The effect of accounting conservatism on information quality in financial reporting : a study on European listed firms

Hele tekst

(1)Amsterdam Business School. The effect of accounting conservatism on information quality in financial reporting A study on European listed firms. 1st supervisor: Ir. Drs. A.C.M. de Bakker 2nd supervisor: Dr. Ir. S.P. van Triest. Name: Stephanie Rijneker Student number: 10358633 Date: August 14, 2016 Word count: 11930 MSc Accountancy & Control, specialisation Accountancy Faculty of Economics and Business, University of Amsterdam.

(2) Statement of Originality This document is written by student Stephanie Rijneker 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.. 2.

(3) Abstract Accounting conservatism is a much criticised topic as standard setters prefer a neutral approach in financial reporting despite a growing number of research providing evidence of a negative association between conservatism and information asymmetry. Conservatism is defined as the differential verifiability required for the recognition of profits versus losses (Watts, 2003). This study investigates the effect of accounting conservatism in financial reporting on information quality for European listed firms. Prior literature suggests that a conservative reporting system limits earnings management and leads to a future decrease in the bid-ask spread and in stock-returns volatility thus decreasing information asymmetry and improving information quality. Information quality will be measured on the basis of two proxies, earnings management by exploring market data and investment efficiency by exploring analyst data. The Khan & Watts (2009) C_score model are applied to the generated data to measure accounting conservatism. Information effects will be measured by association between conservatism and the bid-ask spread, forecast error, analyst following, and returns volatility. The regression analyses conducted for the purpose of this investigation indicate that for European listed firms. A negative association between conservatism and the bid-ask spread was found, however the association was not significant. Between conservatism and forecast error the association was significant but in the unexpected direction. This was also the case for returns volatility. The association between analyst following and conservatism was significantly positive as expected. The conclusion drawn is that more neutrality in financial reporting does not have an effect on the quality of the information environment. Keywords:. accounting conservatism, C_score, information environment, bid-ask spread, forecast error, analyst following, returns volatility. 3.

(4) Contents. 1. Introduction ................................................................................................................................................ 6. 2. Prior literature & hypotheses development............................................................................................ 8 2.1. 2.1.1. Definition of accounting conservatism................................................................................... 8. 2.1.2. Explanations for accounting conservatism............................................................................. 8. 2.1.3. Measuring accounting conservatism...................................................................................... 10. 2.2. The Relation between Conservatism and Earnings Management............................................. 10. 2.3. Indirect effects of Conservatism on the Firm Information Environment .............................. 11. 2.3.1. Quality of information............................................................................................................. 11. 2.3.2. Conservatism and bid-ask spread........................................................................................... 12. 2.3.3. Conservatism and returns volatility ....................................................................................... 12. 2.3.4. Conservatism and forecast errors .......................................................................................... 12. 2.3.5. Conservatism and analyst coverage ....................................................................................... 13. 2.4 3. Hypothesis development ................................................................................................................. 13. Research methodology ............................................................................................................................ 15 3.1. Determination of conservatism ...................................................................................................... 15. 3.2. Measure of the information environment..................................................................................... 17. 3.3. Determination of control variables................................................................................................ 19. 3.3.1. Beta............................................................................................................................................. 19. 3.3.2. Smoothing ................................................................................................................................. 19. 3.4 4. Accounting conservatism .................................................................................................................. 8. Variables overview............................................................................................................................ 20. Data ............................................................................................................................................................ 22 4.1. Sample selection................................................................................................................................ 22. 4.2. Descriptive statistics......................................................................................................................... 23. 4.2.1. Descriptive statistics................................................................................................................. 23 4.

(5) 4.2.2 5. Correlation and multicollinearity measure............................................................................ 23. Results........................................................................................................................................................ 25 5.1. Estimation results from help model Khan & Watts (2009) C_score........................................ 25. 5.2. Hypothesis 1: Bid-ask spread.......................................................................................................... 26. 5.3. Hypothesis 2: Forecast error........................................................................................................... 28. 5.4. Hypothesis 3: Analyst following..................................................................................................... 29. 5.5. Hypothesis 4: Returns volatility...................................................................................................... 30. 5.6. Testing of assumptions for multiple regression ........................................................................... 32. 6. 5.6.1. Linearity and normal distributed errors ................................................................................ 32. 5.6.2. Homoscedasticity ..................................................................................................................... 32. Conclusion and discussion...................................................................................................................... 34. Literature............................................................................................................................................................. 35 World Wide Web............................................................................................................................................... 37 Appendix ............................................................................................................................................................ 38 A.. Analysis of the residuals by histogram........................................................................................... 38. B.. Analysis of the residuals by P-Plot................................................................................................. 40. C.. Homoscedasticity measure .............................................................................................................. 42. D.. Coefficients from estimation regression per year ........................................................................ 44. 5.

(6) 1. Introduction. Since a number of years both the Financial Accounting Standards Board (FASB) and the International Accounting Standards Board (IASB) have expressed a preference towards neutrality in financial reporting over conservatism (prudence) (Hoogervorst, 2012; IASB, 2010). In September 2010 the IASB revised the concept of prudence (conservatism) by neutrality, whereby neutrality can be interpreted as reporting without bias and prudence is no longer a desirable qualitative characteristic (García, 2014; IASB, 2010). Their intent is to increase decision usefulness for users by moving towards an accounting framework that emphasises timeliness and fair value rather than the past focus on historical cost and conservatism (Hellman, 2008). In addition, according to the IFRS Conceptual Framework of 2011, as issued on January 2012 the general purpose of financial accounting is to provide “financial information about the reporting entity that is useful to existing and potential investors, lenders and other creditors in making decisions about providing resources to the entity. Those decisions involve buying, selling or holding equity and debt instruments, and providing or settling loans and other forms of credit. Many existing and potential investors, lenders and other creditors cannot require reporting entities to provide information directly to them and must rely on general purpose financial reports for much of the financial information they need. Consequently, they are the primary users to whom general purpose financial reports are directed (IFRS Conceptual Framework 2011)”. However prior research into accounting conservatism argues that if the FASB was successful in meeting its stated goal of eliminating conservatism, then it would increase information asymmetry between investors, not reduce it (LaFond & Watts, 2008). Basu (1997) defines accounting conservatism as a greater probability of timelier accounting recognition of bad news than good news, capturing accountants' tendency to require a higher degree of verification for recognising good news than bad news in financial statements. A considerable amount of research has been conducted on information asymmetry caused by accounting conservatism in debt-contracting (Wittenberg-Moerman, 2008; Ball, 2008; Watts, 2003). Artiach & Clarkson (2014) provide evidence of an inverse relation between conservatism and the cost of capital. There is a growing amount of research on the association between information asymmetry and accounting conservatism in equity markets (Garcia, 2014; LaFond & Watts, 2008). Most research has been conducted on data using U.S. firms in a rules-based environment. Prior research on information asymmetry in equity markets (Garcia, 2014; LaFond & Watts, 2008) report a negative association between conservatism and the existence of information asymmetries.. 6.

(7) This study will be conducted on European firms in a principle-based environment, making it distinctive from other studies. It will add a European perspective to the debate of the much criticised neutral approach (Hoogervorst, 2012). This study will contribute to existing literature by providing a simplified replication of the Garcia (2014) research by using the Basu (1997) and the Khan & Watts (2009) accounting conservatism measure model in a different geographical area in order to verify if the findings still hold. This study aims to answer the following research question: ‘What is the effect of accounting conservatism in financial reporting on information quality for European listed firms?’ Informational properties of accounting conservatism will be evaluated, focusing on whether conservatism reduces information asymmetries between firm insiders and outside equity-holders. Information quality will be measured on the basis of two proxies, earnings management by exploring market data and investment efficiency by exploring analyst data. Prior literature suggests that a conservative reporting system limits earnings management and leads to a future decrease in the bid-ask spread and in stock-returns volatility thus decreasing information asymmetry and improving information quality (García, 2014, pp. 174-175). This thesis is divided into several chapters. The second chapter of this thesis will provide the reader with a theoretical background in order to understand the interpreted results of the research. It will provide literature review of accounting conservatism and the proxies for the information environment as used for the purposes of this thesis will be defined. Based on the prior literature hypotheses will be developed in chapter three. Chapter four describes the research methodology used to examine this hypothesis, the sample selection used, and provides the descriptive statistics. The fifth chapter provides the results, interpretation and discussion of the main analyses. Finally, in chapter seven, a summary is made together with the conclusion and recommendations for further research.. 7.

