UNIVERSITEIT VAN AMSTERDAM
Amsterdam Business School
MSc Accountancy & Control
Master Thesis
Influence of International Financial Reporting Standards on
conservatism in the Netherlands
Submitted on Friday, 29
thAugust 2014
Marco Bergsma (9377522)
First supervisor: Dr. S.W. Bissessur
Second supervisor: Dr.ir. Sander van Triest
Final version
(wordcount: 15.189)
Table of Contents
1. Introduction ... 5
1.1 Introduction ... 5
1.2 Research objectives and contributions ... 7
2. Conservatism ... 9
2.1 Introduction ... 9
2.2 Overview of conservatism ... 9
2.3 Explanations and measures of conservatism ... 12
2.3.1 Explanations of conservatism ... 12
2.3.2 Measures of conservatism ... 14
2.4 Positive and negative effects of conservatism ... 16
3. IFRS ... 19
3.1 Introduction ... 19
3.2 Overview of IFRS ... 19
3.3 Benefits of IFRS ... 20
3.4 Conservatism under IFRS ... 21
4. Research method ... 23
4.1 Hypotheses development ... 23
4.2 Sample selection and data collection ... 24
4.3 Descriptive statistics and models ... 25
5. Empirical Results and Analysis ... 27
5.1 Empirical results ... 27
5.2 Analysis of results ... 31
6. Conclusions ... 41
References ... 43
Preface
“Even a journey of a thousand miles begins with but a single step” – Lao Tzu (604 bc - 531 bc)
This thesis has been a long while in the making. Having a professional career before graduating can cause one to shift one’s focus. After deciding from the many interesting topics on the subject that lies before you, I started with my writing Odyssey. In hindsight, I found that the hardest part of this endeavour was getting started.
Changes in the status quo have always intrigued me. Therefore, a change in the accounting rules had an immediate appeal to me. During a study trip to London we visited the IASC, the predecessor of the current IASB. I was quite taken by the future outlook they envisioned. Now I am able to incorporate their work in my thesis.
A heartfelt thank you goes out to all the people who supported me and I am grateful for the feedback they gave me which I needed to complete this project. Overcoming many obstacles I am now proud to present you with the final piece of my study. I hope that you enjoy reading this thesis as much as I enjoyed writing it.
Sincerely yours,
Marco Bergsma
Abstract
This paper examines the change in conservatism after the mandatory adoption of International Financial Reporting Standards (IFRS) in 2005 by the Dutch Capital Markets. It combines two subjects with a vast body of literature. The Introduction of IFRS changed the rules of the game. Local Generally Accepted Accounting Principles (GAAP) are now superseded by rules on a higher level. This could have an effect on existing judgemental behaviour of the preparers and verifiers of financial statements. The prior literature on conservatism and the effect of IFRS shows mixed results. The majority of the studies show a decrease in conservatism after the mandatory adoption of IFRS. However, with voluntary adopters conservatism has been shown to increase. Most studies employ the popular Basu (1997) method for measuring conservatism. By using a new method that corrects for two types of correlation I document that conservatism in the Netherlands has not changed after IFRS came into play.
A sample of 285 companies listed at the NYSE Euronext Amsterdam observed over the period 1998-2013 is gathered to investigate the research question. The Beaver and Ryan (2005) general model of conditional and unconditional conservatism under uncertainty is used to operationalize and measure conservatism. The empirical results show that conservatism has changed after IFRS implementation. This paper contributes to the different views about conservatism by providing insights in the mechanics of calculating the accounting income versus the economic income. My empiric findings add to the robustness of this model. I extend the literature by utilizing more recent data and focussing on the Dutch capital markets, a specific geographical region. I Also take into consideration the recent criticism of Gow et al. (2010) about the use of test statistics that do not compensate for time series and/or cross-sectional dependence. Finally, my research contributes to the accounting practice, as more and more countries are adopting IFRS or are converging towards them.
Key words: IFRS, Conservatism, Beaver and Ryan model, Asymmetry, The Netherlands,
Change, two way clustering
1. Introduction
1.1 Introduction
Every day people make financial decisions. In economics the term homo economicus or economic human, is used to describe people who act rational and narrowly self-interested and who have the ability to make decisions about their subjectively defined ends1. These financial decisions are based on information that is available to them. If one has more information available to him at a particular point in time one can make a better informed decision by narrowing his set of alternative options. This way financial information adds to efficacy of the decision making process.
To aid them with these financial decisions people turn to the auditors. Auditors are in the business of providing assurance about the veracity of legal claims made and information provided in financial statements of businesses. To be able to provide this assurance to the reader auditors need to verify the provided information. Through these financial reports both good news and bad news can be disseminated to the public. Using professional judgement, auditors approach the investigation of good news and bad news differently. A greater measure of prudence is assumed when investigating good news than when bad news is investigated. As a rule expenditures will be recognized when they are incurred but the recognition of income will be deferred until it has been verified by supporting documentation.
Basu (1997) is one of the first to investigate empirically this phenomenon of a more conservative attitude in accounting in his critically acclaimed paper ‘the conservatism principle and the asymmetric timeliness of earnings’. He theorizes that the conservatism principle could be interpreted as capturing accountants’ tendency to require a higher degree of verification for recognizing good news as an income than bad news as a loss in companies’ financial statements. Consequently, because of this need for more verification he also postulates that earnings in the financial statements will incorporate bad news faster than good news. Because of this asymmetry in recognizing good news and bad news, systematic differences will arise in the timeliness and persistence of reported earnings.
As research progressed over the years there came a fine tuning of the concept of conservatism. Researchers now recognized two distinct forms, to be known as conditional and unconditional conservatism, each with its own set of characteristics. The great divider between these two is the dependency of conditional conservatism on ‘economic shocks’ or news, whereas unconditional conservatism has no relation with this variable. Beaver and Ryan (2005) developed a model that captures the essence and the interactions between
1 http://www.princeton.edu~achaney/tmve/wiki100k/docs/Homo_economicus.html. The term "Economic Man" was first used in
the late nineteenth century by critics of John Stuart Mill’s work on political economy. They attacked him for creating a monomaniacal economic man concerned only with the accumulation of money.
5
these two forms. The Basu model only focuses on conditional conservatism. Seven limitations have been identified by Ryan (2006) of the use of Basu’s asymmetric timeliness measure. However, he also proposes ‘workarounds’ and retains the stance that asymmetric timeliness should be the primary measure of conditional conservatism.
Prior literature about the accounting conservatism principle shows mixed results. As a starting point, there is empirical evidence that conservatism does exist. For instance US listed firms have conservative financial statements (Basu, 1997), and also countries outside of the US practice accounting conservatism (Ball et al., 2000) and more recently Ball et al. (2013) have also proven this. However, the methods of measuring conservatism and interpreting the results remain a source of ongoing discussion, as evidenced by the many articles found when doing an internet search on accounting conservatism. The accounting literature (e.g., Watts, 2003a) heralded that not only conservatism exists, it is also increasing. Critics note that some of the effects contributed to conservatism could also be explained by non-conservatism causes. The two most cited ones are earnings management (e.g. Hanna, 2002) and the abandonment option (Hayn, 1995).
