect regression variables data from Chinese public companies and come to a similar conclusion that the more distant between auditors and clients the more magnificent the level of absolute discretionary accruals existed in audited financial statements, showing the general existence of the association in various economics, developed and developing entities.

By way of conclusion, the effect of geographic proximity has been widely examined by researchers in many fields, including investment, finance behavior, investors reactions, regulatory enforcements and auditing industry, suggesting that geographic distance between entities does matter in various ways. In financial accounting research strand, association between discretionary accruals and proximity has been proofed significant and meaningful for us to notice and explore more. With the progress of accounting research, many of discretionary accruals measurement methods are built up and developed, those ways providing solid methodological foundation and guidance for the current studies on professional monitoring efficiency and reporting quality. After financial crisis, researches on quality of financial reporting and audit quality are increasing; interestingly, empirical studies on auditor and clients’ geographic proximity document an association with audit quality (Choi et al.2012, Jensen et al. 2013, Liu WenJun 2014), but surprisingly, former researchers do not take the geographic distance between subsidiaries and headquarters into appropriate consideration when examining the impacts of geographic proximity on audit quality.

(12)

12

2.3 Hypothesis Development

The study has a main aim to test whether auditors’ professional monitoring power has a negative and significant association with headquarters and subsidiaries’ average distances. Many researches state that local monitors have gained information advantages over distant monitoring professionals (Covel and Moskowitz 1991, Jensen et al. 2013); several empirical studies have showed that monitoring power of auditors decreases with geographic distance between clients and themselves (Choi et al. 2012, etc.), but no research has been done to take a look at the impact of geographic distance within companies – headquarters and subsidiaries.

A(n) (group) audit is conducted by auditors with different locations and various subsidiaries. Auditors overall aims to provide reasonable assurance for the group financial statements that incorporated separately audited financial information are free from material misstatements (IAASB 2008); in other words, auditors have responsibilities to separate financial information provided by subsidiaries. In practice, auditors who perform audit processes for client’s subsidiaries are most working in headquarter offices, the situation where is similar to the prior researches of Choi et al. (2012) and Jensen et al. (2013) that demonstrate that the auditor–clients geographic distances have a positive association with the magnitude of absolute discretionary accruals in audited financial statements. Thus, audit quality is negatively associated with geographic distance, which is a result of information asymmetry from effects of geographic factors and information disadvantages embedded in longer geographic distances. Thus, equally, auditor office is to headquarters what headquarters is to their subsidiaries. Accordingly, I expect that distance between headquarter and subsidiaries would result in information advantages and disadvantages, which would lead to a negative or a positive impact on professional monitoring powers of auditors.

Citing other evidence about the relation between geographic proximity effect and professional monitoring power, an empirical result documents an positive association between SEC offices-firms geographic proximity and occurrences of corporates misconduct in financial reporting (Kedia & Rajgopal et al. 2011), indicating that investigators are more knowledgeable of proximate firms and that distant cooperates are more likely to conduct opportunistic earning management when they learn from preference of SEC. Same situation, auditors mostly working in headquarters would be more knowledgeable to proximate subsidiaries and those auditors could provide better audit quality than auditors who conduct audit for distant subsidiaries.

(13)

13

companies are always the information centers where much financial information, transactions information and managerial decisions are made, consolidated and interacted from subsidiaries (Coval and Moskowitz 1999) and, importantly, the headquarters are the main places in which auditors perform audit procedure. Based on the offices of companies’ headquarters, auditors collect, investigate and evaluate financial and nonfinancial information from all relevant active business subsidiaries. Information about legal matters, ownership, regulatory compliance of headquarters and their subsidiaries, and the existence and accuracy of inner transactions should be collected and evaluated. Thus, if one firm has many subsidiaries with substantial distance between them, information asymmetry will increase with the geographic distance. Thus, auditor will be less knowledgeable of distant subsidiaries. Consequently, effectiveness and efficiency of professional monitoring of auditor should be influenced by the information disadvantages from geographic proximity and, finally, more discretionary accruals should be undetected by auditor and retained in financial statements.

Taking all theoretical and practical findings into account, I posit headquarters-subsidiaries proximity has an impact on information asymmetry and finally have an influence on the magnitude of discretionary accruals. Inspired by extant literature, the magnitude of discretionary accrual in audited financial statement is linked to professional monitoring power that is dependent on information asymmetry. Higher monitoring power (Higher audit quality) and detecting power is supposed to link to less absolute discretional accruals, vice versa. Thus, I presume that there is a positive association between headquarter-subsidiaries’ geographic proximity and the magnitude of absolute discretionary accruals –an inverse proxy for monitoring power (audit quality).

H: The bigger the average geographic distance between corporate headquarters and their subsidiaries, the higher the magnitude of absolute discretionary accruals.

(14)

14

3 Research Methodology

3.1 Data Sources and Observation Sorting

My research is an expansion of Choi et al. (2012) and Jensen et al. (2013); in the same vein, I follow the similar methodology to proxy for audit quality by using the magnitude of absolute discretionary accruals and also the same theory of geographic distance measurement – Haversine distance formula. I used empirical model to test the association between absolute discretionary accruals and headquarter-subsidiaries’ average distances after controlling Big 4, non-audit fees, loss, auditor-clients’ proximity etc.; sample source and sorting procedures are detailed as follow:

