以分量迴歸檢視影響資金成本的因素

(1)

行政院國家科學委員會專題研究計畫成果報告

以分量迴歸檢視影響資金成本的因素(第 3 年)

研究成果報告(完整版)

計畫類別：個別型計畫編號： NSC 96-2416-H-004-035-MY3 執行期間： 98 年 08 月 01 日至 99 年 07 月 31 日執行單位：國立政治大學會計學系計畫主持人：戚務君計畫參與人員：碩士班研究生-兼任助理人員：顏立婷碩士班研究生-兼任助理人員：林育正博士班研究生-兼任助理人員：劉玳縈報告附件：出席國際會議研究心得報告及發表論文處理方式：本計畫涉及專利或其他智慧財產權，2 年後可公開查詢

中華民國 99 年 12 月 21 日

(2)

Accounting conservatism in a setting of Information

Asymmetry between majority and

minority shareholders

Wuchun Chi

a,

⁎

, Chenchin Wang

b

a_{Department of Accounting, National Chengchi University, Taipei, Taiwan} b_{Department of Accounting, Tamkang University, Taipei, Taiwan}

Received 7 May 2008

Abstract

Following LaFond and Watts (2008), we examine the relation between information asymmetry (as measured by PIN, probability of information-based trading) and accounting conservatism but focus on a country – Taiwan – whose institutional background is different from that of the United States. Due to the disparate degree of conservatism across the world, the conclusions of,LaFond & Watts (2008) might not be universally applicable. Our findings support, in general, the applicability of their conclusion to a Taiwan data set. We find, however, that the effect of PIN appears weaker when auditor tenure is taken into account, thus supplementing their conclusions.

Keywords: Accounting conservatism; Information asymmetry; PIN (Probability of Information based on Trading)

1. Introduction

The recent Conceptual Framework for Financial Reporting project, a joint project of the International Accounting Standards Board (IASB) and the U.S. Financial Accounting Standards Board (FASB), concludes that conservatism (or prudence) should be excluded from the qualitative characteristics of accounting information. This conclusion motivates us to examine the informational role of conservatism (detailed in the literature review section),1the demand for

The International Journal of Accounting 45 (2010) 465–489

⁎ Corresponding author. Tel.: +886 2 29393091x81031. E-mail address:wchi@nccu.edu.tw(W. Chi). 1

Regarding the joint project and the debate of the two boards, refer toGassen (2008). In addition,Ball, Kothari, and Nikolaev (2009)provide a fairly comprehensive list of papers on conservatism. Besides, as we finish this paper, the joint project of IASB and FASB is still ongoing. All materials can be downloaded athttp://www.fasb. org/project/conceptual_framework.shtml.

(3)

Author's personal copy

and measurement of which have long been examined and explained by researchers (e.g.,Basu, 1997; Roychowdhury & Watts, 2007). If carried out, this exclusion will affect the development of future accounting standards. Therefore, this study examines the impact of conditional conservatism on shareholders, focusing mainly on private information and auditor tenure.

Two important features of conservative accounting are asymmetric timeliness in recognition of accounting gains versus losses and systematic understatement of net assets (Givoly, Hayn & Natarajan, 2007; Roychowdhury & Watts, 2007), features that result from asymmetries in information and loss functions among the contracting parties and the inability of the uninformed to verify the information provided by the informed (Watts, 2003a, 2003b). There is a need for conservatism (i.e., prudence) in financial accounting and reporting because, in the face of the uncertainty of business and economic activities, prudence demands a higher degree of verification for the establishment of recognized assets than of liabilities. Accounting convention should encourage such prudence.

LaFond and Watts (2008), using a U.S. data set, find evidence that information asym-metry is positively related to accounting conservatism and, more importantly, that changes in information asymmetry affect conservatism rather than conservatism driving information asymmetry. They conclude that the roles of accounting conservatism not only relate to debt and compensation contracts, but also cover equity investors. However, the disparate degree of accounting conservatism around the world calls for an investigation into the role of conservatism in different countries.

Three reasons justify why this study follows the method used by LaFond and Watts (2008)and why it uses Taiwanese data to reexamine how information asymmetry is related to conservative accounting. First, it is well known that the degree of accounting conservatism differs across countries and according to preparers' incentives (Ball, Kothari, & Robin, 2000; Ball, Robin, & Wu, 2003; Ball & Shivakumar, 2005). Differences in institutional settings may invalidate the conclusions ofLaFond and Watts (2008), which are based only on U.S. data, so research like the present study is needed before their conclusions can be considered generally valid. Thus, it is worth investigating the relation between information asymmetry and accounting conservatism beyond the U.S. data. We will discuss how specific institutional differences may affect the information role of accounting conservatism found inLaFond and Watts (2008)in the next section. Second, the size of the Taiwan stock market, which was the world's twentieth largest during our research period, makes it a significant example of a code-law regime within which to test the information role of conservatism.2 Third, this study further investigates the role of auditor tenure. Note that audit-firm tenure (e.g.,Gul, Fung, & Jaggi, 2009; Myers, Myers, & Omer, 2003; Stanley & DeZoort, 2007) has been commonly used in auditing research. However, to the best of our knowledge, no studies examine how auditor tenure affects the relation between information asymmetry and accounting conservatism discussed above. Thus, we further incorporate this variable into our analysis. Our findings reveal that auditor tenure affects the relation between information asymmetry and accounting conservatism. Specifically, for a case of good news the relation is weaker when the auditor tenure of the

2 _{According to the World Federation of Exchanges (}_{http://www.world-exchanges.org/files/statistics/}_{), during} our research (2000–2008), the rank of market capitalization of shares of domestic companies in Taiwan ranged from 13th (2000, 2001, and 2003) to 20th (2007).

(4)

company is short; for a case of bad news, however, the relation is weaker when the auditor tenure of the company is long.

Our empirical evidence suggests that conservatism mitigates information asymmetry. Specifically, the level of accounting conservatism is positively related to the level of information asymmetry, and information asymmetry in the current period will further drive an increase of conservatism in the next period. Thus, our findings question the current proposal of IASB and FASB that such conservatism should be excluded from the qualitative characteristics of accounting information. We are concerned that more serious academic input is needed before a final decision on such an exclusion of conservatism can be made. Such a decision should take into account the findings of this paper, which conclude that the necessity for accounting conservatism is greater for firms whose auditors have longer tenure. Note that this study does not suggest that the standard setters, either FASB or IASB, will be influenced by evidence that conservatism is useful.3Instead, it suggests that if conservatism is excluded from the qualitative characteristics of accounting information, shareholders should keep a more careful eye on the issue of information asymmetry. Our findings further suggest that, especially in cases of longer auditor tenure, conservatism should be encouraged, not excluded. Even if standard setters such as the FASB and IASB continue to ignore relevant research, shareholders can still keep our results in mind: at the very least, this paper suggests that if conservatism is excluded from the qualitative characteristics of accounting information, shareholders should closely monitor information asymmetry.

The remainder of this paper is organized as follows. Section 2 provides a review of relevant literature on conservative accounting and information asymmetry. Section 3 describes the research design. Section 4 presents the descriptive statistics and reports the empirical findings. Our summary and conclusions are given in Section 5.

2. Literature review 2.1. The joint project

Currently, the IASB uses the Framework for the Preparation and Presentation of Financial Statements (2001) and the FASB uses the Statements of Financial Accounting Concepts (1978, 1980a, 1980b, 1984, 1985, 2000, 2005, 2006), each of which forms the foundation on which the respective Board (collectively referred to as the Boards, hereafter) sets consistent standards. Although the existing frameworks of the FASB and the IASB contain conservatism (or prudence), their positions have been shifting toward excluding conservatism from the qualitative characteristics of accounting information.

3 _{There is little empirical evidence that standard setters care about the usefulness of accounting information. For} instance, the Securities and Exchange Commission (as well as similar institutions) was created to prevent future market crashes, not to improve contracting or investment efficiency. Furthermore, in a passage that some will no doubt find objectionable,Schumpeter (1949, pp. 262–263 )points out that“the typical citizen drops down to a lower level of mental performance as soon as he enters the political field. He argues and analyzes in a way which he would readily recognize as infantile within the sphere of his real interests. He becomes a primitive again. His thinking becomes associative.” We are grateful to an anonymous reviewer for reminding us of this point.