Recently, the methods for correcting for cross-sectional and time-series dependence in the models used have come under question. Gow et al. (2010) take the stance that many studies deal with variables that are cross-sectionally and serially correlated but use test statistics that, as they claim, do not correct for this problem. Where the popular Basu (1997) model explicitly does correct for heteroscedacity by deflating the variables by opening stock price and book value of assets or equity and by using White t-statistics, it does not correct for both types of dependence.
Then an influential economic event took place. International Financial Reporting Standards (IFRS) were introduced in the European Union by the International Accounting Standards Board (IASB) in 2005 as rules and guidelines which publicly listed businesses need to adhere to when they are preparing their financial reports. Up until 2005 European companies followed the reporting laws that were indigenous to their specific countries, also referred to as local Generally Accepted Accounting Principles (GAAP). But then the Introduction of IFRS changed the rules of the game. Local GAAP were now superseded by rules on a higher level. This could have an effect on existing judgemental behaviour of the preparers and verifiers of financial statements.
The IASB prepares the standards but the European Commission has set IFRS as mandatory for the consolidated financial statements of publicly listed companies in the EU starting from January 1st, 20052. The IASB has no authority to impose those Standards. Currently 102 countries have already adopted IFRS, with a further 22 countries having made
2 Regulation (EC) No 1606/2002 of the European Parliament and of the council of 19 July 2002 on the application of
international accounting standards
6
a public commitment to IFRS as the single set of global accounting standards3. IFRS promise more accurate, comprehensive and timely financial statement information, relative to the national standards they replace for public financial reporting in most of the countries adopting them, Continental Europe included (Ball, 2006).
1.2 Research objectives and contributions
As an accountancy student my interests lie in the fields of reporting and the capital markets. Prior papers on conservatism have sparked my interest so the goal of this thesis is to combine these subjects. The objective of this study will be to examine if the mandatory adoption of IFRS in the Netherlands has had an impact on conservatism in the Dutch capital markets. Further, the goal is to provide insights in whether IFRS has had any influence, and if indeed it has, has this been a positive or a negative influence on accounting conservatism. An exposure draft of the new conceptual framework for Financial Reporting of the IASB shows that IFRS is aiming for neutrality of financial reporting as opposed to conservatism. This begs the question if conservatism decreased since the adoption of IFRS in 2005.
Prior literature shows mixed results about the change in conservatism after adoption of IFRS in Europe. There is no consensus about a possible change in conservatism, and will it decrease or increase after IFRS have been adopted. I will use the Beaver and Ryan (2005) model to test for the change of conditional and unconditional conservatism after 2005 in bad and good news firms in the Dutch capital markets.
My research contributes to the different views about conservatism by providing insights in the mechanics of calculating the accounting income versus the economic income. Moreover, because accountants are in the business of testing the accuracy of accounting numbers and not the economic value it will be helpful to them to further understand the impact conservatism has on the relationship between accounting income and economic income. By employing the model of Beaver and Ryan (2005) I also consider the effect of unconditional conservatism. My empiric findings add to the robustness of this model. I extend the literature by utilizing more recent data than in the prior literature and by focussing on the Dutch capital markets, thereby adding very specific knowledge. I also take into consideration the recent criticism of Gow et al. (2010) about the use of test statistics that do not compensate for time series and cross-sectional dependence. They advocate a method described as two-way clustering. I will employ this method to evaluate its merits and the effects of the standards errors. Finally, my research contributes to the accounting practice, as more and more countries are adopting IFRS or are converging towards them. Because the adoption of IFRS in 2005 made for an explicit change in the calculation rules I can make
3
http://www.ifrs.org/Use-around-the-world/Pages/Analysis-of-the-IFRS-jurisdictional-profiles.aspx
7
a clear cut division in the data sample. This division assists in adding to the current evidence of the distinct particulars and interrelated relationship of the two forms of conservatism.
Beaver and Ryan (2005) will be used as a key paper as their empiric model captures the interactions between both conditional and unconditional conservatism. The Basu (1997) paper that has been most cited by academia only focuses on conditional conservatism. My research questions can be stated as follows:
- Does the mandatory adoption of IFRS since 2005 have an impact on conservatism in the Dutch Capital Markets?
- Will conservatism decrease or increase in the Dutch Capital Markets under IFRS?
The remainder of this thesis is organized in the following manner: section 2 starts with an overview of conservatism, including its definitions, different forms, measures and effects that have been identified in prior literature. Section 3 provides an overview of IFRS, their proposed benefits and effects on conservatism stated in prior literature. Section 4 describes the research methodology. First the hypotheses development, followed by the sample selection and data collection. Then the variables of interest are defined and the section will be concluded with the descriptive statistics of the measure employed. Section 5 details the empirical results and the analysis of these results. Finally, section 6 presents the conclusions of this research.
2. Conservatism
2.1 Introduction
In this section, I provide an overview of the main theories and empirical findings about accounting conservatism in the literature. In the first part, I present the contextual factors for a better understanding of the concept of conservatism, this includes definitions. I then present the explanations and measures researches have come up with in prior literature, followed by the effects of conservatism that are identified. The aim of this chapter is to provide a thorough base of the concept of conservatism so the empirical results in section five are easily understood.
2.2 Overview of conservatism
In order to fully capture the discussion on conservatism in the literature and why it has sparked so much research it is important to define what the term entails. It has been argued that there has been a long period of significant influence of conservatism on the accounting practice. According to Basu (1997) this has been the case for at least the last 500 years. The most common and striking definition for accounting conservatism in scientific literature comes from Bliss (1924) who stated that conservatism is to “anticipate no profit and provide for all probable losses”. This adage is one of the earliest definitions that has been employed to describe the concept. It shows extreme prudence and entails that there is a high, but not impossible, verification requirement for making any legal claim to profits. Under conservatism all possible losses are recognized as soon as they are discovered. Revenues however will be deferred to later periods until they are verified. The relating cash inflows need not be received to be recognized but they do have to be verified (Watts, 2003a). One could also state that anticipated future losses are indeed losses but anticipated profits are not profits.
Over time many other definitions have come to be used. For instance, Watts and Zimmerman (1986) view conservatism as ‘reporting the lowest value among possible alternative values for assets and the highest alternative value for liabilities. Revenues should be recognized later rather than sooner and expenses sooner than later’. Feltham and Ohlson (1995) define conservatism as the portion of firm value not recorded in the accounting system. This is referred to as economic goodwill, which comes about by overstating expenses or understating revenues. Kwon et al. (2001) state that conservatism in their model implies that the accounting system is more likely to report ‘low’ (losses) when the outcome is low than to report ‘high’ (gains) when the outcome is high. Watts (2003a) adds to the framework by raising the interchangeable terms ‘differential verifiability’ and ‘asymmetric verifiability’ to define conservatism. This describes the fact that profits require a higher
degree of verification before they are recognized than losses. Following this asymmetry in verification requirement one can also distinguish between varying degrees of conservatism. If there is a greater difference in the degree of verification required for gains as opposed to losses, then conservatism is assumed to be greater (Watts, 2003a).