(8) 2. Prior literature & hypotheses development. To provide a good understanding of the term information quality and the link with accounting conservatism, the proxies used to determine on information quality are defined. Section 2.1 will define accounting conservatism. Section 2.2 will focus on the stewardship role of financial reporting by discussing existing research about the link between conservatism and earnings management. Finally section 2.3 will focus on the benefits of conservatism for equity markets by discussing the link between conservatism and investment policies.. 2.1 2.1.1. Accounting conservatism Definition of accounting conservatism. Accounting conservatism is a widely documented feature in financial reporting and can be defined as the asymmetric recognition of economic gains and losses into earnings (Guay, 2006, p. 150) where there is a greater probability of timelier accounting recognition of bad news than good news, capturing accountants' tendency to require a higher degree of verification for recognising good news than bad news in financial statements (Basu, 1997). This results in a cumulative understatement of net assets or the anticipation of future net present value investments (Feltham & Ohlson, 1995, p. 693). Accounting conservatism can be seen as a prudent approach to financial accounting by using more strict recognition requirements for profits compared to losses. Due to these strict recognition requirements, the matching principle is therefore unleashed. The matching principle is defined as the recognizing of all expenses and revenue incurred during the accounting period. When revenue is realised it is recognised. However with the use of accounting conservatism, revenue is only recognised when all information relating to the transaction is realised. Particularly, in many business decisions where there is increased uncertainty, implementing accounting conservatism reduces the risk by giving the most prudent reaction to this uncertainty. Considering this, the most suitable definition is the definition of the IFRS Conceptual framework 2011 as outlined in the introduction of this thesis. This definition is also the most recent definition and developed by the standard setters of the accounting standards which are used by the companies in this research. 2.1.2. Explanations for accounting conservatism. Watts (2003) argues that there are four conservatism explanations for conservatism, which are contracting, litigation, taxation and accounting regulation. In addition Watts (2003) proved two nonconservative explanations, earnings management and the abandonment option effect. These explanations will be described in the next subparagraphs.. 8.

(9) 2.1.2.1. Contracting. The first explanation is contracting whereby conservative accounting is used as a means of addressing moral hazard problems such as opportunistic behaviour by the firm’s management by taking into account timeliness, verifiability, and asymmetric verifiability. Examples of such contracts are contracts between the firm and debt holders, called a debt contract, and management compensation contracts. The contracting parties will be informed about the financial performance of the firm by the financial reports, indicating the contracting use of accounting. For example, a bank impose a debt covenant as clause in a loan contract before lending an amount of money to the company. When the company is in breach of a debt covenant, the bank has the legal right to withdraw the loan. A result of this covenant could be that the company will behave more conservative in order to retain the loan. In other words, conservatism mitigates the opportunistic behaviour exhibited by management. The contracting use of accounting exists for quite some time (Watts 2003a) and numerous studies investigated the relation between contracting and conservatism and found evidence for a positive association between debt contracts and conservatism, which supports the indication that contracting is a motivation for conservatism (e.g. Beatty et al. 2008). 2.1.2.2. Litigation. The litigation explanation states that litigation costs are likely to be higher when earnings and net assets are overstated as overstatement could produce asymmetric pay-offs. As a result managers and auditors will be encouraged to understate assets in order to reduce the risk of litigation (Watts, 2003, pp 19-20). Several studies were conducted on the litigation explanation for conservatism. The research of Bushman and Piotroski (2006); Liao et al. (2013), Liu and Elayan (2013) and Basu (1997) found a positive relationship between litigation risk and conservatism. 2.1.2.3. Taxation. The income tax explanation whereby the calculation of taxable income has been linked to reported earnings. The higher the reported earnings, the higher the taxable income and thus, the higher the tax to be paid. As a result, income taxes could be a motivation to apply conservative accounting (Watts 2003a). 2.1.2.4. Accounting regulation. The regulatory explanation whereby Watts (1977; 67) states that losses from overvalued assets and overstated income are more observable and usable in the political process. Therefore Watts (2003 pp 21-22) argues that accounting standard setters and regulators favour conservatism because this will lead to a reduction of the political costs that they will face in case of the discovery of an overstatement. 9.

(10) of assets. However as mentioned previously, standard setters and regulators tend to prefer neutrality instead of a conservative approach in recent days. 2.1.2.5. Non-conservative explanations. Apart from the four explanations for conservatism, Watts (2003b) also discussed earnings management and the abandonment option. Earnings management can be used by managers in a good way, to signal important information; however managers can also manage earnings in a bad way, to obtain personal gains. Earnings management can be used to manage earnings downwards, which can be considered as conservative accounting, but Watts (2003b) considers earnings management as a non-conservatism explanation, because earnings management does not understate the assets systematically. The abandonment option is the other non-conservatism explanation. Managers can abandon operations that are not profitable and this will lead to an understatement of assets. Again, the understatement of assets is part of conservatism but the manager’s intention is not based on prudence (Watts, 2003, p. 6). 2.1.3. Measuring accounting conservatism. Researchers use three methods to assess conservatism. These are net asset measures, earnings and accruals measures, and earnings/stock returns relation measures (Watts, 2003a, p. 288). This study will concentrate on conservatism measures by earnings/stock returns relation using the Khan & Watts (2009) model which is based on the Basu (1997) model. This model provides a firm-year measure of conservatism (C_score) by capturing the asymmetric timeliness of earnings across firms within an industry. By substituting the main characteristics –market-to-book, size and leverage – of the model into the estimation regression the C_score is calculated. A high C_score indicates that losses are reported timelier in comparison profits, hence indicating higher conservatism. The firm year measure of good news is indicated by the G_score. A negative G_score indicates that profits in good times are not reflected timely.. 2.2. The Relation between Conservatism and Earnings Management. Earnings management can be used by managers in a good way, to signal important information; however managers can also manage earnings in a bad way, to obtain personal gains. Watts (2003, p. 207) argues that successful elimination of conservatism will change managerial behaviour and impose costs on investors and the economy in general due to inclusion of unverifiable future cash flows and states asymmetric verifiability is critical to constraining manipulation and fraud. Earnings managements occurs when the true economic performance is masked or untrue and. 10.

(11) considered with all the information made available, would cause the reader to change or alter his or her judgment or decision (Dechow & Skinner, 2000, pp 238 - 240). This could result in either undervaluation where a firm’s stock price is lower than its underlying value, while overvaluation is the situation in which a firm’s stock price is higher than its underlying value (Badertscher, 2011). Earnings management is an important factor that influences earnings quality. The main purpose of earnings management is to mislead investors and other stakeholders by window-dressing financial annual statement whereby possible explanations could be an increase in corporate manager compensation, job security, to avoid violating lending contract or to reduce regulatory costs or to increase regulatory benefits (Healy & Wahlen, 1999, p. 367). Forecast level, accuracy and dispersion are the commonly used forecast properties when examining the degree of earnings management and the quality of information environment (Sohn, 2012, p. 312). Prior literature provides evidence that conservatism limits earnings manipulation by providing a reporting system with recognition of difficult-to-verify losses in the financial statements and difficult-to-verify gains through other channels such as disclosures which leads to a better information environment (Gao, 2013; García, 2014; LaFond & Watts, 2008; Guay & Verrecchia, 2007). Ball & Shivakumar, (2006, p. 240) found that conservative accrual models, which incorporate asymmetry in the relation between accruals and economic losses and gains have a superior explanatory power. Chen (2007, p. 545) showed the degree of earnings management is lower in a conservative regime than in an unbiased accounting regime.. 2.3. Indirect effects of Conservatism on the Firm Information Environment. Investment related effects of accounting conservatism will be investigated as it is argued that managers may deviate from optimal investment to increase their self-serving benefits such as compensation and reputation (García, 2014). First the interpretation of quality of information for the purpose of this thesis is outlined, followed by exploration of the relation between conservatism and the bid-ask spread, returns volatility and analyst coverage. 2.3.1. Quality of information. This section focuses on how quality of information for financial markets is interpreted in published research. Veronesi (2000) states that investors are flooded with a variety of information such as earnings reports, revision of macroeconomic indexes, policymakers' statements and political news and that investors use these this information to update their projections on future growth, inflation rate and interest rate. He found better information give more precise signals on the true state of the economy. 11.