(5)

Author's personal copy

On July 6, 2006, the Boards published for public comment an IASB Discussion Paper/ FASB Preliminary Views.4In it, the Boards revealed some decisions relevant to this paper. First, qualitative relevance and faithful representation are fundamental qualitative characteristics that must be present for information to be useful. Second, faithful representation is attained when the substance of an economic phenomenon is depicted completely, accurately, and neutrally. Finally, conservatism/prudence should be excluded from the qualitative characteristics of accounting information, because there is a clash between conservatism and neutrality.5These decisions of the Boards will become final only after extensive due process; the Boards originally planned to issue a Final Draft on Objectives and Qualitative Characteristics in the fourth quarter of 2009, but as of this writing (2010) additional deliberations have been called for, and no Final Draft has yet been issued. According toChristensen and Demski (2004), the overall purpose of external financial accounting– to provide decision-useful information to market participants – is specified by two sub-objectives: to provide valuation-relevant and contract-relevant information. Gassen (2008)notes that these objectives are implied in Sections 34 (decision usefulness), 37 (valuation usefulness), and 50 (contracting usefulness) of Statements of Financial Accounting Concepts, No. 1.6 Gassen (2008) also points out that the joint project of the Boards implicitly narrows the focus of decision usefulness to valuation usefulness and suggests that the Boards consider explicitly stating that the overall objective of decision usefulness includes contract-relevance (stewardship), because the two sub-objectives – valuation reporting and stewardship – are incompatible alternatives.

2.2. Accounting conservatism

Conservatism has been an important accounting principle for centuries (seeBasu, 1997; Watts & Zimmerman, 1986).Watts (2003a, b)explains that the demand for conservatism results from the contracting role of accounting. Two important reporting features of conservative accounting are asymmetric timeliness in recognition of accounting gains versus losses (conditional conservatism) and systematic understatement of net assets (uncondi-tional conservatism) (Givoly et al., 2007; Roychowdhury & Watts, 2007). These two features result from accountants' predisposition to require a higher degree of verification and certainty for the establishment of assets as compared with liabilities (Watts, 2003a).

Conditional conservatism differs from unconditional conservatism in that it is news (or ex-post) dependent (Beaver & Ryan, 2005). That is, book values are written down under

4 _{It is entitled} _{“Conceptual Framework for Financial Reporting: Objective of Financial Reporting and} Qualitative Characteristics of Decision-Useful Financial Reporting Information. Preliminary Views.”

5 _{The term conservatism appears in Statements of Financial Accounting Concepts No. 2 (1980a) and Concepts} Statement No. 6 (1985) of FASB. The term prudence is shown in the IASB's framework (2001). The Preliminary Views on an improved Conceptual Framework for Financial Reporting (refer to footnote 5) repeats the point that “There is a place for a convention such as conservatism – meaning prudence – in financial accounting and reporting, because business and economic activities are surrounded by uncertainty, but it needs to be applied with care.”

6 _{Statements of Financial Accounting Concepts No. 1 (SFAC) specifies (a)}_{“information that is useful to present} and potential investors and creditors and other users in making rational investment, credit, and similar decisions” (SFAC 1.34)”; (b) “information to help […] users in assessing the amounts, timing, and uncertainty of prospective cash receipts (SFAC 1.37)”; (c) “information about how management of an enterprise has discharged its stewardship responsibility to users (stockholders) for the use of enterprise resources entrusted to it (SFAC 1.50).” 468 W. Chi, C. Wang / The International Journal of Accounting 45 (2010) 465–489

(6)

sufficiently adverse circumstances but not written up under favorable circumstances. In addition, Ball and Shivakumar (2005)point out that conditional conservatism can enhance contracting efficiency, while unconditional conservatism seems inefficient or at best neutral in contracting. Therefore, conditional conservatism – timely loss recognition – has a useful earnings attribute from the viewpoint of the contractors of the firms. As for empirical work that examines unconditional conservatism,Ahmed and Duellman (2007)use book-to-market ratio and average accruals and find that the percentage of inside directors is negatively related to conservatism, while the percentage of outside directors' shareholdings is positively related to conservatism.7On the other hand, using measures of conditional conservatism,LaFond and Watts (2008)find evidence that accounting conservatism reduces information asymmetry.8

To measure conditional conservatism, Basu (1997)focuses on how economic income, measured by market returns, is asymmetrically associated with accounting earnings.9 He also finds that the litigation climate affects the degree of conditional conservatism. Since the publication of Basu's paper, many researchers have begun to pay attention to the incremental coefficient of negative returns (i.e., bad news) as the asymmetric timeliness coefficient— that is, the measure of conservatism in accounting. To demonstrate the validity of this model, Pope and Walker (1999), Ryan and Zarowin (2003), Beaver and Ryan (2005), and Ball, Kothari and Nikolaev (2009)analyze the econometrics of the Basu asymmetric timeliness coefficient through a model in which stock prices lead accounting earnings.

There are many studies that investigate conservative accounting. Basu (1997), Givoly and Hayn (2000), Holthausen and Watts (2001), and Ryan and Zarowin (2003)analyze the change in conditional conservatism over time.Pope and Walker (1999), Ball et al. (2000), Giner and Rees (2001), Ball et al. (2003), and Bushman and Piotroski (2006) make international comparisons of the degrees of conservatism across different accounting systems. Other papers that study conditional conservatism include Krishnan (2005a, b), Huijgen and Lubberink (2005), and Sivakumar and Waymire (2003). Recently, a great deal of research examines how corporate governance affects reporting conservatism (e.g., Ahmed & Duellman, 2007; LaFond & Roychowdhury, 2008). In addition, Zhang (2008) shows that conditional conservatism benefits lenders ex-post through the timely signaling of default risk, as manifested by accelerated covenant violations, and that it benefits borrowers ex-ante through lower interest rates.10

7

Other papers that use measures of unconditional conservatism includeAhmed, Morton, and Schaefer (2000), Beaver and Ryan (2000), Feltham and Ohlson (1996), Givoly and Hayn (2000), Penman and Zhang (2002), and Zhang (2000).

8 _{Other papers that use measures of conditional conservatism include}_{Ball et al. (2000), Ball and Shivakumar}

(2005), Chi, Liu, and Wang (2009), Giner and Rees (2001), Pope and Walker (1999), and Zhang (2008). 9 _{Ball and Shivakumar (2005) and Beaver and Ryan (2005)} _{describe it as a measure of conditional} conservatism, because of the emphasis on asymmetric timeliness in the Basu model.Basu (1995, 1997)introduces several other measures, but the market-based measure is used most often.

10 _{Ball et al. (2009)}_{list papers that investigate conditional conservatism under the following headings: (i) across} countries and legal regimes, (ii) over time, (iii) across fiscal quarters, (iv) a function of litigation, auditors' legal liability, earnings management or auditor malfeasance, (v) a function of the different demands on financial reporting of public companies and family ownership, (vi) the association between conditional conservatism and the equity cost of capital, (vii) the role of accounting accruals in conservatism, (viii) the relation between conditional conservatism and the debt, and (ix) in the context of the management compensation and corporate governance.

(7)

Author's personal copy

While conventional wisdom explains that agency costs generated by managers' private information and asymmetric loss function mainly from the viewpoints of debt and compensation, LaFond and Watts (2008) provide evidence that the level of information asymmetry between inside and outside investors is positively related to the degree of conservatism and that information asymmetry leads conservatism. LaFond and Watts (2008)explain why conservatism responds promptly– in the same year or the next year – to changes in information asymmetry. Specifically, they point out that information asymmetry between inside and outside investors not only generates agency costs that reduce the future cash flows of firms, but also increases the required rate of return on the stock. Since information asymmetry between security traders will reduce the firm value, the practice of fulfilling demand for credible information, as conservatism has traditionally done, is potentially cost effective. LaFond and Watts (2008) explain how accounting conservatism reduces information asymmetry. First, it provides the best possible non-stock price “hard” information on current performance for uninformed investors. Second, that hard information provides a benchmark that makes it possible for alternative“soft” sources to generate credible information on unverifiable gains.

LikeLaFond and Watts (2008), we also use unconditional probability of information-based trading (hereafter referred to as PIN) in this study to measure the level of information asymmetry. The empirical findings of LaFond and Watts (2008) on how PIN affects conservatism are as follows. A greater information asymmetry between insiders and uninformed equity investors results in (1) lower gains and greater losses and (2) larger asymmetric recognition of gains and losses as reflected in current financial statements. In addition, information asymmetry leads to conservatism.