To extend the definition, two distinct forms of conservatism have been identified in the research. The first form is unconditional conservatism (or ex ante or news independent), meaning that aspects of the accounting process determined at the inception of assets and liabilities yield expected unrecorded goodwill (Pope and Walker, 2003; Richardson and Tinaikar, 2004; Chandra et al., 2004; Ball and Shavikumar, 2005; Beaver and Ryan, 2005). This reducing of earnings happens independently of any current economic news. Examples are the immediate expensing of R&D and accelerated depreciation. Empiric research has been directed towards the balance sheet to identify this type of conservatism. It is more difficult to detect on the income statement as the effects are more likely to be consistent across reporting periods (Ruch and Taylor, 2011).
The second is conditional conservatism (or ex post or news dependent), meaning that book values are written down under sufficiently adverse circumstances but not written up under favourable circumstances, with the latter being the conservative behaviour (Pope and Walker, 2003; Richardson and Tinaikar, 2004; Chandra et al., 2004; Ball and Shavikumar, 2005; Beaver and Ryan, 2005). Basu (1997) cites conservatism as causing bad news to be reported more quickly than good news. Ruch and Taylor (2011) define conditional conservatism as a system of reporting whereby the understatement of accounting value occurs dependent of economic events. By this they refer to the news about these economic events and not the events themselves. They note that news can be partially or fully recorded. News is considered fully recorded if all available information is used in determining the value of a balance sheet item. This usually holds for the liabilities. If the value of a balance sheet item has been determined by not using all available information, the news is considered partially recorded. This usually holds for the assets. Beaver and Ryan (2005) find that the literature on conditional conservatism puts greater emphasis on improving contracting efficiency given managers’ incentives to report upward-biased accounting numbers. The dependence on economic news events makes for the primary distinction between the two forms. This news can be either positive or negative.
Ruch & Taylor (2011) propose two reasons why it is important to make the distinction between the different forms of conservatism. Reason one is that unconditional conservatism will have a relatively consistent impact on the income statement from one period to the next, whereas the impact of conditional conservatism on the income statement is more likely to be of a transitory nature. This is because the news that gives rise to the understatements will differ both in content and timing from one period to the next.
Their second reason is that the presence of one type of conservatism affects the presence of the other type. Beaver and Ryan (2005) examine the interactions between the two forms of conservatism. They find that “accounting slack” created by unconditional conservatism can give rise to conditional conservatism. When measured over a short period of time Roychowdhury and Watts (2007) find a negative association between conditional and unconditional conservatism.
An important distinction to be made relating to the research is the one between accounting income and economic income, also referred to as return. Accounting income consists of the cash flow generated by the operations of a firm, and accruals adjustments on cash flow from operations based on expectations of cash flows (Bissessur, 2008, p. 27). Economic income is the change of the market value of equity, adjusted for dividends and capital contributions (Ball and Shivakumar, 2005). These two different calculation concepts drive the market models of conservatism used in the literature. For a description of the different measures used in the literature I refer to paragraph 2.3.1. The focus on accounting income is due to it being a sensitive barometer of financial reporting in general, in that income statement variables are structurally correlated with changes in balance sheet variables. Income statement timeliness thus is an indicator of financial reporting timeliness generally (Ball et al., 2013).
Timeliness is also a factor to be considered when looking at conservatism. This refers to the degree to which contemporaneous earnings reflect current economic income (Ball et al., 2000). Asymmetric timeliness of result recognition is the most used instrument of choice for the measurement of conservatism. One could take the strong stance that earnings should only be recognised when they are received and not when they are verifiable. However, this delay would lead to asymmetrical cost because it requires overinvestment in the firm. This overinvestment can be by shareholders because possible dividends will be restricted until a later point in time and it can also be by managers because their compensation plans are postponed until verification has been obtained.
In the U.S. the Financial Accounting Standards Board (FASB) considers conservatism undesired because accounting information should be neutral in their opinion (Watts, 2003a/FASB, 2010). The bias that is otherwise introduced in the decision making process can result in suboptimal decisions. (Gigler et al., 2009; Guay and Verrecchia, 2006). However, this steering away from conservatism could bring about unforeseen consequences and be damaging to the quality of financial reporting. Academic research concludes that conservatism benefits users of the firm’s accounting reports (Watts, 2003a).
2.3 Explanations and measures of conservatism
2.3.1 Explanations of conservatism
Accounting conservatism is viewed as causing the persistent understatement of net asset values through its asymmetric treatment of gains and losses (Watts, 2003a). Ruch and Taylor (2011) speak of an incomplete and/or inconsistent recognition of economic value in accounting earnings resulting in a downward bias in accounting net asset value relative to economic net asset value. Ball and Shivakumar (2005) argue that loss recognition timeliness, which is an attribute of earnings conservatism, is an important measure of earnings quality.
Accounting research has produced four alternative explanations for conservatism. Watts (2003a). This is later re-examined by Qiang (2007) who investigates if the explanations apply to conditional conservatism, unconditional conservatism, or perhaps both.
The first explanation is contracting. Watts notes that this the most developed explanation in the literature. Conservatism is a means of counteracting firms’ incentives for aggressive accounting induced by the moral hazard and this lowers potential losses to investors of the firm who rely on accounting numbers in contracts. Without conservatism the values of the firm would be biased upwards by the managers and thus in the long run have a negative effect of the welfare of investors. To induce conservatism for contracting purposes, the investors can impose on firms certain contracting costs. contracts with a firm are subject to verification. But the payoffs to the relevant parties are asymmetric. These relevant parties can be investors in the firm’s debt (debt holders) or members of management. Debt holders can only be adversely affected if the net assets at the maturity of the provided loan are less than the contracted sum. Should the value of the net assets surpass the face value of the contracted debt, no additional compensation will be granted. Therefore, the debt holders are concerned with the verifiable lower bound measures of the firm’s net assets. Managers on the contrary have the incentive to overstate future cash flows and consequently receive (large) earnings-based compensation payments. Therefore, if earnings are measured conservatively, this will give some timely
incentives and deferred compensation will reward
managers for currently unverifiable future cash flows (Smith and Watts, 1982). To put it another way, conservatism reduces agency costs (LaFond and Watts, 2008). Qiang (2007) concludes that contracting induces conditional conservatism because unconditional conservatism does not provide new information.The second explanation is the possibility of shareholder litigation. By overstating the net assets a firm and its auditors are more likely to be sued by new shareholders (investors) and thus generate litigation costs than when the net assets are understated and current shareholders want to sell their shares. This was evidenced empirically already by Kellogg 12
(1984). The new shareholders feel duped and want compensation. To prevent this, managers are induced to report conservative numbers. Also, the legal system surrounding the firms has to grow to keep in check the possible opportunistic payments to managers and other parties to the firm. In this sense, litigation costs are viewed as a necessary evil and by their asymmetric nature induce conservatism. Qiang (2007) proclaims that depending on the circumstances either form can emerge. sometimes due to the possibility of litigation firms will practice unconditional conservatism to be on the safe side and take early losses. However, sometimes conditional conservatism is also called for because of a particular demand by some potential plaintiff.
Then the third explanation as to why conservatism emerges is taxation. Simply put, high profits make for higher income taxes. A firm by nature has the incentive to minimize the income tax liability. By delaying the recognition of gains and expediting the loss recognitions (i.e. conservatism) taxes can be deferred to a later period, causing the present value of the taxes to diminish and the value of the firm to be increased. This falls under the definition of unconditional conservatism (Qiang, 2007).