(12) and therefore tend to increase the equity premium, and found a negative relation between information quality and the required return to capital. Brevik and d'Addona (2010) argue that high-quality signals will enable investors to make high-quality forecasts on the state of the economy. In the next sub-paragraphs the link between quality of information as measured for the purpose of this thesis and the association with accounting conservatism are outlined. 2.3.2. Conservatism and bid-ask spread. The bid-ask spread is seen as a measure that reflects information asymmetry and market liquidity. The variable used in this study, the bid-ask spread, studies the effects of investor and analyst beliefs on security trading. Prior literature provides evidence that an increase in firm-level conservatism leads to a future decrease in the bid-ask spread, in stock-returns volatility and more accurate analysts’ forecast (Garcia, 2014, p 175). This is in line with the work of Francis & Martin (2010) who show that investment inefficiencies are reduced by accounting conservatism by providing evidence that conservatism does not allow to defer the recognition of losses of investments in negative net present value projects. 2.3.3. Conservatism and returns volatility. Prior research found a negative association between accounting conservatism and returns volatility (García, 2014; Suijs, 2008). Atilgan (2014) argues that trading activity of informed investors is an important driver of volatility spreads and that the degree of earnings announcement predictability is stronger when the degree of information asymmetry for a stock is greater. Suijs (2008, p. 1312) also found a negative association between volatility of future stock prices and the cost of capital by means of informative disclosures. This view is supported by the research of Healy et al (1999) who find that expanded voluntary disclosure is followed by improved stock performance, less volatility in forecast revisions and a greater number of analyst following. 2.3.4. Conservatism and forecast errors. Lang and Lundholm (1996, pp 467-468) argue that financial analysts are an essential part of capital markets as component of the information environment. This thesis uses forecast error as main element to measure the information environment. A forecast error measures the precision of analyst forecasts (García, 2014, pp 180-181) and forecast dispersion reflects differences of opinion across analysts (Pae & Thornton, 2004, p. 4). Garcia (2014, p. 175) found that increases in firm-level conservatism are followed by more accurate and less dispersed analysts’ forecasts and a greater number of analyst following. However Sohn (2012, p. 312) found that forecast errors of more conservative firms do not significantly differ from less 12.

(13) conservative firms but he does argue that forecasting earnings is more difficult for less conservative firms. 2.3.5. Conservatism and analyst coverage. This section will concentrate on the interlinking of accounting conservatism and analyst coverage. The research of García (2014) measures analyst coverage as the number of analysts following the firm. However due to data limitations this measure was completed by substituting the number of analysts following the firm by the number of recommendations expressed for one firm in a year. Analysts are primary information providers and function as an external monitoring mechanism (Lobo, 2012, p. 507; Gentry, 2013, p. 123). In section 2.2 existing literature was discussed that provides evidence of reduced earnings management when accounting conservatism is present. Yu (2008, p. 268) provides evidence that more analyst coverage leads to less earnings management which in turn leads to a better information environment. O’Brien and Bhushan (1990, p. 75) found that analysts prefer firms with less earnings volatility but do not find a positive association between firm size and analyst following. However García (2014) did not report significant results for the association between conservatism and forecast dispersion or analyst following.. 2.4. Hypothesis development. In the previous chapters, several theories and prior research related to accounting conservatism have been discussed. Based on these theories and the results of prior research the hypotheses will be drawn. To the author’s knowledge, no research has examined the impact of accounting conservatism on the quality of the information environment in Europe. Therefore, this thesis will extend the literature by investigating the impact of accounting conservatism on the quality of information for European listed companies. Based on accounting theory and prior research, which found a higher level of accounting conservatism leading to a better information environment, it is expected that the impact of accounting conservatism on the information environment to be positive. The first objective of this paper is to examine the initial market reaction to high-level conservative firms, when they report good and bad news respectively. Basu (1997) stated that the market reacts stronger to good news than to bad news of conservative firms, as under conservatism, good news is more likely to be persistent in contrast to bad news that is more likely to be momentary. Francis et al. (2013) found a significant positive association between conservatism and stock performance. On the contrary, the research of Basu (1997) states that stock performance does not necessarily has to be. 13.

(14) positively associated with conservatism, and in addition, profitability and stock performance are not perfectly correlated. After measuring whether conservatism, measured as the C_score, exists an association is made between conservatism and the information environment. The previous section outlined the proxies that form the information environment. The first proxy, the bid-ask spread, is seen as a measure that captures information asymmetry in the form of market data. Consistent with the results of García (2014) the hypothesis regarding the association between conservatism and the bid-ask spread is formed as follows: H1:. A negative association exists between conservatism and the bid-ask spread. The next three hypotheses will measure whether an association exists between accounting conservatism and analyst performance. The main element of the first two hypotheses are the forecasts of earningsper-share (EPS). Below hypotheses are formed in line with the results of prior research as outlined in section two and are as follows: H2:. A negative association exists between conservatism and forecast errors. In the paper of García (2014) analyst coverage is measured by the number of analysts following the firm. However due to data limitations we decided to perform this measure by substituting the number of analysts following the firm by the number of recommendations expressed for one firm in a year. H3:. A positive association exists between conservatism and analyst coverage. In the paper of García (2014) the researchers found that a negative association between accounting conservatism and returns volatility, which is measured as an estimate of the average variability of one year of daily stock returns. H4:. A negative association exists between conservatism and returns volatility. 14.

(15) 3. Research methodology. This chapter will provide the research design and is several sections. This first section will provide a description of models used in accounting literature used to capture firm-year specific accounting conservatism followed by an outline of the models used to capture the information environment.. 3.1. Determination of conservatism. The accounting literature uses several models to measure accounting conservatism. In order to select the most appropriate model for this thesis the characteristics of several models. The Basu (1997) and Khan & Watts (2009) models use stock returns to distinguish between good and bad news, while the AACF model from Ball and Shivakumar (2005) uses the cash flow from operations to distinguish between good and bad news. Both models use a dummy variable to measure the asymmetric verifiability of bad news versus good news, indicating the level of conservatism. The Basu model and Khan & Watts (2009) model can only be used to measure conservatism for listed companies and is appropriate to use for large samples. The Khan & Watts (2009) model is based on the Basu model but altered to include the market-to-book, size and leverage variables into the regression. In order to measure accounting conservatism the Khan & Watts (2009) C_score model as a firm-yearspecific accounting conservatism measure that will reflect earnings’ tendency to recognise bad news as losses timelier than to recognise good news as gains will be applied to the selected data. The Khan & Watts (2009) C_score model is based on the Basu (1997) model but altered to include variation in conservatism and the tendency described above. This model provides a firm-year measure of conservatism (C_score) by capturing the asymmetric timeliness of earnings across firms within an industry. By substituting the main characteristics –market-to-book, size and leverage – of the model into the estimation regression the C_score is calculated. Khan & Watts (2009) argue that the marketto-book ratio is related to accounting conservatism due to asymmetric verification requirement which causes a cumulative understatement of net assets relative to market values. The second argument is that larger firms tend to be more mature and to have a richer information environment such as more analyst following. The argument for the use of leverage is that to prevent debt covenant violations a higher contracting demand for highly leveraged firms stimulates conservatism. The firm year measure of good news is indicated by the Score. The following equation is Basu’s (1997) measure on asymmetric timeliness: ‫݊ݎܽܧ‬௝ = ߚ଴ + ߚଵ݀ܰ݁݃௝ + ߚଶ ܴ݁‫ݐ‬௝ + ߚଷ ܰ݁݃௝ ܴ݁‫ݐ‬௝ + ߝ௝. 15.