2.3. Why this study uses Taiwanese data

This subsection discusses three points. First, we survey the literature on institutional settings. Next, we provide the summary of the institutional differences between the United States and Taiwan. Finally, we explain why these institutional differences might affect the information role of accounting conservatism found in LaFond and Watts (2008). We believe that if such a role, which has been evidenced in common-law countries with strong investor protection, can be extended to a code-law country with relatively weak investor protection, then the application of their conclusions to countries with different legal regimes and information environments is more justifiable.

The empirical evidence shows that the degree of accounting conservatism differs across countries and according to preparers' incentives (Ball & Shivakumar, 2005; Ball et al., 2000; Ball et al., 2003). Differences in institutional settings may invalidate the conclusions of LaFond and Watts (2008), which are based only on U.S. data, so research such as the present study is needed before their conclusions can be considered generally valid. Although Basu, Huang, Mitsudome, and Weintrop (2005) find that Taiwanese firms' earnings exhibit conservatism, the question of how these earnings attribute is affected by or affects information asymmetry outside the United States is not answered in prior literature. Thus, it is worth investigating the relation between information asymmetry and accounting conservatism beyond the U.S. data.

(8)

The striking differences between the U.S. system and the Taiwan system make Taiwanese data an excellent candidate for such an investigation. In particular, the United States employs a common-law regime and a unitary-board system, while Taiwan employs a civil-law (i.e., code-law) regime and a two-tier board system.11 These institutional differences may well affect the quality of financial reporting, which, as some researchers have found, is determined by factors such as legal systems (Ball et al., 2003; Leuz, 2003; Schipper, 2005). Prior research has also suggested that civil-law and common-law countries differ in financial-reporting transparency (e.g., Bushman, Chen, Engel, & Smith, 2004; Leuz, 2003). Specifically, in civil-law countries, the demand for accounting earnings is influenced more by the payout preferences of capital and government and less by the demand for public disclosure. As a result, accounting standards under this system tend to give greater discretion to managers, who may therefore vary the application of accounting standards to achieve intended results. Likewise, investor protection in civil-law countries has been found to be weak (Ball et al., 2000; La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 1998). In contrast, in common-law countries, protection for investors is assumed to be stronger, implying greater transparency in financial reporting under this system.

In addition to the protection for investors, the soundness of corporate governance also affects earnings quality and the degree of information asymmetry. In the United States, which uses a unitary-board system, the functioning of board and of management is separate: large shareholders monitor the management team, and independent board members enhance the monitoring role of the board. In contrast, Taiwanese firms use a two-tier board system in which, according to the Company Act, the supervisory tier takes responsibility for supervision, while board members take charge of the overall strategy of the firm. However, like board members, members of the supervisory tier are commonly large shareholders or related parties; thus, the supervisory tier's monitoring role in Taiwan is weaker than that of the board in the United States. In addition, the power of the supervisory tier in Taiwan is lower, and hence its monitoring role is weaker, than those of its counterparts in Germany and Japan (Yeh 2005; Yeh, Lee, & Woidtke, 2001). If our study supports the conclusions of LaFond and Watts (2008) about the information role of accounting conservatism even under such relatively adverse conditions, the application of their conclusions to countries with different legal regimes and information environments is more justifiable.

Besides, compared with U.S. firms, Taiwanese firms are owned through closely held shares, so the interests of managers and majority shareholders are more likely to be aligned, but the problem of asymmetry exists between the controlling parties (i.e., majority shareholders) and outsiders (i.e., minority shareholders). For example,Claessens, Djankov, and Lang (2000) point out that in Taiwan the alignment of ownership and control is rather pronounced in family-controlled firms and small firms. It is normal in Taiwanese companies for the ownership of shares to be highly concentrated and for board members to

11 _{Under both unitary-board and two-tier board systems, governance structures have a supervisory function and a} managerial function, but the distinction between them is more formalized in a two-tier system. Using data on European countries, some of which use a unitary-board system and others of which use a two-tier board system,

Dargenidou, McLeay, and Raonic (2007)show that accounting behavior differs among the countries. In Taiwan, a two-tier system, the supervisory tier, not board members, take sole responsibility for the supervision.

(9)

Author's personal copy

be the managers themselves or to be selected by those managers, who in turn are controlled by the large shareholders (Yeh et al., 2001). In other words, the ownership structure in Taiwan is whatLa Porta et al. (1998)call“controlling-ownership.” Prior research points out that entrenchment is more serious when the ownership structure is dispersed among minority shareholders (i.e., there is no additional large blockholder to offset the power of the controlling owner-shareholder), and information asymmetry between the two types of shareholders is more severe (e.g.,Claessens et al., 2000, Claessens, Djankov, Fan, & Lang, 2002; La Porta, Lopez-de-Silanes, Shleifer, & Vishny, 2002; Fan & Wong 2002). In addition, Fan and Wong (2002) and Leuz (2003)find that controlling shareholders have discretion over the policy of financial reporting and disclosures.

Due to these institutional differences, we need further empirical evidence to examine the impact of information asymmetry and conservatism in cases where such asymmetry exists not between management and shareholders (as studied by LaFond & Watts, 2008), but between majority and minority shareholders. Therefore, the conclusion of LaFond and Watts (2008, page 48) that “conservatism is an equilibrium response to mitigate value reductions resulting from information asymmetries between managers and outside equity investors” still needs to be extended to a majority-versus-minority shareholders situation. This study does exactly that and finds that, when applied to Taiwanese data, the conclusions reached byLaFond and Watts (2008)are strongly supported when firms are faced with bad news.

Several unique capital market characteristics offer a further reason to use a research sample from Taiwan. According toBaber, Lee, Liu, and Odean (2009), the average annual stock turnover rate on the Taiwan Stock Exchange Corporation of almost 300% is remarkably higher than the 97% on the New York Stock Exchange. In addition, they also point out that individual investors account for roughly 90% of all trading volume in Taiwan and that stocks are broadly held in Taiwan and constitute an important asset class for many Taiwanese households. Thus, the question of whether accounting conservatism alleviates information asymmetry is of wide interest in Taiwan. In addition, the size of the Taiwan stock market, which was the world's twentieth largest during our research period, makes it a significant example of a code-law regime within which to test the information role of conservatism.12

2.4. Information-based trading

Easley, Kiefer, O'Hara, and Paperman (1996), building on Easley and O'Hara (1987, 1992), develop and estimate the PIN score, which is a proxy for the degree of information asymmetry. Specifically, firms with bigger PIN scores represent a higher unconditional probability of information-based trading. The formula for the calculation of PIN provided byEasley, Engle, O'Hara, and Wu (2008) includes estimates of the following parameters: (1) the probability of information events; (2) the probability of different asset valuations from buyers and sellers given the occurrence of an information event; (3) the arrival rate of

12 _{According to the World Federation of Exchanges (}_{http://www.world-exchanges.org/files/statistics/}_{), during} our research (2000–2008), the rank of market capitalization of shares of domestic companies in Taiwan ranged from 13th (2000, 2001, and 2003) to 20th (2007).

(10)

uninformed buyers and uninformed sellers; and (4) the arrival rate of informed traders when an information event takes place. To reduce the potential convergence problem of numerical maximization,Easley et al. (2008) suggest a rearranged log-likelihood function and a PIN approximation, which can be easily calculated.

One of the benefits of using PIN score is that it is a direct measure of information asymmetry rather than an indirect one such as spread-based proxies (Leuz & Verrecchia, 2000; Welker, 1995). This type of direct measure has proved its usefulness in a number of studies.13 To simplify the calculation of PIN, we follow the approach proposed byAktas, Bodt, Declerck, and Oppens (2007), who study the PIN anomaly around mergers and acquisitions announcements in the French stock market, to compute PIN.