The fourth explanation that arises in prior literature is accounting regulation. Standard setters and regulators (e.g. the Security and Exchange commission (SEC), FASB and IASB) have a higher chance of facing public criticism if investors suffer losses due to alleged net asset overstatements. To combat this conservatism is highly preferable and thus induced by confronting firms with regulation costs.
Basu (1997) offers four other alternative explanations for the result he ascribed to conservatism, results that are later echoed to some degree by Watts (2003b). The first non-conservatism explanation offered is shareholders’ liquidation or abandonment option. The premise is that investors do not want to lose money and consequently, will leave the sinking ship by selling their interest in the company. The critique of Watts (2003b) on this explanation is that it is unable to explain, by itself, the systematic understatement of net assets. The second explanation is that opportunistic managers wish to delay the disclosure of bad news to the market and leak good news ahead of schedule. Explanation three is that managers on the other hand could wish to disclose bad news earlier to reduce the odds of being sued by shareholders. Last, Basu also discusses the effect of mandatory accounting changes, although he is quick to dismiss this explanation.
The information perspective shows us that trough conservatism estimates of net assets and earnings are downward-biased in present years but upwardly biased in future years. Because in the future these assets and earnings are indeed verifiable and will be realized one could think that these future earnings are not conservative. Critics of conservatism have thought this way. However, conservatism is not about the size and scope of the future earnings but about the degree of certainty. So, even though future earnings can 13
turn out to be higher than estimated, they are indeed conservative because by then there will be verifiable evidence to those numbers.
Qiang (2007) echoes the findings by other researchers (Pope and Walker, 2003; Beaver and Ryan, 2005; Roychowdhury and Watts, 2007) that unconditional and conditional conservatism are negatively interrelated. unconditional conservatism lowers book values, thereby lowering the subsequent threshold for conditional conservatism. Unconditional conservatism induces “a form of ‘accounting slack’ that preempts the application of conditional conservatism unless news is sufficiently bad to use up that slack.” (Beaver and Ryan, 2005, p. 270).
2.3.2 Measures of conservatism
Measuring is knowing. This old adage is simple, true enough but in the case of the underlying subject of conservatism it poses some challenges as there is more than one way to measure conservatism with each measure having its advantages and drawbacks. In order for researchers to make any empiric inferences they must first make a selection of the many models available and then operationalize their measures so as to be able to draw conclusions about the sample being investigated. The measure employed is the instrument or the yardstick by which the empiric data is tested. Watts (2003b) classifies three types of measures that have been used to assess conservatism in three categories: Net asset measures, earnings and accrual measures and earnings/stock returns relation measures.
Due to conservatism there is an understatement or carrying value below market value of the net assets. To measure this understatement models are employed that estimate the extent of undervaluation of the firm’s equity shares (valuation model measures) and/or represent the ratio of the firm’s book value of its net assets to the market value of its equity (book-to-market measures). For valuation measures the Feltham-Ohlson models (1995, 1996) are usually used. The book-to-market ratios will be lower because of the lower numerator.
Under the second category Watts classifies measures that look at the particulars of the accrual and earnings relationships under conservatism. Because expected losses tend to have only a one period negative impact on earnings whereas the upward effects of expected gains on earnings are spread out over multiple periods, the latter are said to be persistent. This persistence provides a measure for conservatism. Under conservatism, if news of an oncoming loss reaches the firm this tends to be fully accrued for whereas if news of an expected gain reaches the firm this has the tendency not to be accrued for. That is why negative periodic net accruals and negative cumulative accruals are used as measures of conservatism. Moreover, under conservatism the news of expected losses make for more 14
very large accruals than news of expected gains. This in turn causes negative skewness of distribution of accruals and earnings which provide measures for conservatism.
Referring to the third category, under conservatism accounting losses are expected to be more contemporaneous with stock returns than accounting gains. This is also found by employing the Basu model (1997).
Ruch and Taylor (2011) use the categories of ‘market measures’ and ‘book measures’ to classify the different measures that have evolved in empiric research. They state that market measures employ relationships between accounting and market data and book measures employ relationships between accounting earnings, accruals and cash flows. This is a simplification of the categories Watts proposed. In accounting research various methods have been used to measure conservatism. According to Wang et al. (2009) there are five measures that are commonly used in the literature.
First and most used is asymmetric timeliness of earnings (Basu, 1997). This measure defines conditional conservatism as the extent to which the correlation of bad news (represented by negative market returns) and the accounting earnings is greater than the correlation of good news (represented by positive market returns) and the accounting earnings. Basu (1997) provides several conditional conservatism estimators (Ball et al. 2013).
Second, the asymmetric accruals-to-cash flow measure (Ball and Shivakumar, 2005). This measure is also about the same news-dependent asymmetric timeliness as the Basu-measure, except the news is proxied by operating cash flow instead of the market returns.
Third, the Market-to-book bias ratio (Beaver and Ryan, 2000). This one deconstructs the market-to-book ratio into a recognition lag component and a bias component, which is to be interpreted as accounting unconditional conservatism. It is classified as unconditional because it does not react to economic news.
Fourth, the hidden reserve measure (Penman and Zhang, 2002). This measure estimates hidden reserves of unrecognized assets on the balance sheet. This one is also classified as unconditional because of the same disregard for economic news as the market-to-book ratio.
Fifth, the negative accruals measure (Givoly and Hayn, 2000). This measure captures the accumulation of negative nonoperating accruals over time. This measure does not make a distinction between conditional and unconditional conservatism.
Since then, other measures of conservatism have emerged out of the accounting research. Dichev and Tang (2008) constructed a measure that defines conservative accounting as the correlation between current period revenues and prior period expenses. Put another way, conservatism can also be seen as a lack of matching between current period revenues and current period expenses.
Khan and Watts (2009) take the Basu measure and substitute the cross-sectional asymmetric timeliness coefficients with functions of firm-specific characteristics. The outcome is referred to as the C-score.
Callen et al. (2010) constructed a measure that captures the proportion of an economic news shock that is recognized in earnings. The outcome is referred to as the conservatism ratio and measures conditional conservatism because it depends on economic news, good or bad.
As the asymmetric timeliness measure is most used in empiric research for gauging conditional conservatism it is also worth noting that Ryan (2006) identifies at least seven limitations to the use of his measure.
First, returns are not equivalent to non-earnings news. Second, conditional conservatism is preempted (i.e. applied less frequently and/or in smaller amounts) by unconditional conservatism. Third, bad news may not be immediately recognized in earnings Fourth, it can be difficult to observe asymmetric timeliness empirically when multiple shocks are aggregated in returns and earnings. Fifth, certain economic phenomena yield asymmetric timeliness. Sixth, certain types of discretionary accounting behaviour also yield asymmetric timeliness. Seventh, there is little time-series consistency in estimates of asymmetric timeliness at the firm level and even for cross sections of firms grouped based on the level of asymmetric timeliness. Yet, despite these limitations Ryan keeps the stance that the asymmetric timeliness measure of Basu is the best option because in some manner or form both good news and bad news have to be taken in consideration.