(16) Where ‫݊ݎܽܧ‬௝. =. ܴ݁‫ݐ‬௝ ݀ܰ݁݃௝. = =. Net income before extraordinary items, deflated by market value of equity at beginning of year ݆ The annual stock return of year ݆ Indicator dummy variable that equals 1 if Ret݆is negative (thus when there is bad news) and zero when Ret݆is positive. The dummy variable ܰ݁݃௝ measures the level of conservatism by measuring the additive response of. earnings to bad news over good news. The higher the coefficient β1, the higher the level of conservatism. The Khan & Watts model measures the level of conservatism based on earnings. This implies that the Khan & Watts model measures earnings conservatism, which is qualified as conditional conservatism in this thesis. Khan and Watts incorporate the theoretical relation between conservatism and three firm characteristics and express ߚଶ and ߚଷ as the following linear function of Size, MTB and Leverage. whereby G_score is the good news timeliness measure and the C_score is the incremental timeliness of the correlation of accountancy surplus and stock price of bad news with respect to good news: ‫ߚ = ݁ݎ݋ܿܵ_ܩ‬ଶ = ߤଵ+ ߤଶ ܵ݅‫݁ݖ‬௝ + ߤଷ ‫ܤܶ ܯ‬௝ + ߤସ ‫ݒ݁ܮ‬௝ ‫ߚ = ݁ݎ݋ܿܵ_ܥ‬ଷ = λଵ+ λଶ ܵ݅‫݁ݖ‬௝ + λଷ ‫ܤܶ ܯ‬௝ + λସ ‫ݒ݁ܮ‬௝. Where ߤ௜, λ௜ =. Empirical estimators of ߤ and ߣ, i= 1–4, are constant across firms, but vary over time since they are estimated from annual cross-sectional regressions. The natural log of market value of equity Market-to-book ratio Long-term and short-term debt deflated by market value of equity. ܵ݅‫݁ݖ‬௝ = ‫ܤܶ ܯ‬௝ = ‫ݒ݁ܮ‬௝ =. This annual cross section model is as follows: ‫݊ݎܽܧ‬௝ = ߚ଴ + ߚଵ݀ܰ݁݃௝ + ߤଵܴ݁‫ݐ‬௝ + ߤଶ ܴ݁‫݁ݖ݅ܵݐ‬+ ߤଷ ܴ݁‫ ܤܶ ܯݐ‬+ ߤସ ܴ݁‫ݒ݁ܮݐ‬ λଵ݀ܰ݁݃ ܴ݁‫ ݐ‬+ λଶ ݀ܰ݁݃ ܴ݁‫ ݁ݖ݅ܵݐ‬+ λଷ ݀ܰ݁݃ ܴ݁‫ ܤܶ ܯݐ‬+. +. λସ ݀ܰ݁݃ ܴ݁‫ ݒ݁ܮݐ‬+ δଵܵ݅‫݁ݖ‬௝ + δଶMTB௝+ δଷ‫ݒ݁ܮ‬௝ + δସNeg ௝ܵ݅‫݁ݖ‬௝ +. δହNeg ௝ ‫ܤܶ ܯ‬௝ + δ଺݀Neg ௝ ‫ݒ݁ܮ‬௝ + ߝ௝. The compounded returns are calculated from the beginning till the end of the fiscal year. Unlike the tests of Khan and Watts (2009) used in the paper of García (2014) which are based on returns from 9 months before fiscal year-end to three months after fiscal year-end. In addition, due to data limitations, we substituted the number of analysts following the firm by the number of recommendations. This might cause differences in the results compared to those of the study of García (2014). 16.

(17) 3.2. Measure of the information environment. 3.2.1.1. Information effects of conservatism. The market based-proxies for information asymmetry between managers and equity-holders are the bid-ask spread and returns vulnerability and performance of financial analysts. The models described below are similar to the models used in the work of García (2014). 3.2.1.2. Information effects and the bid-ask spread. García (2014) uses the following model to test the association between changes in firm-level conservatism and future changes in the bid-ask spread whereby leverage, size, market-to-book and beta are used as controls of which the main coefficient, The main coefficient ߚଵ is expected to be significantly negative.. ‫݀ܽ݁ݎ݌ݏ݇ݏܣ݀݅ܤ‬௧ାଵ =. ߙ + ߚଵ‫݁ݎ݋ܿݏ_ܥ‬௧ + ߚଶܵ݅‫݁ݖ‬௧+ ߚଷ‫ݒ݁ܮ‬௧ + ߚସ‫ܤܶ ܯ‬௧ + ߚହܵ݉ ‫ݐ݋݋‬ℎ݅݊݃௧ + ߚ଺‫ܽݐ݁ܤ‬௧ + ߝ௧ାଵ. The Bid-Ask spread is defined as the natural log of one plus the average daily bid-ask spread over the fiscal year scaled by the midpoint of the spread. To estimate the bid-ask spread the daily bid price and daily ask price were extracted from DataStream. Using this data the average daily bid-ask spread estimation was calculated. First the percentage spread and first the yearly average bid price and ask price per company per year were calculated. For the bid-ask spread midpoint the daily midpoint was calculated as a basis for the average yearly midpoint. The bid-ask spread estimation is used as a dependent variable for the bid-ask spread hypothesis and is also used as a control variable in the hypotheses concerning analyst following. ‫ = ݊݋݅ݐܽ ݉݅ݐݏ݁ ݀ܽ݁݌ݏ ݇ݏܣ݀݅ܤ‬ln( % ‫)݀ܽ݁ݎ݌ݏ‬ % ‫ = ݀ܽ݁ݎ݌ݏ‬100 ‫ݔ‬ ‫=ݐ݊݅݋݌݀݅ ܯ‬. ‫݁ܿ݅ݎ݌ ݇ݏܣ‬− ‫݁ܿ݅ݎ݌ ݀݅ܤ‬ ‫݁ܿ݅ݎ݌ݐ݊݅݋݌݀݅ ܯ‬. ‫ ݁ܿ݅ݎ݌ ݇ݏܣ‬+ ‫݁ܿ݅ݎ݌ ݀݅ܤ‬ 2. The control variables Beta and Smoothing are outlined in 3.3.1 and 3.3.2 respectively. 3.2.1.3. Information effects and returns volatility. The second proxy will capture whether increases in current conservatism leads future changes in stockreturns volatility whereby stock-return volatility is measured as the natural log of one plus the standard 17.