My geographic data and financial data are gathered from Compustat Capital IQ – Simplified Financial Statement Extracts file and Fundamentals Annual file, Audit Analytics audit opinion databases and other public information channels in North American settings (Canada and the USA). Financial data involved in empirical models are from Compustat database simplified financial extracts (Sales, Loss, Total asset, etc.) during the year 2012-2014 in order to make sure the timeliness and good match to subsidiaries’ geographic information. Financial data for the calculation of absolute discretionary accruals is from Compustat – North America-Fundamentals Annual database, I use all available companies (excluding financial companies SIC code from 6000-6999) as my observations during the year 2012 to 2014 for the measurement of coefficients in Modified Jones Model (Dechow et al. 1995). Financial data (Year 2012 – 2014) for DeAngelo model is from Simplified Financial Extracts database. Auditor name, auditor address, audit fees etc. are gathered from Audit Analytics-Audit opinion database during the year 2013-2014. Subsidiaries number and address’ information is found in 10-K forms, Exhibit 21. I collect subsidiaries geographic distance and the number of subsidiaries in just one year (2015) and I assume that subsidiaries of my observation do not change during a short period (three years). At the same time, in order to decline impact of M&A activities, I limit time range for my samples (2013, 2014). Finally, I had 112 observations that I can find useful geographic information out of 1000 firms that are randomly selected from Compustat North American public firms lists.

3.2 Measurement of Audit Quality

Following Choi et al. (2012) and Jensen et al. (2013), I choose the measurement of the magnitude of absolute discretionary accruals (ADA1 and ADA2) as a dependent variable in my

(15)

15

empirical model and as a inverse proxy for audit quality. Many measurements of discretionary accruals are developed in past decades as mentioned in literature review, there are used according to studies’ objectives and features. In the study, two types of discretionary accruals measurement model are used. I use DeAngelo expectation model (DeAngelo et al. 1985 and Jones et al. 1991) for two main reasons: firstly, since the difficulty and vigorous selection standards for observation sample, I have small sample group (112 observations) for this empirical analysis, whereas Modified Jones model is normally used for big volume of observations; secondly, for the consideration of timeliness and availability of subsidiaries information, I used samples only for 2 years from 2013-2014 so that economic circumstances are stable; thus, DeAngelo model is appropriate for this research where modifies jones model takes a long time series into consideration, but the DeAndelo model is less accurate in capturing discretionary accrual (ADA1). In addition, in order to verify my empirical study, Modified Jones model (Dechow et al. 1995) that considers sales in credit is used in my empirical model to estimate absolute discretionary accruals (ADA2). Thus, I use both DeAngelo model (DeAngelo et al.1986 and Jones 1991) and Modified Jones Model (Dechow et al. 1995, Choi et al. 2012 etc.) to estimate discretionary accruals (ADA1 and ADA2).

3.2.1 DeAngelo Model

In DeAngelo expectation model (Jones et al.1991 and DeAngelo et al. 1986), discretionary accruals are calculated by the formula:

∆TAt = (TAt – TAt−1) = (DAt – DAt−1) – (NAt – NAt−1)

TAjt = [∆Current Assets - ∆Cash and short-term investments- [∆Current Liabilities (5) - ∆ Debt in Current liabilities (44) - ∆Income Taxes Payable] - Depreciation and Amortization Expense, where the change (TA) is computed between time t and time t – 1. TA is scaled by total asset in year t-1. Abnormal total accruals are difference between current total accrual and total accruals in the past year. In this model, total accruals are used as a benchmark for current normal total accruals, which is under the condition that nondiscretionary accrual doesn’t change from year to year. After measurement of discretionary accruals, I use the absolute value of those results and denote ADA1 as absolute discretionary accruals. Accordingly, ADA1 is measured as one of my proxies for the magnitude of opportunistic earnings management in audited financial statements, meaning lower audit quality.

3.2.2 Modified Jones Model

(16)

16

samples volumes, I use Modified Jones model that is widely used in big volumes observations and longer year range in former studies on estimating the magnitude of discretionary accruals (Dechow et al.1995, and Jones 1991). Data input for discretionary accruals calculations are all available companies (10638 firm-years) after deleting observations that have blank SIC code, blank and negative assets, blank PPE, etc. in Compustat - North - American- Financial Annual database, ranging from the year 2012-2014 and excluding financial companies, SIC Code from 6000-6999. The model is to estimate accruals based on every industry group with at a minimum of 20 observations. 2-digit SIC code are used to identify the industry. Modified Jones model is separated in three step detailed as following:

I. Estimation model

𝑇𝑇𝑇𝑇𝑡𝑡 = _{𝑇𝑇𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴}𝛼𝛼1

𝑗𝑗𝑡𝑡−1+ 𝑎𝑎2(∆𝑅𝑅𝑅𝑅𝑅𝑅 − ∆𝑅𝑅𝑅𝑅𝑅𝑅)𝑡𝑡+ 𝑎𝑎3𝑃𝑃𝑃𝑃𝑅𝑅𝑗𝑗𝑡𝑡+ 𝜀𝜀𝑗𝑗𝑡𝑡

TA is defined as total accrual, consisting of discretionary and nondiscretionary accruals: TA = [∆Current Assets - ∆Cash and short-term investments - [∆Current Liabilities - ∆ Debt in Current liabilities - ∆Income Taxes Payable] - Depreciation and Amortization. Asset is total assets in year t-1; ∆𝑅𝑅𝑅𝑅𝑅𝑅 is the change of revenue between year t and year t – 1; ∆REC is the change of receivables between year t and year t – 1; PPE is gross amount of Property Plant and Equipment in year t; all items defined in modified model are scaled by lagged asset (total asset in year t - 1).

II. Event Period Model:

𝑁𝑁𝑁𝑁𝑇𝑇𝑡𝑡 =_{𝑇𝑇𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴}𝛽𝛽1

𝑡𝑡−1+ 𝛽𝛽2(∆𝑅𝑅𝑅𝑅𝑅𝑅 − ∆𝑅𝑅𝑅𝑅𝑅𝑅)𝑗𝑗𝑡𝑡+ 𝛽𝛽3𝑃𝑃𝑃𝑃𝑅𝑅𝑗𝑗𝑡𝑡

NDA is expected nondiscretionary accruals in events years (2012-2014).