To calculate PIN, each trade must be classified as either buyer-initiated or seller-initiated. A transaction is buyer-initiated if its price is higher than the mid-quote (i.e., the average of the corresponding best-selling and buying limit orders). If the price is lower than the mid-quote, then the transaction is seller-initiated. The data come from Taiwan Economics Journal intraday database. In addition, we adopt the PIN approximation suggested by Easley et al. (2008). This PIN approximation is easily implemented and less subject to the potential convergence problem of numerical maximization. The latter advantage is important because it removes the need to further reduce the sample size to avoid problems resulting from numerical maximization (Aktas et al., 2007). The annual PIN approximation (hereafter also designated PIN for the sake of simplicity) is calculated by:

PIN = E½jB−S j

E½jB + S j ð1Þ

where E is an expectation operator, B is the number of buyers, and S is the number of sellers. For each firm, we first calculate daily PIN and then compute their mean on a yearly basis.

2.5. Auditor tenure

The information role of auditors is to alleviate potential conflicts of interests between owners, managers, and other security holders.14 High audit quality reduces the cost of capital by contributing to the credibility of financial disclosure (Jensen & Meckling, 1976; Watts, 1977; Watts & Zimmerman, 1986). Such credibility reduces the cost of contracting with the firm for all stakeholders and potential stakeholders. We are interested in whether auditor tenure affects the relation between PIN and accounting conservatism, because the Sarbanes-Oxley Act (SOX) of 2002 mandates auditor partner rotation (Section 203 of SOX) and requires the General Accounting Office to study the issue on audit firm rotation.

13 _{Easley, Hvidkjaer, and O'Hara (2004)}_{, for example, provide evidence that the level of PIN score is positively} associated with the firms' cost of capital.Brown, Hillegeist, and Lo (2004)further point out that the number of conference calls held in one quarter is negatively related to PIN score during the subsequent year. Ellul and Pagano (2006)find that IPO underpricing and PIN are related.

14 _{Two valuable roles of auditing to the users of financial statements are an information role and an insurance} role (Dye, 1993).

(11)

Author's personal copy

Many papers on auditing have examined the relation between auditor tenure and earning quality (e.g., Carcello & Nagy, 2004; Ghosh & Moon, 2005; Jenkins & Velury, 2008; Johnson, Khurana, & Reynolds, 2002; Mansi, Maxwell, & Miller, 2004; Myers, Myers and Omer, 2003; Stanley & DeZoort, 2007).

According to the auditor expertise perspective, longer auditor tenure allows an auditor to accumulate client-specific experience, which is an important factor in enhancing audit quality. This perspective assumes that auditor competence is positively related to an accumulated client-auditor relationship. Against this perspective is the argument that long auditor tenure reduces perceived audit quality due to the increased possibility of a lack of independence of the auditor from the client. This viewpoint holds that a new auditor– with a fresh and skeptical eye – is a critical factor to enhance audit quality.

Prior studies have investigated whether long auditor tenure leads to complacency over time and thus to a reduction in audit quality or whether it increases quality through more comprehensive client-specific knowledge. Numerous papers have investigated the relation between audit firm tenure and earnings quality (e.g., Carcello & Nagy, 2004; Ghosh & Moon, 2005; Jenkins & Velury, 2008; Johnson et al., 2002; Mansi et al., 2004; Myers et al., 2003; Stanley & DeZoort, 2007). We extend this line of research by examining whether the relation between PIN and accounting conservatism is a function of auditor tenure, since, to our best knowledge, no prior studies have analyzed such a question.

Many studies focusing on auditor tenure consider a U-shaped (or inverse U-shaped) relation between auditor tenure and the variables of interest. For example, in their investigation of the relation between audit firm tenure and absolute discretionary accruals,Johnson et al. (2002) classify audit firm tenure into three categories: short (two to three years), medium (four to eight years), and long (nine or more years). Using medium tenure as a benchmark, they find that short tenure is associated with larger absolute discretionary accruals, but long tenure is not— a finding that suggests that long audit firm tenures are not associated with a decline in earnings quality. In addition, Boone, Khurana, and Raman (2008)find a U-shaped relation between auditor tenure and the equity risk premium. Therefore, we analyze not only whether auditor tenure affects the relation between PIN and accounting conservatism, but also whether such a relation is non-linear.

3. Basic empirical model

3.1. brief description of the LaFond and Watts (2008) model — the level of PIN

Asymmetric timeliness in recognition of accounting gains and losses and systematic understatement of net assets are two important features of conservative accounting (Givoly et al., 2007; Roychowdhury & Watts, 2007). The Basu-based model focuses on the former, whereas market-to-book ratio centers on the latter. This study examines the relation between PIN and accounting conservatism, as do LaFond and Watts (2008), using the Basu-based coefficient.

LaFond and Watts (2008)find a positive correlation between concurrent asymmetries in information and accounting conservatism. In addition, they further point out that both prior and current changes in information asymmetries affect current accounting conservatism.

(12)

They conclude that stronger accounting conservatism is a reaction to an environment with more severe asymmetries in information, not a practice that creates information asymmetry. LaFond and Watts (2008) use the following two regressions to support their findings. Eq (2) is run for a good news sample and a bad news sample (classified by positive or negative stock returns) separately, and Eq.(3) is a combined model.15

NIt = a0 + a1RETt + a2PINt + a3PINt⋅RETt + et ð2Þ

NIt = β0 + β1DRt + β2RETt + β3RETt⋅DRt + β4PINt + β5PINt⋅DRt

+ β6PINt⋅RETt + β7PINt⋅RETt⋅DRt + ut ð3Þ

where NI, PIN, and RET (subscripts for firm are omitted) denote earnings, probability of informed trading, and stock return, respectively; the dummy variable DR equals one (bad news) when RET is negative, and zero (good news) otherwise.16

Regarding Eq.(2), we focus only on the coefficient a3, which shows how PIN affects the

relation between accounting earnings and stocks returns. We cannot predict the sign of a1

and a2due to a lack of theoretical foundation for them if we use the original value of PIN.17

To provide an economic meaning for the a1when running our regression model, the variable

PIN has already been centered at its mean. In addition, since LaFond and Watts (2008) explain and find that the lower gains (greater losses) are reflected in current financial statements for a good (bad) news sample when the level of PIN is greater, the coefficient a3

in Eq. (2) is expected to have a negative (positive) sign on a positive (negative) return sample.

Eq. (3) analyzes a combined sample, through the dummy DR, to examine how PIN affects the asymmetric recognition of gains versus losses. In the case of good news (DR equals zero), the coefficientβ6represents the marginal effect of PIN on the reverse relation

of return-and-earnings. The coefficientβ7shows how PIN impacts that relation when firms

are faced with bad news. According to LaFond and Watts (2008), the higher the PIN, the lower the gains and the greater the losses that are recognized. Therefore, we expect a negative β6and a positive β7.

In the case of bad news (DR equals one), the coefficient β3is used to measure the

difference in sensitivity of earnings to negative and positive returns (Basu, 1997). We

15 _Although_{Roychowdhury and Watts (2007)}_{show that asymmetric timeliness regression appears to measure} conservatism more efficiently when it is estimated cumulatively over multiple years, this study only uses a one-year return because the data needed to estimate PIN was not made publicly available in Taiwan until 2000.

16

Note that there are algebraic relations between the α-coefficients (in Eq. (2)) and their corresponding β -coefficients (in Eq.(3)). For example,β0equals a0when the good news sample is run, andβ0+β1equals a0 when the bad news sample is run.

17 _{Taking the first derivative of Eq.}₍₂₎_{with respect to RET, the result will be:} dNI= dRET = a1 + a3PIN

Apparently, the meaning of the coefficient a1is the incremental effect of RET on NI for observations in which PIN equals zero. However, a firm's PIN can neither theoretically nor empirically be zero (actually it must be greater than zero). Thus, we cannot predict the sign of the estimated coefficient of RET, a1. If the PIN is a mean-center value, however, the meaning of coefficient a1becomes the incremental effect of RET on NI for observations in which PIN equals its mean. As for a2, no accounting theory predicts the relation between earnings and the level of PIN. In fact, our empirical results do not suggest any significance of estimated a2.

(13)

Author's personal copy

expect this larger asymmetric recognition will be greater as PIN increases (LaFond & Watts, 2008). Thus, the sign of the coefficient β7in Eq. (3) will be positive. In addition,

several control variables (CVj) are added into Eq. (3) to form Eq. (4) as a further check.