(18) deviation of one year of daily stock returns ending at the end of the fiscal year. The main coefficient ߚଵ is expected to be significantly negative. This association is empirically tested by the following model whereby the same of controls is used as in the previous model: ܴ݁‫ݕݐ݈݅݅ݐ݈ܽ݋ݒݏ݊ݎݑݐ‬௧ାଵ =. ߙ + ߚଵ‫݁ݎ݋ܿݏ_ܥ‬௧ + ߚଶܵ݅‫݁ݖ‬௧+ ߚଷ‫ݒ݁ܮ‬௧ + ߚସ‫ܤܶ ܯ‬௧ + ߚହܵ݉ ‫ݐ݋݋‬ℎ݅݊݃௧ + ߚ଺‫ܽݐ݁ܤ‬௧ + ߝ௧ାଵ. Returns volatility is the natural log of one plus the standard deviation of one year of daily stock returns. For each firm the daily return index (RI) was extracted from DataStream on which standard deviation per year was calculated followed by the natural log of the standard deviation. The control variables Beta and Smoothing are outlined in chapter 3.3.1 and 3.3.2 respectively. 3.2.1.4. Information effects and analyst forecasting. The analyst related proxies attempt to analyse the link between accounting conservatism and the precision and dispersion of analysts’ work whereby three elements are taken into account. These elements are precision and dispersion of forecasts of earnings per share and the number of analysts following the firm. The association between conservatism and analysts’ forecast errors is captured by the following two models which include the same set of variables (Leverage; Size, Market-to-Book; Beta; Smoothing) and the main coefficient ߚଵ is expected to be significantly negative. ‫ݎ݋ݎݎܧݐݏܽܿ݁ݎ݋ܨ‬௧ାଵ =. ߙ + ߚଵ‫݁ݎ݋ܿݏ_ܥ‬௧ + ߚଶܵ݅‫݁ݖ‬௧+ ߚଷ‫ݒ݁ܮ‬௧ + ߚସ‫ܤܶ ܯ‬௧ + ߚହܵ݉ ‫ݐ݋݋‬ℎ݅݊݃௧ + ߚ଺‫ܽݐ݁ܤ‬௧ + ߝ௧ାଵ. Forecast error is the earnings per share analysts’ forecast error measured as the absolute value of the difference between the mean forecast of annual EPS and the actual EPS, scaled by the actual EPS. The mean earnings forecast per fiscal year (EPS#MN) and EPS, calculated as net income (WC01751)/common shares outstanding (WC05301), per entity were extracted from DataStream. The downloaded forecast error is the yearly mean value of all estimates for a company derived by the majority of contributing analysts. The control variables Beta and Smoothing are outlined in chapter 3.3.1 and 3.3.2 respectively.. 18.

(19) 3.2.1.5. Information effects and analyst following. The last test for the analyst related proxy is to analyse whether a positive coefficient exists between timely recognition of bad news and increases in the number of analyst recommendations, as outlined in chapter 3.4. This differs from the proxy used in García (2014) as García measured analyst following, however due to data limitations the analyst following has been substituted for the number of recommendations expressed for one company in one year as a similar outcome is expected. The below model includes the following controls; Leverage; Size; Market-to-Book; Beta and Smoothing. The main coefficient ߚଵ is expected to be significantly positive. ‫ݏ݊݋݅ݐܽ݀݊݁ ݉ ݉݋ܿ݁ݎݐݏݕ݈ܽ݊ܣ‬௧ାଵ =. ߙ + ߚଵ‫݁ݎ݋ܿݏ_ܥ‬௧ + ߚଶܵ݅‫݁ݖ‬௧+ ߚଷ‫ݒ݁ܮ‬௧ + ߚସ‫ܤܶ ܯ‬௧ + ߚହܵ݉ ‫ݐ݋݋‬ℎ݅݊݃௧ + ߚ଺‫ܽݐ݁ܤ‬௧ + ߝ௧ାଵ. The control variables Beta and Smoothing are outlined in chapter 3.3.1 and 3.3.2 respectively.. 3.3. Determination of control variables. Due to the extensive calculations below calculation were performed in separate databases in Microsoft Excel. A detailed explanation for each database is provided. 3.3.1. Beta. A different approach has been taken for the Beta control variable. For the purpose of this investigation Beta is the slope coefficient from the regression of a firm’s daily excess returns over a rolling 12 month window ending in the current fiscal year. As part of the downloaded company information the local market index of an equity (INDXL) was downloaded. 27 unique index codes were found, for 24 firms no market index was found. For both the stock market indices and the individual firms the daily return index (RI) was extracted from DataStream for the period January 1st 2006 to December 31st 2014. First the daily change was calculated for the indices and the individual firms, followed by calculating the excess return when comparing individual firms to the stock index for that particular day. The slope coefficient is calculated from the regression of a firm’s daily excess returns on the daily excess returns as calculated earlier. 3.3.2. Smoothing. Smoothing is the ratio of earnings volatility to cash flow from operations (CFO) volatility.. 19.

(20) Earnings volatility is the standard deviation of the firm’s rolling five-year earnings before extraordinary items (WC01751) scaled by average firm’s rolling five-year total assets (WC02999). CFO volatility is the standard deviation of the firm’s rolling five-year cash flows from operations (WC04860) scaled by average total assets.. 3.4. Variables overview. Bid–Ask spread. The natural log of one plus the average daily bid–ask spread over the fiscal year scaled by the midpoint of the spread, as a percentage.. Returns volatility Forecast error. The natural log of one plus the standard deviation of one year of daily stock returns (RI), as a percentage. We use annual data to calculate the EPS forecast error. It is measured as the absolute value of the difference between the mean forecast of annual. EPS. (EPS#MN). and. the. actual. EPS. (income. (WC01751)/common shares outstanding (WC05301)), scaled by the actual EPS. The forecast error is the natural log of one plus the yearly mean value of all estimates for a company derived by the majority of contributing analysts. Analyst. The natural log of the number of analysts following the firm. Recommendations. measured by the number of analysts’ recommendations expressed per firm per year (RECNO).. C_score. The firm-year specific conservatism proxy developed by Khan and Watts (2009), C_score. It measures the incremental timeliness of earnings to bad news over good news. See 5.1 for the calculation of this variable.. Size. The market value of equity (Common Shares Outstanding WC05301 multiplied by Market Price - Year End WC05001).. Leverage (Lev). The natural log of one plus the ratio of interest-bearing debt (WC03255) to total assets (WC02999).. Market-to-Book (MTB). The natural log of one plus the ratio of market value of equity to book value of equity. (Total assets WC02999 – total liabilities WC03351 scaled by Common Shares Outstanding WC05301 multiplied by Market Price - Year End WC05001).. 20.

(21) Smoothing. Smoothing is the natural log of the ratio of earnings volatility to cash flow from operations (CFO) volatility. Earnings volatility is the natural log of one plus the standard deviation of the firm’s rolling five-year earnings before extraordinary items (WC01751) scaled by average firm’s rolling five-year total assets (WC02999). CFO volatility is the standard deviation of the firm’s rolling five-year cash flows from operations (WC04860) scaled by average total assets.. Beta. Beta is the slope coefficient from the regression of a firm’s daily excess returns over a rolling 12 month window ending in the current fiscal year.. 21.

(22) 4. Data. This chapter describes the sample selection. First, the criteria for inclusion in the sample are discussed. The next chapter outlines the descriptive statistics of the sample, including the correlation matrix and multicollinearity measure.. 4.1. Sample selection. The sample originally consists of 21.628 firm years listed on European stock exchanges (G#LTOTMKEU) from the years 2006 to 2014. The market related proxies are used from the returns file from I/B/E/S from Thomson Reuters and the accounting data are acquired from the DataStream database, resulting in 14.202 firm-year observations. Stocks that from non-European countries are removed from the sample. This resulted in 141 observations were deleted from the sample as this did not concern European countries. Observations with incomplete data (i.e. earnings before extraordinary items and discontinued operations, cash flow from operating activities, cash flows from investing activities, common shares outstanding and ending stock prices of previous fiscal years) are excluded from the regressions. In line with the data sample exclusion in the study of Badertscher (2011) firmyear observations with negative book values were deleted as return of equity for these firms cannot be interpreted in economic terms. Firm-year observations with a dividend pay-out ratios higher than 100%, and firm-year observations with a stock price of less than 1 (local currency). After exclusion, the total sample consists of 14.061 firm years observed, consisting of 1.969 firms spread over 32 countries. Table 1 shows the number of observations per year. The year coincides with the fiscal year of the firm. The second column shows the frequency of observations per year. As can be seen, the sample is evenly spread out over the years. Table 1 - Number of observations per fiscal year 14,061 Firm-year observations, distribution per fiscal year Year. # observations. 2006 2007 2008 2009 2010 2011 2012 2013 2014. 1613 1600 1452 1527 1603 1587 1588 1603 1488. 22.