III. Discretionary Accruals (DA) are the difference between actual accruals and nondiscretionary accruals following the same theory as Deangelo model.

𝑁𝑁𝑇𝑇𝑗𝑗𝑡𝑡=𝑇𝑇𝑇𝑇𝑗𝑗𝑡𝑡− 𝑁𝑁𝑁𝑁𝑇𝑇𝑗𝑗𝑡𝑡

Lastly, I use absolute value of discretionary accruals (ADA2) calculated by modified Jones model (Dechow et al. 1995) as my second proxy for audit quality along with ADA1 from DeAngelo model (DeAngelo 1986, Jones et al.1991).

3.3 Measurement of Subsidiaries’ Distance and the Number of Subsidiaries.

Learning from prior literature, I use the average geographic distance between headquarters and their subsidiaries as my proxy for geographic proximity. Distance measurement method in prior

(17)

17

researches (Sinnott 1984, Jensen et al.2013, Liu Wenjun et al.2014, etc.) is based on Haversine distance formula, which requires latitude and longitude data of agencies’ addresses. Haversine Distance Formula (more details in Sinnott 1984):

𝑑𝑑𝑖𝑖𝐴𝐴𝐴𝐴𝑎𝑎𝑖𝑖𝑖𝑖𝐴𝐴𝑖𝑖𝑗𝑗 = 𝑎𝑎𝑖𝑖𝑎𝑎𝑎𝑎𝐴𝐴𝐴𝐴{cos(𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑖𝑖) cos(𝑙𝑙𝑎𝑎𝑖𝑖𝑙𝑙𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑖𝑖) cos�𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑗𝑗� cos�𝑙𝑙𝑎𝑎𝑖𝑖𝑙𝑙𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑗𝑗�

+ cos(𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑖𝑖) sin(𝑙𝑙𝑎𝑎𝑖𝑖𝑙𝑙𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑖𝑖) cos�𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑗𝑗� sin�𝑙𝑙𝑎𝑎𝑖𝑖𝑙𝑙𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑗𝑗�

+ sin(𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑖𝑖) sin�𝑙𝑙𝑎𝑎𝐴𝐴𝑖𝑖𝐴𝐴𝑙𝑙𝑑𝑑𝐴𝐴𝑗𝑗� ∗ 2𝜋𝜋𝑎𝑎/360

With the recent technology development, Google map has launched Distance Measurement tool in 2014. The new tool calculates geographic distances by using the same theory of Haversine distance formula, but simplifies procedures to work out straight distances between two geographic points. This tool works with just addresses information, such as street address or post codes instead of latitude and longitude information. Distance measurement tool in Google Map seems more effective and efficient to get geographic distance and reduce errors involved in hand-calculations; thus, I do not follow prior calculation based on latitude and longitude and I use distance measurement tool in Google map.

Subsidiaries names are found in 10-K forms, Exhibit 21. I selected appropriate samples by various criteria that I use to choose observation observations. Based on subsidiaries’ name, I search for address information via American company information website OpenCorporates.com, which is an open database of the corporate world in the USA. I input subsidiaries names into this website, and then find out subsidiaries’ address information and record them in an excel file. After that, I use headquarters’ address information (can be found in Compustat) and subsidiaries’ address information to calculate the distances by google map distance measurement function. After distance collection, I use average distances as proxies for geographic proximity.

In order to exclude influence of geographic proximity between auditors and clients (Choi et al. 2012, Jensen et al. 2013 and Liu Wenjun 2014), I measure the distances between auditor and clients via the same method following former studies (Coval and Moskowitz 1999; Pirinsky and Wang 2006) as a control variable. Borrowing from prior researches, if the distance that is less than 100 KM, as a control variable; 100 KM is considered as an appropriate geographic boundary where most social and economic information transferred, exchanged and integrated within a community (Choi et al.2012, Coval and Moskowirtz 2001, Kedia and Rajgopal 2011), and Malloy (2005). If the distance between headquarters and auditors is less than 100 km, I give D100 1, otherwise 0. Finally, I use the geographic distance indicator (1 or 0) between auditor and clients as a control variable in the

(18)

18

regression model. In prior research of Choi et al. (2012), the number of business segments and operations including suppliers, service suppliers have been tested in the audit quality research. Thus, I collect the number of subsidiaries from 10K-Froms, Exhibit 21 files and use as control variable. 3.4 Empirical Model

By using the results Angelo model as of proxies for audit quality, measuring the geographic distance between headquarters and subsidiaries, and considering several control variables defined below, I use an empirical model to examine the association between audit quality and geographic distance between headquarters and their subsidiaries. To test my hypotheses, I formulate the following regression model that links the magnitude of absolute abnormal accruals, average distances between headquarters and subsidiaries in addition to other control variables known to affect the extent of earnings management were included:

ADA = α0+ α1ADjt+ α2Njt+ α3D100jt+ α4BIG4jt+ α5NASjt+ α6CHGSALEjt+ α7LOSSjt

+ α8CFOjt+ α9DYjt+ εjt

Where in this empirical model,

ADA: Absolute discretionary accruals based on (ADA1) Deangelo

model(Deangelo et al.1986 and (ADA2) Modified Jones model (Dechow 1995)

j represents: Firm j, 84 firm samples after selection out of 1000 North American public companies from Compustat and SEC website.

t represent: Year t, since the stability of companies structures, average distance and the number of subsidiaries are kept same in year 2013,2014, meanwhile, other variables are subject to annual changes.