Recently, Gow et al. (2010) have criticised many studies that deal with variables that are cross-sectionally and serially correlated but use test statistics that do not correct for this problem. They also specifically address the conservatism literature. Where the popular Basu (1997) model explicitly does correct for heteroscedacity by deflating the variables by opening stock price and book value of assets or equity and by using White t-statistics, it does not correct for both types of dependence. It is one-way cluster robust. Gow et al. (2010) advocate an extension of the one-way clustered method, which is the two-way cluster robust standard errors method4. The added benefit of this method is that two way clustering allows for both time-series correlation and cross-sectional correlation. In this thesis this method will be employed to examine the effect on conservatism.
2.4 Positive and negative effects of conservatism
In this section I describe the effects of conservatism that have come out of the prior literature.
4 Gow et al. Refer in their article to two-way clustering as CL-2. One way clustering is denoted either by CL-i, for clustering a
cross-sectional variable like firm or country or by CL-t when clustering a time-series variable like year. For clarity purposes I will use the full description.
16
There has been much research done on accounting conservatism. Some studies argue that conservatism benefits the financial reporting quality (Ball and Sivakumar, 2005; Krishnan, 2005; Visvanathan and Krishnan, 2008). Other studies have conflicting findings (Ahmed et al, 2002; Bandyopadhyay, 2010; Garcia Lara et al., 2011).
Until recently, the judgement call that came out of earlier academic research is that conservatism benefited users of the firm’s accounting reports (e.g. Watts, 2003a). Conditional conservatism has been found to reduce earnings persistence and future earnings predictability because of timelier loss recognition (Kim and Kross, 2005; Dichev and Tang, 2008; Paek et al., 2007; Bandyopadhyay, 2010). Here earnings persistence5 is the extent to which earnings in one period persist to the next period and earnings predictability6 is the ability of earnings to predict future earnings and cash flows (Ruch and Taylor, 2011). There is a natural relationship between these two concepts as higher earnings persistence signifies more predictable earnings (Dichev and Tang, 2009).
As evidenced by Paek et al. (2007) conditional conservatism has a negative effect on the persistence of earnings and this in turn makes for lower pricing multiples. This is concluded after regressing one-period-ahead earnings on current period earnings.
Dichev and Tang (2008) claim that the increase in accounting conservatism, as evidenced by Givoly and Hayn (2000) can have a causal effect on the deterioration of the matching between contemporaneous revenues and expenses. This can have a declining effect on the persistence of earnings and this in turn results in a diminishing ability of contemporaneous earnings to predict future earnings.
Research by Kim and Kross (2005) and Bandyopadhyay (2010) has indicated that conditional conservatism enhances the ability of earnings to predict future cash flows, but it also weakens the ability of current earnings to predict future earning (Bandyopadhyay, 2010). This is referred to as the relevance-reliability trade-off, which was noted by the FASB in Statements of Financial Accounting Concepts (SFAC) No. 2 for business enterprises. Since then, SFAC no. 8 was issued in September 2010 as a replacement of SFAC 1 and 2. In here, the FASB replaced ‘reliability’ with ‘faithful representation’7.
Findings by Penman and Zhang (2002) are indicative of the fact that unconditional conservatism can diminish the ability of earnings to predict future earnings. Their finding is that unconditional conservatism causes reserves “not recorded on the balance sheet, they are unrecorded "hidden" reserves that are released into earnings when investment growth slows. We refer to them as unrecorded reserves”. When these reserves are released into the earnings it makes for a temporary change. This reduces the predictive ability of current earnings for future earnings. Moreover, this can also help facilitate earnings management.
5 For a more detailed understanding of the econometrics of earnings persistence refer to Koremndi and Lipe (1987). 6 For the extreme form of earnings Predictability which is perfect foresight’ earnings refer to S.A. Richardson et al. (2005). 7
www.fasb.org
17
LaFond and Watts (2008) claim that information asymmetry, which entails that management has private information relevant for assessing future firm performance that is unavailable to outside equity investors, can cause conservatism.
Research has suggested that reporting conservatism results in a lower debt cost of capital for firms (i.e. lower interest rates) (Zhang, 2008). However, though this may be the case, they forgo some flexibility in future access to capital, and this ultimately affects their financial decisions (Lee, 2010). Taken these effects together it seems there is much to be discussed before reaching a consensus. Further research is what is called for to be able to provide a well balanced opinion.
3. IFRS
3.1 Introduction
In this chapter the history of IFRS will be described. From the facts leading to their conception, and their (perceived) advantages, followed by the term fair value.
3.2 Overview of IFRS
IFRS have had a long history in the making. The International Accounting Standards Committee (IASC) was established in 1973. Their mandate was to develop a set of high quality financial reporting standards, that were internationally acceptable (Barth et al. 2008). These were later known as International Accounting Standards (IASs).
There came increasing demands for comparability and transparency of annual reports between different European countries. Harmonisation of accounting standards is what was called for (Alfredson, 2007, p.7). He states that over time the ‘calls for harmonisation of accounting standards’ were replaced by ‘calls for a single global set of financial reporting standards that could replace national accounting standards’. So, comparability and an improvement of the overall accounting quality were the reasons for the development of the IASs.
The successor of the IASC, the IASB was founded on April 1st, 2001, headquartered in London and consists of an independent group of experts with much experience in setting accounting standards; in preparing, auditing, or using financial reports; and in accounting education. The members are also culturally diverse, bringing distinct insights to the table. They are appointed by the Trustees through an open and rigorous process that includes advertising vacancies and consulting relevant organizations. The IASB has 16 full-time members. Following the foundation, the IASs that were already in place are slowly being superseded by the new IFRS. The IASB has a Conceptual Framework that aids in the development of these standards. The Framework is not a set of standards in itself, but rather it conceptualizes he building blocks that are used to form the standards. The IASB issues standards that are principle based whereas their US counterpart the FASB issues standards that are rule based. Although, fundamentally different in their views how to go about their work, a joint meeting between the IASB and the FASB was held in October 2004 where a roadmap was formed to develop a common conceptual framework. This would be based on and built on both the existing IASB and FASB frameworks, on which in turn their accounting standards, IFRS and US GAAP would be based. The objective of this common conceptual framework project is to create a sound foundation for future accounting standards that are principles-based, internally consistent and internationally converged. Up till now the IASB
has revised portions of its framework; while the FASB has issued ‘Concepts Statement 8’ to replace ‘Concepts Statements 1 and 2’. In 2010 both Boards announced the completion of the first phase A on objectives and qualitative characteristics of their joint project. There were to be 8 phases, titled A through H but the planning was tabled. Near the end of 2012 the IASB started to continue the project on its own.
The IASB has four principle objectives:
- Objective 1 to develop, in the public interest, a single set of high quality,
understandable, enforceable and globally accepted international financial reporting standards (IFRSs) based upon clearly articulated principles
- Objective 2 to promote the use and rigorous application of those standards
- Objective 3 to take account of the needs of a range of sizes and types of entities in diverse economic settings
- Objective 4 to promote and facilitate adoption of IFRSs through the convergence of national accounting standards and IFRSs.