Specifically, to test the robustness of the relation between information asymmetry and conservatism, based on Roychowdhury and Watts (2007), we further control for factors related to other demands for conservatism by using leverage (Leverage) and market-to-book ratio (MTB) as our control variables.18

NIt =β0 + β1DRt + β2RETt + β3RETt⋅DRt + β4PINt + β5PINt⋅DRt

+ β6PINt⋅RETt + β7PINt⋅RETt⋅DRt + ∑ J j = 1 γ jCVj + ∑ J j = 1 KjCVj⋅DRt + ∑ J j = 1 λ jCVj⋅RETt + ∑ J j = 1 ω jCVj⋅RETt⋅DRt + υt ð4Þ

3.2. A brief description of the LaFond and Watts (2008) model — changes in PIN This section discusses whether changes in asymmetries in information (prior, concurrent, and future) affect current accounting conservatism, or vice versa. This difference is very important, because if the former is correct (increases in information asymmetry lead to conservatism), then conservatism likely reduces information asymmetry. On the other hand, if the evidence reveals that increases in information asymmetry follow conservatism, conservatism engenders information asymmetry.

Eq. (5) is the regression model. Differences in subscripts x (−1, 0 or +1) change the focus of Eq. (5), and the coefficient of interest in this equation isδ7 andδ11, the former of

which is used to test how the changes in information asymmetry (for the prior, the current, and the next year) are related to current information asymmetry, and the latter of which examines the level of information asymmetry (for the two years before, the previous year, and the current year).

NIt =δ0 + δ1DRt + δ2RETt + δ3RETt⋅DRt + δ4ΔPINt + x + δ5ΔPINt + x⋅DRt

+ δ6ΔPINt + x⋅RETt + δ7ΔPINt + x⋅RETt⋅DRt + δ8PINt−1 + x

+ δ9PINt− 1 + x⋅DRt + δ10PINt−1 + x⋅RETt + δ11PINt−1 + x⋅RETt⋅DRt + εt

ð5Þ

In the case of x =−1 (hereafter, the lag analysis), following the theory of the information role of conservatism (LaFond & Watts, 2008), the coefficients of ΔPINt + x⋅RETt⋅DRt(δ7,

where x is taken to be −1) and PINt− 1 + x⋅RETt⋅DRt(δ11, where x is taken to be −1) are

both predicted to be positive. In particular, the former (δ7) examines how changes in PIN

over the previous year affect current conservatism, and the latter (δ11) analyzes how the

levels of PIN in the previous two years affect current conservatism.

18 _{Lev and Sunder (1979)} _{discuss problems with negative denominators in accounting ratios. Because our} sample does not include any companies with negative book value, and for reasons of consistency withLaFond and Watts (2008)who use MTB in their regression models, we report the results on MTB.

(14)

The second case is for x = + 0 (hereafter, the concurrent analysis). Again following the model ofLaFond and Watts (2008), the coefficients ofΔPINt + x⋅RETt⋅DRt(δ7, where x is

taken to be 0) and PINt− 1 + x⋅RETt⋅DRt(δ11, where x is taken to be 0) are both predicted to

be positive. In particular, the former (δ7) examines how changes in PIN over the current

year affect current conservatism, and the latter (δ11) analyzes how the levels of PIN in the

previous year affect current conservatism.

The last case is for x = + 1 (hereafter, the forward analysis), where the coefficient of PINt− 1 + x⋅ RETt⋅DRt(δ11, where x is taken to be + 1) is expected to be positive, due to

the positive relation between concurrent PIN and conservatism. However, the coefficient of ΔPINt + x⋅RETt⋅ DRt(δ7, where x is taken to be + 1) should not be significantly different

from zero, becauseLaFond and Watts (2008)demonstrate that conservatism does not drive information asymmetry. In summary, in Eq.(5), a positiveδ7in a forward analysis implies

that conservatism engenders information asymmetry, in a concurrent analysis, it reveals that conservatism and information asymmetries move together during the same period; and in a lag analysis, it reveals that information asymmetry leads conservatism.

3.3. How auditor tenure affects the relation between PIN and conservatism

In our basic model, Eq. (3), there are eight explanatory variables (including the intercept). Because this study focuses not only on how auditor tenure affects the relation between PIN and conservatism but also on whether the effect of tenure on this relation is non-linear or not, we extend our model by classifying auditor tenure into short, medium, and long, including the dummy variables SHORT (indicating that tenure is less than four years) and LONG (indicating that tenure is greater than ten years).19 To examine whether tenure affects the findings of LaFond and Watts (2008), we multiply all PIN-related variables in Eq.(3) by these two variables, SHORT and LONG, to produce Eq. (6):

NIt =β0 +β1DRt +β2RETt+β3RETt⋅DRt +β4PINt +β5PINt⋅DRt +β6PINt⋅RETt

+β7PINt⋅RETt⋅DRt +β8SHORTt⋅PINt+β9SHORTt⋅PINt⋅DRt

+β10SHORTt⋅PINt⋅RETt+β11SHORTt⋅PINt⋅RETt⋅DRt +β12LONGt⋅PINt

+β13LONGt⋅PINt⋅DRtβ14LONGt⋅PINt⋅RETt +β15LONGt⋅PINt⋅RETt⋅DRt

+∑J_{j = 1}γjCVj+∑Jj = 1KjCVj⋅DRt+∑j = 1J λjCVj⋅RETt +∑Jj = 1ωjCVj⋅RETt⋅DRt +υt

ð6Þ A positive (negative) β10 shows that the relation between PIN and conservatism

decreases (increases) with short auditor tenure (i.e., SHORT = 1) when firms are faced with good news. In contrast, a positive (negative) β11reveals that the relation between PIN and

conservatism increases (decreases) with short auditor tenure when firms are faced with bad news. Similarly, a positive (negative) β14 shows that the relation between PIN and

con-servatism decreases (increases) with long auditor tenure (i.e., LONG = 1) when firms are faced with good news. In contrast, a positive (negative)β15reveals that the relation between

PIN and conservatism increases (decreases) with long auditor tenure when firms are faced with bad news.

19 _{Our results are qualitatively similar if the cutoff points of four and ten years are replaced by three and nine} years, respectively.

(15)

Author's personal copy

4. Sample selection and data descriptions

Our sample, obtained from the Taiwan Economics Journal Database (referred to as TEJ database, hereafter), consists of all companies listed on the Taiwan Stock Exchange Corporation (TWSE) and Gre Tai Securities Market (GTSM) from 2000 to 2007, according to year-end data.20

The procedure for selecting samples begins with the identification of 8456 observations for the research period. Next, we delete observations from our sample for the following reasons: (1) 169 firm-year observations in financial-related industries had unique industry characteristics; (2) 29 firm-year observations used non-calendar years; (3) 559 firm-year observations were missing data on net income or on stock returns; (4) 2108 firm-year observations were missing data needed to compute PIN; and (5) 218 firm-year observations were missing data on auditor tenure and the control variables used in this study, firm size, leverage, and beginning-of-year's market-to-book ratio. As Table 1 indicates, the above process leaves 5373 companies from the original sample.

To reduce the potential influence of extreme observations, all the NI and RET related variables are winsorized at 1% and 99% of their respective distributions.

Table 2 provides the mean, median, the first and the third quartiles, and the standard deviation of PIN and all other variables examined in this paper. For reasons explained in footnote 16, we use mean-centered PIN by year for each industry to control for potential industry effects (although this variable is still labeled PIN in our study) as the explanatory variable. InTable 2 we provide two descriptive statistics of PIN–PIN_ original (i.e., non-centered) and PIN (i.e., mean-non-centered). In addition, RET is yearly market-adjusted return. The statistics of PIN_original reveal that our mean (0.114) and median (0.092) are smaller than those reported by LaFond and Watts (2008).21 Like the firms they studied,

Table 1

Sample selection.

Selection model (by firm-years) Number of observations

Listed companies from year 2000 to 2007 8456

Less

Financial-related industries (169)

Non-calendar year (29)

Missing data on net income or on stock return (559)

Missing data while computing PIN (2108)

Missing data of auditor tenure, leverage, and beginning-of-year's market-to-book (218)

Total available data 5373

20 _{The TWSE and GTSM in Taiwan are analogous to the NYSE and NASDAQ in the United States.} 21 _{One possible reason that our descriptive statistics of PIN are smaller than those in}_{LaFond and Watts (2008)} (0.195 and 0.182 inTable 1) is that we use a PIN-approximation formula (see Eq.(1)). In addition, the Taiwan Stock Exchange has set the highest and lowest prices that a stock price is permitted to reach in a given trading session. The daily price limit is +/−7% of the previous day's closing price. This will limit the variation of the stock return and may lower the estimated level of PIN. Therefore, it is possible that the daily price limit in Taiwan affects the usefulness of the PIN measure relative to the older bid-ask spread-based measures. We appreciate an anonymous reviewer's reminder of this point.