AD

represents: Log (1+average distance). Average distances between headquarters and their subsidiaries. N

represents: Log of total number of all active, 100 percent owned subsidiaries of observations from 10-k forms Exhibit 21. D100

represents: Indicator variable for auditor location where the distance between client headquarters and auditor is less than 100 KM; if distance between auditors and headquarters is less than 100 km, D100 is 1; otherwise 0.(Choi et al 2012, Jensen et al.2013, Liu Wenjun 2014).

BIG4

represents: Auditor of firm jt. BIG4 is 1 if auditor if one of BIG 4, and otherwise 0. BIG 4(PwC, Delloite, Ernst & Young and KPMG). NAS

represents Proportion of the difference between audit fees and total audit fees over total audit fees; calculated as the difference between audit fees and total audit fees scaled by total assets for year t-1.

CHGSALES

(19)

19

LOSS

represent: Loss report in annual financial statements; LOSS is 1 if a loss incurs, and otherwise 0. LOSS incurs when income before taxes are less then 0. CFO

represents: Cash flow from operating activities from cash flow statements; scaled by total asset of year t-1. DY

represents Dummy year. If ADA is for year 2013, Dummy year is 0; If ADA is for 2014, Dummy year is 1. 𝜀𝜀 represents The error terms in regression analysis

Distances between auditor’s offices and client’s headquarters have been tested to be a factor that influences opportunistic earnings management (Choi et al. 2013, Jensen et al. 2013 etc.) The distance between SEC offices and firms’ headquarters is also known to be associated with earnings management according to other geographic research in finance study (Kedia & Rajgopal 2011); thus, I used the distance between auditors and firm’s headquarters as a control variable in the regression model. D100 represents that the distance between auditors and headquarters is less than 100km.

Referring to empirical findings Choi et al. (2012), Jensen et al. (2013), the number of business segments have positive and significant impacts on the association between audit quality and geographic locality, a negative association between geographic distance and absolute discretionary accruals; thus, I take the number of subsidiaries as a control variable in my model.

Auditor brand names are taken into the model since several researches show that prominent accounting firms are more effective in deterring management from reporting discretionary accruals by means of leveraging opportunistic earnings management. Becker (1998) performed empirical research by using the Jones model to test the association between audit quality and earnings management and find that prominent accounting firms Big 6 (previously) have better audit quality than non-Big 6 accounting firms. Evidence that prominent auditors are more efficient in decreasing discretionary accrual is found in Kirshnan (2003). In addition, clients of Big 6 firms represent absolute discretionary accruals that are 1.2 percent lower than those of non-big 6 firms (Kirshnan et al 2003). Currently, since the collapses of 2 of Big 6, I use Big 4 (PwC, Deloitte, Ernst & Young and KPMG) as a control variable in the empirical test.

Similar to former researches, I use non audit related fees, which are the difference between total fees involved and audit related fees, as control variables. Prior researches have tested impact of nonaudit fees on discretionary accruals (Ashbaugh et al.2003, Frankel et al 2002, Larcker and Richardson 2004). Change of sales, Loss and CFO cash flow from operating activities were taken into consideration in the regression model with a purpose of controlling those impacts on discretionary accruals driven by incentives of managements.

(20)

20

4 Sample and Descriptive Statistics

4.1 Sample and Sample Selection

Financial data in my study is from databases (Compustat and AuditAnalytics) and geographic information, including geographic information necessary to distance measurement and the number of subsidiaries, is from public information websites. Collection of geographic information and steps of distance measurement are the most time-consuming, difficult and innovative part in my study. Procedures are detailed as follow: firstly, I download an entire company list form Compustat Capital IQ Simplified Financial Statements Extract database, then I randomly select 1000 public companies that are listed in the USA, that are active, and have available financial reports data. After that, I use CIK number of those companies to search for 10-K reports on SEC website; subsidiaries’ names and the number of subsidiaries are found in Exhibit 21 in 10-K forms. During this procedure, I exclude samples based on several criteria: I delete observations that have foreign subsidiaries out of North America, that are out of core business geographic areas of the USA such as Alaska, Hawaii etc. Since all subsidiaries located in Canada are all very near to border line of the USA, thus, I keep those observations; besides, firms located in North America but not in USA and Canada are deleted from observations; Mexican Companies and Caribbean companies are all excluded. Secondly I exclude those companies that have inactive subsidiaries, finally, I have 182 observations from 1000 total samples. In the second phase, I try to find out address information via website Opencorporates.com with the names of subsidiaries; I delete observations that I cannot find out address information for their subsidiaries, that their subsidiaries are not active and that their subsidiaries are not hundred percent controlled by the parents. Consequently, I delete 70 unqualified samples and get 112 samples that have various subsidiaries’ address information and the number of the subsidiaries, the companies that are all in north America (USA and Canada) and meet my sorting criteria.

Next step is to calculate the distance between headquarters and subsidiaries; to simplify my research, I use Google map distance measurement tool to calculate the geographic distance. Google map has a very effective function Distance Measure in which I input addresses of headquarters (from SEC files or and Compustat database) and addresses of subsidiaries (from Opencorpetes website) and then I can work out straight distances between two points. I use average distance of all distances as proxy for geographic proximity, total distances between headquarters and subsidiaries

(21)

21

divided by number of subsidiaries. Furthermore, I extend prior research Choi et al 2012 that demonstrated diversity of segment has impact on the association between auditor-clients geographic proximity and absolute discretionary accruals. In this case, I take the number of subsidiaries into consideration when I was modeling regression tests, and use as a control variable. In all, I had 112 samples out 1000 companies for average distance and the number of subsidiaries (can be found at appendix 3).