Currently there are 43 separate Standards in effect. These are comprised of 15 IFRS issued, of which the latest two will come into effect in 2016 and 2017. Of the 41 original IAS 11 have now been either withdrawn or superseded by other standards. Of the 30 remaining still active IASs two will be superseded by the recently issued IFRS 15.
3.3 Benefits of IFRS
IFRS were introduced for several reasons. Their raisons d’être are considered to be the following.
- Comparability
- Transparency (Jeanjean and Stolowy, 2008) - Improving quality
IFRS stands on fair value as opposed to other valuation methods like historical costing8. Because this requires a (timely) assessment of the company assets and liabilities it should represent a more accurate value investors are willing to pay for it.
8 This applies only for a particular set of line items and not the entire balance sheet. That is why it the IFRS model is considered
a mixed measurement model.
20
3.4 Conservatism under IFRS
A distinction to be made is the one between conservatism and prudence. Prior literature in accounting research most often uses these interchangeably. Conservatism is to be considered a characteristic of the accounting system and prudence describes the preferences of a Decision Maker (Kirschenheiter and Ramakrishnan, 2012). They note that FASB (2010) in paragraphs BC3.27 and BC3.28, ‘seems to conflate prudence and conservatism, interpreting them as interchangeable concepts’.
In 2010, the IASB revised its Conceptual Framework and rejected the concept of conservatism (or prudence), arguing it to be inconsistent with neutrality, an aspect of faithful representation (Barker and McGeachin, 2013). However on May 21, 2014 the IASB tentatively decided9:
(a) to reintroduce a reference to prudence in the Conceptual Framework;
(b) to describe prudence as the exercise of caution when making judgments under
conditions of uncertainty. The exercise of prudence is consistent with neutrality and should not allow the overstatement or understatement of assets, liabilities, income or expenses;
This text proves to demonstrate that the IASB now indeed considers prudence and conservatism to be two different concepts. The FASB states in SFAC 8, BC3.27 “Chapter 3 does not include prudence or conservatism as an aspect of faithful representation because including either would be inconsistent with neutrality”. From this it can be concluded that the FASB has the same stance as IASB, in that they are two different concepts.
Ruch and Taylor (2011) argue that fair value standards are more consistent with neutrality as they reduce the gain/loss recognition asymmetry by reducing the delay in recognizing gains. As IFRS are aimed at the fair values it stands to reason that conservatism would have diminished since their mandatory adaption.
To build the case for this thesis next will be the findings of researchers stating conservatism increases after IFRS adoption in Europe and literature detailing conservatism showing a decrease post-IFRS. Garcia Lara et al. (2008) investigate the voluntary adoption of IFRS by companies. By using the Basu (1997) regression they found that, when focusing in on Europe, companies using IFRS voluntarily show higher conservatism than companies from the same countries that use their local GAAPs.
Manganaris et al. (2011) examine three code-law European countries (Germany, France and Greece) and one common-law European country (UK). They question whether conservatism exists during the last decade (1999-2008) and whether its level has changed over this period. They also employed the Basu (1997) model to run regressions. The results
9 http://www.ifrs.org/Current-Projects/IASB-Projects/Conceptual-Framework/Documents/Effect-of-Board-decisions-on-DP-May-
2014.pdf
21
provide evidence that conservatism exists in all countries before the IFRS adoption and that its level has decreased after 2005 only in France and Germany. Also, they found a decrease in the adjusted-R2 of the Basu (1997) model in the post-IFRS period. They take it to imply that the model may not be working effectively in the later years.
Piot et al. (2011) researched the mandatory adoption in the EU. They examine more than 5,000 IFRS adopters from 22 EU countries, observed over 2001-2008 by using the Basu (1997) regression to measure the conditional conservatism and the bias component of the market-to-book ratio to proxy for unconditional conservatism. Their finding is that conditional conservatism has decreased after the mandatory adoption.
Andre et al. (2013) examine the impact of mandatory IFRS adoption on conditional conservatism in Europe from a final sample of 7.251 firm-year observations drawn from 16 European countries, observed over 2002-2007. They also find that conditional conservatism has decreased. So, the findings about conservatism in Europe together show mixed results. Most research shows that conservatism has decreased after the adoption of IFRS. Voluntary adoption of IFRS seems to increase conservatism. As my research question concerns the mandatory adoption of IFRS, based on these findings I suspect that conservatism in the Netherlands also decreases.
4. Research method
In this section the hypotheses development is provided, which is followed by the method of sample selection and data collection. Lastly, the descriptive statistics, the model and measure employed are described in detail.
4.1 Hypotheses development
In statistical analyses it is common for the null hypothesis to state that there is no relationship. In statistical terms, the technique of using two different hypotheses to test is also called confirmatory data analysis. I will use the Beaver and Ryan (2005) model to measure conservatism both before the adoption of the IFRS standards in 2005 and afterwards. The model simulates the complex interactions between conditional and unconditional conservatism, and implicitly the role of accounting standards and the judgments and estimates involved in applying those Standards (Basu, 2005). The model assumes a firm has only has two types of economic assets, tangible and intangible assets and makes six assumptions about the market value of those assets, the market value of owners’ equity and dividends (Beaver and Ryan, 2005, p. 275)10:
Assumption 1: The market values of tangible and intangible assets at time t sum to the
market value of common owners’ equity. There are no financial assets, operating or financial liabilities .
Assumption 2: The owners’ equity is subject to the no arbitrage condition.
Assumption 3: tangible assets depreciate economically at a random annual rate 1-γ(1+ει), where 0 < γ < 1 and ει > -1 is a shock to the economic depreciation rate of those assets.
Assumption 4: the market value of new investment in tangible assets (IT) has grown over an infinite history and continues to grow at a constant expected rate g, with 1 < g < r.
Assumption 5: the market value of intangible assets on average grows at the same expected rate g, though with different shocks η to that rate.
Assumption 6: the dividend yield is constant.
What we have to bear in mind is what is the dependent variable, or what are we trying to explain and what is the independent variable, or what do we input as a given in our model. We want to explain the conservatism in earnings by considering the returns as a given. Basu (1997) considers returns as the independent variable in his model. He concludes, based on prior researchers that stock prices “reflect information received from sources other than current earnings” and that they “lead accounting earnings by up to four years” (p. 11). The
10
For a full algebraic derivation of the assumptions underlying the model please refer to the article by Beaver and Ryan (2005). 23
second thing that we have to be aware of is that there is a different response by returns when earnings are negative than when they are positive. These relates to the earnings persistence, also proclomated by Basu. Accounting research (e.g. Ryan and Zarowin, 2003 and Pope and Walker, 1999) has pointed out that on overage annual earnings respond more to the first three lagged annual returns when they are negative then when they are positive.
Based on the outcomes of prior literature studying Europe I expect that IFRS has an impact on the Dutch capital markets which leads me to come to the following hypothesis.
Hypothesis 1:
Hypothesis 1A (H0): Conservatism does not change after the mandatory adoption of IRS in the Netherlands
Hypothesis 1B (H1): Conservatism changes after the mandatory adoption of IRS in the Netherlands
Prior research with European samples have indicated that conservatism decreases in the post-IFRS periods (Andre et al., 2013;Piot et al., 2011). Based on this research with European samples I predict that conservatism in the Dutch capital markets decreases after the mandatory adoption in 2005. This leads me to the second hypothesis.