(16)

most of our sample firms are profitable and have positive returns, because the central tendencies of net income before extraordinary items, scaled by beginning-of-year's market value of equity (NI, the mean and median are 0.004 and 0.058, respectively) and yearly market-adjusted return (RET, the mean and median are 0.037 and 0.023, respectively), are both positive.22In addition, NI and RET skew to the left and right, respectively, consistent

Table 2

Descriptive statistics.

Panel A: Basic statistics (2000–2007; Obs. =5373)

Variables Mean Median Q1 Q3 Std

PIN −0.001 −0.018 −0.064 0.022 0.102 PIN_original 0.114 0.092 0.048 0.137 0.104 NI 0.004 0.058 0.001 0.105 0.327 RET 0.037 0.023 −0.272 0.340 0.501 Tenure 10.851 10.000 7.000 15.000 5.674 MTB 1.402 1.101 0.732 1.713 1.184 Leverage 0.183 0.122 0.000 0.309 0.195

Panel B: correlation matrix

PIN PIN_original NI RET Tenure MTB Leverage

PIN 0.959 −0.099 −0.084 −0.177 −0.076 −0.011 (0.000) (0.000) (0.000) (0.000) (0.000) (0.426) PIN_original 0.984 −0.089 −0.058 −0.160 −0.097 −0.009 (0.000) (0.000) (0.000) (0.000) (0.000) (0.524) NI −0.049 −0.025 0.347 0.003 0.420 −0.036 (0.000) (0.061) (0.000) (0.828) (0.000) (0.008) RET −0.030 −0.022 0.123 0.038 0.118 −0.045 (0.019) (0.098) (0.000) (0.005) (0.000) (0.001) Tenure −0.119 −0.115 0.055 0.024 −0.203 0.090 (0.000) (0.000) (0.000) (0.065) (0.000) (0.000) MTB −0.089 −0.082 0.122 0.169 −0.168 −0.067 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Leverage 0.010 0.010 0.024 −0.050 0.070 −0.053 (0.445) (0.469) (0.069) (0.000) (0.000) (0.000)

Note: definition of variables.

PIN = the probability of an information based trade at the end of the year, calculated by Eq.(1)and adjusted by its industry average (which results in a negative PIN in the research sample).

PIN_original = the probability of an information-based trade as the end of the year, calculated by Eq.(1). NI = equal to net income before extraordinary items divided by beginning-of-year's market value of equity. RET = the yearly market-adjusted stock return.

Tenure = number of consecutive years since 1983 that the company has retained the audit firm.

MTB = the market-to-book ratio defined as market value of equity divided by book value of equity at the end of the year.

Leverage = total debt divided by total assets at the end of the year.

Upper (lower) triangular contains Spearman (Pearson) correlation coefficients, and numbers in parentheses are two-tailed p-values.

22 _{Our results are qualitatively similar when NI is defined as earnings per share divided by beginning-of-year} stock price.

(17)

Author's personal copy

with prior studies on conservatism (Ball et al., 2000; Basu, 1995; Basu et al., 2005). The mean of auditor tenure (Tenure) is 10.851 years. Finally, the sample firms have an average market-to-book ratio (MTB) of 1.402 and leverage (Leverage) of 0.183, both of which are smaller than the corresponding numbers reported in LaFond and Watts (2008).

Panel B of Table 2 represents the pairwise correlation matrix between the variables, where the upper right portion exhibits Spearman rank-order correlations and the lower left portion exhibits Pearson correlations. The variables that have a significantly negative relations to PIN, both in Pearson and Spearman, are NI, RET, Tenure, and MTB. The signs of all the correlations are qualitatively similar to the corresponding ones in LaFond and Watts (2008), although this study includes the variable Tenure, which their study does not. Finally, NI and RET are positively correlated, suggesting that accounting earnings capture the information of returns.

5. Empirical findings

5.1. The relation between PIN and conservatism

We first examine the relation between PIN scores and accounting conservatism by using Eq.(2), as used byLaFond and Watts (2008), separately for the positive and negative return samples. We predict that PIN will be highly correlated with asymmetric recognition of gains and losses in financial statements. In other words, we predict that the coefficient of PIN⋅RET, a3will be negative when Eq.(2)is used with a positive return sample, and vice

versa. The corresponding result is reported in Panel A ofTable 3.

Next, we report the estimated coefficients of the Basu basic model for comparison with the results of related studies in both Panel A and Panel B of Table 3. The Good News columns show that the estimated coefficient of RET (0.018, p-value = 0.476) is insignificant, but the Bad News columns reveal that the coefficient of RET (0.195, p-valueb0.01) is significantly positive. Thus, for earnings reflected in current financial statements, we find evidence that earnings are more sensitive to bad news than to good news.

Then, we discuss how PIN affects conservatism through the PIN-extended Basu model (i.e., Eq. (3), for the positive return sample (Panel A), then fill in the evidence for the negative return sample (Panel A), and finally address the results of the full sample (Panel B). The Good News columns show that the estimated coefficient of PIN⋅RET, a3(0.006,

p-value = 0.972) is insignificant, but the Bad News columns reveal that the coefficient (0.292, p-valueb0.10) is significantly positive. Thus, for earnings reflected in current financial statements, we find no evidence that greater gains are reported when a firm with higher PIN receives good news, but we find strong evidence that greater losses are reported when it is faced with bad news. Therefore, Panel A shows that the greater the information asymmetry between insiders and uninformed equity investors, the larger the conservatism in current financial statements — but just in cases of bad news.

The “without control variables” (Without CV) column of Panel B also supports our hypothesis that the greater the information asymmetry between insiders and uninformed equity investors, the greater the losses (due toβ7= 0.286, p-valueb0.10) reflected in current

financial statements. To test the robustness of the relation between information asymmetry and conservatism, the “with control variables” (with CV) column further controls for

(18)

factors related to other demands for conservatism, leverage (Leverage), and market-to-book ratio (MTB). In short, the negative β7 (0.239, p-value = 0.168) turns into a one-tailed

significant result, which is still modestly consistent with our expectations. Regarding the coefficient of PIN⋅RET, β6, we do not find a significant result in either the“Without CV”

column (0.006) or the “With CV” column (−0.017).

We further use the empirical findings in the“With CV” column to provide an economic interpretation of the estimated results in this study. Since the variable PIN is mean-centered, the coefficientβ3(0.104) can be interpreted as the traditional Basu coefficient when PIN is

set at its mean. The coefficient β7 (0.239) then reports how “one-unit” increases in PIN

affect the level of the Basu coefficient. According toTable 2, the range of the first and third quartiles (Q1 and Q3) is just 0.086; since we never see an actual“one-unit” increase in PIN,

Table 3

The relation between PIN and conservatism.

Panel A: separate positive and negative return sample NIt= a0+ a1RETt+ a2PINt+ a3PINt⋅RETt+ et

Variable Positive return (good news) Negative return (bad news)

Intercept 0.024 (0.007) 0.029 (0.001) 0.040 (0.000) 0.043 (0.000) RET 0.018 (0.476) 0.033 (0.188) 0.195 (0.000) 0.143 (0.000) PIN −0.082 (0.158) −0.023 (0.606) PIN*RET (a3) 0.006 (0.972) 0.292 (0.051) Adj. R2(%) 0.01 0.23 2.56 4. 57 n 2886 2487

Panel B: combined sample NIt=β0+β1DRTt +β2RETt+β3RETt⋅DRt+β4PINt +β5PINt⋅DRt+

β6PINt⋅RETt+β7PINt⋅RETt⋅DRt+ ∑ J j = 1γ jCVj + ∑ J j = 1 KjCVj⋅DRt + ∑ J j = 1 λ jCVj⋅RETt+ ∑ J j = 1ω jCVj⋅RETt⋅DRt+υt

Variable Basu basic model Without CV With CV

Intercept 0.024 (0.007) 0.029 (0.001) −0.024 (0.174) DR 0.016 (0.000) 0.014 (0.185) −0.004 (0.865) RET 0.018 (0.476) 0.033 (0.188) 0.030 (0.461) RET*DR 0.177 (0.000) 0.110 (0.000) 0.104 (0.046) PIN −0.082 (0.158) −0.053 (0.357) PIN*DR 0.059 (0.606) 0.059 (0.419) PIN*RET (β6) 0.006 (0.972) −0.017 (0.916) PIN*RET*DR(β7) 0.286 (0.051) 0.239 (0.168) Leverage 0.047 (0.230) Leverage*DR 0.018 (0.053) Leverage*RET −0.049 (0.690) Leverage*DR*RET 0.043 (0.789) MTB 0.036 (0.000) MTB *DR 0.009 (0.000) MTB *RET −0.005 (0.640) MTB *DR*RET 0.012 (0.878) Adj. R2(%) 2. 15 2. 79 5. 6 n 5373

Notes: DR, a dummy variable, is equal to one if RET is negative; otherwise, it is equal to zero. Other variables are as defined inTable 2, and reported numbers in parentheses are the two-tailed p-values, which have been adjusted by firm-cluster standard error.