After geographic distance measurement and records of the number of subsidiaries, I start to collect data from databases. Compustat Capital IQ and Audit Analytics Audit opinion. Financial data involved in empirical models and DeAngelo model are from Compustat North America Simplified Financial Extracts database (Sales, Loss, Total asset, etc.). Financial data involved in Modified Jones model is from Compustat – North America - financial annual during the year 2013-2015. Current assets, cash and short-term investments, current liabilities, debt in current liabilities, income taxes payable and depreciation and amortization expense are used in the calculation of discretionary accruals. During this process, I delete samples that had blanks; the number of observations that have inadequate financial information in databases amounts to 25. Consequently, I have 87 observations for the year 2013 and the year 2014, respectively.

In next process, I collected auditor related information (BIG 4, nonaudit fees, auditor address). Auditor name, auditor address, audit fees etc. are gathered from Audit Analytics - Audit opinion database during the year 2013-2015. I face a big data loss, because, Audit Analytics database is less updated compared to Compustat Capital IQ database and has inadequate auditor-related information for the year 2015, I only have 24 samples in 2015 that meet my criteria so that I give up my samples in 2015 and keep data in 2013 and 2014 as observations for final regression analysis. As a response to prior researches (Choi et al.2012, Jensen et al.2013), I measured distance between auditors and clients’ headquarter offices as a control variable. I collected data auditor address information Jensen et al. (2013) from Audit Analytics audit opinion database. During this process, since the lack of address information of several observations, I delete 3 observations that were belong to the year 2014 samples but could not be found in observations of the year 2013. Finally, after operating all data sorting procedures including distance measurement within companies, discretionary accrual calculation and distance measurement between auditor and clients’ headquarters and collection other financial data that were used as control variables, I have observation range from 2013-2014, with observations samples 87 and 84 respectively; after matching firm-year formality, the number of my observations amount to firm-year 168.

(22)

22

4.2 Descriptive Statistics and Pearson Correlation

Based on 168 observations(firm-year) after all sorting procedures following abovementioned description, I produce descriptive statistics table as below Table 1 by using the SPSS data analysis tool. ADA represents for the absolute value of discretionary accruals, which is used to be a proxy for audit quality. ADA1 denotes absolute discretionary accruals by using Deangelo model; (DeAngelo 1986), mean and standard deviation of ADA1 are 0.0733 and 0.0979 respectively; ADA2 denotes absolute discretionary accrual that is calculated by using modified Jones Model (Dechow 1995), ADA2 has a mean 0.7519 that is very close to ADA1 but ADA 2 has a higher standard deviation 0.1652. With reference to prior researches, means of ADA1 and ADA2 are lower than those of discretionary accruals of observations in Choi et al. (2012) in which mean is in the range of 0.0846 to 0.1046, and standard deviation is in range of 0.1062 to 0.1333. The differences can be explained by using different data including years, firms, the special sorting criteria, the smaller number of observations and different accruals calculation model, etc..

When it comes to average distance, I consider and exclud outliers (corporate headquarters and subsidiaries located in Hawaii, Alaska, Caribbean areas, Mexico); the maximum is 3943.33 km, and the minimum is 0.00 km, which means that headquarters and subsidiaries are all located at a same building or block (average geographic distance and the number of subsidiaries can be found at appendix 3). Considering the huge difference between the maximum and the minimum, I use log of the sum of average distance and 1 as independent variables, log(1+AD). After calculation, the main independent variable in the regression model, standard deviation is 0.9800 and mean of average distance is 2.4620. Similar method log is used to decrease the big difference with maximum and minimum of the number of subsidiaries. Before use log function, the maximum number of subsidiaries is 30 and minimum is 1, denoting that one company has minimum 1 and maximum 30 subsidiaries. Brand names of auditors are included as a control variable; Big 4 auditors are regarded as professional auditing specialists that provide better audit services for clients. 53 companies out of my 84 observation hired Big 4 auditors (PwC, Deloitte, Ernst & Young and KPMG); I use 1 or 0 as indicators for hiring big 4 or the other accounting firms as their auditors. Mean is 65 percent with a standard deviation 0.479.

Table 1 Descriptive Statistics

Table 1 Descriptive Statistics

N Minimum Maximum Mean Std. Deviation

ADA1 168 .0004 .5629 .0733 .0979

(23)

23 AD 168 .0000 3.5960 2.4620 .9800 N 168 .0000 1.3010 .7493 .3609 BIG 4 168 0 1 .6590 .4788 NAS 168 .0000 1.0000 .1503 .1524 D100 168 0 1 .7860 .4116 CHGSALES 168 -2.6661 1.3467 .0314 .2936 LOSS 168 0 1 .1670 .3738 CFO 168 -.6173 4.5412 .1003 .3738 DY 168 0 1 .5000 .5015

In order to control auditor – client proximity, D100(Distance between headquarters of observation and auditor office is less than 100 KMs) is 1 or otherwise 0. The statistics table shows that 79 percent observations hired their auditors near to headquarter offices (within 100 km). The component of hiring Big 4 and local auditors are consistent with descriptive statistics from prior researches (Choi et al. 2012 and Jensen et al.2013). Only 17 percentages in my observation report losses in two years in their financial statements. Nonaudit fees, change of sales and cash flow from operating activities are all scared by the total asset of the previous year (t-1). As to NAS, the maximum is 1 and the minimum is 0, meaning that the percentage of non-audit fees out of total fees paid to auditors. Change of sales has a minimum -2.666 and maximum 1.347, while CFO (cash flow from operating activities) has a minimum of 0.6173 and a maximum of 4.5412, leading to biggest range 5.1585 among all variables. In order to exclude of possible fluctuation from different year, I use dummy year as a control variable. Since I have only two years, there is the same number of 1 and 0; the mean is 1; standard deviation is 0.501.

Before regression analysis, I conduct Pearson correlation test in order to make sure that my dependent variables, independent variables and control variables have significant correlations. In table 2, the Pearson correlation tests are based on two-tailed test and I have three categories of significance level, 10% level where p-value is less than 0.1, 5% level where p-value is less than 0.05 and 1% level where p-value is less than 0.01.