(23) 4.2. Descriptive statistics. This section describes the sample selection, including the number of observations per fiscal year, the correlation matrix and the multicollinearity measure. 4.2.1. Descriptive statistics. Table 2 sets out the descriptive statistics for the entire sample population including the mean, minimum, median, and maximum values, and the standard deviation of the main variables. The sample includes 14.061 firm year observations between 2006 and 2014. Table 2 descriptive statistics Descriptive Statistics N Bid-ask spread Forecast error. Mean. Std. Deviation. Minimum. Maximum. 12654. -3,57179. 9,14529. 0,73762. 0,84230. 8913. -6,59278. 7,98830. 0,82557. 1,99100. Analyst following. 10399. 0,69315. 3,98898. 2,25866. 0,82037. Returns volatility. 9919. 0,00002. 0,34232. 0,02198. 0,01248. 14061 14061 13170 14031 13949. -9,48091 5,83323 0,00000 -2,65655 0,00068. 5,73057 21,62517 1,26353 3,67021 2,26424. 0,22300 13,85823 0,21151 0,57508 0,34276. 0,48876 2,03079 0,13892 0,36289 0,31763. 9592. -0,48074. 2,35794. 0,83310. 0,25194. C_score Size Leverage Market-to-Book Smoothing Beta. 4.2.2. Correlation and multicollinearity measure. As the variables used are more or less continuous the first correlation measure used is the Pearson’s product-moment correlation coefficient. Table 3 below shows the correlation coefficients for all dependent and independent variables. The correlation coefficients are always between -1 and 1, indicating negative or positive correlation. The Pearson correlation indicates the strength whereby the further the correlation is from zero the stronger the correlation gets. The significance level indicates the probability of whether there is strong correlation due to chance. In Table 3 we found one correlation coefficient, namely the ‫ ݁ݎ݋ܿݏ_ܥ‬and ‫ݎ݋ݎݎܧݐݏܽܿ݁ݎ݋ܨ‬௧ାଵ, where the sign of the coefficient is not in line with our prediction. However the correlation coefficient is not significant.. The absolute values of the correlations are relatively low. So there is no indication of multicollinearity. Multicollinearity is discussed in more detail below.. 23.

(24) Table 3 Pearson Correlation ‫ ݀݅ܤ‬−. Bid − ask spread ୲ାଵ. ‫ݐݏܽܿ݁ݎ݋ܨ‬. ܽ‫ ݀ܽ݁ݎ݌ݏ ݇ݏ‬௧ା ݁‫ ݎ݋ݎݎ‬௧ାଵ 1. Forecast error ୲ାଵ. Analyst following ୲ାଵ Returns volatility ୲ାଵ. C_score Size. Leverage. Market-to-Book Beta. -,171. **. **. -,580. ,341. **. ,053. Size. **. -,517. Market-. Leverage. -,013. to-Book. **. ,267. Smoothing. **. -,045. ,113**. -,037**. ,008. -,095**. -,030*. -,194**. ,043**. ,144**. -,580**. ,113**. 1. -,055**. -,081**. ,646**. ,026*. -,158**. -,014. ,252**. ,341**. -,037**. -,055**. 1. ,081**. -,176**. ,049**. ,114**. ,016. ,340**. ,053**. ,008. -,081**. ,081**. 1. -,101**. ,240**. ,078**. ,013. -,065**. -,517**. -,095**. ,646**. -,176**. -,101**. 1. ,006. -,254**. -,055**. ,176**. -,013. -,030*. ,026*. ,049**. ,240**. ,006. 1. ,036**. -,017*. -,020. **. **. **. **. **. **. -,254. **. ,036. 1. **. -,024*. -,194. -,158. ,114. ,078. ,098. -,045**. ,043**. -,014. ,016. ,013. -,055**. -,017*. ,098**. 1. ,051**. -,228**. ,144**. ,252**. ,340**. -,065**. ,176**. -,020. -,024*. ,051**. 1. chosen as indicator of collinearity diagnostic is the Variance Inflation Factor (VIF). The VIF indicates whether a predictor has a strong linear relationship with the other predictor(s). There are no set cut-off values for determining significant collinearity however Myers (1990) states that values obtained in the 5-10 range indicate significant collinearity. The VIF score of the predictor variables used in this thesis are below five. The C_score model incorporates the variables size, leverage and market-to-book therefore moderate multicollinearity was assumed. Table 4 Collinearity measure VIF 1,107. Size. 1,095. Leverage. 1,090. Market-toBook. 1,058. Smoothing. 1,016. Beta. 1,045. -,228**. 1. In addition, Table 4 below outlines the correlation measure of independent variables. The method. C_score. Beta. -,171**. ,267. Smoothing. **. ‫ݐݏݕ݈ܽ݊ܣ‬ ܴ݁‫ݏ݊ݎݑݐ‬ ݂‫ ݃݊݅ݓ݋݈݈݋‬௧ା ‫ ݕݐ݈݅݅ݐ݈ܽ݋ݒ‬௧ାଵ C_score. 24.

(25) 5. Results. This section describes the empirical findings of the estimation of the models describes above. First, the results of the main accounting conservatism measure, indicated as C_score, from the estimation regression as set out in section 3.1. Second, the results of the estimation equations that test the association between accounting conservatism and information asymmetry as measured by returns volatility, the bid-ask spread, forecast error and analyst following. Third, the assumptions and criteria for inclusion of data used in order to use multiple regression will be outlined.. 5.1. Estimation results from help model Khan & Watts (2009) C_score. Before testing the first hypothesis the original Khan & Watts (2009) model will be run on the whole sample to investigate if conservatism exists at all. As mentioned before, under conservatism the coefficient ߚଵ should be significantly positive. The results provided by the Khan & Watts (2009) model are presented in table 5. These results are the calculation of the average figures based on the. independent conservatism variables per fiscal year as outlined in Table 10 and Table 11 in appendix D. Table 5 shows the results of the average results of the annual analysis on the regression analysis measured for 2006 to 2014. This test contains an average of estimates separate regressions for every fiscal year and estimates the slope coefficients per year. It shows coefficients from the C_score estimation as outlined in section 3.1 by showing the relation between earnings and returns as represented by ܴ݁‫ݐ‬and ܰ݁݃ ‫ ݐܴ݁ ݔ‬on a sample of 14,061 firm-year observations, from 2006 to 2014.. Earnings is the dependent variable scaled by the lagged market price of equity. The significantly positive asymmetric timeliness coefficient ܰ݁݃ ‫ ݐܴ݁ ݔ‬suggests firms are conservative on average. In line with. the prediction the coefficient ܰ݁݃ ‫ ݁ݖ݅ܵ ݔݐܴ݁ ݔ‬is significantly negative, suggesting larger firms have. lower asymmetric timeliness. This is in line with the results of Khan & Watts (2009). The coefficient ܰ݁݃ ‫ ܤܶ ܯ ݔݐܴ݁ ݔ‬is significantly positive suggesting a cumulative understatement of net assets caused. by asymmetric verification requirements. However when looking at the individual years the ܰ݁݃ ‫ ܤܶ ܯ ݔݐܴ݁ ݔ‬coefficient has a positive association with conservatism for a couple of which is likely to be caused by the ‘buffer problem’ (Roychowdhury and Watts, 2007). The buffer problem. causes that unrecognised increases in asset values reduce the necessity to recognise asset value losses (Khan & Watts, 2009). The coefficient ܰ݁݃ ‫ ݒ݁ܮ ݔݐܴ݁ ݔ‬is significantly positive as predicted. This suggests that more levered firms have a higher asymmetric timeliness.. 25.