Hypothesis 2:
Hypothesis 2A (H0): Conservatism does not decrease after the mandatory adoption of IRS in the Netherlands
Hypothesis 2B (H1): Conservatism decreases after the mandatory adoption of IRS in the Netherlands
I test the models in different time frames, with one period falling before the introduction of IFRS and the next starting with the mandatory adoption as of January 1st, 2005.
4.2 Sample selection and data collection
As input for the model, a time series sample is obtained of 285 firms observed over a 16 year timeframe, the period 1998 – 2013 of companies listed at the NYSE Euronext Amsterdam. This results in 2613 firm-year observations. The data are collected from the Worldscope database through Datastream. A first glance at the sample tells us that the number of firms listed has been steadily declining since 1998, dropping to below 100 in 2013.
We first have to correct for any possible duplicate observations. Next, to control for the effect of extreme observations, equity, returns and earnings were windsorized at the 99% level. Truncation is considered undesired because we do not want to disregard the observations, merely censor the impact on the sample. The next step is to create the one period lagged variables. Although data was collected of 285 companies, not all of the data items required as input for the model could be retrieved as they were unavailable. As a consequence, the following data were dropped: 60 observations of negative equity, 200 observations with missing information on the share price at the yearend and/or missing outstanding number of shares, 293 observations missing information about the return, 254 observations missing information about the lagged returns controlled for by a dummy and 215 observations that missed information about a two period lag, also controlled for by a dummy. This resulted in the deletion of 1.027 observations.
Because we are interested in examining the effect of the adoption of IFRS a dummy is added to the model. IFRS were introduced in 2005 so for the years 2004 and earlier this dummy naturally equals zero. Also, the firms with a voluntary adoption of IFRS were discarded. The resulting sample consists of 192 firms observed over 2001 – 2013. This results in 1576 firm-year observations over a 13 year period. The sample is subsequently split up in two subsamples. One subsample is for the period 2001-2004, the pre -IFRS period and the other subsample is for the period 2005-2013, the post-IFRS period. The pre-IFRS sample consists of 666 firm-year observations and the post-IFRS sample consists of 910 firm-year observations.
4.3 Descriptive statistics and models
This section provides a description of the variables used in the model. The six variables selected are analyzed in the form of descriptives and through the method of simple linear regression. Table 1, inserted in the next page, shows the results and their corresponding descriptions. Also, the two models that are used in this study are specified. The results of the regressions are presented in the next section. In my analysis I will first run the regressions for the entire sample. I will do this first using the Beaver and Ryan (model) en then the Basu (1997) model. This enables me to compare results based on the same samples. Next I will again run the regressions using the Beaver and Ryan (2005) model first and then the Basu model, however this time the samples will be split in two subsamples. One subsample is for the period pre IFRS (2001-2004) and the other subsample will be the post IFRS period (2005-2013). I will also use different techniques of coping with the standard errors. So, in total I will be running 27 regressions.
Table 1 - Descriptive analysis of the variables used variable N mean sd p25 p50 p75 p 1.576.000 24.503 78.899 6.089 13.457 24.227 eps 1.576.000 0.610 15.861 0.023 0.721 1.667 earn 1.576.000 0.023 0.173 0.004 0.065 0.101 ret 1.545.000 0.087 0.411 -0.167 0.059 0.300 D_ret 1.576.000 0.426 0.495 0.000 0.000 1.000 IFRS 1.576.000 0.577 0.494 0.000 1.000 1.000
p = market price of share at the end of the calender year (DataStream data item WC01551). eps = earnings per common share outstanding (Common share is DataStream data item WC05301). earn = net Income Before Extra Items/Preferred Dividends (DataStream data item WC01551).
ret = return including dividends over the financial year, measured as the change in price market (p) over the financial year plus cash dividends (DataStream data item WC04551).
D_ret = dummy variable equal to 1 if return is negative and 0 otherwise.
IFRS = dummy variable equal to 1 if IFRS applies (i.e. year 2005 and later) and 0 otherwise.
At first glance it seems the standard deviations are quite high. Standard deviations are used as a measure of dispersion around the mean. If the dispersion is higher the measured values tend to vary more, triggering greater uncertainty in the reader because the mean becomes less representative. In the analyses this is an item that is taken into consideration.
The Beaver and Ryan (2005) model also with an IFRS dummy is specified as follows:
BM
jt= α
j+
α
t+
∑
2𝑇𝑇=0βτ
*RET
jt-τ+ 2∑ τ
=0γ0DR
jt-τ*RET
jt-τ+
∑
2𝑇𝑇=0IFRS*
γτDR
jt-τ*RET
jt-τ+
ε
jtWhere BMjt is book-to-market for firm j at the end of fiscal year t and is book value of equity / market value and DRjt is a dummy variable that equals 1 if RETjt is negative and 0
otherwise. As required by the model two lags have been created in STATA. A lag refers to the previous period.
Next, as a robustness test I will repeat the regressions with the Basu (1997) model. I specify the Basu (model) with the IFRS dummy. For the years 2001 – 2004 this dummy is 0.
X
ιτ/P
ι(τ−1)= α
0+ β
1DR
ιτ+ β
2RET
ιτ+ β
3RET
ιτ*DR
ιτ+ β
4IFRS
ιτ+
β
5IFRS
ιτ*DR
ιτ+
β
6IFRS
ιτ*RET
ιτ+
β
7IFRS
ιτ*RET
ιτ*DR
ιτWhere the value of β3 denotes the conditional conservatism.
5. Empirical Results and Analysis
5.1 Empirical results
I want to see the influence of IFRS on conservatism. To do this I use the model of Beaver and Ryan (2005) to involve both conditional and unconditional conservatism and a dummy variable will be added to simulate the impact of the mandatory adoption of IFRS. Before using the empiric model I would like to know is there is a dependency or correlation between any two of the variables I use in my model. To find this out I first conduct two preliminary tests to check for any linear and non-lineair relationships. First, to investigate if there is a linear relationship between any of the variables I examine the Pearson product-moment correlation coefficient also known as Pearson's correlation coefficient. This coefficient is obtained by dividing the covariance of two variables by the product of their standard deviations. The results are shown in table 2.
Table 2 - Results of Pearson correlation test for a linear relationship
Var. p eps earn ret D_ret IFRS
p 1.0000 eps -0.0260 1.0000 earn 0.0420 0.2346 1.0000 ret 0.0153 0.0974 0.2545 1.0000 D_ret -0.0020 -0.0717 -0.2828 -0.7156 1.0000 IFRS -0.0270 -0.0056 0.0201 0.0545 0.0713 1.0000
For the definitions of the variables used please refer to table 1.
As we can see most of the numbers are fairly centered around 0, meaning they are not statistically significant. The dummy for returns is negatively associated with all other variables, but with the possible exception of the relationship to returns the associations between the other variables are not statistically significant.
Second, to investigate if there is a non-linear relationship between any of the variables I examine the Spearman's rank correlation coefficient. This measures the extent to which, as one variable increases, the other variable tends to increase. The results are shown in table 3.