As can be seen in table 2, AD, average distance between headquarters and subsidiaries is positively and significantly and positively associated with absolute discretionary accruals (ADA1) at 10% significance level (correlation = 0.143; P value = 0.065) and with ADA2 at 5% significance level (correlation = 0.173, p-value = 0.025). Absolute value of discretionary accruals (ADA1 and ADA2) are significantly and negatively correlated to log of the number of subsidiaries at 5 percent level (correlation = -0.197 and p value = 0.01 for ADA1, and correlation = -0.192, p value = 0.013). At the same time, the number of subsidiaries has positive correlation with AD at 1 % significance level (correlation = 0.207; p value = 0.007). Big 4 has negative association with ADA1 at 1% significance level (correlation = 0.300 and p value = 0.000) and with ADA2 at 1% level ( correlation = -0.327

(24)

24

and p-value = 0.000); in addition, Big 4 has a positive correlation with the number of subsidiaries at 1% significance level (Correlation = 0.027; P = 0.007). Nonaudit fees (NAS) has a significant and positive correlation with ADA1 but a not significant association with ADA2. The distance between auditor and clients (D100) are not correlated with other variables at significance level. Change of sales between year t and year t-1 has a negative and significant correlation with absolute discretionary accruals (ADA1) based on DeAngelo Model with coefficient – 0.256 and p-value =0.001. LOSS (when income before taxed < 0) has a significant and positive association with both ADA1 at 1% significance level with p value = 0.000 and ADA 2 at 1% level with a p-value = 0.000. Furthermore, LOSS has a negative correlation (correlation = with number of subsidiaries and Big 4 at 5% (P-value = 0.043) and at 1% (P value is 0.000) significance level respectively. CFO is positively and significantly correlated with absolute discretionary accruals (ADA1) and negatively and significant related to ADA2 at 5% level. CFO also has a negative and significant association with the LOSS (both p values are 0.000, 1% significance level). DY (dummy years) does not show any indicator that it is correlated with any other variables including (dependent variables and independent variables) at significant level; P-values of DY with other variables are from 0.473 to 1, which are not in any significant level, suggesting that those variables are not subject to year’s influence.

Table 2. Pearson Correlation Matrix for ADA1 and ADA2

Table 2

Pearson Correlation Matrix

ADA1 ADA2 AD N Big 4 NAS D100 CHGS

ALES LOSS CFO DY ADA1 1 ADA2 1 AD .143* (.065) .173** (.025) 1 N -.197** (.010) -.192** (.013) 0.207*** (.007) 1 Big 4 -.300*** (.000) -.327*** (.000) .038 (.629) .0270*** (.000) 1 NAS .162** (.036) .086 (.629) .061 (.435) .045 (.566) -.101 (.192) 1 D100 -.096 (.214) -.007 (.929) -.120 (.122) -.042 (.592) -.020 (.802) -.088 (.257) 1 CHGS ALES -.256*** (.001) -.105 (.177) .049 (.526) -.008 (.526) -.059 (.917) -.007 (.924) .110 (.156) 1 LOSS .283*** (.000) .439*** (.000) -.156** (.043) .224*** (.004) -.273*** (.000) .103 (.186) -.156** (.044) -.275 (.000) 1 CFO .291*** (.000) -.164** (.034) -.061 (.431) -.120 (.121) (.849) -.015 .018 (.821) .058 (.454) -.018 (.821) -.266*** (.000) 1 DY -.012 (.881) .021 (.784) .000 (1.000) .000 (1.000) -.037 (6.30) .036 (.642) .029 (.709) .017 (.830) .000 (1.000) .056 (.473 ) 1

Observation Sample: 168 firm year.

(25)

25 **. Correlation is significant at 0.05 level (Pearson Correlation, Two-Tailed) .

***. Correlation is significant at the 0.01 level (Pearson Correlation, Two-Tailed). Two Tailed p Value are presented in parentheses.

In accordance with expectation, the average distance (AD) between clients’ headquarters and their subsidiaries are significantly associated with absolute discretionary accruals (ADA1)at 10% level and ADA2 at 5% level. Control variables (N, Big 4, CHGSALES, LOSS, and CFO) are correlated with absolute discretionary accrual at significant level, which are in consistent with prior studies of opportunistic earnings managements (Kirshman et al. 2003, Frankel et al. 2012, Kim et al. 2003, etc.). DY has no significant correlations with any other variables.

(26)

26

5 Empirical Results

5.1 Regression Analysis and Coefficients

With a purpose of examining the association between the geographic distance of subsidiaries and audit quality proxied by absolute discretionary accruals (ADA1 and ADA2), I perform collinearity tests before regression analyses and the findings show that there is no collinearity association among all variables in both regression analysis (ADA1 and ADA2). The variance inflation factors (VIF) that measure the severity of multicollinearity are all less than 10 ( Table 5, in appendix 1 and appendix 2); according to Kennedy. (1992), VIFs that are less than 10 means that no significant collinearity exists among variables - both independent and independent variables; thus, all variables are appropriate to be included in the empirical model for current regression analysis. Then, I conducted a regression analysis via SPSS and finally got empirical results that could give a proper support to my hypotheses and details of regression results are described as follows:

For regression analysis based on absolute discretionary accruals (ADA1) calculated by DeAngelo model (DeAngelo et al.1986 and Jones et al. 1991), referring to AVONA table (Table 3), I have the first empirical result F = 8.665 with a significance value (p = 0.000). As we all know, once significance in AVONA table is less than 0.05, indicating that at least one of independent variables has a significant association with the dependent variable (ADA1). Thus, I confirm that my regression model is effective and independent variables have a collectively explanatory effect. Similarly, F value of regression analysis based on absolute discretionary accruals (ADA2) calculated by Modified Jones model is 6.628, less than that of regression based on DeAngelo model (ADA1); P-value of the ADA2 regression result is 0.000, indicating that the empirical model is overall significant (Table3, and Appendix 2). In all, the empirical model for both dependent variables (ADA1 and ADA2) are all significant.