(26) Table 5. Mean coefficients from estimation regression. Independent variable Intercept Neg. Predicted sign. Coefficient -,152 ,182. p-value -1,686 1,028. Ret Ret x Size Ret x MTB Ret x Lev. + + -. ,362 -,049 ,072 -,117. 3,597 -3,363 3,480 -1,001. Neg x Ret Neg x Ret x Size Neg x Ret x MTB Neg x Ret x Lev. + + +. ,385 -,062 -,135 ,722. 1,393 -1,338 -,231 2,048. Size MTB Lev. ,035 ,033 -,116. 2,809 1,154 -1,724. Neg x Size Neg x MTB Neg x Lev. -,028 -,072 ,103. -,882 -2,144 ,977. Adjusted R square. 0,252. The next sections outline the results and interpretation of four separate multiple linear regressions between accounting conservatism, measured as the C_score and the dependent variables bid-ask spread; forecast errors; analyst following; and returns volatility.. 5.2. Hypothesis 1: Bid-ask spread. As discussed in chapter 2, an increase in accounting conservatism is expected to lead to a decrease in the bid-ask spread. In this section the following hypothesis is tested: H1:. A negative association exists between conservatism and the bid-ask spread. In short, will accounting conservatism in the current year lead to a smaller bid-ask spread in the succeeding year. Prior theory indicated that an increase in accounting conservatism leads to a decrease in the bid-ask spread. The results in Table 6 below present the outcome of a possible association between accounting conservatism and the bid-ask spread.. 26.

(27) Table 6 Effects of change in accounting conservatism on future change in bid-ask spread Coefficients β. Std. Error. t-value. p-value. (Constant). 3,263. ,051. 64,098. 0,000***. C_score Size Leverage Market-to-Book Smoothing Beta. -,018 -,177 ,030 ,176 -,061 -,321. ,015 ,003 ,044 ,017 ,024 ,023. -1,175 -52,836 ,677 10,084 -2,553 -13,920. ,240 0,000*** ,499 ,000*** ,011 ,000***. * Two-tailed significance: p < ,10 ** Two-tailed significance: p < ,05 *** Two-tailed significance: p < ,01. The β-values outline the relationship between the ‫ ݀݅ܤ‬− ܽ‫݀ܽ݁ݎ݌ݏ ݇ݏ‬௧ାଵ and the independent variables as described in 2.3.2. The presented values indicate the individual contribution of each predictor to the model. The standard error is an indicator about how well the mean represents the sample data, a small standard error indicates that there are not many outliers. Table 6 above shows that there are no big standard errors present, this indicates that the sample mean represents the sample data. The t-value is a measure to see whether the predictor is making a significant contribution to the model. If the t-test associated with a β-value is significant then the predictor is making a significant contribution to the model. The negative coefficient for the C_score indicates that an increase in conservatism leads to a reduction in the bid-ask spread. This is in line with prior literature, however the coefficient is not significantly negative. This suggests that an increase in accounting conservatism does not have to lead to a decrease in the bid-ask spread. The control variables size, market-to-book and beta do have a significant impact on the bid-ask spread meaning the results are sensitive to the inclusion of these additional control variables in the model. Size has a negative significant association with the bid-ask spread, meaning the larger the firm the smaller the bid-ask spread. The market-to-book variable has a positive significant association meaning the when the market-to-book ratio increase the bid-ask spread increases. The variable beta has a negative significant association meaning that an increase in the slope coefficient from the regression of a firm’s daily excess returns leads to a decrease in the bid-ask spread decreases. The variable leverage indicated a positive association and the variable smoothing indicated a negative association but both are not significant. The conclusion based on the information provided by the linear regression is that the first hypothesis is rejected.. 27.

(28) 5.3. Hypothesis 2: Forecast error. As discussed in chapter 2, an increase in accounting conservatism is expected to lead to a decrease in forecast errors. In this section the following hypothesis is tested: H2:. A negative association exists between conservatism and forecast errors. In short, will accounting conservatism in the current year lead to a smaller forecast errors in the succeeding year. Prior theory indicated that an increase in accounting conservatism leads to a decrease in forecast errors. The results in Table 7 below present the outcome of a possible association between accounting conservatism and forecast errors. Table 7 Effects of change in accounting conservatism on future change in forecast errors Coefficients β (Constant) C_score Size Leverage Market-to-Book Smoothing Beta. Std. Error. t-value. p-value. 2,646 ,135 -,149 -,131 -1,571 ,594. ,240 ,075 ,015 ,200 ,092 ,116. 11,042 1,810 -9,589 -,653 -17,077 5,104. ,000 ,070* ,000*** ,514 ,000*** ,000***. 1,300. ,103. 12,563. ,000***. * Two-tailed significance: p < ,10 ** Two-tailed significance: p < ,05 *** Two-tailed significance: p < ,01. The β-values outline the relationship between the ‫ݎ݋ݎݎ݁ݐݏܽܿ݁ݎ݋ܨ‬௧ାଵ and the independent variables as described in 2.3.2. The presented values indicate the individual contribution of each predictor to the. model. The standard error is an indicator about how well the mean represents the sample data, a small standard error indicates that there are not many outliers. Table 7 above shows that there are no big standard errors present for the C_score, this indicates that the sample mean represents the sample data. The t-value is a measure to see whether the predictor is making a significant contribution to the model. If the t-test associated with a β-value is significant then the predictor is making a significant contribution to the model. García (2014) found a negative significant association between accounting conservatism and future forecast errors. The negative coefficient for the C_score indicates that an increase in conservatism leads to an increase in the future forecast errors. The coefficient is significantly positive at a ten percent significance level. This suggests that an increase in accounting conservatism leads to an increase in the number of forecast errors. This is in line with the findings of García (2014) however Garcia (2014) found his result to be significant.. 28.

(29) In this investigation the control variables size, market-to-book, smoothing and beta have a significant impact on the number of analyst recommendations meaning that the control variables are likely to have an effect on the forecast error coefficient. This indicates that the results are sensitive to the inclusion of these additional control variables in the model. The control variable size indicates that smaller forecast errors are exist for larger firms. For the control variable beta it means that the higher the slope coefficient from the regression of a firm’s daily excess returns, the higher the number of analyst recommendations. The second hypothesis is rejected, because of the observed positive association between accounting conservatism and forecast errors instead of the expected negative association.. 5.4. Hypothesis 3: Analyst following. As discussed in chapter 2, an increase in accounting conservatism is expected to lead to an increase in analyst following. In this section the following hypothesis is tested: H3:. A positive association exists between conservatism and analyst following. In short, will accounting conservatism in the current year lead to a bigger number of analyst following in the succeeding year. Prior theory indicated that an increase in accounting conservatism leads to an increase in the number of analyst following. The results in Table 8 below present the outcome of a possible association between accounting conservatism and the number of analyst recommendations expressed in the succeeding fiscal year. Table 8 Effects of change in accounting conservatism on future change in analyst following Coefficients β (Constant) C_score Size Leverage Market-to-Book Smoothing Beta. Std. Error. t-value. p-value. -47,905. ,682. -70,219. ,000. ,549 4,048 ,429 ,405 ,075 3,596. ,212 ,045 ,572 ,242 ,311 ,301. 2,589 90,829 ,751 1,675 ,241 11,959. ,010*** ,000*** ,453 ,094* ,810 ,000***. * Two-tailed significance: p < ,10 ** Two-tailed significance: p < ,05 *** Two-tailed significance: p < ,01. The β-values outline the relationship between the ‫ݏ݊݋݅ݐܽ݀݊݁ ݉ ݉݋ܿ݁ݎݐݏݕ݈ܽ݊ܣ‬௧ାଵ and the. independent variables as described in 2.3.5. The presented values indicate the individual contribution of each predictor to the model. The standard error is an indicator about how well the mean represents. 29.