(19)

Author's personal copy

the coefficient β7 itself appears relatively meaningless in economic terms. If we use its

standard deviation (0.102), however, the economic significance of the increase in PIN becomes clearer. Specifically, at its mean, when PIN increases one standard deviation, 0.102, the percentage increase in the Basu coefficient is 23% (i.e., from 0.104,which isβ3,

to 0.239*0.102). Thus, we believe the estimated result for the coefficient of PIN*RET*DR is, despite the small range of PIN, significant in economic terms.

In sum, in Panel B ofTable 3, the insignificant coefficient of the two-way interaction between PIN⋅RET (β6) indicates that the level of information asymmetry between

insiders and uninformed investors does not alter the speed with which good news is incorporated into earnings. However, the positive three-way interaction term, RET⋅DR (β7), indicates that earnings exhibit greater conservatism when the level of PIN score

increases.

LaFond and Watts (2008)demonstrate that earnings contain less information on gains but more information on losses as the information asymmetry of a firm increases. Combining the findings shown inTable 3, we offer evidence from a Taiwanese data set to support the assertion of LaFond and Watts (2008) that information asymmetry and accounting conservatism are correlated, but only in cases of bad news. Of particular significance is that our findings suggest that their conclusion– that, information asymmetry between insiders and outside investors is an important determinant of accounting conservatism in common-law systems – is applicable to a code-law country only when firms are faced with bad news, which is measured as negative returns.

5.2. Effect of PIN changes on conservatism

Table 4presents the estimated results for Eq.(5)using the full sample. The two results of particular interest to this analysis are the coefficients ofδ7andδ11. Based on the empirical

model described in Section 3, we provide the following predictions. For the case of lag analysis, x =−1, we expect the signs of both δ7 (in this case, the effect from the previous

year's changes in PIN) and δ11 (in this case, the effect from PIN of the year before the

previous year) to be positive. For the case of the concurrent analysis, x = 0, we expect the signs of bothδ7(in this case, the effect from the current year's changes in PIN) and δ11(in

this case, the effect from the previous PIN) to be positive. As for the case of the forward analysis, x = 1, we expect the sign ofδ7(in this case, the effect on the next year's changes in

PIN) to be negative, and that of δ11 (in this case, the effect from the current PIN) to be

positive.

Consistent with our expectations, in the x =−1 column, the estimated coefficient δ7 is

significantly positive (1.013, p-valueb0.05), implying that conservatism is related to previous changes in information asymmetry. The estimated coefficient δ11 is also

significantly positive (1.003, p-valueb0.05), implying that PIN score of the year before the previous year is related to current conservatism.

In the concurrent analysis column, the estimated coefficientδ7is significantly positive

(0.743, p-valueb0.10), implying that conservatism is related to current changes in information asymmetry. The estimated coefficientδ11is also significantly positive (1.258,

p-valueb0.01), implying that previous PIN score is related to current conservatism.

(20)

However, the results in the forward analysis column are not fully consistent with our expectations. Specifically, a one-tailed, significantly negative δ7(–0.649, p-value=0.117)

reveals that current conservatism can marginally decrease future information asymmetry. However, the insignificant result forδ11(–0.235, p-value=0.567) reveals that current PIN

score is in fact unrelated to current conservatism.

In sum, Table 4 reveals that contemporaneous increases in information asymmetry are associated with increased conservatism (δ7 of the lag and current analysis) and that an

increased conservatism can reduce the following period's information asymmetry (δ7of the

forward analysis). Regarding how the level of information asymmetry is related to conservatism, the positive coefficient, δ11, in the lag and concurrent analysis reveal that

accounting conservatism is positively related to the previous level of information asymmetry. In other words, a firm's increases in information asymmetry may lead to greater accounting conservatism. Hence, we find that higher levels of information asymmetry between informed and uninformed equity investors are associated with more conservative earnings. The results are consistent with the conjecture of LaFond and Watts (2008) that “conservatism is an equilibrium response by equity market participants to reduce agency cost arising from information asymmetry among equity market participants.”

5.3. Effect of auditor tenure

Table 5 reports how auditor tenure affects the relation between information asymmetry and conservatism. Column 1 includes SHORT-related and LONG-related variables, but

Table 4

A further examination on the effect of PIN and PIN changes on conservatism.

NIt=δ0+δ1DRt+δ2RETt+δ3RETt⋅DRt+δ4ΔPINt + x+δ5ΔPINt + x⋅DRt+δ6ΔPINt + x⋅RETt+δ7ΔPINt + x⋅ RETt⋅DRt+δ8PINt− 1 + x+δ9PINt - 1 + x⋅DRt+δ10PINt− 1 + x⋅RETt+δ11PINt− 1 + x⋅RETt⋅DRt+εt

Variable Lag Concurrent Forward

x =−1 x = 0 x = + 1

Coefficient (p-value) Coefficient (p-value) Coefficient (p-value)

Intercept 0.048 (0.002) 0.065 (0.000) 0.036 (0.007) DRt 0.035 (0.131) −0.024 (0.262) 0.021 (0.311) RETt 0.105 (0.000) 0.051 (0.050) 0.016 (0.548) RETt*DRt −0.008 (0.872) −0.037 (0.447) 0.115 (0.010) ΔPINt + x 0.106 (0.495) −0.271 (0.059) 0.086 (0.510) ΔPINt + x*DRt −0.468 (0.042) 0.424 (0.045) −0.252 (0.193) ΔPINt + x*RETt −0.658 (0.010) 0.184 (0.423) 0.113 (0.655) ΔPINt + x*RETt*DRt(δ7) 1.013 (0.021) 0.743 (0.076) −0.649 (0.117) PINt-1 + x −0.082 (0.578) −0.398 (0.003) −0.009 (0.947) PINt-1 + x*DRt −0.396 (0.081) 0.437 (0.035) −0.158 (0.400) PINt-1 + x*RETt −0.625 (0.007) 0.011 (0.959) 0.076 (0.774) PINt-1 + x*RETt*DRt(δ11) 1.003 (0.023) 1.258 (0.005) −0.235 (0.567) Adj R2(%) 4.0 3.8 2.2 n 3877

All variables are as defined inTable 2 and Table 3, and reported numbers in parentheses are the two-tailed p-values, which have been adjusted by firm-cluster standard error.

(21)

Author's personal copy

excludes the control variables; and column 2 includes both the SHORT-related and LONG-related variables and the control variables. In each of the two columns, the estimated coefficient of RET⋅DR is significantly positive (all p-valuesb0.10). The estimated coefficient of PIN⋅RET⋅DR is also significantly positive in each column (p-valuesb0.05). Our discussion centers on the estimated coefficients of four tenure-related variables (β10,

β11,β14, andβ15,). These four coefficients can be grouped according to whether or not their

variables interact with DR or not. The examined variables that do not interact with DR (β10

and β14) capture how the relation examined in this study (i.e., between information

asymmetry and conservatism) is affected by auditor tenure when firms get good news. A positive (negative) β10 shows that the relation between PIN and conservatism decreases

Table 5

A further examination of whether auditor tenure affects the information role of conservatism.