Table 3 ADA1 & ADA2 ANOVA Tables

Model Sum of

Squares df Mean Square F Sig.

ADA1 Regression .529 9 .059 8.665 .000 Residual 1.072 158 .007 Total 1.601 167 ADA2 Regression 1.250 9 .139 Residual 3.310 158 .021 6.628 .000 Total 4.559 167

a. Dependent Variable: ADA1 (DeAngelo Model) & ADA2 (Modified Jones Model)

b. Predictor: Constant, AD, N, BIG 4, D100, NAS, CHGSALES, CFO, LOSS, DY

(27)

27

ADA1, R is 0.575 and adjusted R Square is 0.292 with standard error of the estimate 0.0824 , which means, in all, that independent variables in my research are substantive to explain 29.2 percent change of dependent variables (ADA1); moving to regression result based on dependent variables ADA2, R is 0.524 with a adjusted R square 0.233, which is less than that of ADA1, meaning that my empirical results could explain 23.3 percent changes of dependent variables(ADA2). Thus, the empirical model can be considered meaningful to the research question. The small adjusted R Square can be explained by the small number of my observation (84 observations and for only two years, 2014 and 2015).

Table 4 ADA1 & ADA2 Model Summary

Model R R Square Adjusted R

Square Std Error of the estimate

ADA1 .575 .330 .292 .0824

ADA2 .524 .274 .233 .1447

a. Dependent Variable: ADA1 (DeAngelo Model) & ADA2 (Modified Jones Model) b. Predictor: Constant, AD, N, BIG 4, D100, NAS, CFO, CHGSALES, LOSS,DY

Table 5 reports coefficients of variables included in the empirical model that is built up to test the association between dependent variables (ADA1 and ADA2) and independent variables the number of subsidiaries (N), distance between headquarters and subsidiaries (AD), prominent accountants firms (BIG 4), non-audit fees (NAS), distance between auditors and clients (D100), change of sales over year (CHGSALES), LOSS in financial statements (LOSS) and cash flow from operating activities (CFO). All definitions of independent variables (control variables) and dependent variable are given in empirical model explanation with reference to methodology part in this paper.

Table 5

Association between Headquarter – Subsidiaries Distance and Absolute Discretionary Accruals (ADA1, ADA2)

Expected Sign ADA1 DeAngelo Model Year 2013 & 2014 Collinearity

Statistics ADA2 Modified Jones Model Year 2013 & 2014

Collinearity Statistics Using ADA1 as

Dependent Variable Tolerance VIF Dependent Variable Using ADA2 as Tolerance VIF Coefficients Sig. Coefficients Sig.

Constant ? .073 .010 .089 .270 AD + .014 .047* .905 1.104 .026 .032* .905 1.104 N - -.023 .084 .814 1.228 -.049 .156 .814 1.228 BIG 4 - -.046 .002** .849 1.178 -.076 .003** .849 1.178 NAS + .067 .120 .965 1.036 .037 .618 .965 1.036 D100 - -.011 .503 .952 1.051 .025 .375 .952 1.051 CHGSALES + -.065 .006** .887 1.128 -.019 .634 .887 1.128 LOSS + .052 .012* .684 1.462 .134 .000** .684 1.462 CFO - .089 .000** .873 1.145 -.042 .189 .873 1.145 DY ? -.007 .560 .993 1.007 .005 .993 .993 1.007

*,** Denote p-value less than 5 percent, and 1 percent, respectively, indicating a significant association. a. Dependent Variable: ADA1 (DeAngelo Model) & ADA2 (Modified Jones Model)

(28)

28 b. Predictor: Constant, AD, N, BIG 4, D100, NAS,CHGSALES, CFO, LOSS,DY

In table 5, coefficients of AD (log of average distance between headquarters and their subsidiaries) for both ADA1 and ADA2 are, respectively, 0.14 with significance value (p = 0.047) and 0.026 with significance value of (p = 0.032), which both are less than 0.05, indicating there are significant associations between log of average distance (AD) and dependent variables absolute discretionary accruals (ADA1 & ADA2). Besides, positivity of both coefficients is in accordance with my initial expectation, a positive sign. Clearly, the empirical results based on both two models DeAngelo Model and modified Jones Model show that the average distances between headquarters and subsidiaries has a positive and significant association with the magnitude of absolute discretionary accruals (both ADA1 & ADA2); accordingly, my hypothesis that the bigger the average geographic distance between corporate headquarters and their subsidiaries, the higher the magnitude of absolute discretionary accruals, can be supported by the result. Positive and significant coefficients of AD suggest that the longer average distance between headquarters to their subsidiaries, the higher level of absolute discretionary accruals, indicating that other things being equal, audit quality will be impaired by the longer distance between headquarters of clients and their subsidiaries. Similar to former research findings, my empirical finding offers a practical evidence to support the studies that geographic proximity does matter to the magnitude of absolute discretionary accruals, and finally, to audit quality (Choi et al. 2012, Jensen et al. 2013, and Liu Wenjun 2014).