(30) the sample data, a small standard error indicates that there are not many outliers. Table 8 above shows that there are standard errors present, this indicates that the sample mean does not represent the sample data. The t-value is a measure to see whether the predictor is making a significant contribution to the model. If the t-test associated with a β-value is significant then the predictor is making a significant contribution to the model. The positive coefficient for the C_score indicates that an increase in conservatism leads to an increase in the future analyst recommendations. The coefficient is significantly positive at a one percent significance level. This suggests that an increase in accounting conservatism leads to an increase in the number of analyst recommendations. This is in line with research findings of O’Brien and Bhushan (1990, p. 75) and the findings of García (2014). These researchers found a positive association, however these were not significant. The control variables size, market-to-book and beta have a significant impact on the number of analyst recommendations meaning that the control variables are likely to have an effect on the future number of analyst recommendations. The control variable size indicates that more analyst recommendations are expressed from larger firms. For the variable market-to-book it means that the closer the book value is to the market value of a firm, hence the higher the market-to-book ratio the greater the number of analyst recommendations expressed. For the control variable beta it means that higher the slope coefficient from the regression of a firm’s daily excess returns, the higher the number of analyst recommendations. Based on these findings, the third hypothesis should not be rejected.. 5.5. Hypothesis 4: Returns volatility. As discussed in chapter 2, an increase in accounting conservatism is expected to lead to a decrease in returns volatility. In this section the following hypothesis is tested: H4:. A negative association exists between conservatism and returns volatility. In brief, will accounting conservatism in the current year lead to a lesser returns volatility in the succeeding year. Prior theory indicated that an increase in accounting conservatism leads to a decrease in returns volatility. The results in Table 9 below present the outcome of a possible association between accounting conservatism and the level of returns volatility expressed in the succeeding fiscal year. The association of ܴ݁‫ݕݐ݈݅݅ݐ݈ܽ݋ݒݏ݊ݎݑݐ‬௧ାଵ and the calculated C_score is measured by a multiple linear regression analysis. Hereby the control variables size, leverage, market-to-book, smoothing and beta are also taken into account.. 30.

(31) Table 9 Effects of change in accounting conservatism on future change returns volatility Coefficients β (Constant) C_score Size Leverage Market-to-Book Smoothing Beta. Std. Error. t-value. p-value. ,025 ,003 -,001 ,003 ,001 ,000. ,001 ,000 ,000 ,001 ,000 ,000. 24,317 9,827 -19,490 3,285 3,303 -,180. ,000 ,000*** ,000*** ,001*** ,001*** ,857. ,016. ,000. 34,396. ,000***. * Two-tailed significance: p < ,10 ** Two-tailed significance: p < ,05 *** Two-tailed significance: p < ,01. The β-values outline the relationship between the ܴ݁‫ݕݐ݈݅݅ݐ݈ܽ݋ݒݏ݊ݎݑݐ‬௧ାଵ and the independent variables as described in 2.3.3. The presented values indicate the individual contribution of each predictor to the. model. The standard error is an indicator about how well the mean represents the sample data, a small standard error indicates that there are not many outliers. Table 9 above shows that there are no big standard errors present, this indicates that the sample mean represents the sample data. The t-value is a measure to see whether the predictor is making a significant contribution to the model. If the t-test associated with a β-value is significant then the predictor is making a significant contribution to the model. The positive coefficient for the C_score indicates that an increase in conservatism leads to an increase in the future returns volatility. The coefficient is significantly positive at a one percent significance level. This suggests that an increase in accounting conservatism leads to an increase in returns volatility. This contradicts the findings of García (2014) as García found a negative significant association. For the control variables leverage, market-to-book and beta a positive significant association was found. For the variable leverage it means that the higher the debt to equity ratio is, the larger returns volatility. For the variable market-to-book it means that the closer the book value is to the market value of a firm, hence the higher the market-to-book ratio the greater the returns volatility. For the control variable beta it means that higher the slope coefficient from the regression of a firm’s daily excess returns, the higher the returns volatility. For the control variable size a significant negative association was found, meaning the smaller the firm the higher the future returns volatility. To summarise, it is concluded that the fourth hypothesis should be rejected as it can be said that a fluctuation in accounting conservatism significantly impacts the level of future returns volatility. However the fluctuation is significantly positive instead of negative.. 31.

(32) 5.6. Testing of assumptions for multiple regression. This section entails a discussion of criteria for inclusion in the sample as in order to use multiple regression the data used has to meet several assumptions. When these assumptions are not satisfied, the result provided by the regression analysis could be misleading. When the assumptions are met, the model that is produced by the sample can be accurately applied to the population of interest (Field, 2009, p. 221). The assumptions that are tested are linearity, normally distributed errors and homoscedasticity. 5.6.1. Linearity and normal distributed errors. In this investigation multiple linear regressions is used. For this method the relationship between the dependent variable and the independent variables should be linear. Linearity may be tested by analysing the residuals, which are the differences between the predicted values and the observed values. The mean values of the outcome variable for each increment of the predictor should lie along a straight line to be considered linear (Field, 2009, p. 221). Under the assumption of normality, it is assumed that the residuals in the model are random, normally distributed variables with a mean of zero meaning that the differences between the model and the observed data are close to zero. As a result, the residuals should follow a normal distribution. Figure 1, Figure 2, Figure 3 and Figure 4 in appendix A show that the histogram of the residuals have bell curved distributions and a peak can be found only around zero for all dependent variables. Figure 1, Figure 2, Figure 3 and Figure 4 in appendix A in support this conclusion. In addition, Figure 5, Figure 6, Figure 7 and Figure 8 in appendix B show that the residuals are distributed normally and lye around the diagonal. Figure 5, Figure 6, Figure 7 and Figure 8 in appendix B shows the residuals arisen from for each dependent variable and the Khan & Watts (2009) model including the control variables as independent variables. The figures show that the residuals are not extremely deviating from the mean of zero indicating a linear relationship and normal distribution between the dependent and independent variables, so the first condition for multiple regression has been met. 5.6.2. Homoscedasticity. When applying multiple regression the data used has to meet homoscedasticity assumption whereby at each level of the predictor variable, the variance of residuals should be constant for each of the dependent variable (Field, 2009, p. 220). To test whether the residuals are homoscedastic, a scatterplot is analysed for linearity.. 32.

(33) Figure 9, Figure 10, Figure 11 and Figure 12 in appendix C present the results and show that the magnitude of the error is not strongly influenced by the value of the dependent variable so the results indicate that the residuals are homoscedastic. Hence, this condition for multiple regression has been met.. 33.

(34) 6. Conclusion and discussion. Recently, the IASB has expressed a preference for neutrality in financial reporting over accounting conservatism. However, there has been an amount criticism on this preference. This investigation adds to this debate by examining the relation between conservatism in financial reporting and the quality of information as measured by the bid-ask spread, forecast error, analyst following and returns volatility. Conservatism in financial reporting is defined as accountants’ tendency to require a higher verification for recognizing good news as gains than for recognizing bad news as losses (Basu, 1997). In this investigation accounting conservatism is measured as the C_score, a help model of Khan & Watts (2009). The results provided in the previous chapters do not correspond with the hypotheses presented in chapter two. Three out of four hypotheses have been rejected. The first hypothesis, expected to show a significant negative association between accounting conservatism and the bid-ask spread was rejected. A negative association was found, however this was not significant. The second hypothesis, expected to show a significant negative association between accounting conservatism and the forecast errors was rejected due to a contradicting result. A significant positive association was found between accounting conservatism and forecast errors. The third hypothesis, expected to show a significant positive association between accounting conservatism and the analyst following was not rejected. A positive significant association between accounting conservatism and the number of analyst recommendation expressed was found. Lastly, the fourth hypothesis, expected to show a significant negative association between accounting conservatism and the returns volatility was rejected due to a contradicting result. A positive significant association was found. The results of the multiple linear regressions performed for the purpose of this thesis show that accounting conservatism is not likely to lead to a significant change in the information environment. Therefore, it can be argued that more neutrality in financial reporting does not have an effect on the quality of the information environment. There are a couple of limitations to this research. The most important limitation is the determination of the quality of information. Quality of information could possibly be measured in a different way. There could be other aspects involved that impact the information environment. In addition, the financial crisis that started in 2008 could have impacted the results, also a split could be made for the results in different industries.. 34.

No results found