Table 3 - Results of Spearman correlation test for a non-linear relationship
Var. p eps earn ret D_ret IFRS
p 1.0000 eps 0.6338 1.0000 earn 0.2743 0.7818 1.0000 ret 0.2358 0.2633 0.4263 1.0000 D_ret -0.2361 -0.2718 -0.3800 -0.8564 1.0000 IFRS 0.0549 0.0257 -0.0084 0.0853 -0.0713 1.0000
For the definitions of the variables used please refer to table 1.
The 1.0000 values are off course the perfect linear relationships of the variables with themselves. The minus signs are indicative of a negative relationship and the closer the coefficient is to either −1 or 1, the stronger the correlation between these variables. As we can see most of the numbers are higher than in table 2, but this is not surprising, as a non-linear relationship is easier to find than a non-linear relationship. The dummy for returns is again negatively associated with all other variables. There are significant indications for a relationship between the return dummy and the returns, between the earnings and EPS and between the EPS and the market price p. Of course, these are consistent with our model, because we specified the dummy of the returns as being equal to 1 if return is negative and 0 otherwise. Earnings and EPS are also related because both sides of the equation use the variable earnings. There is also a relation between the market price and EPS, because if the EPS go up, investors will want to pay more for that particular stock and so the price will go up.
Before running the regressions I perform a two sample t-test on each of the variables to test if the differences in the means from the subsamples significally differ from 0. The results are shown in table 4. The statistics of the combined groups are also presented in the table. Group one concerns the pre IFRS period of 2000-2004 and contains 666 observations and group two concerns the post IFRS period of 2005-2013 and contains 910 observations.
Table 4 - Two-sample t test with equal variances to test the means
and two-sample Wilcoxon rank-sum (Mann-Whitney) test to test the medians Panel A: t-test and ranksum of p
Year Obs Mean Rank sum
2001-2004 666 26.98951 505688
2005-2013 910 22.68277 736988
2001-2013 1576 24.50275 1242676
diff 4.306739
diff = mean(0) - mean(1) Pr(|T|>|t|) = 0,2846 t = 1.0705
Panel B: t-test and ranksum of eps
Year Obs Mean Rank sum
2001-2004 666 7140971 516026
2005-2013 910 0.5335479 726650
2001-2013 1576 0.6098459 1242676
diff 0.1805492
diff = mean(0) - mean(1) Pr(|T|>|t|) = 0,8234 t = 0.2232
Panel C: t-test and ranksum of earn
Year Obs Mean Rank sum
2001-2004 666 0.019309 528106.5
2005-2013 910 0.0263387 714569.5
2001-2013 1576 0.023368 1242676
diff -0.0070297
diff = mean(0) - mean(1) Pr(|T|>|t|) = 0,4249 t = -0.7982
Panel D: t-test and ranksum of ret
Year Obs Mean Rank sum
2001-2004 666 0.0658888 494941
2005-2013 910 0.1155736 747735
2001-2013 1546 0.0945773 1242676
diff -0.0496848
diff = mean(0) - mean(1) Pr(|T|>|t|) = 0,0304 t = -2.167
Panel E: t-test and ranksum of
D_ret
Year Obs Mean Rank sum
2001-2004 666 0.466967 546766
2005-2013 910 0.3956044 695910
2001-2013 1576 0.4257614 1242676
diff 0.0713626
diff = mean(0) - mean(1) Pr(|T|>|t|) = 0,0046 t = 2.8356
Panel A shows that the difference in the means of p of the subsample pre-IFRS and the subsample post-IFRS is EUR 4,31. However, because of the small t-value and the large standard errors this is statistically not significant, even at the 99% confidence level. Panels B through D show very small differences in means of the two subsamples. And even if they did show a large difference, their corresponding t-values of 0.2232, -0.7982 and t = -2.167 are to small to fall in the rejection region and therefore the differences are statistically not significant. Panel E shows a high t-value so the difference in means is statistically significant. This means that there is enough evidence to reject the null-hypothesis and state that there is a genuine observable difference between the mean value of the return dummy before IFRS and post IFRS.
Next, I also perform a two-sample Wilcoxon rank-sum (Mann-Whitney) test, also called a median test for parallel groups. This is a nonparametric test of the null hypothesis that two populations are the same against an alternative hypothesis, that a particular population tends to have larger values than the other. The primary assumptions underlying this test are that the sampling is done at random an that there is at least an ordinal measurement scale. In this case both assumptions apply. The results are also presented in table 4.
In Panel A the Wilcoxon ranked sum test is used to check for possible differences in the median level of the price before IFRS and post IFRS. The z-statistic reports a value of –2.180 for p which falls in the rejection area at the 99% confidence level and makes it plausible for us to assume that p does not remain stable after IFRS based on the
observations of the median levels in the years 2001 – 2013. In Panel B the differences are checked for the variable eps pre- and post-IFRS. The presented z-value of -1.021 is not significant at any confidence level so the value of eps is assumed to remain stable between the before and after IFRS period. In Panel C there is a check for a difference in the variable earnings between pre- and post-IFRS. Based on the low z-value no such difference is assumed. Panel D tells us there is a difference between the before and after IFRS periods. The z-value of -3.384is statistically significant at the 99% confidence level. Last, in panel E the Wilcoxon ranked sum test checks if the variable D_ret differs significally in the pre-IFRS period from the post-IFRS period. Based on the z-value of 2.829 I reject the null-hypothesis and conclude with a 99% confidence level that there is a statistically significant difference. These findings, when compared with the results I present in the descriptive statistics table and the performed t-tests merit further research as to the underlying causes.
After the preliminary testing I can now run the regressions. As part of my analysis I use two models and four methods. This way I can compare the outputs that are produced using the different methods. This will make it easier to interpret the results and see if the two way clustered method indeed is better. In all I run 27 regressions. First, I run six regressions for the entire period 2001 – 2013 with the different methods for correcting for standard errors using the Beaver and Ryan (2005) model. Next I perform seven regressions with the same sample but I now I use the Basu (1997) model. Then I use the Beaver and Ryan (2005) model to run six regressions, three for the pre IFRS period (2001-2004) and three for the post IFRS period (2005-2013). Last, I use the Basu model to run eight regressions, also with pre and post IFRS samples. With each regression a separate method is used to correct for standard errors.
When running a regression choosing the right method is key. The popular phrase garbage in is garbage out comes to mind. When reading the article of Gow et al. (2010) they state that most examined studies in the economics literature use statistics and standard errors that overstate their values, dangerously compromising the end results. They critize the Z2 statistic for being “highly correlated with Fama-MacBeth tstatistics, producing Type I error rates that are nearly identical to Fama-MacBeth, rejecting a true null hypothesis more than 30 percent of the time at the 1 percent level” (p. 485). Their method of two way clustering the data along a time dimension and a cross-sectional dimension, like firm will shield the regression from such overstatements and making the results more reliable. One of the first things that has to be decided before running the regression is how to determine the significance or confidence level. This is dependent on the method of choice. Most studies use the popular Basu model. But, as stated, this method only clusters the data one way. So, for samples that exhibit both time series and cross-sectional correlation this method is less reliable. Gow et al. (2010) reason that you have to use two way cluster robust standard 30