NIt =β0 +β1DRt +β2RETt +β3RETt⋅DRT +β4PINt +β5PINt⋅DRt+β6PINt⋅RETt +β7PINt⋅RETt⋅DRt

+β8SHORTt⋅PINt+β9SHORTt⋅PINt⋅DRt+β10SHORTt⋅PINt⋅RETt+β11SHORTt⋅PINt⋅RETt⋅DRt

+β12LONGt⋅PINt +β13LONGt⋅PINt⋅DRt +β14LONGt⋅PINt⋅RETt+β15LONGt⋅PINt⋅RETt⋅DRt

+∑J

j = 1γjCVj+∑Jj = 1KjCVj⋅DRt +∑Jj = 1λjCVj⋅RETt +∑Jj = 1ωjCVj⋅RETt⋅DRt +t

Variable Column 1 Column 2

Intercept 0.043(0.000) −0.027(0.133) DR 0.015(0.155) −0.004(0.870) RET 0.038(0.118) 0.038(0.303) RET*DR 0.104(0.000) 0.091(0.053) PIN 0.009(0.861) 0.047(0.363) PIN*DR 0.120(0.089) 0.154(0.009) PIN*RET −0.312(0.012) −0.320(0.012) PIN*RET*DR 0.643(0.008) 0.942(0.011) SHORT* PIN −0.638(0.073) −0.605(0.076) SHORT * PIN *DR 0.303(0.062) 0.240(0.033) SHORT * PIN*RET (β10) 1.730(0.040) 1.662(0.037) SHORT * IN*RET*DR (β11) −2.127(0.345) −2.069(0.304) LONG* PIN −0.089(0.382) −0.109(0.261) LONG* PIN *DR −0.130(0.019) −0.179(0.003) LONG* PIN*RET (β14) 0.402(0.182) 0.374(0.189) LONG*P IN*RET*DR (β15) −1.065(0.025) −1.110(0.013) Leverage 0.052(0.174) Leverage*DR 0.017(0.044) Leverage*RET −0.065(0.580) Leverage*DR*RET 0.068(0.978) MTB 0.036(0.000) MTB *DR 0.011(0.000) MTB *RET −0.006(0.567) MTB *DR*RET 0.015(0.849) Adj. R2(%) 4.7 8.3 n 5373

Notes: SHORT, a dummy variable, is equal to one if tenure is shorter than four years, otherwise, it is equal to zero; LONG, a dummy variable, is equal to one if Ttenure is longer than 10 years, zero otherwise. Other variables are as defined in Table 2 and Table 3, and reported numbers in parentheses are the two-tailed p-, which have been adjusted by firm-cluster standard error.

(22)

(increases) with short auditor tenure when firms are faced with good news. In other words, a positive (negative)β10implies that, all other things being equal, the observation ofLaFond

and Watts (2008)that greater (lower) gains will be reflected in earnings if the level of PIN is greater, is actually weaker (stronger) when auditor tenure increases. Our column 1 shows a significantly estimated coefficient of SHORTt⋅PINt⋅RETt (1.730, p-valueb0.05) and

Column (2) also shows a significant result (1.662, p-valueb0.05), meaning that, in the case of good news, short tenure significantly affects reported earnings when their PIN scores increase.

A positive (negative) β14, on the other hand, shows that the relation between PIN and

conservatism decreases (increases) with long auditor tenure when firms are faced with good news. The estimated coefficients of LONGt⋅ PINt⋅RETt in column 1 (0.402,

p-value = 0.182) and in column 2 (0.374, p-p-value = 0.189), however, are both insignificantly positive. To sum up, when firms face good news, auditor tenure affects the observation of LaFond and Watts (2008)that lower gains will be reflected in earnings if the level of PIN is greater. However, their claim is weaker when auditor tenure is shorter, at least according to Taiwanese data.

The results in cases of bad news, however, are not identical. The examined variables – SHORT and LONG – that do interact with DR (β11 andβ15) capture how the relationship

between information asymmetry and accounting conservatism is affected by auditor tenure when firms face bad news. Using β11as an example, we can see that a positive (negative)

estimated result for the estimated coefficients of those variables shows that the observation of LaFond and Watts (2008)that greater (lower) losses will be reflected in earnings if the level of PIN is greater is stronger (weaker) when SHORT equals one. The interpretation of β15 is similar to that ofβ11, except for focusing on cases where LONG equals one.

InTable 5, the estimated coefficient of SHORTt⋅PINt⋅DRt(β15) is insignificant in both

column 1 (–2.127, p-value=0.345) and column 2 (–2.069, p-value=0.304). This evidence implies that the effect of PIN score on accounting conservatism is not affected when audit tenure is short and firms face bad news.

The estimated coefficient of LONGt⋅PINt⋅RETt⋅DRt(β15), however, is significant both

in column 1 (–1.065, p-valueb0.05) and in column 2 (–1.110, p-value b 0.05). Taken together, these findings for auditor tenure extend LaFond and Watts (2008)by examining whether and how this factor influences the relation between information asymmetry and conservatism. Specifically, using the middle tenure group as a benchmark, we find that short and long tenure have disparate impacts. When tenure is short, we find support for the contention that good news is incorporated into earnings earlier, but this effect is not significant with long auditor tenure. For the case of bad news, however, it is long auditor tenure that shows significant results: with long auditor tenure, bad news is incorporated into earnings later; with short auditor tenure, this effect is not significant.

6. Conclusion

LaFond and Watts (2008) explain and find evidence that information asymmetry between firm insiders and outside equity investors generates conservatism in financial statements. They conclude that accounting conservatism increases firm value. However, the disparate degree of accounting conservatism around the world calls for an investigation into

(23)

Author's personal copy

the role of conservatism in different countries and into the factors that affect this role. Therefore, by using the LaFond and Watts (2008) method on Taiwanese data, this study investigates the information role of conservatism and refines the conclusions of LaFond and Watts by examining whether auditor tenure affects the information role of conservatism. We find evidence that the information role of conservatism is also supported in Taiwan, but only in cases of bad news. In addition, our evidence reveals that the effect of PIN– which according to LaFond and Watts (2008) causes good (bad) news to be incorporated into earnings later (earlier)– is actually weaker in cases of good news when tenure is short as well as in cases of bad news when tenure is long, at least according to Taiwanese data.

This study is useful because the differences in institutional settings may invalidate the conclusions of LaFond and Watts (2008), which are based only on U.S. data. Using a Taiwanese data set, this study investigates the cause-and-effect relation between information asymmetry and accounting conservatism. Our empirical evidence suggests that conservatism mitigates information asymmetries.

Specifically, we find evidence that accounting conservatism is positively related to the level of and changes in information asymmetry. In sum, conservatism, one of the most prominent characteristics of financial accounting, is actually an efficient way to address the moral hazard problem arising from the asymmetric information among interested stakeholders. In this regard, we suggest that the IASB/FASB Conceptual Framework Project should examine carefully the economic role of conservatism further before making their final decision. One immediate concern is this: if conservatism is no longer structured into the financial reporting standards, how will the IASB and FASB technically deal with the uncertainty that a company faces when preparing its accounts? The advantage of conservatism as a useful earning attribute for creditors and outside equity holders will vanish after the removal of accounting conservatism in financial statements. However, the economic demand for conservatism (e.g., Watts, 2003a, b) still exists in all stakeholders. Although we hesitate to conclude that the IASB's and FASB's statement was hastily made without serious academic input, we believe that the IASB and FASB have not formally taken into consideration how to compensate lenders and borrowers (e.g., Zhang, 2008) as well as stockholders (e.g., LaFond & Watts, 2008) for the lost benefits of conservatism.

Thus, our findings are inconsistent with the current proposal of IASB and FASB that such conservatism should be excluded from the qualitative characteristics of accounting information. Indeed, it is our opinion that a final decision on such an exclusion should be made only after serious consideration of academic input. Even if standard setters such as the FASB and IASB continue to ignore relevant research, shareholders can still keep our results in mind: at the very least, this paper suggests that if conservatism is excluded from the qualitative characteristics of accounting information, shareholders should closely monitor information asymmetry.

Acknowledgements

We gratefully acknowledge the comments and suggestions of Fan-Hua Kung, Fujiing Shiue, T.J. Sellari, Jianling Wang, Kevin Chen (Co-Editor), Taychang Wang, the participants at the twelfth (2008) Accounting and Management Symposium across the Taiwan Strait (Sian, China; where this paper won the Best Paper Award), as well as