Coefficients of the number of subsidiaries (N) for both ADA1 (p-value=0.084) and ADA2 (p-value = 0.156) have a p-value bigger than 0.05 (significant level) so that absolute discretionary accruals have no significant association with N (the number of subsidiaries). My empirical result demonstrates that coefficient of N is not significant with a negative sign, which is in accordance with to my prior expectation of a negative sign, where the other research on association between auditor-clients’ proximity and audit quality find that log number of business segments (LNBGS) is negatively and significantly associated with absolute discretionary accrual Choi et al. (2012), suggesting that diversified firms (have more business operations segments) have the higher earning quality. The insignificance in my study can be explained by the reasons that the different definitions of the number the business segments used in empirical model could explain the insignificant result. Business segments (LNBGS, Choi et al. 2012 and Jensen et al. 2013) including registered branches, operations, and other business centers as a proxy for business diversity of clients and as a control variable, but in my empirical study, only the number of subsidiaries is used in the regression model

(29)

29

as a control variable. Secondly, the number of business segments in former studies are from Compustat database in Choi et al. (2012) and Jensen (2013); in my research, the number of subsidiaries is collected all by hands and with consideration of excluding observations that have too many subsidiaries, more than 30. Thus, the coefficient of N in my finding has a negative sign like Choi et al. (2012), but is not significant. In addition, hand-made data would possibly incur that lead to the insignificant coefficient.

The coefficients of BIG 4 for ADA1 and ADA2 are, respectively, -0.046 with at a significant level (p-value = 0.002) and -0.076 at a significant level (p-value = 0.003), indicating that the absolute discretionary accrual has a negative and significant association with a choice of prominent auditors (Big 4). The negativity for both ADA1 and ADA2 are the same to my expectation. Thus, the empirical results are in accordance with prior conclusions. DeAngelo (1981) and Dopuch and Simunic (1980) find evidence that BIG 4 auditors are better than non-BIG4 auditors. (Krishnan et al. 2003 and Becker et al. 1998) document that prominent accounting firms are more effective in investigating opportunistic earnings managements and deterring management from reporting discretionary accruals in financial statements, resulting in better audit quality compared to other auditors. Khurana and Raman (2004) find that the BIG 4 auditors are considered as more effective auditors in detecting fraud and misstatement, but this research is unique to the USA setting, they do not find same evidence in the UK and Australia backgrounds. While my research provides other evidence to the quality audit work of BIG 4 over non-BIG4 accounting firms by using DeAngelo model and modified Jones model in US settings.

The coefficient of NAS is 0.067 at insignificant level (p-alue is 0.120) for ADA1; the coefficient of NAS for ADA2 is 0.038 with a p-value 0.618, this empirical results are consistent with (Ashbaugh et al.2003), suggesting that nonaudit fees do not have significant impacts on absolute discretionary accruals. Looking back forwards to former studies on the association between audit quality and nonaudit fees, there is a debate on this topic. Larcker and Richardson (2002) document that ratio of non-audit fees to total fees have a positive and significant association with abnormal accruals. Frankel (2002) states that nonaudit fees have a positive association with magnitude of discretionary accruals; on the contrary, different conclusions have been reached. Ashbaugh et al. (2003) find no proof offering a support the positive and significant association between nonaudit fees and absolute discretionary accrual; at the same time. Ashbaugh et al.(2003) argue against Franker et al. (2002) that their study results are sensitive to research design and subject to weakness of research methodology. Consequently, Ashbaugh et al. (2003) come to a conclusion that purchase of nonaudit services do not

(30)

30

impair independence of auditors in audit engagement and also do not influence absolute discretionary accruals.

The coefficient of D100 for ADA1 is negative (-0.011) and not significant with a P value 0.503, suggesting that the association auditor and headquarters is not significant. Referring back to prior studies, the coefficient of proximity proxy between auditors and clients were negative and significant (Choi et al. 2012, Jensen et al.2013, Liu Wenjun et al. 2014), indicating that geographic proximity between auditors and clients results in less absolute discretionary accruals. In my research, the insignificant p-value could be explained by my small samples. In my research, all data in my study was on a two years’ basis, where Jensen et al. (2013) was based on seven years’ observation and Choi et al (2012) based on 4 years’ observations. Smaller year range and volume of observation could explain the insignificance of D100. Coefficient of D100 for ADA2 is 0.25 at insignificant level (p-value = 0.370), meaning that association between D100 and absolute discretionary accruals is not significant and negativity is not same to the coefficient of ADA1, the difference is caused by the limited volume of data and way of measurement accruals.

The variable change of sales (CHGSALES) is used to control the potential association between discretionary and sales change. Change of sales is used to proxy for growing speed of firms (Hribar and Craig 2007, Choi et al. 2012, etc.). Hribar and Craig (2007) companies that have better growth prospects are likely to report a higher level of absolute discretionary accruals. Besides, other studies also document that changing sales performance provides fewer incentives for corporate management to manipulate discretionary accruals (Dechouw et al. 1995; Frankel et al 2002; Kothari et al.2002), but when it comes to my research finding, there is a different result. Coefficients of change of sales (CHGSALES) for ADA1 and ADA2 are -0.065 with a p-value 0.006, and the other -0.019 with a p-value = 0.634, respectively. The difference and the inconsistence of my empirical finding can be explained by the short of observation year, thus, change of sales in only two years cannot possibly and accurately capture the sales volatility and growing prospect of observations, whereas, Hribar and Craig (2007) used their observation at least 6 years, Choi et al. (2012) used 4 years periods, etc.

LOSS is an indicator that when firms report a loss in their financial reporting; it controls for the impact of loss and profit observations and considers the influence of possible big bath actions in loss years (Choi et al. 2007). As can be found in table 3 and appendix 1.3 and appendix 2.3, control variable LOSS has a positive coefficient 0.052 with a significance value (p-value= 0.012) for ADA1, and a positive coefficient 0.134 at a significant level (p-value = 0.000) for ADA2, which means that

Geographic distances between corporate headquarters and their subsidiaries : how do they affect audit quality?